Version 1.0

.codeocean/environment.json

@@ -0,0 +1,131 @@
+{
+	"version": 1,
+	"base_image": "registry.codeocean.com/codeocean/py-r:python3.10.12-R4.3.2-JupyterLab4.0.10-RStudiorstudio-server-2023.12.0-369-ubuntu22.04",
+	"options": {},
+	"installers": {
+		"pip": {
+			"packages": [
+				{
+					"name": "matplotlib",
+					"version": "3.10.0"
+				},
+				{
+					"name": "numpy",
+					"version": "1.26.4"
+				},
+				{
+					"name": "pandas",
+					"version": "2.2.3"
+				},
+				{
+					"name": "rpy2",
+					"version": "3.5.17"
+				},
+				{
+					"name": "seaborn"
+				},
+				{
+					"name": "stargazer",
+					"version": "0.0.7"
+				}
+			],
+			"options": {},
+			"pre_install_options": {}
+		},
+		"rcran": {
+			"packages": [
+				{
+					"name": "car",
+					"version": "3.1-3"
+				},
+				{
+					"name": "corrplot",
+					"version": "0.95"
+				},
+				{
+					"name": "covr",
+					"version": "3.6.4"
+				},
+				{
+					"name": "doParallel",
+					"version": "1.0.17"
+				},
+				{
+					"name": "fastDummies",
+					"version": "1.7.5"
+				},
+				{
+					"name": "feather",
+					"version": "0.3.5"
+				},
+				{
+					"name": "ggtext",
+					"version": "0.1.2"
+				},
+				{
+					"name": "janitor",
+					"version": "2.2.1"
+				},
+				{
+					"name": "lubridate",
+					"version": "1.9.4"
+				},
+				{
+					"name": "mockr",
+					"version": "0.2.1"
+				},
+				{
+					"name": "psych",
+					"version": "2.4.12"
+				},
+				{
+					"name": "randomizr",
+					"version": "1.0.0"
+				},
+				{
+					"name": "sandwich",
+					"version": "3.1-1"
+				},
+				{
+					"name": "stargazer",
+					"version": "5.2.3"
+				},
+				{
+					"name": "systemfonts",
+					"version": "1.2.1"
+				},
+				{
+					"name": "textshaping",
+					"version": "1.0.0"
+				},
+				{
+					"name": "tidyverse",
+					"version": "2.0.0"
+				}
+			]
+		},
+		"vscode": {
+			"packages": [
+				{
+					"name": "REditorSupport.R"
+				},
+				{
+					"name": "continue.continue"
+				},
+				{
+					"name": "ms-python.python"
+				},
+				{
+					"name": "ms-toolsai.jupyter"
+				},
+				{
+					"name": "reageyao.bioSyntax"
+				},
+				{
+					"name": "saoudrizwan.claude-dev"
+				}
+			],
+			"version": "4.95.3"
+		}
+	}
+}

.gitignore

@@ -0,0 +1,4 @@
+/data/
+.Rproj.user
+
+.vscode

code/04_postprocessing_exploration_issues12.R

@@ -0,0 +1,972 @@
+cat(rep('=', 80),
+    '\n\n',
+    'OUTPUT FROM: 04_postprocessing_exploration_issues12.R',
+    '\n\n',
+    sep = ''
+    )
+
+## YouTube Algorithms and Minimum Wage Opinions
+## Data collected May-June 2022 via MTurk/CloudResearch
+
+## Preamble ----------------------------
+library(tidyverse)
+library(janitor)
+library(lubridate)
+library(stargazer)
+library(broom)
+library(patchwork)
+
+# plotting w/ custom colors (optional)
+red_mit = '#A31F34'
+red_light = '#A9606C'
+blue_mit = '#315485'
+grey_light= '#C2C0BF'
+grey_dark = '#8A8B8C'
+black = '#353132'
+vpurple = "#440154FF"
+vyellow = "#FDE725FF"
+vgreen = "#21908CFF"
+
+## edited 13 june 2024 at request of reviewers ---------------------------------
+
+understanding_1 <-
+  read_csv('../results/intermediate data/gun control (issue 1)/guncontrol_understanding_basecontrol_pretty.csv') %>%
+  mutate(
+    layer2_treatmentcontrast = recode(
+      layer2_treatmentcontrast,
+      "31 pro - 22 pro" = "con 31 - con 22",
+      "anti 31 - anti 22" = "lib 31 - lib 22",
+      "31 neutral anti - 22 neutral anti" = "neutral lib 31 - neutral lib 22",
+      "22 neutral pro - 22 neutral anti" = "neutral con 22 - neutral lib 22",
+      "31 neutral pro - 31 neutral anti" = "neutral con 31 - neutral lib 31",
+      "31 neutral pro - 22 neutral pro" = "neutral con 31 - neutral con 22"
+    )
+  )
+
+
+understanding_2 <-
+  read_csv('../results/intermediate data/minimum wage (issue 2)/understanding_basecontrol_pretty.csv')
+understanding_2 <- understanding_2 %>%
+  mutate(
+    layer2_treatmentcontrast = recode(
+      layer2_treatmentcontrast,
+      "31 pro - 22 pro" = "con 31 - con 22",
+      "anti 31 - anti 22" = "lib 31 - lib 22",
+      "31 neutral anti - 22 neutral anti" = "neutral lib 31 - neutral lib 22",
+      "22 neutral anti - 22 neutral pro" = "neutral con 22 - neutral lib 22",
+      "31 neutral anti - 31 neutral pro" = "neutral con 31 - neutral lib 31",
+      "31 neutral pro - 22 neutral pro" = "neutral con 31 - neutral con 22"
+    )
+  )
+
+
+understanding_3 <- read_csv('../results/intermediate data/minimum wage (issue 2)/understanding_basecontrol_pretty_yg.csv')
+understanding_3 <- understanding_3 %>%
+  mutate(
+    layer2_treatmentcontrast = recode(
+      layer2_treatmentcontrast,
+      "31 pro - 22 pro" = "con 31 - con 22",
+      "anti 31 - anti 22" = "lib 31 - lib 22",
+      "31 neutral anti - 22 neutral anti" = "neutral lib 31 - neutral lib 22",
+      "22 neutral anti - 22 neutral pro" = "neutral con 22 - neutral lib 22",
+      "31 neutral anti - 31 neutral pro" = "neutral con 31 - neutral lib 31",
+      "31 neutral pro - 22 neutral pro" = "neutral con 31 - neutral con 22"
+    )
+  )
+
+understanding_1$Study <- 1
+understanding_2$Study <- 2
+understanding_3$Study <- 3
+
+understanding <- rbind(understanding_1,
+                       understanding_2,
+                       understanding_3
+)
+understanding$Study <- factor(understanding$Study,
+                              levels = 3:1,
+                              labels = c('Minimum Wage\n(YouGov)',
+                                         'Minimum Wage\n(MTurk)',
+                                         'Gun Control\n(MTurk)'
+                              )
+)
+
+understanding <- understanding %>%
+  mutate(outcome =
+           recode(layer3_specificoutcome,
+                  'right_to_own_importance_w2' = 'Question 1:\nRight to own more important than regulation (Gun Control)\nRestricts business freedom to set policy (Minimum Wage)',
+                  'concealed_safe_w2' = 'Question 2:\nMore concealed carry makes US safer (Gun Control)\nRaising hurts low-income workers (Minimum Wage)',
+                  'mw_restrict_w2' = 'Question 1:\nRight to own more important than regulation (Gun Control)\nRestricts business freedom to set policy (Minimum Wage)',
+                  'mw_help_w2' = 'Question 2:\nMore concealed carry makes US safer (Gun Control)\nRaising hurts low-income workers (Minimum Wage)'
+           )
+  )
+
+understanding <- understanding %>%
+  mutate(ci_lo_99 = est + qnorm(0.001)*se,
+         ci_hi_99 = est + qnorm(0.999)*se,
+         ci_lo_95 = est + qnorm(0.025)*se,
+         ci_hi_95 = est + qnorm(0.975)*se,
+         ci_lo_90 = est + qnorm(0.05)*se,
+         ci_hi_90 = est + qnorm(0.95)*se
+  )
+
+understanding <- understanding %>%
+  mutate(
+    contrast = ifelse(
+      layer2_treatmentcontrast %in% c("neutral con 31 - neutral lib 31",
+                                      "neutral con 22 - neutral lib 22"
+      ),
+      yes = 'seed',
+      no = 'algorithm'
+    )
+  )
+
+understanding$layer2_treatmentcontrast <- factor(
+  understanding$layer2_treatmentcontrast,
+  levels = c('lib 31 - lib 22',
+             'neutral lib 31 - neutral lib 22',
+             'neutral con 31 - neutral con 22',
+             'con 31 - con 22',
+             'neutral con 31 - neutral lib 31',
+             'neutral con 22 - neutral lib 22'
+  ),
+  labels = c('Liberal respondents,\nliberal seed',
+             'Moderate respondents,\nliberal seed',
+             'Moderate respondents,\nconservative seed',
+             'Conservative respondents,\nconservative seed',
+             'Moderate respondents,\n3/1 algorithm',
+             'Moderate respondents,\n2/2 algorithm'
+  ),
+  ordered = TRUE
+)
+
+understanding_plot_algo <- ggplot(
+  understanding %>% filter(contrast == 'algorithm'),
+  aes(x = layer2_treatmentcontrast,
+      group = Study,
+      color = p.adj < 0.05
+  )
+) +
+  geom_errorbar(aes(ymin=ci_lo_95,ymax=ci_hi_95),
+                position=position_dodge(width=0.5),
+                width=0,
+                lwd=0.5
+  ) +
+  geom_errorbar(aes(ymin=ci_lo_90,ymax=ci_hi_90),
+                position=position_dodge(width=0.5),
+                width=0,
+                lwd=1
+  ) +
+  geom_point(aes(y=est,shape=Study),
+             position=position_dodge(width=0.5),
+             size=2
+  ) +
+  geom_hline(yintercept = 0,lty=2) +
+  facet_wrap( ~ outcome,scales="free") +
+  scale_color_manual(breaks=c(F,T),values = c("black","blue"),guide="none") +
+  coord_flip(ylim=c(-0.1,0.2)) +
+  theme_bw(base_family = "sans") +
+  theme(strip.background = element_rect(fill="white"),legend.position = "none") +
+  ylab('Treatment effect of 3/1 vs. 2/2 algorithm (95% and 90% CIs)') +
+  xlab(NULL)
+understanding_plot_algo
+
+
+understanding_plot_seed <- ggplot(
+  understanding %>% filter(contrast == 'seed'),
+  aes(x = layer2_treatmentcontrast,
+      group = Study,
+      color = p.adj < 0.05
+  )
+) +
+  geom_errorbar(aes(ymin=ci_lo_95,ymax=ci_hi_95),
+                position=position_dodge(width=0.5),
+                width=0,
+                lwd=0.5
+  ) +
+  geom_errorbar(aes(ymin=ci_lo_90,ymax=ci_hi_90),
+                position=position_dodge(width=0.5),
+                width=0,
+                lwd=1
+  ) +
+  geom_point(aes(y=est,shape=Study),
+             position=position_dodge(width=0.5),
+             size=2
+  ) +
+  geom_hline(yintercept = 0,lty=2) +
+  facet_wrap(~ outcome,scales="free") +
+  scale_color_manual(breaks=c(F,T),values = c("black","blue"),guide="none") +
+  coord_flip(ylim=c(-0.1,0.2)) +
+  theme_bw(base_family = "sans") +
+  theme(strip.background = element_rect(fill="white"),legend.position = "bottom",legend.margin = margin(0,0,0,-3,"lines")) +
+  ylab('Treatment effect of conservative seed vs. liberal seed video (95% and 90% CIs)') +
+  xlab(NULL)
+
+understanding_plot <- (understanding_plot_algo / understanding_plot_seed) +
+  plot_layout(heights = c(2, 1))
+
+ggsave(understanding_plot,
+       filename = "../results/understanding_3studies.png",width=12,height=8.5)
+
+## Base-control Figures ----------------------------------------------------
+
+coefs_basecontrol_guns <- read_csv("../results/intermediate data/gun control (issue 1)/guncontrol_padj_basecontrol_pretty.csv") %>%
+  mutate(est = case_when(layer3_specificoutcome=="pro_fraction_chosen" ~ -1*est,
+                         layer3_specificoutcome!="pro_fraction_chosen" ~ est),
+         layer2_treatmentcontrast = dplyr::recode(layer2_treatmentcontrast,
+                                                  "pro 31 - pro 22"="con 31 - con 22",
+                                                  "anti 31 - anti 22"="lib 31 - lib 22",
+                                                  "neutral anti 31 - neutral anti 22"="neutral lib 31 - neutral lib 22",
+                                                  "neutral pro 22 - neutral anti 22"="neutral con 22 - neutral lib 22",
+                                                  "neutral pro 31 - neutral anti 31"="neutral con 31 - neutral lib 31",
+                                                  "neutral pro 31 - neutral pro 22"="neutral con 31 - neutral con 22"
+         ))
+coefs_basecontrol <- read_csv("../results/intermediate data/minimum wage (issue 2)/padj_basecontrol_pretty.csv") %>%
+  mutate(layer2_treatmentcontrast = dplyr::recode(layer2_treatmentcontrast,
+                                                  "pro 31 - pro 22"="lib 31 - lib 22",
+                                                  "anti 31 - anti 22"="con 31 - con 22",
+                                                  "neutral anti 31 - neutral anti 22"="neutral con 31 - neutral con 22",
+                                                  "neutral anti 22 - neutral pro 22"="neutral con 22 - neutral lib 22",
+                                                  "neutral anti 31 - neutral pro 31"="neutral con 31 - neutral lib 31",
+                                                  "neutral pro 31 - neutral pro 22"="neutral lib 31 - neutral lib 22"
+  ))
+coefs_basecontrol_yg <- read_csv("../results/intermediate data/minimum wage (issue 2)/padj_basecontrol_pretty_yg.csv") %>%
+  mutate(layer2_treatmentcontrast = dplyr::recode(layer2_treatmentcontrast,
+                                                  "pro 31 - pro 22"="lib 31 - lib 22",
+                                                  "anti 31 - anti 22"="con 31 - con 22",
+                                                  "neutral anti 31 - neutral anti 22"="neutral con 31 - neutral con 22",
+                                                  "neutral anti 22 - neutral pro 22"="neutral con 22 - neutral lib 22",
+                                                  "neutral anti 31 - neutral pro 31"="neutral con 31 - neutral lib 31",
+                                                  "neutral pro 31 - neutral pro 22"="neutral lib 31 - neutral lib 22"
+  ))
+coefs_basecontrol <- bind_rows(mutate(coefs_basecontrol_guns,Sample="Gun Control\n(MTurk)"),
+                               mutate(coefs_basecontrol,Sample="Minimum Wage\n(MTurk)"),
+                               mutate(coefs_basecontrol_yg,Sample="Minimum Wage\n(YouGov)")) %>%
+  mutate(Sample = factor(Sample,levels=c("Minimum Wage\n(YouGov)","Minimum Wage\n(MTurk)","Gun Control\n(MTurk)"),ordered=T)) %>%
+  mutate(layer1_hypothesisfamily = recode(layer1_hypothesisfamily,
+                                          "gunpolicy"="policy",
+                                          "mwpolicy"="policy"),
+         layer3_specificoutcome = recode(layer3_specificoutcome,
+                                         "gun_index_w2"="policyindex",
+                                         "mw_index_w2"="policyindex"))
+
+# look at significant effects:
+coefs_basecontrol %>% filter(!str_detect(layer2_treatmentcontrast,"neutral") & p.adj < .05 & layer3_specificoutcome != 'overall')
+
+
+coefs_basecontrol %>% filter(str_detect(layer2_treatmentcontrast,"neutral") & p.adj < .05 & layer3_specificoutcome != 'overall' &
+                               ((str_detect(layer2_treatmentcontrast,"lib") & !str_detect(layer2_treatmentcontrast,"con")) |
+                                  !(str_detect(layer2_treatmentcontrast,"lib") & str_detect(layer2_treatmentcontrast,"con"))))
+
+outcome_labels <- data.frame(outcome = c(
+  "Liberal videos\nchosen (fraction)",
+  "Likes & saves\nminus dislikes (#)",
+  "Total watch\ntime (hrs)",
+  "Policy\nindex",
+  "Trust in\nmajor news",
+  "Trust in\nYouTube",
+  "Never fabrication\nby major news",
+  "Never fabrication\nby YouTube",
+  "Perceived intelligence",
+  "Feeling thermometer",
+  "Comfort as friend"),
+  specificoutcome = c(
+    "pro_fraction_chosen",
+    "positive_interactions",
+    "platform_duration",
+    "policyindex",
+    "trust_majornews_w2",
+    "trust_youtube_w2",
+    "fabricate_majornews_w2",
+    "fabricate_youtube_w2",
+    "affpol_smart_w2",
+    "affpol_ft_w2",
+    "affpol_comfort_w2"),
+  family = c(
+    rep("Platform Interaction",3),
+    rep("Policy Attitudes\n(unit scale, + is more conservative)",1),
+    rep("Media Trust\n(unit scale, + is more trusting)",4),
+    rep("Affective Polarization\n(unit scale, + is greater polarization)",3))
+)
+
+##### Liberals #####
+coefs_third1_basecontrol <- coefs_basecontrol %>%
+  filter(layer2_treatmentcontrast == "lib 31 - lib 22" &
+           layer3_specificoutcome != "overall")
+
+coefs_third1_basecontrol$outcome = outcome_labels$outcome[match(coefs_third1_basecontrol$layer3_specificoutcome,
+                                                                outcome_labels$specificoutcome)]
+
+
+coefs_third1_basecontrol$family = outcome_labels$family[match(coefs_third1_basecontrol$layer3_specificoutcome,
+                                                              outcome_labels$specificoutcome)]
+
+
+coefs_third1_basecontrol <- mutate(coefs_third1_basecontrol,
+                                   family = factor(family,
+                                                   levels = c(
+                                                     "Policy Attitudes\n(unit scale, + is more conservative)",
+                                                     "Platform Interaction",
+                                                     "Media Trust\n(unit scale, + is more trusting)",
+                                                     "Affective Polarization\n(unit scale, + is greater polarization)"),ordered = T))
+
+## manipulate to get all unit scales:
+coefs_third1_basecontrol$est[coefs_third1_basecontrol$layer3_specificoutcome=="platform_duration"] <- coefs_third1_basecontrol$est[coefs_third1_basecontrol$layer3_specificoutcome=="platform_duration"]/3600
+coefs_third1_basecontrol$se[coefs_third1_basecontrol$layer3_specificoutcome=="platform_duration"] <- coefs_third1_basecontrol$se[coefs_third1_basecontrol$layer3_specificoutcome=="platform_duration"]/3600
+
+coefs_third1_basecontrol$est[coefs_third1_basecontrol$layer3_specificoutcome=="affpol_ft_w2"] <- coefs_third1_basecontrol$est[coefs_third1_basecontrol$layer3_specificoutcome=="affpol_ft_w2"]/100
+coefs_third1_basecontrol$se[coefs_third1_basecontrol$layer3_specificoutcome=="affpol_ft_w2"] <- coefs_third1_basecontrol$se[coefs_third1_basecontrol$layer3_specificoutcome=="affpol_ft_w2"]/100
+
+coefs_third1_basecontrol <- coefs_third1_basecontrol %>%
+  mutate(ci_lo_99 = est + qnorm(0.001)*se,
+         ci_hi_99 = est + qnorm(0.999)*se,
+         ci_lo_95 = est + qnorm(0.025)*se,
+         ci_hi_95 = est + qnorm(0.975)*se,
+         ci_lo_90 = est + qnorm(0.05)*se,
+         ci_hi_90 = est + qnorm(0.95)*se,
+         plotorder = rep((nrow(coefs_third1_basecontrol)/3):1,3),
+         alpha = ifelse(p.adj<0.05, T, F),
+         alpha = as.logical(alpha),
+         alpha = replace_na(alpha,F),
+         Sample_color = as.character(Sample),
+         Sample_color = replace(Sample_color,alpha==F,"insig")
+  )
+tabyl(coefs_third1_basecontrol,Sample_color)
+
+(coefplot_third1_basecontrol <- ggplot(filter(coefs_third1_basecontrol),aes(x=plotorder,group=Sample,col=Sample,alpha=alpha)) +
+    geom_errorbar(aes(ymin=ci_lo_95,ymax=ci_hi_95,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=0.5,alpha=0.25) +
+    geom_errorbar(aes(ymin=ci_lo_90,ymax=ci_hi_90,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=1,alpha=0.25) +
+    geom_point(aes(y=est,shape=Sample,col=Sample_color),position=position_dodge(width=0.5),size=3,alpha=0.25) +
+    geom_text(data=filter(coefs_third1_basecontrol,layer1_hypothesisfamily=="policy"),aes(y=est+0.006,label=Sample),alpha=1,position=position_dodge(width=0.5),size=3) +
+    geom_hline(yintercept = 0,lty=2) +
+    facet_wrap(~family,ncol=1,scales="free") +
+    scale_x_continuous("",
+                       breaks = coefs_third1_basecontrol$plotorder,labels = coefs_third1_basecontrol$outcome) +
+    scale_y_continuous("Treatment effect of 3/1 vs. 2/2\nalgorithm, all liberal seed\n(95% and 90% CIs)") +
+    scale_color_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)","insig"),values=c(vgreen,red_mit,blue_mit,"black")) +
+    scale_shape_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)"),values=c(16,17,18)) +
+    scale_alpha_manual(breaks=c(F,T),values=c(0.25,1)) +
+    coord_flip() +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"),
+          legend.position = "none",
+    )
+)
+ggsave(coefplot_third1_basecontrol,
+       filename = "../results/coefplot_third1_basecontrol_3studies.png",width=5,height=8.5)
+ggsave(coefplot_third1_basecontrol,
+       filename = "../results/coefplot_third1_basecontrol_3studies.pdf",width=5,height=8.5)
+
+(coefplot_third1_basecontrol_empty <- ggplot(filter(coefs_third1_basecontrol),aes(x=plotorder,group=Sample,alpha=alpha,col=Sample)) +
+    geom_blank(aes(ymin=ci_lo_95,ymax=ci_hi_95),position=position_dodge(width=0.5),width=0,lwd=0.5) +
+    geom_blank(aes(ymin=ci_lo_90,ymax=ci_hi_90),position=position_dodge(width=0.5),width=0,lwd=1) +
+    geom_blank(aes(y=est,shape=Sample),position=position_dodge(width=0.5),size=3) +
+    geom_blank(data=filter(coefs_third1_basecontrol,layer1_hypothesisfamily=="policy"),aes(y=est+0.006,label=Sample),position=position_dodge(width=0.5),size=3) +
+    geom_hline(yintercept = 0,lty=2) +
+    facet_wrap(~family,ncol=1,scales="free") +
+    scale_x_continuous("",
+                       breaks = coefs_third1_basecontrol$plotorder,labels = coefs_third1_basecontrol$outcome) +
+    scale_y_continuous("Treatment effect of 3/1 vs. 2/2\nalgorithm, all liberal seed\n(95% and 90% CIs)") +
+    scale_color_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)","insig"),values=c(vgreen,red_mit,blue_mit,"black")) +
+    scale_shape_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)"),values=c(16,17,18)) +
+    coord_flip() +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"),legend.position = "none")
+)
+ggsave(coefplot_third1_basecontrol_empty,
+       filename = "../results/coefplot_third1_basecontrol_empty_3studies.png",width=5,height=8.5)
+
+(coefplot_third1_basecontrol_3studies_toptwo <- ggplot(filter(coefs_third1_basecontrol,layer1_hypothesisfamily %in% c("policy","platform")),aes(x=plotorder,group=Sample,col=Sample,alpha=alpha)) +
+    geom_errorbar(aes(ymin=ci_lo_95,ymax=ci_hi_95,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=0.5,alpha=0.25) +
+    geom_errorbar(aes(ymin=ci_lo_90,ymax=ci_hi_90,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=1,alpha=0.25) +
+    geom_point(aes(y=est,shape=Sample,col=Sample_color),position=position_dodge(width=0.5),size=3,alpha=0.25) +
+    geom_text(data=filter(coefs_third1_basecontrol,layer1_hypothesisfamily=="policy"),aes(y=est+0.006,label=Sample),alpha=1,position=position_dodge(width=0.5),size=3) +
+    geom_hline(yintercept = 0,lty=2) +
+    facet_wrap(~family,ncol=1,scales="free") +
+    scale_x_continuous("",
+                       breaks = coefs_third1_basecontrol$plotorder,labels = coefs_third1_basecontrol$outcome) +
+    scale_y_continuous("Treatment effect of 3/1 vs. 2/2\nalgorithm, all liberal seed\n(95% and 90% CIs)") +
+    scale_color_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)","insig"),values=c(vgreen,red_mit,blue_mit,"black")) +
+    scale_shape_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)"),values=c(16,17,18)) +
+    scale_alpha_manual(breaks=c(F,T),values=c(0.25,1)) +
+    coord_flip() +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"),
+          legend.position = "none",
+    )
+)
+ggsave(coefplot_third1_basecontrol_3studies_toptwo,
+       filename = "../results/coefplot_third1_basecontrol_3studies_toptwo.png",width=5,height=4.75)
+ggsave(coefplot_third1_basecontrol_3studies_toptwo,
+       filename = "../results/coefplot_third1_basecontrol_3studies_toptwo.pdf",width=5,height=4.75)
+
+
+##### Conservatives #####
+
+coefs_third3_basecontrol <- coefs_basecontrol %>%
+  filter(layer2_treatmentcontrast == "con 31 - con 22" &
+           layer3_specificoutcome != "overall")
+
+coefs_third3_basecontrol$outcome = outcome_labels$outcome[match(coefs_third3_basecontrol$layer3_specificoutcome,
+                                                                outcome_labels$specificoutcome)]
+
+coefs_third3_basecontrol$family = outcome_labels$family[match(coefs_third3_basecontrol$layer3_specificoutcome,
+                                                              outcome_labels$specificoutcome)]
+
+coefs_third3_basecontrol <- mutate(coefs_third3_basecontrol,
+                                   family = factor(family,levels = c("Policy Attitudes\n(unit scale, + is more conservative)","Platform Interaction","Media Trust\n(unit scale, + is more trusting)","Affective Polarization\n(unit scale, + is greater polarization)"),ordered = T))
+
+## manipulate to get all unit scales:
+coefs_third3_basecontrol$est[coefs_third3_basecontrol$layer3_specificoutcome=="platform_duration"] <- coefs_third3_basecontrol$est[coefs_third3_basecontrol$layer3_specificoutcome=="platform_duration"]/3600
+coefs_third3_basecontrol$se[coefs_third3_basecontrol$layer3_specificoutcome=="platform_duration"] <- coefs_third3_basecontrol$se[coefs_third3_basecontrol$layer3_specificoutcome=="platform_duration"]/3600
+
+coefs_third3_basecontrol$est[coefs_third3_basecontrol$layer3_specificoutcome=="affpol_ft_w2"] <- coefs_third3_basecontrol$est[coefs_third3_basecontrol$layer3_specificoutcome=="affpol_ft_w2"]/100
+coefs_third3_basecontrol$se[coefs_third3_basecontrol$layer3_specificoutcome=="affpol_ft_w2"] <- coefs_third3_basecontrol$se[coefs_third3_basecontrol$layer3_specificoutcome=="affpol_ft_w2"]/100
+
+coefs_third3_basecontrol <- coefs_third3_basecontrol %>%
+  mutate(ci_lo_99 = est + qnorm(0.001)*se,
+         ci_hi_99 = est + qnorm(0.999)*se,
+         ci_lo_95 = est + qnorm(0.025)*se,
+         ci_hi_95 = est + qnorm(0.975)*se,
+         ci_lo_90 = est + qnorm(0.05)*se,
+         ci_hi_90 = est + qnorm(0.95)*se,
+         plotorder = rep((nrow(coefs_third3_basecontrol)/3):1,3),
+         alpha = ifelse(p.adj<0.05, T, F),
+         alpha = as.logical(alpha),
+         alpha = replace_na(alpha,F),
+         Sample_color = as.character(Sample),
+         Sample_color = replace(Sample_color,alpha==F,"insig")
+  )
+
+
+(coefplot_third3_basecontrol <- ggplot(filter(coefs_third3_basecontrol),aes(x=plotorder,group=Sample,col=Sample,alpha=alpha)) +
+    geom_errorbar(aes(ymin=ci_lo_95,ymax=ci_hi_95,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=0.5) +
+    geom_errorbar(aes(ymin=ci_lo_90,ymax=ci_hi_90,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=1) +
+    geom_point(aes(y=est,shape=Sample,col=Sample_color),position=position_dodge(width=0.5),size=3) +
+    geom_text(data=filter(coefs_third3_basecontrol,layer1_hypothesisfamily=="policy"),aes(y=est+0.006,label=Sample),alpha=1,position=position_dodge(width=0.5),size=3) +
+    geom_hline(yintercept = 0,lty=2) +
+    facet_wrap(~family,ncol=1,scales="free") +
+    scale_x_continuous("",
+                       breaks = coefs_third3_basecontrol$plotorder,labels = coefs_third3_basecontrol$outcome) +
+    scale_y_continuous("Treatment effect of 3/1 vs. 2/2\nalgorithm, all conservative seed\n(95% and 90% CIs)") +
+    scale_color_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)","insig"),values=c(vgreen,red_mit,blue_mit,"black")) +
+    scale_shape_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)"),values=c(16,17,18)) +
+    scale_alpha_manual(breaks=c(F,T),values=c(0.25,1)) +
+    coord_flip() +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"),legend.position="none")
+)
+
+ggsave(coefplot_third3_basecontrol,
+       filename = "../results/coefplot_third3_basecontrol_3studies.png",width=5,height=8.5)
+ggsave(coefplot_third3_basecontrol,
+       filename = "../results/coefplot_third3_basecontrol_3studies.pdf",width=5,height=8.5)
+
+(coefplot_third3_basecontrol_empty <- ggplot(filter(coefs_third3_basecontrol),aes(x=plotorder,group=Sample,col=ifelse(p.adj<0.05,T,F))) +
+    geom_blank(aes(ymin=ci_lo_95,ymax=ci_hi_95),position=position_dodge(width=0.5),width=0,lwd=0.5) +
+    geom_blank(aes(ymin=ci_lo_90,ymax=ci_hi_90),position=position_dodge(width=0.5),width=0,lwd=1) +
+    geom_blank(aes(y=est,shape=Sample),position=position_dodge(width=0.5),size=2) +
+    geom_blank(data=filter(coefs_third3_basecontrol,layer1_hypothesisfamily=="policy"),aes(y=est+0.006,label=Sample),position=position_dodge(width=0.5),size=3) +
+    geom_hline(yintercept = 0,lty=2) +
+    facet_wrap(~family,ncol=1,scales="free") +
+    scale_x_continuous("",
+                       breaks = coefs_third3_basecontrol$plotorder,labels = coefs_third3_basecontrol$outcome) +
+    scale_y_continuous("Treatment effect of 3/1 vs. 2/2\nalgorithm, all conservative seed\n(95% and 90% CIs)") +
+    scale_color_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)","insig"),values=c(vgreen,red_mit,blue_mit,"black")) +
+    scale_shape_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)"),values=c(16,17,18)) +
+    coord_flip(ylim=c(-0.17,0.17)) +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"),legend.position="none")
+)
+ggsave(coefplot_third3_basecontrol_empty,
+       filename = "../results/coefplot_third3_basecontrol_empty_3studies.png",width=5,height=8.5)
+
+(coefplot_third3_basecontrol_toptwo <- ggplot(filter(coefs_third3_basecontrol,layer1_hypothesisfamily %in% c("policy","platform")),aes(x=plotorder,group=Sample,col=Sample,alpha=alpha)) +
+    geom_errorbar(aes(ymin=ci_lo_95,ymax=ci_hi_95,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=0.5) +
+    geom_errorbar(aes(ymin=ci_lo_90,ymax=ci_hi_90,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=1) +
+    geom_point(aes(y=est,shape=Sample,col=Sample_color),position=position_dodge(width=0.5),size=3) +
+    geom_text(data=filter(coefs_third3_basecontrol,layer1_hypothesisfamily=="policy"),aes(y=est+0.006,label=Sample),alpha=1,position=position_dodge(width=0.5),size=3) +
+    geom_hline(yintercept = 0,lty=2) +
+    facet_wrap(~family,ncol=1,scales="free") +
+    scale_x_continuous("",
+                       breaks = coefs_third3_basecontrol$plotorder,labels = coefs_third3_basecontrol$outcome) +
+    scale_y_continuous("Treatment effect of 3/1 vs. 2/2\nalgorithm, all conservative seed\n(95% and 90% CIs)") +
+    scale_color_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)","insig"),values=c(vgreen,red_mit,blue_mit,"black")) +
+    scale_shape_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)"),values=c(16,17,18)) +
+    scale_alpha_manual(breaks=c(F,T),values=c(0.25,1)) +
+    coord_flip() +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"),legend.position="none")
+)
+
+ggsave(coefplot_third3_basecontrol_toptwo,
+       filename = "../results/coefplot_third3_basecontrol_3studies_toptwo.png",width=5,height=4.75)
+ggsave(coefplot_third3_basecontrol_toptwo,
+       filename = "../results/coefplot_third3_basecontrol_3studies_toptwo.pdf",width=5,height=4.75)
+
+
+##### Moderates (algorithm) #####
+
+coefs_third2_pro_basecontrol <- coefs_basecontrol %>%
+  filter(layer2_treatmentcontrast == "neutral lib 31 - neutral lib 22" &
+           layer3_specificoutcome != "overall")
+
+coefs_third2_pro_basecontrol$outcome = outcome_labels$outcome[match(coefs_third2_pro_basecontrol$layer3_specificoutcome,
+                                                                    outcome_labels$specificoutcome)]
+
+coefs_third2_pro_basecontrol$family = outcome_labels$family[match(coefs_third2_pro_basecontrol$layer3_specificoutcome,
+                                                                  outcome_labels$specificoutcome)]
+
+coefs_third2_pro_basecontrol <- mutate(coefs_third2_pro_basecontrol,
+                                       family = factor(family,levels = c("Policy Attitudes\n(unit scale, + is more conservative)","Platform Interaction","Media Trust\n(unit scale, + is more trusting)","Affective Polarization\n(unit scale, + is greater polarization)"),ordered = T))
+
+## manipulate to get all unit scales:
+coefs_third2_pro_basecontrol$est[coefs_third2_pro_basecontrol$layer3_specificoutcome=="platform_duration"] <- coefs_third2_pro_basecontrol$est[coefs_third2_pro_basecontrol$layer3_specificoutcome=="platform_duration"]/3600
+coefs_third2_pro_basecontrol$se[coefs_third2_pro_basecontrol$layer3_specificoutcome=="platform_duration"] <- coefs_third2_pro_basecontrol$se[coefs_third2_pro_basecontrol$layer3_specificoutcome=="platform_duration"]/3600
+
+coefs_third2_pro_basecontrol$est[coefs_third2_pro_basecontrol$layer3_specificoutcome=="affpol_ft_w2"] <- coefs_third2_pro_basecontrol$est[coefs_third2_pro_basecontrol$layer3_specificoutcome=="affpol_ft_w2"]/100
+coefs_third2_pro_basecontrol$se[coefs_third2_pro_basecontrol$layer3_specificoutcome=="affpol_ft_w2"] <- coefs_third2_pro_basecontrol$se[coefs_third2_pro_basecontrol$layer3_specificoutcome=="affpol_ft_w2"]/100
+
+coefs_third2_pro_basecontrol <- coefs_third2_pro_basecontrol %>%
+  mutate(ci_lo_99 = est + qnorm(0.001)*se,
+         ci_hi_99 = est + qnorm(0.999)*se,
+         ci_lo_95 = est + qnorm(0.025)*se,
+         ci_hi_95 = est + qnorm(0.975)*se,
+         ci_lo_90 = est + qnorm(0.05)*se,
+         ci_hi_90 = est + qnorm(0.95)*se,
+         plotorder = rep((nrow(coefs_third2_pro_basecontrol)/3):1,3),
+         alpha = ifelse(p.adj<0.05, T, F),
+         alpha = as.logical(alpha),
+         alpha = replace_na(alpha,F),
+         Sample_color = as.character(Sample),
+         Sample_color = replace(Sample_color,alpha==F,"insig")
+  )
+writeLines(as.character(abs(round(filter(coefs_third2_pro_basecontrol,layer3_specificoutcome=="platform_duration" & Sample=="Minimum Wage\n(YouGov)")$est*60,1))),
+           con = "../results/beta_minutes_recsys_duration_third2_proseed_study3.tex",sep="%")
+
+(coefplot_third2_pro_basecontrol <- ggplot(filter(coefs_third2_pro_basecontrol),aes(x=plotorder,group=Sample,col=Sample,alpha=alpha)) +
+    geom_errorbar(aes(ymin=ci_lo_95,ymax=ci_hi_95,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=0.5) +
+    geom_errorbar(aes(ymin=ci_lo_90,ymax=ci_hi_90,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=1) +
+    geom_point(aes(y=est,shape=Sample,col=Sample_color),position=position_dodge(width=0.5),size=3) +
+    geom_text(data=filter(coefs_third2_pro_basecontrol,layer1_hypothesisfamily=="policy"),aes(y=est+0.006,label=Sample),alpha=1,position=position_dodge(width=0.5),size=3) +
+    geom_hline(yintercept = 0,lty=2) +
+    facet_wrap(~family,ncol=1,scales="free") +
+    scale_x_continuous("",
+                       breaks = coefs_third2_pro_basecontrol$plotorder,labels = coefs_third2_pro_basecontrol$outcome) +
+    scale_y_continuous("Treatment effect of 3/1 vs. 2/2\nalgorithm, all liberal seed\n(95% and 90% CIs)") +
+    scale_color_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)","insig"),values=c(vgreen,red_mit,blue_mit,"black")) +
+    scale_shape_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)"),values=c(16,17,18)) +
+    scale_alpha_manual(breaks=c(F,T),values=c(0.25,1)) +
+    coord_flip() +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"),legend.position="none")
+)
+ggsave(coefplot_third2_pro_basecontrol,
+       filename = "../results/coefplot_third2_pro_basecontrol_3studies.png",width=5,height=8.5)
+ggsave(coefplot_third2_pro_basecontrol,
+       filename = "../results/coefplot_third2_pro_basecontrol_3studies.pdf",width=5,height=8.5)
+
+(coefplot_third2_pro_basecontrol_empty <- ggplot(filter(coefs_third2_pro_basecontrol),aes(x=plotorder,group=Sample,col=Sample,alpha=alpha)) +
+    geom_blank(aes(ymin=ci_lo_95,ymax=ci_hi_95),position=position_dodge(width=0.5),width=0,lwd=0.5) +
+    geom_blank(aes(ymin=ci_lo_90,ymax=ci_hi_90),position=position_dodge(width=0.5),width=0,lwd=1) +
+    geom_blank(aes(y=est,shape=Sample),position=position_dodge(width=0.5),size=3) +
+    geom_blank(data=filter(coefs_third2_pro_basecontrol,layer1_hypothesisfamily=="policy"),aes(y=est+0.006,label=Sample),position=position_dodge(width=0.5),size=3) +
+    geom_hline(yintercept = 0,lty=2) +
+    facet_wrap(~family,ncol=1,scales="free") +
+    scale_x_continuous("",
+                       breaks = coefs_third2_pro_basecontrol$plotorder,labels = coefs_third2_pro_basecontrol$outcome) +
+    scale_y_continuous("Treatment effect of 3/1 vs. 2/2\nalgorithm, all liberal seed\n(95% and 90% CIs)") +
+    scale_color_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)","insig"),values=c(vgreen,red_mit,blue_mit,"black")) +
+    scale_shape_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)"),values=c(16,17,18)) +
+    scale_alpha_manual(breaks=c(F,T),values=c(0.25,1)) +
+    coord_flip() +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"),legend.position="none")
+)
+ggsave(coefplot_third2_pro_basecontrol_empty,
+       filename = "../results/coefplot_third2_pro_basecontrol_empty_3studies.png",width=5,height=8.5)
+
+(coefplot_third2_pro_basecontrol_toptwo <- ggplot(filter(coefs_third2_pro_basecontrol,layer1_hypothesisfamily %in% c("policy","platform")),aes(x=plotorder,group=Sample,col=Sample,alpha=alpha)) +
+    geom_errorbar(aes(ymin=ci_lo_95,ymax=ci_hi_95,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=0.5) +
+    geom_errorbar(aes(ymin=ci_lo_90,ymax=ci_hi_90,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=1) +
+    geom_point(aes(y=est,shape=Sample,col=Sample_color),position=position_dodge(width=0.5),size=3) +
+    geom_text(data=filter(coefs_third2_pro_basecontrol,layer1_hypothesisfamily=="policy"),aes(y=est+0.006,label=Sample),alpha=1,position=position_dodge(width=0.5),size=3) +
+    geom_hline(yintercept = 0,lty=2) +
+    facet_wrap(~family,ncol=1,scales="free") +
+    scale_x_continuous("",
+                       breaks = coefs_third2_pro_basecontrol$plotorder,labels = coefs_third2_pro_basecontrol$outcome) +
+    scale_y_continuous("Treatment effect of 3/1 vs. 2/2\nalgorithm, all liberal seed\n(95% and 90% CIs)") +
+    scale_color_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)","insig"),values=c(vgreen,red_mit,blue_mit,"black")) +
+    scale_shape_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)"),values=c(16,17,18)) +
+    scale_alpha_manual(breaks=c(F,T),values=c(0.25,1)) +
+    coord_flip() +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"),legend.position="none")
+)
+ggsave(coefplot_third2_pro_basecontrol_toptwo,
+       filename = "../results/coefplot_third2_pro_basecontrol_3studies_toptwo.png",width=5,height=4.75)
+ggsave(coefplot_third2_pro_basecontrol_toptwo,
+       filename = "../results/coefplot_third2_pro_basecontrol_3studies_toptwo.pdf",width=5,height=4.75)
+
+coefs_third2_anti_basecontrol <- coefs_basecontrol %>%
+  filter(layer2_treatmentcontrast == "neutral con 31 - neutral con 22" &
+           layer3_specificoutcome != "overall")
+
+coefs_third2_anti_basecontrol$outcome = outcome_labels$outcome[match(coefs_third2_anti_basecontrol$layer3_specificoutcome,
+                                                                     outcome_labels$specificoutcome)]
+
+coefs_third2_anti_basecontrol$family = outcome_labels$family[match(coefs_third2_anti_basecontrol$layer3_specificoutcome,
+                                                                   outcome_labels$specificoutcome)]
+
+coefs_third2_anti_basecontrol <- mutate(coefs_third2_anti_basecontrol,
+                                        family = factor(family,levels = c("Policy Attitudes\n(unit scale, + is more conservative)","Platform Interaction","Media Trust\n(unit scale, + is more trusting)","Affective Polarization\n(unit scale, + is greater polarization)"),ordered = T))
+
+## manipulate to get all unit scales:
+coefs_third2_anti_basecontrol$est[coefs_third2_anti_basecontrol$layer3_specificoutcome=="platform_duration"] <- coefs_third2_anti_basecontrol$est[coefs_third2_anti_basecontrol$layer3_specificoutcome=="platform_duration"]/3600
+coefs_third2_anti_basecontrol$se[coefs_third2_anti_basecontrol$layer3_specificoutcome=="platform_duration"] <- coefs_third2_anti_basecontrol$se[coefs_third2_anti_basecontrol$layer3_specificoutcome=="platform_duration"]/3600
+
+coefs_third2_anti_basecontrol$est[coefs_third2_anti_basecontrol$layer3_specificoutcome=="affpol_ft_w2"] <- coefs_third2_anti_basecontrol$est[coefs_third2_anti_basecontrol$layer3_specificoutcome=="affpol_ft_w2"]/100
+coefs_third2_anti_basecontrol$se[coefs_third2_anti_basecontrol$layer3_specificoutcome=="affpol_ft_w2"] <- coefs_third2_anti_basecontrol$se[coefs_third2_anti_basecontrol$layer3_specificoutcome=="affpol_ft_w2"]/100
+
+coefs_third2_anti_basecontrol <- coefs_third2_anti_basecontrol %>%
+  mutate(ci_lo_99 = est + qnorm(0.001)*se,
+         ci_hi_99 = est + qnorm(0.999)*se,
+         ci_lo_95 = est + qnorm(0.025)*se,
+         ci_hi_95 = est + qnorm(0.975)*se,
+         ci_lo_90 = est + qnorm(0.05)*se,
+         ci_hi_90 = est + qnorm(0.95)*se,
+         plotorder = rep((nrow(coefs_third2_anti_basecontrol)/3):1,3),
+         alpha = ifelse(p.adj<0.05, T, F),
+         alpha = as.logical(alpha),
+         alpha = replace_na(alpha,F),
+         Sample_color = as.character(Sample),
+         Sample_color = replace(Sample_color,alpha==F,"insig")
+  )
+
+writeLines(as.character(abs(round(filter(coefs_third2_anti_basecontrol,layer3_specificoutcome=="platform_duration" & Sample=="Gun Control\n(MTurk)")$est*60,1))),
+           con = "../results/beta_minutes_recsys_duration_third2_antiseed_study1.tex",sep="%")
+
+(coefplot_third2_anti_basecontrol <- ggplot(filter(coefs_third2_anti_basecontrol),aes(x=plotorder,group=Sample,col=Sample,alpha=alpha)) +
+    geom_errorbar(aes(ymin=ci_lo_95,ymax=ci_hi_95,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=0.5) +
+    geom_errorbar(aes(ymin=ci_lo_90,ymax=ci_hi_90,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=1) +
+    geom_point(aes(y=est,shape=Sample,col=Sample_color),position=position_dodge(width=0.5),size=3) +
+    geom_text(data=filter(coefs_third2_anti_basecontrol,layer1_hypothesisfamily=="policy"),aes(y=est+0.006,label=Sample),alpha=1,position=position_dodge(width=0.5),size=3) +
+    geom_hline(yintercept = 0,lty=2) +
+    facet_wrap(~family,ncol=1,scales="free") +
+    scale_x_continuous("",
+                       breaks = coefs_third2_anti_basecontrol$plotorder,labels = coefs_third2_anti_basecontrol$outcome) +
+    scale_y_continuous("Treatment effect of 3/1 vs. 2/2\nalgorithm, all conservative seed\n(95% and 90% CIs)") +
+    scale_color_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)","insig"),values=c(vgreen,red_mit,blue_mit,"black")) +
+    scale_shape_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)"),values=c(16,17,18)) +
+    scale_alpha_manual(breaks=c(F,T),values=c(0.25,1)) +
+    coord_flip() +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"),legend.position="none")
+)
+ggsave(coefplot_third2_anti_basecontrol,
+       filename = "../results/coefplot_third2_anti_basecontrol_3studies.png",width=5,height=8.5)
+ggsave(coefplot_third2_anti_basecontrol,
+       filename = "../results/coefplot_third2_anti_basecontrol_3studies.pdf",width=5,height=8.5)
+
+(coefplot_third2_anti_basecontrol_empty <- ggplot(filter(coefs_third2_anti_basecontrol),aes(x=plotorder,group=Sample,col=ifelse(p.adj<0.05,T,F))) +
+    geom_blank(aes(ymin=ci_lo_95,ymax=ci_hi_95),position=position_dodge(width=0.5),width=0,lwd=0.5) +
+    geom_blank(aes(ymin=ci_lo_90,ymax=ci_hi_90),position=position_dodge(width=0.5),width=0,lwd=1) +
+    geom_blank(aes(y=est,shape=Sample),position=position_dodge(width=0.5),size=2) +
+    geom_hline(yintercept = 0,lty=2) +
+    facet_wrap(~family,ncol=1,scales="free") +
+    scale_x_continuous("",
+                       breaks = coefs_third2_anti_basecontrol$plotorder,labels = coefs_third2_anti_basecontrol$outcome) +
+    scale_y_continuous("Treatment effect of 3/1 vs. 2/2\nalgorithm, all conservative seed\n(95% and 90% CIs)") +
+    scale_color_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)","insig"),values=c(vgreen,red_mit,blue_mit,"black")) +
+    scale_shape_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)"),values=c(16,17,18)) +
+    scale_alpha_manual(breaks=c(F,T),values=c(0.25,1)) +
+    coord_flip() +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"),legend.position="none")
+)
+ggsave(coefplot_third2_anti_basecontrol_empty,
+       filename = "../results/coefplot_third2_anti_basecontrol_empty_3studies.png",width=5,height=8.5)
+
+(coefplot_third2_anti_basecontrol_toptwo <- ggplot(filter(coefs_third2_anti_basecontrol,layer1_hypothesisfamily %in% c("policy","platform")),aes(x=plotorder,group=Sample,col=Sample,alpha=alpha)) +
+    geom_errorbar(aes(ymin=ci_lo_95,ymax=ci_hi_95,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=0.5) +
+    geom_errorbar(aes(ymin=ci_lo_90,ymax=ci_hi_90,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=1) +
+    geom_point(aes(y=est,shape=Sample,col=Sample_color),position=position_dodge(width=0.5),size=3) +
+    geom_text(data=filter(coefs_third2_anti_basecontrol,layer1_hypothesisfamily=="policy"),aes(y=est+0.006,label=Sample),alpha=1,position=position_dodge(width=0.5),size=3) +
+    geom_hline(yintercept = 0,lty=2) +
+    facet_wrap(~family,ncol=1,scales="free") +
+    scale_x_continuous("",
+                       breaks = coefs_third2_anti_basecontrol$plotorder,labels = coefs_third2_anti_basecontrol$outcome) +
+    scale_y_continuous("Treatment effect of 3/1 vs. 2/2\nalgorithm, all conservative seed\n(95% and 90% CIs)") +
+    scale_color_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)","insig"),values=c(vgreen,red_mit,blue_mit,"black")) +
+    scale_shape_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)"),values=c(16,17,18)) +
+    scale_alpha_manual(breaks=c(F,T),values=c(0.25,1)) +
+    coord_flip() +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"),legend.position="none")
+)
+ggsave(coefplot_third2_anti_basecontrol_toptwo,
+       filename = "../results/coefplot_third2_anti_basecontrol_3studies_toptwo.png",width=5,height=4.75)
+ggsave(coefplot_third2_anti_basecontrol_toptwo,
+       filename = "../results/coefplot_third2_anti_basecontrol_3studies_toptwo.pdf",width=5,height=4.75)
+
+
+##### Moderates (seed) #####
+coefs_third2_31_basecontrol <- coefs_basecontrol %>%
+  filter(layer2_treatmentcontrast == "neutral con 31 - neutral lib 31" &
+           layer3_specificoutcome != "overall")
+
+coefs_third2_31_basecontrol$outcome = outcome_labels$outcome[match(coefs_third2_31_basecontrol$layer3_specificoutcome,
+                                                                   outcome_labels$specificoutcome)]
+
+coefs_third2_31_basecontrol$family = outcome_labels$family[match(coefs_third2_31_basecontrol$layer3_specificoutcome,
+                                                                 outcome_labels$specificoutcome)]
+
+coefs_third2_31_basecontrol <- mutate(coefs_third2_31_basecontrol,
+                                      family = factor(family,levels = c("Policy Attitudes\n(unit scale, + is more conservative)","Platform Interaction","Media Trust\n(unit scale, + is more trusting)","Affective Polarization\n(unit scale, + is greater polarization)"),ordered = T))
+
+## manipulate to get all unit scales:
+coefs_third2_31_basecontrol$est[coefs_third2_31_basecontrol$layer3_specificoutcome=="platform_duration"] <- coefs_third2_31_basecontrol$est[coefs_third2_31_basecontrol$layer3_specificoutcome=="platform_duration"]/3600
+coefs_third2_31_basecontrol$se[coefs_third2_31_basecontrol$layer3_specificoutcome=="platform_duration"] <- coefs_third2_31_basecontrol$se[coefs_third2_31_basecontrol$layer3_specificoutcome=="platform_duration"]/3600
+
+coefs_third2_31_basecontrol$est[coefs_third2_31_basecontrol$layer3_specificoutcome=="affpol_ft_w2"] <- coefs_third2_31_basecontrol$est[coefs_third2_31_basecontrol$layer3_specificoutcome=="affpol_ft_w2"]/100
+coefs_third2_31_basecontrol$se[coefs_third2_31_basecontrol$layer3_specificoutcome=="affpol_ft_w2"] <- coefs_third2_31_basecontrol$se[coefs_third2_31_basecontrol$layer3_specificoutcome=="affpol_ft_w2"]/100
+
+coefs_third2_31_basecontrol <- coefs_third2_31_basecontrol %>%
+  mutate(ci_lo_99 = est + qnorm(0.001)*se,
+         ci_hi_99 = est + qnorm(0.999)*se,
+         ci_lo_95 = est + qnorm(0.025)*se,
+         ci_hi_95 = est + qnorm(0.975)*se,
+         ci_lo_90 = est + qnorm(0.05)*se,
+         ci_hi_90 = est + qnorm(0.95)*se,
+         plotorder = rep((nrow(coefs_third2_31_basecontrol)/3):1,3),
+         alpha = ifelse(p.adj<0.05, T, F),
+         alpha = as.logical(alpha),
+         alpha = replace_na(alpha,F),
+         Sample_color = as.character(Sample),
+         Sample_color = replace(Sample_color,alpha==F,"insig")
+  )
+
+dummy_df <- data.frame(family=c("Platform Interaction","Platform Interaction"),est=c(-0.5,0.5),plotorder=c(9,9),Sample=c("Gun Control\n(MTurk)","Gun Control\n(MTurk)"),alpha=c(FALSE,FALSE)) %>% mutate(family=factor(family))
+
+(coefplot_third2_31_basecontrol <- ggplot(filter(coefs_third2_31_basecontrol),aes(x=plotorder,group=Sample,col=Sample,alpha=alpha)) +
+    geom_errorbar(aes(ymin=ci_lo_95,ymax=ci_hi_95,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=0.5) +
+    geom_errorbar(aes(ymin=ci_lo_90,ymax=ci_hi_90,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=1) +
+    geom_point(aes(y=est,shape=Sample,col=Sample_color),position=position_dodge(width=0.5),size=3) +
+    geom_blank(data=dummy_df,aes(y=est)) +
+    geom_text(data=filter(coefs_third2_31_basecontrol,layer1_hypothesisfamily=="policy"),aes(y=est+0.006,label=Sample),alpha=1,position=position_dodge(width=0.5),size=3) +
+    geom_hline(yintercept = 0,lty=2) +
+    facet_wrap(~family,ncol=1,scales="free") +
+    # facet_grid(rows = "family",scales="free",space = "free_y",switch = "y") +
+    scale_x_continuous("",
+                       breaks = coefs_third2_31_basecontrol$plotorder,labels = coefs_third2_31_basecontrol$outcome) +
+    scale_y_continuous("Treatment effect of conservative seed vs.\nliberal seed video, all 3/1 algorithm\n(95% and 90% CIs)") +
+    scale_color_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)","insig"),values=c(vgreen,red_mit,blue_mit,"black")) +
+    scale_shape_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)"),values=c(16,17,18)) +
+    scale_alpha_manual(breaks=c(F,T),values=c(0.25,1)) +
+    coord_flip() +
+    # coord_flip(ylim=c(-0.3,0.3)) +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"),legend.position="none")
+)
+ggsave(coefplot_third2_31_basecontrol,
+       filename = "../results/coefplot_third2_31_basecontrol_3studies.png",width=5,height=8.5)
+ggsave(coefplot_third2_31_basecontrol,
+       filename = "../results/coefplot_third2_31_basecontrol_3studies.pdf",width=5,height=8.5)
+
+(coefplot_third2_31_basecontrol_empty <- ggplot(filter(coefs_third2_31_basecontrol),aes(x=plotorder,group=Sample,col=ifelse(p.adj<0.05,T,F))) +
+    geom_blank(aes(ymin=ci_lo_95,ymax=ci_hi_95),position=position_dodge(width=0.5),width=0,lwd=0.5) +
+    geom_blank(aes(ymin=ci_lo_90,ymax=ci_hi_90),position=position_dodge(width=0.5),width=0,lwd=1) +
+    geom_blank(aes(y=est,shape=Sample),position=position_dodge(width=0.5),size=2) +
+    geom_hline(yintercept = 0,lty=2) +
+    facet_wrap(~family,ncol=1,scales="free") +
+    scale_x_continuous("",
+                       breaks = coefs_third2_31_basecontrol$plotorder,labels = coefs_third2_31_basecontrol$outcome) +
+    scale_y_continuous("Treatment effect of conservative seed vs.\nliberal seed video, all 3/1 algorithm\n(95% and 90% CIs)") +
+    scale_color_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)","insig"),values=c(vgreen,red_mit,blue_mit,"black")) +
+    scale_shape_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)"),values=c(16,17,18)) +
+    scale_alpha_manual(breaks=c(F,T),values=c(0.25,1)) +
+    coord_flip(ylim=c(-0.3,0.3)) +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"),legend.position="bottom",legend.margin = margin(0,0,0,-3,"lines"))
+)
+ggsave(coefplot_third2_31_basecontrol_empty,
+       filename = "../results/coefplot_third2_31_basecontrol_empty_3studies.png",width=5,height=8.5)
+
+
+# create DF to set axis limits:
+dummy_df <- data.frame(family=c("Platform Interaction","Platform Interaction"),est=c(-0.5,0.5),plotorder=c(9,9),Sample=c("Gun Control\n(MTurk)","Gun Control\n(MTurk)"),alpha=c(FALSE,FALSE)) %>% mutate(family=factor(family))
+
+(coefplot_third2_31_basecontrol_toptwo <- ggplot(filter(coefs_third2_31_basecontrol,layer1_hypothesisfamily %in% c("policy","platform")),aes(x=plotorder,group=Sample,col=Sample,alpha=alpha)) +
+    geom_errorbar(aes(ymin=ci_lo_95,ymax=ci_hi_95,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=0.5) +
+    geom_errorbar(aes(ymin=ci_lo_90,ymax=ci_hi_90,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=1) +
+    geom_point(aes(y=est,shape=Sample,col=Sample_color),position=position_dodge(width=0.5),size=3) +
+    geom_blank(data=dummy_df,aes(y=est)) +
+    geom_text(data=filter(coefs_third2_31_basecontrol,layer1_hypothesisfamily=="policy"),aes(y=est+0.006,label=Sample),alpha=1,position=position_dodge(width=0.5),size=3) +
+    geom_hline(yintercept = 0,lty=2) +
+    facet_wrap(~family,ncol=1,scales="free") +
+    # facet_grid(rows = "family",scales="free",space = "free_y",switch = "y") +
+    scale_x_continuous("",
+                       breaks = coefs_third2_31_basecontrol$plotorder,labels = coefs_third2_31_basecontrol$outcome) +
+    scale_y_continuous("Treatment effect of conservative seed vs.\nliberal seed video, all 3/1 algorithm\n(95% and 90% CIs)") +
+    scale_color_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)","insig"),values=c(vgreen,red_mit,blue_mit,"black")) +
+    scale_shape_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)"),values=c(16,17,18)) +
+    scale_alpha_manual(breaks=c(F,T),values=c(0.25,1)) +
+    coord_flip() +
+    # coord_flip(ylim=c(-0.4,0.4)) +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"),legend.position="none",plot.margin = margin(5,10,5,5))
+)
+ggsave(coefplot_third2_31_basecontrol_toptwo,
+       filename = "../results/coefplot_third2_31_basecontrol_3studies_toptwo.png",width=5,height=4.75)
+ggsave(coefplot_third2_31_basecontrol_toptwo,
+       filename = "../results/coefplot_third2_31_basecontrol_3studies_toptwo.pdf",width=5,height=4.75)
+
+
+
+coefs_third2_22_basecontrol <- coefs_basecontrol %>%
+  filter(layer2_treatmentcontrast == "neutral con 22 - neutral lib 22" &
+           layer3_specificoutcome != "overall")
+
+coefs_third2_22_basecontrol$outcome = outcome_labels$outcome[match(coefs_third2_22_basecontrol$layer3_specificoutcome,
+                                                                   outcome_labels$specificoutcome)]
+
+coefs_third2_22_basecontrol$family = outcome_labels$family[match(coefs_third2_22_basecontrol$layer3_specificoutcome,
+                                                                 outcome_labels$specificoutcome)]
+
+coefs_third2_22_basecontrol <- mutate(coefs_third2_22_basecontrol,
+                                      family = factor(family,levels = c("Policy Attitudes\n(unit scale, + is more conservative)","Platform Interaction","Media Trust\n(unit scale, + is more trusting)","Affective Polarization\n(unit scale, + is greater polarization)"),ordered = T))
+
+## manipulate to get all unit scales:
+coefs_third2_22_basecontrol$est[coefs_third2_22_basecontrol$layer3_specificoutcome=="platform_duration"] <- coefs_third2_22_basecontrol$est[coefs_third2_22_basecontrol$layer3_specificoutcome=="platform_duration"]/3600
+coefs_third2_22_basecontrol$se[coefs_third2_22_basecontrol$layer3_specificoutcome=="platform_duration"] <- coefs_third2_22_basecontrol$se[coefs_third2_22_basecontrol$layer3_specificoutcome=="platform_duration"]/3600
+
+coefs_third2_22_basecontrol$est[coefs_third2_22_basecontrol$layer3_specificoutcome=="affpol_ft_w2"] <- coefs_third2_22_basecontrol$est[coefs_third2_22_basecontrol$layer3_specificoutcome=="affpol_ft_w2"]/100
+coefs_third2_22_basecontrol$se[coefs_third2_22_basecontrol$layer3_specificoutcome=="affpol_ft_w2"] <- coefs_third2_22_basecontrol$se[coefs_third2_22_basecontrol$layer3_specificoutcome=="affpol_ft_w2"]/100
+
+coefs_third2_22_basecontrol <- coefs_third2_22_basecontrol %>%
+  mutate(ci_lo_99 = est + qnorm(0.001)*se,
+         ci_hi_99 = est + qnorm(0.999)*se,
+         ci_lo_95 = est + qnorm(0.025)*se,
+         ci_hi_95 = est + qnorm(0.975)*se,
+         ci_lo_90 = est + qnorm(0.05)*se,
+         ci_hi_90 = est + qnorm(0.95)*se,
+         plotorder = rep((nrow(coefs_third2_22_basecontrol)/3):1,3),
+         alpha = ifelse(p.adj<0.05, T, F),
+         alpha = as.logical(alpha),
+         alpha = replace_na(alpha,F),
+         Sample_color = as.character(Sample),
+         Sample_color = replace(Sample_color,alpha==F,"insig")
+  )
+
+(coefplot_third2_22_basecontrol <- ggplot(filter(coefs_third2_22_basecontrol),aes(x=plotorder,group=Sample,col=Sample,alpha=alpha)) +
+    geom_errorbar(aes(ymin=ci_lo_95,ymax=ci_hi_95,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=0.5) +
+    geom_errorbar(aes(ymin=ci_lo_90,ymax=ci_hi_90,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=1) +
+    geom_point(aes(y=est,shape=Sample,col=Sample_color),position=position_dodge(width=0.5),size=3) +
+    geom_text(data=filter(coefs_third2_22_basecontrol,layer1_hypothesisfamily=="policy"),aes(y=est+0.006,label=Sample),alpha=1,position=position_dodge(width=0.5),size=3) +
+    geom_hline(yintercept = 0,lty=2) +
+    facet_wrap(~family,ncol=1,scales="free") +
+    scale_x_continuous("",
+                       breaks = coefs_third2_22_basecontrol$plotorder,labels = coefs_third2_22_basecontrol$outcome) +
+    scale_y_continuous("Treatment effect of conservative seed vs.\nliberal seed video, all 2/2 algorithm\n(95% and 90% CIs)") +
+    scale_color_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)","insig"),values=c(vgreen,red_mit,blue_mit,"black")) +
+    scale_shape_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)"),values=c(16,17,18)) +
+    scale_alpha_manual(breaks=c(F,T),values=c(0.25,1)) +
+    coord_flip() +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"),legend.position="none")
+)
+ggsave(coefplot_third2_22_basecontrol,
+       filename = "../results/coefplot_third2_22_basecontrol_3studies.png",width=5,height=8.5)
+ggsave(coefplot_third2_22_basecontrol,
+       filename = "../results/coefplot_third2_22_basecontrol_3studies.pdf",width=5,height=8.5)
+
+(coefplot_third2_22_basecontrol_empty <- ggplot(filter(coefs_third2_22_basecontrol),aes(x=plotorder,group=Sample,col=ifelse(p.adj<0.05,T,F))) +
+    geom_blank(aes(ymin=ci_lo_95,ymax=ci_hi_95),position=position_dodge(width=0.5),width=0,lwd=0.5) +
+    geom_blank(aes(ymin=ci_lo_90,ymax=ci_hi_90),position=position_dodge(width=0.5),width=0,lwd=1) +
+    geom_blank(aes(y=est,shape=Sample),position=position_dodge(width=0.5),size=2) +
+    geom_hline(yintercept = 0,lty=2) +
+    facet_wrap(~family,ncol=1,scales="free") +
+    scale_x_continuous("",
+                       breaks = coefs_third2_22_basecontrol$plotorder,labels = coefs_third2_22_basecontrol$outcome) +
+    scale_y_continuous("Treatment effect of conservative seed vs.\nliberal seed video, all 2/2 algorithm\n(95% and 90% CIs)") +
+    scale_color_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)","insig"),values=c(vgreen,red_mit,blue_mit,"black")) +
+    scale_shape_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)"),values=c(16,17,18)) +
+    scale_alpha_manual(breaks=c(F,T),values=c(0.25,1)) +
+    coord_flip(ylim=c(-0.6,0.6)) +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"),legend.position="bottom",legend.margin = margin(0,0,0,-3,"lines"))
+)
+ggsave(coefplot_third2_22_basecontrol_empty,
+       filename = "../results/coefplot_third2_22_basecontrol_empty_3studies.png",width=5,height=8.5)
+
+(coefplot_third2_22_basecontrol_toptwo <- ggplot(filter(coefs_third2_22_basecontrol,layer1_hypothesisfamily %in% c("policy","platform")),aes(x=plotorder,group=Sample,col=Sample,alpha=alpha)) +
+    geom_errorbar(aes(ymin=ci_lo_95,ymax=ci_hi_95,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=0.5) +
+    geom_errorbar(aes(ymin=ci_lo_90,ymax=ci_hi_90,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=1) +
+    geom_point(aes(y=est,shape=Sample,col=Sample_color),position=position_dodge(width=0.5),size=3) +
+    geom_text(data=filter(coefs_third2_22_basecontrol,layer1_hypothesisfamily=="policy"),aes(y=est+0.006,label=Sample),alpha=1,position=position_dodge(width=0.5),size=3) +
+    geom_hline(yintercept = 0,lty=2) +
+    facet_wrap(~family,ncol=1,scales="free") +
+    scale_x_continuous("",
+                       breaks = coefs_third2_22_basecontrol$plotorder,labels = coefs_third2_22_basecontrol$outcome) +
+    scale_y_continuous("Treatment effect of conservative seed vs.\nliberal seed video, all 2/2 algorithm\n(95% and 90% CIs)") +
+    scale_color_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)","insig"),values=c(vgreen,red_mit,blue_mit,"black")) +
+    scale_shape_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)"),values=c(16,17,18)) +
+    scale_alpha_manual(breaks=c(F,T),values=c(0.25,1)) +
+    coord_flip() +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"),legend.position="none")
+)
+ggsave(coefplot_third2_22_basecontrol_toptwo,
+       filename = "../results/coefplot_third2_22_basecontrol_3studies_toptwo.png",width=5,height=4.75)
+ggsave(coefplot_third2_22_basecontrol_toptwo,
+       filename = "../results/coefplot_third2_22_basecontrol_3studies_toptwo.pdf",width=5,height=4.75)
+
+##### All respondents, attitudinal DV only #####
+coefs_policyindex <- filter(coefs_third2_22_basecontrol,layer1_hypothesisfamily=="policy") %>% mutate(contrast="Seed, 2/2",subset="Moderates") %>%
+  bind_rows(filter(coefs_third2_31_basecontrol,layer1_hypothesisfamily=="policy") %>% mutate(contrast="Seed, 3/1",subset="Moderates")) %>%
+  bind_rows(filter(coefs_third2_pro_basecontrol,layer1_hypothesisfamily=="policy") %>% mutate(contrast="Algorithm, lib. seed",subset="Moderates (liberal seed)")) %>%
+  bind_rows(filter(coefs_third2_anti_basecontrol,layer1_hypothesisfamily=="policy") %>% mutate(contrast="Algorithm, cons. seed",subset="Moderates  (conservative seed)")) %>%
+  bind_rows(filter(coefs_third1_basecontrol,layer1_hypothesisfamily=="policy") %>% mutate(contrast="Algorithm, lib. seed",subset="Liberals  (liberal seed)")) %>%
+  bind_rows(filter(coefs_third3_basecontrol,layer1_hypothesisfamily=="policy") %>% mutate(contrast="Algorithm, cons. seed",subset="Conservatives  (conservative seed)")) %>%
+  mutate(subset = factor(subset,levels=c("Liberals  (liberal seed)","Conservatives  (conservative seed)","Moderates (liberal seed)","Moderates  (conservative seed)"),ordered = T))
+
+(coefplot_policyindex_basecontrol <- ggplot(filter(coefs_policyindex,str_detect(contrast,"Algorithm")),aes(x=plotorder,group=Sample,col=Sample,alpha=alpha)) +
+    geom_errorbar(aes(ymin=ci_lo_95,ymax=ci_hi_95,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=0.5) +
+    geom_errorbar(aes(ymin=ci_lo_90,ymax=ci_hi_90,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=1) +
+    geom_point(aes(y=est,shape=Sample,col=Sample_color),position=position_dodge(width=0.5),size=3) +
+    geom_text(data=filter(coefs_policyindex,subset=="Liberals  (liberal seed)"),aes(y=est+0.006,label=Sample),alpha=1,position=position_dodge(width=0.5),size=3) +
+    geom_hline(yintercept = 0,lty=2) +
+    facet_wrap(~subset,ncol=2,scales="free") +
+    scale_x_continuous("",breaks = 8,labels="") +
+    scale_y_continuous("Treatment effect of more extreme 3/1 vs. 2/2\nalgorithm on policy index (95% and 90% CIs)") +
+    scale_color_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)","insig"),values=c(vgreen,red_mit,blue_mit,"black")) +
+    scale_shape_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)"),values=c(16,17,18)) +
+    scale_alpha_manual(breaks=c(F,T),values=c(0.25,1)) +
+    coord_flip(ylim=c(-0.11,0.11)) +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"),legend.position="bottom",legend.margin = margin(0,0,0,-3,"lines"),
+          axis.ticks.y = element_blank())
+)
+ggsave(coefplot_policyindex_basecontrol,
+       filename = "../results/coefplot_policyindex_basecontrol_3studies.png",width=4.5,height=4.5)
+
+(coefplot_policyindex_seed_basecontrol <- ggplot(filter(coefs_policyindex,str_detect(contrast,"Seed")),aes(x=plotorder,group=Sample,col=Sample,alpha=alpha)) +
+    geom_errorbar(aes(ymin=ci_lo_95,ymax=ci_hi_95,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=0.5) +
+    geom_errorbar(aes(ymin=ci_lo_90,ymax=ci_hi_90,col=Sample_color),position=position_dodge(width=0.5),width=0,lwd=1) +
+    geom_point(aes(y=est,shape=Sample,col=Sample_color),position=position_dodge(width=0.5),size=2) +
+    geom_text(data=filter(coefs_policyindex,contrast=="Seed, 2/2"),aes(y=est+0.006,label=Sample),alpha=1,position=position_dodge(width=0.5),size=3) +
+    geom_hline(yintercept = 0,lty=2) +
+    facet_wrap(~contrast,ncol=2,scales="free") +
+    scale_x_continuous("",breaks = 8,labels="") +
+    scale_y_continuous("Treatment effect of conservative vs. liberal\nseed on policy index (95% and 90% CIs)") +
+    scale_color_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)","insig"),values=c(vgreen,red_mit,blue_mit,"black")) +
+    scale_shape_manual("Study:",breaks = c("Gun Control\n(MTurk)","Minimum Wage\n(MTurk)","Minimum Wage\n(YouGov)"),values=c(16,17,18)) +
+    scale_alpha_manual(breaks=c(F,T),values=c(0.25,1)) +
+    coord_flip(ylim=c(-0.11,0.11)) +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"),legend.position="none",
+          axis.ticks.y = element_blank())
+)
+ggsave(coefplot_policyindex_seed_basecontrol,
+       filename = "../results/coefplot_policyindex_seed_basecontrol_3studies.png",width=4.5,height=2.5)
+
+rm(list = ls())

code/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2025 Dean Knox
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

code/gun control (issue 1)/01_trt_assign.R

@@ -0,0 +1,164 @@
+cat(rep('=', 80),
+    '\n\n',
+    'OUTPUT FROM: gun control (issue 1)/01_trt_assign.R',
+    '\n\n',
+    sep = ''
+    )
+    
+library(tidyverse)
+library(janitor)
+library(lubridate)
+library(randomizr)
+
+# create directory to hold cached intermediate files
+dir.create("../results/intermediate data/gun control (issue 1)/",
+           recursive = TRUE, showWarnings = FALSE)
+
+w1 <- read_csv("../data/gun control (issue 1)/wave1_final.csv")[-c(1,2),] %>%
+  clean_names() %>%
+  filter(finished == "True", q62 == "I agree to participate\u2028")
+
+# Recodes ======================================================================
+
+w1 <- w1 %>% mutate(start_date = as_datetime(start_date),
+                    end_date = as_datetime(end_date),
+                    survey_time = as.numeric(end_date-start_date))
+
+print('wave 1 survey time:')
+summary(w1$survey_time)
+
+w1 <- w1 %>%
+  mutate(man = ifelse(q26 == "Man", 1, 0),
+         black = ifelse(str_detect(q29, "Black"), 1, 0),
+         white = ifelse(str_detect(q29, "White"), 1, 0),
+         college = ifelse(str_detect(q30, "college ") | str_detect(q30, "Post"), 1, 0),
+         income_gt50k = ifelse(q31 %in% names(table(w1$q31))[c(2,3,5,10:13)], 1, 0)
+  )
+
+# PID:
+w1$pid <- ifelse(w1$pid1=="Democrat",-1,NA)
+w1$pid <- ifelse(w1$pid1=="Republican",1,w1$pid)
+w1$pid[w1$pid4=="Closer to the Republican Party"] <- 1
+w1$pid[w1$pid4=="Closer to the Democratic Party"] <- -1
+w1$pid[w1$pid4=='Neither'] <- 0
+
+print('wave 1 party id:')
+round(table(w1$pid) / sum(table(w1$pid)), digits=2)
+
+w1$ideo <- ifelse(w1$ideo1=="Liberal",-1,NA)
+w1$ideo <- ifelse(w1$ideo1=="Conservative",1,w1$ideo)
+w1$ideo[w1$ideo4=="Closer to liberals"] <- -1
+w1$ideo[w1$ideo4=="Closer to conservatives"] <- 1
+w1$ideo[w1$ideo4=="Neither"] <- 0
+
+print('wave 1 ideology:')
+round(table(w1$ideo) / sum(table(w1$ideo)), digits=2)
+
+w1$age <- 2021 - as.numeric(w1$q27)
+
+
+
+# A/V check ====================================================================
+
+print("audio check:")
+length(which(w1$q87 == "Quick and easy")) / length(w1$q87)
+
+print("video check:")
+length(which(w1$q89 == "wikiHow")) / length(w1$q89)
+
+w1$audio_ok <- 1*(w1$q87 == "Quick and easy")
+w1$video_ok <- 1*(w1$q89 == "wikiHow")
+
+w1 <- w1 %>%
+  mutate(gun_own = dplyr::recode(q15, "Yes" = 1, "No" = 0))
+
+# Convert pre-treatment DV to numeric unit scale -------------------------------
+
+w1 <- w1 %>%
+  mutate( # higher = more pro-gun
+    right_to_own_importance = recode(q79, "Protect the right to own guns" = 1, "Regulate gun ownership" = 0),
+    assault_ban = (match(q81, names(table(q81))[c(5,3,1,2,4)])-1)/4,
+    handgun_ban = (match(q82, names(table(q82))[c(5,3,1,2,4)])-1)/4,
+    concealed_safe = 1-(match(q83, names(table(q83))[c(2,5,3,4,1)])-1)/4,
+    stricter_laws = (match(q23, names(table(q23))[c(5,3,1,2,4)])-1)/4
+  )
+
+w1 <- w1 %>%
+  rowwise() %>%
+  mutate(gun_index = sum(c(right_to_own_importance,assault_ban,handgun_ban,concealed_safe,stricter_laws), na.rm=T),
+         gun_index_2 = mean(c(right_to_own_importance,assault_ban,handgun_ban,concealed_safe), na.rm=T))
+
+# Cronbach's alpha -------------------------------------------------------------
+
+index_fa <- psych::alpha(select(w1, right_to_own_importance, assault_ban, handgun_ban, concealed_safe, stricter_laws), check.keys = TRUE)
+alpha <- index_fa$total["raw_alpha"]
+writeLines(as.character(round(alpha,2)),
+           con = "../results/guncontrol_outcomes_alpha.tex",sep = "%")
+
+
+
+# trim sample -------------------------------------------------------------
+
+# We exclude respondents who took less than 120 seconds to complete the Wave 1 survey, failed either
+# an audio check or a video check, as well as those whose gun policy opinions fall within the most
+# extreme 5% of the gun policy index outcome (i.e. < 0.25 or > 4.75 on the 0-5 scale, to guard
+# against eventual ceiling/floor effects; in a pilot study this was 15% of the sample).
+
+w1 <- w1 %>% filter(audio_ok == 1, video_ok == 1)
+w1 <- w1 %>% filter(survey_time >= 2)
+w1 <- w1 %>% filter(gun_index >= 0.25, gun_index <= 4.75)
+
+print('gun index:')
+summary(w1$gun_index)
+
+
+
+# Block random assignment ======================================================
+
+# We randomly assign respondents to both a seed video type (pro-gun vs. anti-gun) and a recommendation system (3/1 vs. 2/2)
+# blocking on Wave 1 gun policy opinions. In the sample of respondents
+# who will be invited to Wave 2, we form terciles of the Wave 1 gun policy opinion index, referring
+# to the lower, middle and upper terciles as anti-gun, moderate and pro-gun respectively
+
+w1$tercile <- cut(w1$gun_index, breaks = quantile(w1$gun_index, c(0, 1/3, 2/3, 1)), include.lowest = TRUE, labels = 1:3)
+tapply(w1$gun_index, w1$tercile, mean)
+table(w1$tercile)
+
+#  pure control (with 1/5 probability), anti-gun 2/2 (with 2/5 probability), or anti-gun 3/1 (with 2/5 probability).
+# seed position (pro-gun or anti-gun), recommendation system (2/2 or 3/1), or a
+# pure control group (i.e. one of five possible conditions) with equal probability
+
+set.seed(2021)
+
+w1$trt_system <- block_ra(blocks = w1$tercile, prob_each = c(2/5, 2/5, 1/5), conditions = c("2/2", "3/1", "pure control"))
+
+w1$seed <- rep("", nrow(w1))
+w1[w1$tercile == 1,]$seed <- "anti-gun seed"
+w1[w1$tercile == 3,]$seed <- "pro-gun seed"
+w1[w1$tercile == 2,]$seed <- complete_ra(N = length(which(w1$tercile == 2)), prob = 0.5, conditions = c("pro-gun seed",
+                                                                                                        "anti-gun seed"))
+with(w1[w1$tercile == 1,], round(prop.table(table(seed, trt_system)), digits = 3))
+with(w1[w1$tercile == 2,], round(prop.table(table(seed, trt_system)), digits = 3))
+with(w1[w1$tercile == 3,], round(prop.table(table(seed, trt_system)), digits = 3))
+
+w1 <- w1 %>% mutate(trt_assign = case_when(seed == "anti-gun seed" & trt_system == "2/2" ~ 1,
+                                           seed == "anti-gun seed" & trt_system == "3/1" ~ 2,
+                                           seed == "pro-gun seed" & trt_system == "2/2" ~ 3,
+                                           seed == "pro-gun seed" & trt_system == "3/1" ~ 4,
+                                           trt_system == "pure control" ~ 5))
+
+print('treatment assignment:')
+table(w1$trt_assign)
+print('seed assignment:')
+table(w1$seed)
+print('system assignment:')
+table(w1$trt_system)
+print('seed & system assignment:')
+table(w1$trt_system, w1$seed)
+
+w1$batch <- sample(c(rep(1:floor(nrow(w1)/500), 500), rep(6, nrow(w1)-500*5)))
+
+# sent to Qualtrics ------------------------------------------------------------
+
+# write_csv(data.frame(trt = w1$trt_assign, id = w1$worker_id, batch = w1$batch),
+#           "guncontrol_wave1_assignments.csv")

code/gun control (issue 1)/02_clean_merge.R

@@ -0,0 +1,560 @@
+## YouTube Algorithms and Gun Control Opinions
+## Data collected June 2021 via MTurk/CloudResearch
+
+cat(rep('=', 80),
+    '\n\n',
+    'OUTPUT FROM: gun control (issue 1)/02_clean_merge.R',
+    '\n\n',
+    sep = ''
+    )
+
+## Preamble ----------------------------
+library(tidyverse)
+library(janitor)
+library(lubridate)
+library(stargazer)
+library(broom)
+library(corrplot)
+
+a <- read_csv("../data/gun control (issue 1)/wave1_final.csv")[-c(1,2),] %>%
+  clean_names()
+
+# Wave 1 =======================================================================
+
+# Recodes:
+a <- a %>% mutate(start_date = as_datetime(start_date),
+                  end_date = as_datetime(end_date),
+                  survey_time = as.numeric(end_date-start_date))
+
+print('wave 1 survey time:')
+summary(a$survey_time)
+
+# Demographics -----------------------------------------------------------------
+
+a <- a %>%
+  mutate(female = ifelse(q26 == "Woman", 1, 0),
+         male = ifelse(q26 == "Man", 1, 0),
+         black = ifelse(str_detect(q29, "Black"), 1, 0),
+         white = ifelse(str_detect(q29, "White"), 1, 0),
+         college = ifelse(str_detect(q30, "college ") | str_detect(q30, "Post"), 1, 0),
+         # dk: confirmed
+         income_gt50k = ifelse(q31 %in% names(table(a$q31))[c(2,3,5,10:13)], 1, 0)
+         )
+a$income_gt50k[is.na(a$q31)] <- NA
+
+# PID:
+
+a <- a %>%
+  mutate(pid = case_when(pid1=="Democrat" ~ -1,
+                         pid1=="Republican" ~ 1,
+                         pid4=="Closer to the Republican Party" ~ 1,
+                         pid4=="Closer to the Democratic Party" ~ -1,
+                         pid4=="Neither" ~ 0))
+
+a <- a %>%
+  mutate(ideo = case_when(ideo1=="Liberal" ~ -1,
+                          ideo1=="Conservative" ~ 1,
+                          ideo4=="Closer to conservatives" ~ 1,
+                          ideo4=="Closer to liberals" ~ -1,
+                          ideo4=="Neither" ~ 0))
+
+a$age <- 2021 - as.numeric(a$q27)
+
+# age categories: 18-29;  30-44;  45-64;  65+
+a <- a %>%
+  mutate(age_cat = case_when(age>=18 & age<=29 ~ "18-29",
+                             age>=30 & age<=44 ~ "30-44",
+                             age>=45 & age<=64 ~ "45-64",
+                             age>=65 ~ "65+"
+  ))
+a <- a %>%
+  fastDummies::dummy_cols(select_columns = "age_cat")
+
+## Need:
+# political  interest  (5-point  scale:   1=Not  atall interested, 5=Extremely interested),
+# self-reported YouTube usage frequency (7-pointscale: 0=None, 6=More than 3 hours per day),
+# number of self-reported favorite YouTubechannels (count coded from open-ended question: “Who/what are your favorite YouTubebroadcasters or channels?”; 0 if blank),
+# indicator for having watched videos from popularchannels (1 if any selected:  “In the past week, have you watched videos from any of thefollowing  YouTube  broadcasters  or  channels?”),
+# video  vs.   text  preference  (1=Alwaysprefer  videos,  10=Always  prefer  text),
+# gun  enthusiasm  (additive  index  of  “Do  you  ordoes anyone in your household own a gun?”  with yes=1 and “How often, if ever, do youvisit websites about guns, hunting or other shooting sports?”  from 0=Never or Hardlyever to 1=Sometimes or Often),
+# gun policy issue importance (4-point scale:  1=Not atall important, 4=Very important)
+
+a <- a %>%
+  mutate(pol_interest = recode(q91,"Extremely interested"=5,"Very interested"=4,"Somewhat interested"=3,"Not very interested"=2,"Not at all interested"=1),
+         freq_youtube = recode(q77,"More than 3 hours per day"=6,"2–3 hours per day"=5,"1–2 hours per day"=4,"31–59 minutes per day"=3,"10–30 minutes per day"=2,"Less than 10 minutes per day"=1,"None"=0),
+         fav_channels = str_count(q8,"\n"), # should be one per line but this might not be right - need to hand-code
+         popular_channels = ifelse(is.na(q78),0,1),
+         vid_pref = recode(q9,"Always prefer videos\n1\n"=1,"2"=2,"3"=3,"4"=4,"5"=5,"6"=6,"7"=7,"8"=8,"9"=9,"Always prefer text\n10\n"=10),
+         visit_shooting_sites = recode(q18,"Never"=0,"Hardly ever"=0,"Sometimes"=1,"Often"=1),
+         gun_own = recode(q15, "Yes" = 1, "No" = 0),
+         gun_enthusiasm = visit_shooting_sites + gun_own,
+         gun_importance = recode(q76_8,"Very important"=4,"Somewhat important"=3,"Not too important"=2,"Not at all important"=1)
+  )
+
+descr_data <- as.data.frame(select(a,
+                                   female,
+                                   white,
+                                   black,
+                                   age,
+                                   college,
+                                   income_gt50k))
+descr_data <- descr_data %>% filter(rowSums(is.na(.)) != ncol(.))
+descriptive_tab <- stargazer(descr_data,
+                             summary = T, digits=2,
+                             summary.stat=c("mean","sd","median","min","max","n"),
+                             covariate.labels = c("Female",
+                                                  "White",
+                                                  "Black",
+                                                  "Age",
+                                                  "College educated",
+                                                  "Income \\textgreater 50k"),
+                             float = F,
+                             out = "../results/guncontrol_descriptive.tex"
+                             )
+
+summary_tab <- a %>%
+  dplyr::summarize(female = mean(female,na.rm=T),
+                   white = mean(white,na.rm=T),
+                   black = mean(black,na.rm=T),
+                   age1829 = mean(`age_cat_18-29`,na.rm=T),
+                   age3044 = mean(`age_cat_30-44`,na.rm=T),
+                   age4564 = mean(`age_cat_45-64`,na.rm=T),
+                   age65p = mean(`age_cat_65+`,na.rm=T),
+                   college = mean(college,na.rm=T),
+                   income_gt50k = mean(income_gt50k,na.rm=T),
+                   democrat = mean(pid==-1,na.rm=T),
+                   republican = mean(pid==1,na.rm=T))
+
+summary_tab <- pivot_longer(summary_tab,
+                            cols=c(female,
+                                   white,
+                                   black,
+                                   age1829,
+                                   age3044,
+                                   age4564,
+                                   age65p,
+                                   college,
+                                   income_gt50k,
+                                   democrat,
+                                   republican),
+                            names_to = "outcome",values_to = "survey_avg")
+outcome_labels <- data.frame(outcome_pretty = c("Female",
+                                                "White",
+                                                "Black",
+                                                "Age 18-29",
+                                                "Age 30-44",
+                                                "Age 45-64",
+                                                "Age 65+",
+                                                "College educated",
+                                                "Income >$50k",
+                                                "Democrat",
+                                                "Republican"),
+                             outcome = c("female",
+                                         "white",
+                                         "black",
+                                         "age1829",
+                                         "age3044",
+                                         "age4564",
+                                         "age65p",
+                                         "college",
+                                         "income_gt50k",
+                                         "democrat",
+                                         "republican"))
+summary_tab$outcome_pretty <-  outcome_labels$outcome_pretty[match(summary_tab$outcome,outcome_labels$outcome)]
+summary_tab <- summary_tab %>%
+  mutate(outcome_pretty = factor(outcome_pretty,levels = c("Republican",
+                                                           "Democrat",
+                                                           "Income >$50k",
+                                                           "College educated",
+                                                           "Age 65+",
+                                                           "Age 45-64",
+                                                           "Age 30-44",
+                                                           "Age 18-29",
+                                                           "Female",
+                                                           "Black",
+                                                           "White"
+  ),ordered=T))
+
+(descrip_fig <- ggplot(summary_tab) +
+    geom_point(aes(y=outcome_pretty,x=survey_avg)) +
+    geom_text(aes(y=outcome_pretty,x=survey_avg,label=paste0(round(100*survey_avg,0),"%")),nudge_x = 0.1) +
+    scale_y_discrete("") +
+    scale_x_continuous("",labels=scales::percent_format(),limits=c(0,1)) +
+    theme_bw()
+)
+ggsave(descrip_fig,filename = "../results/guncontrol_demographics.pdf",height=5,width=4)
+
+
+
+#### Outcomes ####
+
+# policy opinions, convert to numeric unit scale:
+a <- a %>%
+  mutate( # higher = more pro-gun
+    right_to_own_importance = recode(q79, "Protect the right to own guns" = 1, "Regulate gun ownership" = 0),
+    assault_ban = recode(q81, "Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4,
+    handgun_ban = recode(q82, "Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4,
+    concealed_safe = recode(q83,"Much safer"=4,"Somewhat safer"=3,"No difference"=2,"Somewhat less safe"=1,"Much less safe"=0)/4,
+    stricter_laws = recode(q23, "Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4
+  )
+
+a <- a %>%
+  rowwise() %>%
+  mutate(gun_index = mean(c(right_to_own_importance,assault_ban,handgun_ban,concealed_safe,stricter_laws), na.rm=T),
+         gun_index_2 = mean(c(right_to_own_importance,assault_ban,handgun_ban,concealed_safe), na.rm=T)) %>%
+  ungroup()
+
+# Cronbach's alpha
+index_fa <- psych::alpha(select(a, right_to_own_importance, assault_ban, handgun_ban, concealed_safe, stricter_laws), check.keys = TRUE)
+write.csv(data.frame(cor(select(a, right_to_own_importance, assault_ban, handgun_ban, concealed_safe, stricter_laws), use = "complete.obs")),row.names = T,
+          file = "../results/guncontrol_cormat_gun_index_w1.csv")
+pdf("../results/guncontrol_corrplot_gunindex_w1.pdf")
+w1_corrplot <- corrplot::corrplot(cor(select(a, right_to_own_importance, assault_ban, handgun_ban, concealed_safe, stricter_laws), use = "complete.obs"),method = "shade")
+dev.off()
+
+alpha <- index_fa$total["raw_alpha"]
+writeLines(as.character(round(alpha,2)),con = "../results/guncontrol_outcomes_alpha.tex",sep = "%")
+
+# FACTOR ANALYSIS WITH VARIMAX ROTATION (PRE)
+pca2 <- psych::principal(select(a, right_to_own_importance, assault_ban, handgun_ban, concealed_safe, stricter_laws),
+                         rotate="varimax",
+                         nfactors=1
+)
+pc2 <- pca2$Vaccounted[2]
+writeLines(as.character(round(pc2, 2)),con = "../results/outcomes_pc2_study1_pre.tex",sep = "%")
+
+# media trust
+a <- a %>%
+  mutate( # higher = more trusting
+    trust_majornews = recode(q58_1,"A lot"=3,"Some"=2,"Not too much"=1,"Not at all"=0)/3,
+    trust_localnews = recode(q58_2,"A lot"=3,"Some"=2,"Not too much"=1,"Not at all"=0)/3,
+    trust_social = recode(q58_3,"A lot"=3,"Some"=2,"Not too much"=1,"Not at all"=0)/3,
+    trust_youtube = recode(q58_4,"A lot"=3,"Some"=2,"Not too much"=1,"Not at all"=0)/3,
+    fabricate_majornews = recode(q89_1,"Never"=4,"Once in a while"=3,"About half the time"=2,"Most of the time"=1,"All the time"=0)/4,
+    fabricate_youtube = recode(q90,"Never"=4,"Once in a while"=3,"About half the time"=2,"Most of the time"=1,"All the time"=0)/4
+  ) %>%
+  rowwise() %>%
+  mutate(media_trust = mean(trust_majornews,trust_localnews,fabricate_majornews,na.rm=T)) %>%
+  ungroup()
+
+media_trust_fa <- psych::alpha(select(a, trust_majornews,trust_localnews,fabricate_majornews), check.keys = TRUE)
+
+print('media trust alpha:')
+media_trust_fa$total["raw_alpha"]
+
+
+# affective polarization
+print('check affpol feeling thermometers:')
+a %>%
+  group_by(pid) %>%
+  summarize(mean_2=mean(as.numeric(q5_2),na.rm=T),
+            mean_5=mean(as.numeric(q5_5),na.rm=T),
+            mean_11=mean(as.numeric(q5_11),na.rm=T),
+            mean_12=mean(as.numeric(q5_12),na.rm=T))
+
+a <- a %>%
+  mutate(
+    smart_dems = recode(q61, "Extremely"=4,"Very"=3,"Somewhat"=2,"A little"=1,"Not at all"=0)/4,
+    smart_reps = recode(q62_1, "Extremely"=4,"Very"=3,"Somewhat"=2,"A little"=1,"Not at all"=0)/4,
+    comfort_dems = recode(q87_1,"Extremely comfortable"=3,"Somewhat comfortable"=2,"Not too comfortable"=1,"Not at all comfortable"=0)/3,
+    comfort_reps = recode(q88,"Extremely comfortable"=3,"Somewhat comfortable"=2,"Not too comfortable"=1,"Not at all comfortable"=0)/3,
+    ft_dems = as.numeric(q5_11),
+    ft_reps = as.numeric(q5_12),
+    affpol_smart = case_when(
+      pid==-1 ~ smart_dems-smart_reps,
+      pid==1 ~ smart_reps-smart_dems
+    ),
+    affpol_comfort = case_when(
+      pid==-1 ~ comfort_dems-comfort_reps,
+      pid==1 ~ comfort_reps-comfort_dems
+    ),
+    affpol_ft = case_when(
+      pid==-1 ~ ft_dems-ft_reps,
+      pid==1 ~ ft_reps-ft_dems
+    )
+  )
+
+
+
+## for reinvitations:
+w1_reinvited <- a %>% filter(q87 == "Quick and easy", q89 == "wikiHow") # AV checks
+w1_reinvited <- w1_reinvited %>% filter(survey_time >= 2)
+w1_reinvited <- w1_reinvited %>% filter(gun_index >= 0.05, gun_index <= 0.95)
+
+
+w1_reinvited$thirds <- cut(w1_reinvited$gun_index, breaks = quantile(w1_reinvited$gun_index, c(0, 1/3, 2/3, 1)), include.lowest = TRUE, labels = 1:3)
+a$thirds <- w1_reinvited$thirds[match(a$worker_id,w1_reinvited$worker_id)]
+
+write_csv(a, "../results/intermediate data/gun control (issue 1)/guncontrol_qualtrics_w1_clean.csv")
+
+
+# Wave 2 (main survey) =========================================================
+
+w2 <- read_csv("../data/gun control (issue 1)/wave2_final.csv")[-c(1,2),] %>%
+  clean_names() %>%
+  select(-thirds) # remove all-NA column
+
+w2 <- w2 %>% mutate(start_date_w2 = as_datetime(start_date),
+                    end_date_w2 = as_datetime(end_date),
+                    survey_time_w2 = as.numeric(end_date_w2-start_date_w2))
+
+print('wave 2 survey time:')
+summary(w2$survey_time_w2)
+
+print('audio ok:')
+length(which(w2$q81 == "Quick and easy"))/length(w2$q81)
+print('video ok:')
+length(which(w2$q82 == "wikiHow"))/length(w2$q82)
+
+
+#### Outcomes ####
+
+##### policy opinions ######
+# convert to numeric unit scale:
+w2 <- w2 %>%
+  mutate( # higher = more pro-gun
+    right_to_own_importance_w2 = recode(q19, "Protect the right to own guns" = 1, "Regulate gun ownership" = 0),
+    assault_ban_w2 = recode(q20, "Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4,
+    handgun_ban_w2 = recode(q21, "Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4,
+    concealed_safe_w2 = recode(q22,"Much safer"=4,"Somewhat safer"=3,"No difference"=2,"Somewhat less safe"=1,"Much less safe"=0)/4,
+    stricter_laws_w2 = recode(q23, "Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4
+  )
+
+# Cronbach's alpha
+index_fa <- psych::alpha(select(w2, right_to_own_importance_w2, assault_ban_w2, handgun_ban_w2, concealed_safe_w2, stricter_laws_w2), check.keys = T)
+write.csv(data.frame(cor(select(w2, right_to_own_importance_w2, assault_ban_w2, handgun_ban_w2, concealed_safe_w2, stricter_laws_w2), use = "complete.obs")),row.names = T,
+          file = "../results/guncontrol_cormat_gun_index_w2.csv")
+pdf("../results/guncontrol_cormat_gun_index_w2.pdf")
+w2_corrplot <- corrplot::corrplot(cor(select(w2, right_to_own_importance_w2, assault_ban_w2, handgun_ban_w2, concealed_safe_w2, stricter_laws_w2), use = "complete.obs"),method = "shade")
+dev.off()
+
+print('wave 2 policy opinion alpha:')
+alpha <- index_fa$total["raw_alpha"]
+print(alpha)
+
+# FACTOR ANALYSIS WITH VARIMAX ROTATION (POST)
+pca2 <- psych::principal(select(w2, right_to_own_importance_w2, assault_ban_w2, handgun_ban_w2, concealed_safe_w2, stricter_laws_w2),
+                         rotate="varimax",
+                         nfactors=1
+)
+pc2 <- pca2$Vaccounted[2]
+writeLines(as.character(round(pc2, 2)),con = "../results/outcomes_pc2_study1_post.tex",sep = "%")
+
+w2 <- w2 %>%
+  rowwise() %>%
+  mutate(gun_index_w2 = mean(c(right_to_own_importance_w2,assault_ban_w2,handgun_ban_w2,concealed_safe_w2,stricter_laws_w2), na.rm=T),
+         gun_index_2_w2 = mean(c(right_to_own_importance_w2,assault_ban_w2,handgun_ban_w2,concealed_safe_w2), na.rm=T))
+
+
+# media trust
+w2 <- w2 %>%
+  mutate( # higher = more trusting
+    trust_majornews = recode(q96_1,"A lot"=3,"Some"=2,"Not too much"=1,"Not at all"=0)/3,
+    trust_localnews = recode(q96_2,"A lot"=3,"Some"=2,"Not too much"=1,"Not at all"=0)/3,
+    trust_social = recode(q96_3,"A lot"=3,"Some"=2,"Not too much"=1,"Not at all"=0)/3,
+    trust_youtube = recode(q96_4,"A lot"=3,"Some"=2,"Not too much"=1,"Not at all"=0)/3,
+    fabricate_majornews = recode(q98,"Never"=4,"Once in a while"=3,"About half the time"=2,"Most of the time"=1,"All the time"=0)/4,
+    fabricate_youtube = recode(q100_1,"Never"=4,"Once in a while"=3,"About half the time"=2,"Most of the time"=1,"All the time"=0)/4
+  ) %>%
+  rowwise() %>%
+  mutate(media_trust = mean(trust_majornews,trust_localnews,fabricate_majornews,na.rm=T)) %>%
+  ungroup()
+
+##### affective polarization #####
+# check FTs:
+w2 <- w2 %>%
+  mutate(
+    smart_dems = recode(q61, "Extremely"=4,"Very"=3,"Somewhat"=2,"A little"=1,"Not at all"=0)/4,
+    smart_reps = recode(q62_1, "Extremely"=4,"Very"=3,"Somewhat"=2,"A little"=1,"Not at all"=0)/4,
+    comfort_dems = recode(q92,"Extremely comfortable"=3,"Somewhat comfortable"=2,"Not too comfortable"=1,"Not at all comfortable"=0)/3,
+    comfort_reps = recode(q94,"Extremely comfortable"=3,"Somewhat comfortable"=2,"Not too comfortable"=1,"Not at all comfortable"=0)/3,
+    ft_dems = as.numeric(q90_11),
+    ft_reps = as.numeric(q90_12)
+  )
+
+
+write_csv(w2, "../results/intermediate data/gun control (issue 1)/guncontrol_qualtrics_w2_clean.csv")
+
+
+# join to W1 by MT worker ID:
+w12 <- left_join(a, filter(w2,!is.na(worker_id)), by = "worker_id",suffix=c("_w1","_w2"))
+names(w12)
+
+w12 <- w12 %>%
+  mutate(
+         affpol_smart_w2 = case_when(
+           pid==-1 ~ smart_dems_w2-smart_reps_w2,
+           pid==1 ~ smart_reps_w2-smart_dems_w2
+         ),
+         affpol_comfort_w2 = case_when(
+           pid==-1 ~ comfort_dems_w2-comfort_reps_w2,
+           pid==1 ~ comfort_reps_w2-comfort_dems_w2
+         ),
+         affpol_ft_w2 = case_when(
+           pid==-1 ~ ft_dems_w2-ft_reps_w2,
+           pid==1 ~ ft_reps_w2-ft_dems_w2
+         ))
+
+write_csv(w12, "../results/intermediate data/gun control (issue 1)/guncontrol_qualtrics_w12_clean.csv")
+
+
+# Wave 3 (post survey) =========================================================
+
+w3 <- read_csv("../data/gun control (issue 1)/wave3_final.csv")[-c(1,2),] %>%
+  clean_names()
+
+w3 <- w3 %>% mutate(start_date_w3 = as_datetime(start_date),
+                    end_date_w3 = as_datetime(end_date),
+                    survey_time_w3 = as.numeric(end_date_w3-start_date_w3))
+
+print('wave 3 survey time:')
+summary(w3$survey_time_w3)
+
+
+#### Outcomes ####
+
+# policy opinions, convert to numeric unit scale:
+w3 <- w3 %>%
+  mutate( # higher = more pro-gun
+    right_to_own_importance_w3 = recode(q79, "Protect the right to own guns" = 1, "Regulate gun ownership" = 0),
+    assault_ban_w3 = recode(q81, "Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4,
+    handgun_ban_w3 = recode(q82, "Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4,
+    concealed_safe_w3 = recode(q83,"Much safer"=4,"Somewhat safer"=3,"No difference"=2,"Somewhat less safe"=1,"Much less safe"=0)/4,
+    stricter_laws_w3 = recode(q23, "Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4
+  )
+write.csv(data.frame(cor(select(w3, right_to_own_importance_w3, assault_ban_w3, handgun_ban_w3, concealed_safe_w3, stricter_laws_w3), use = "complete.obs")),row.names = T,
+          file = "../results/guncontrol_cormat_gun_index_w3.csv")
+pdf("../results/guncontrol_corrplot_gunindex_w3.pdf")
+corrplot(cor(select(w3, right_to_own_importance_w3, assault_ban_w3, handgun_ban_w3, concealed_safe_w3, stricter_laws_w3), use = "complete.obs"),method = "shade")
+dev.off()
+
+w3 <- w3 %>%
+  rowwise() %>%
+  mutate(gun_index_w3 = mean(c(right_to_own_importance_w3,assault_ban_w3,handgun_ban_w3,concealed_safe_w3,stricter_laws_w3), na.rm=T),
+         gun_index_2_w3 = mean(c(right_to_own_importance_w3,assault_ban_w3,handgun_ban_w3,concealed_safe_w3), na.rm=T))
+
+##### media trust #####
+w3 <- w3 %>%
+  mutate( # higher = more trusting
+    trust_majornews_w3 = recode(q88_1,"A lot"=3,"Some"=2,"Not too much"=1,"Not at all"=0)/3,
+    trust_localnews_w3 = recode(q88_2,"A lot"=3,"Some"=2,"Not too much"=1,"Not at all"=0)/3,
+    trust_social_w3 = recode(q88_3,"A lot"=3,"Some"=2,"Not too much"=1,"Not at all"=0)/3,
+    trust_youtube_w3 = recode(q88_4,"A lot"=3,"Some"=2,"Not too much"=1,"Not at all"=0)/3,
+    fabricate_majornews_w3 = recode(q90,"Never"=4,"Once in a while"=3,"About half the time"=2,"Most of the time"=1,"All the time"=0)/4,
+    fabricate_youtube_w3 = recode(q92,"Never"=4,"Once in a while"=3,"About half the time"=2,"Most of the time"=1,"All the time"=0)/4
+  ) %>%
+  rowwise() %>%
+  mutate(media_trust_w3 = mean(trust_majornews_w3,trust_localnews_w3,fabricate_majornews_w3,na.rm=T)) %>%
+  ungroup()
+
+
+# affective polarization
+
+w3<- w3 %>%
+  mutate(
+    smart_dems_w3 = recode(q61, "Extremely"=4,"Very"=3,"Somewhat"=2,"A little"=1,"Not at all"=0)/4,
+    smart_reps_w3 = recode(q62_1, "Extremely"=4,"Very"=3,"Somewhat"=2,"A little"=1,"Not at all"=0)/4,
+    comfort_dems_w3 = recode(q94,"Extremely comfortable"=3,"Somewhat comfortable"=2,"Not too comfortable"=1,"Not at all comfortable"=0)/3,
+    comfort_reps_w3 = recode(q96,"Extremely comfortable"=3,"Somewhat comfortable"=2,"Not too comfortable"=1,"Not at all comfortable"=0)/3,
+    ft_dems_w3 = as.numeric(q5_11),
+    ft_reps_w3 = as.numeric(q5_12)
+  )
+
+write_csv(w3, "../results/intermediate data/gun control (issue 1)/guncontrol_qualtrics_mturk_w3_clean.csv")
+
+w123 <- left_join(w12, filter(w3,!is.na(worker_id)), by = "worker_id",suffix=c("","_w3"))
+names(w123)
+
+w123 <- w123 %>%
+  mutate(
+    affpol_smart_w3 = case_when(
+      pid==-1 ~ smart_dems_w3-smart_reps_w3,
+      pid==1 ~ smart_reps_w3-smart_dems_w3
+    ),
+    affpol_comfort_w3 = case_when(
+      pid==-1 ~ comfort_dems_w3-comfort_reps_w3,
+      pid==1 ~ comfort_reps_w3-comfort_dems_w3
+    ),
+    affpol_ft_w3 = case_when(
+      pid==-1 ~ ft_dems_w3-ft_reps_w3,
+      pid==1 ~ ft_reps_w3-ft_dems_w3
+    ))
+
+
+
+
+## YTRecs session data: -------------------------------------------------------
+
+ytrecs <- read_rds("../data/gun control (issue 1)/Wave2_video_June_2021_interactions.rds") %>%
+  clean_names() %>%
+  as_tibble() %>%
+  mutate(duration = end_time2 - start_time2) %>% # have to recalculate this
+  select(topic_id,urlid,pro,anti,duration,pro_up,pro_down,anti_up,anti_down,pro_save,anti_save,start_time2, end_time2) %>%
+  filter(nchar(urlid)==5 & !is.na(pro))
+
+ytrecs <- ytrecs %>%
+  group_by(topic_id,urlid) %>%
+  mutate(dupes = n(),
+         max_duration = ifelse(duration==max(duration),1,0)
+  ) %>%
+  filter(max_duration==1) # using longest session as valid one
+
+ytrecs <- ytrecs %>%
+  rowwise() %>%
+  mutate(
+    pro_up = replace_na(pro_up,0),
+    pro_down = replace_na(pro_down,0),
+    anti_up = replace_na(anti_up,0),
+    anti_down = replace_na(anti_down,0),
+    pro_save = replace_na(pro_save,0),
+    anti_save = replace_na(anti_save,0),
+
+    total_likes = sum(pro_up,anti_up,na.rm=T),
+    total_dislikes = sum(pro_down,anti_down,na.rm=T),
+    total_thumbs = sum(pro_up,pro_down,anti_up,anti_down,na.rm=T),
+    total_saved = sum(pro_save,anti_save,na.rm=T),
+    total_interactions = sum(pro_up,pro_down,anti_up,anti_down,pro_save,anti_save,na.rm=T),
+    positive_interactions = total_likes + total_saved - total_dislikes
+    )
+
+ytrecs <- ytrecs %>%
+  mutate(seed = str_replace(topic_id,".*_([p,a])$","\\1")) %>%
+  mutate(pro_fraction_chosen = case_when(
+    seed=="a" ~ pro/(pro+anti-1),
+    seed=="p" ~ (pro-1)/(pro+anti-1)
+  ))
+# adjust for zeros:
+ytrecs$pro_fraction_chosen[ytrecs$pro==0 & ytrecs$anti==0] <- NA
+
+
+w123 <- w123 %>%
+  ungroup() %>%
+  mutate(topic_id = str_replace(video_link_w2,".*&topicid=(.*)&allowDupe=1&id=(.*)$","\\1"),
+         urlid = str_replace(video_link_w2,".*&topicid=(.*?)&allowDupe=1&id=(.*)$","\\2"),
+  )
+
+
+w123 <- left_join(w123,ytrecs,by=c("topic_id","urlid"))
+
+w123 <- w123 %>%
+  arrange(worker_id, start_time2) %>%
+  group_by(worker_id) %>%
+  slice(1) %>% # Keep first resp
+  ungroup()
+
+print("ISSUE 2 NUMBERS (MTURK):")
+print(paste('count w/ valid ytrecs data:', sum(!is.na(w123$pro))))
+print(paste('count w/ valid ytrecs interactions:', sum(!is.na(w123$total_thumbs))))
+print('interactions:')
+summary(w123$total_interactions)
+
+# create numeric dosage version of treatment:
+w123 <- w123 %>%
+  mutate(treatment_dose = recode(treatment_arm,
+                                 "anti_31"= 1, "anti_22" = 0,
+                                 "pro_31"= 1, "pro_22" = 0,
+                                 "control"=NA_real_),
+         treatment_seed = str_replace(treatment_arm,"(.*)\\_\\d{2}","\\1")
+  )
+
+write_csv(w123, "../results/intermediate data/gun control (issue 1)/guncontrol_qualtrics_w123_clean.csv")

code/gun control (issue 1)/03_analysis_multipletesting.R

@@ -0,0 +1,1293 @@
+cat(rep('=', 80),
+    '\n\n',
+    'OUTPUT FROM: gun control (issue 1)/03_analysis_multipletesting.R',
+    '\n\n',
+    sep = ''
+    )
+
+library(data.table)
+library(car)
+library(sandwich)
+library(lmtest)
+library(ggplot2)
+library(assertthat)
+library(foreach)
+library(doParallel)
+registerDoParallel(cores = detectCores() - 1)
+
+
+
+###############
+## functions ##
+###############
+
+`%.%` <- paste0
+
+simes <- function(ps){
+  min(sort(length(ps) * ps / rank(ps)))
+}
+
+### functions to handle inconsistent interaction ordering of mlm() ###
+
+## convert interaction terms of form 'b#:a#' to 'a#:b#'
+reorder.interaction.names <- function(x, prefix = ''){
+  x <- gsub('^' %.% prefix, '', x)
+  sapply(strsplit(x, ':'),
+         function(y){
+           paste(sort(y), collapse = ':')
+         })
+}
+
+## take term of form 'a1:b1', look up in vector of form 'b#:a#, return 'b1:a1'
+convert.interaction.names <- function(x, y, prefix.y = ''){
+  ind <- match(reorder.interaction.names(x),
+               reorder.interaction.names(y, prefix = prefix.y)
+               )
+  return(y[ind])
+}
+
+## modified from print.linearHypothesis.mlm to use alternate df & return pvals
+##   (print method is responsible for doing the actual computation of pvals)
+extract.lht <- function(x,
+                        SSP = TRUE,
+                        SSPE = SSP,
+                        digits = getOption('digits'),
+                        df.residual = x$df.residual
+                        ){
+  test <- x$test
+  if (!is.null(x$P) && SSP) {
+    P <- x$P
+    cat("\n Response transformation matrix:\n")
+    attr(P, "assign") <- NULL
+    attr(P, "contrasts") <- NULL
+    print(P, digits = digits)
+  }
+  if (SSP) {
+    cat("\nSum of squares and products for the hypothesis:\n")
+    print(x$SSPH, digits = digits)
+  }
+  if (SSPE) {
+    cat("\nSum of squares and products for error:\n")
+    print(x$SSPE, digits = digits)
+  }
+  if ((!is.null(x$singular)) && x$singular) {
+    warning("the error SSP matrix is singular; multivariate tests are unavailable")
+    return(invisible(x))
+  }
+  SSPE.qr <- qr(x$SSPE)
+  eigs <- Re(eigen(qr.coef(SSPE.qr, x$SSPH), symmetric = FALSE)$values)
+  tests <- matrix(NA, 4, 4)
+  rownames(tests) <- c("Pillai", "Wilks", "Hotelling-Lawley",
+                       "Roy")
+  if ("Pillai" %in% test)
+    tests[1, 1:4] <- car:::Pillai(eigs, x$df, df.residual)
+  if ("Wilks" %in% test)
+    tests[2, 1:4] <- car:::Wilks(eigs, x$df, df.residual)
+  if ("Hotelling-Lawley" %in% test)
+    tests[3, 1:4] <- car:::HL(eigs, x$df, df.residual)
+  if ("Roy" %in% test)
+    tests[4, 1:4] <- car:::Roy(eigs, x$df, df.residual)
+  tests <- na.omit(tests)
+  ok <- tests[, 2] >= 0 & tests[, 3] > 0 & tests[, 4] > 0
+  ok <- !is.na(ok) & ok
+  tests <- cbind(x$df, tests, pf(tests[ok, 2], tests[ok, 3],
+                                 tests[ok, 4], lower.tail = FALSE))
+  colnames(tests) <- c("Df", "test stat", "approx F", "num Df",
+                       "den Df", "Pr(>F)")
+  tests <- structure(as.data.frame(tests),
+                     heading = paste("\nMultivariate Test",
+                                     if (nrow(tests) > 1)
+                                       "s", ": ", x$title, sep = ""),
+                     class = c("anova",
+                               "data.frame"
+                               )
+                     )
+  return(tests)
+}
+
+
+
+###############
+## load data ##
+###############
+
+d <- fread('../results/intermediate data/gun control (issue 1)/guncontrol_qualtrics_w123_clean.csv')
+
+## drop pure control
+d <- d[treatment_arm != 'control',]
+
+## drop NA video counts
+d <- d[!is.na(pro) & !is.na(anti),]
+
+
+
+##############
+## controls ##
+##############
+
+platform.controls <- c('age_cat',
+                       'male',
+                       'pol_interest',
+                       'freq_youtube',
+                       'fav_channels',
+                       'popular_channels',
+                       'vid_pref',
+                       'gun_enthusiasm',
+                       'gun_importance'
+                       )
+
+gunpolicy.controls <- 'gun_index'
+
+media.controls <- c('trust_majornews_w1',
+                    'trust_youtube_w1',
+                    'fabricate_majornews_w1',
+                    'fabricate_youtube_w1'
+                    )
+
+affpol.controls <- c('affpol_ft',
+                     'affpol_smart',
+                     'affpol_comfort'
+                     )
+
+controls.raw <- unique(c(platform.controls,
+                         gunpolicy.controls,
+                         media.controls,
+                         affpol.controls
+                         )
+                       )
+
+## transform control variables by creating dummies and demeaning
+controls.trans <- list()
+for (j in controls.raw){
+  ## convert to dummies if needed
+  controls.j <- model.matrix(as.formula('~ 0 + ' %.% j),
+                             model.frame(as.formula('~ 0 + ' %.% j),
+                                         data = d,
+                                         na.action = 'na.pass'
+                                         )
+                             )
+  ## demean by column
+  controls.j <- sweep(controls.j,
+                      MARGIN = 2,
+                      STATS = colMeans(controls.j, na.rm = TRUE),
+                      FUN = `-`,
+                      )
+  colnames(controls.j) <- make.names(colnames(controls.j))
+  ## remove control from original data
+  d[[j]] <- NULL
+  ## reinsert transformed control
+  d <- cbind(d, controls.j)
+  ## keep track of which original controls map to which transformed controls
+  controls.trans[[j]] <- colnames(controls.j)
+}
+
+## map original control variables to transformed versions
+platform.controls <- unlist(controls.trans[platform.controls])
+gunpolicy.controls <- unlist(controls.trans[gunpolicy.controls])
+media.controls <- unlist(controls.trans[media.controls])
+affpol.controls <- unlist(controls.trans[affpol.controls])
+
+
+
+##############
+## outcomes ##
+##############
+
+### hypothesis family 1: platform interactions ###
+
+## platform interaction time: compute windorized usage time
+warning('diverges from pap, 95% windsorized due to extreme outliers')
+d[, platform_duration := duration]
+d[platform_duration <= quantile(d$duration, .025),
+  platform_duration := quantile(d$duration, .025)
+  ]
+d[platform_duration >= quantile(d$duration, .975),
+  platform_duration := quantile(d$duration, .975)
+  ]
+## all platform interaction outcomes
+platform.outcomes <- c('pro_fraction_chosen',
+                       'positive_interactions',  # positive - negative (dislike)
+                       'platform_duration'
+                       )
+
+
+
+### hypothesis family 2: gun policy attitudes ###
+
+## only one preregistered outcome in this family
+gunpolicy.outcomes <- 'gun_index_w2'
+## added 4 jun 2024 by reviewer request
+gunpolicy.outcomes.understanding <- c('right_to_own_importance_w2',
+                                      'concealed_safe_w2'
+                                      )
+
+
+
+### hypothesis family 3: media trust ###
+media.outcomes <- c('trust_majornews_w2',
+                    'trust_youtube_w2',
+                    'fabricate_majornews_w2',
+                    'fabricate_youtube_w2'
+                    )
+
+
+
+### hypothesis family 4: affective polarization ###
+affpol.outcomes <- c('affpol_ft_w2',
+                     'affpol_smart_w2',
+                     'affpol_comfort_w2'
+                     )
+
+outcomes <- unique(c(platform.outcomes,
+                     gunpolicy.outcomes,
+                     media.outcomes,
+                     affpol.outcomes
+                     )
+                   )
+
+
+
+################
+## treatments ##
+################
+
+## create attitude dummies
+d[, attitude := c('anti', 'neutral', 'pro')[thirds]]
+d[, attitude.anti := as.numeric(attitude == 'anti')]
+d[, attitude.neutral := as.numeric(attitude == 'neutral')]
+d[, attitude.pro := as.numeric(attitude == 'pro')]
+
+## create seed dummies
+d[, seed.anti := as.numeric(treatment_seed == 'anti')]
+d[, seed.pro := as.numeric(treatment_seed == 'pro')]
+
+## create recsys dummies
+d[, recsys.22 := as.numeric(treatment_arm %like% '22')]
+d[, recsys.31 := as.numeric(treatment_arm %like% '31')]
+
+## manually define coefficients to estimate
+treatments <- c('attitude.anti:recsys.22',
+                'attitude.anti:recsys.31',
+                'attitude.neutral:seed.anti:recsys.22',
+                'attitude.neutral:seed.pro:recsys.22',
+                'attitude.neutral:seed.anti:recsys.31',
+                'attitude.neutral:seed.pro:recsys.31',
+                'attitude.pro:recsys.22',
+                'attitude.pro:recsys.31'
+                )
+
+contrasts <- rbind(
+  i = c(treat = 'attitude.pro:recsys.31',
+        ctrl = 'attitude.pro:recsys.22'
+        ),
+  ii = c(treat = 'attitude.anti:recsys.31',
+         ctrl = 'attitude.anti:recsys.22'
+         ),
+  iii = c(treat = 'attitude.neutral:seed.pro:recsys.31',
+          ctrl = 'attitude.neutral:seed.pro:recsys.22'
+          ),
+  iv = c(treat = 'attitude.neutral:seed.anti:recsys.31',
+         ctrl = 'attitude.neutral:seed.anti:recsys.22'
+         ),
+  v = c(treat = 'attitude.neutral:seed.pro:recsys.31',
+        ctrl = 'attitude.neutral:seed.anti:recsys.31'
+        ),
+  vi = c(treat = 'attitude.neutral:seed.pro:recsys.22',
+         ctrl = 'attitude.neutral:seed.anti:recsys.22'
+         )
+)
+
+## check that contrasts are valid
+assert_that(all(unlist(contrasts) %in% treatments))
+
+## check that specifications are equivalent
+coefs.v1 <- coef(lm(gun_index_w2 ~ 0 + attitude:treatment_arm, d))
+coefs.v2 <- coef(
+  lm(gun_index_w2 ~
+       0 +
+       attitude.anti:recsys.22 +
+       attitude.anti:recsys.31 +
+       attitude.neutral:seed.anti:recsys.22 +
+       attitude.neutral:seed.pro:recsys.22 +
+       attitude.neutral:seed.anti:recsys.31 +
+       attitude.neutral:seed.pro:recsys.31 +
+       attitude.pro:recsys.22 +
+       attitude.pro:recsys.31,
+     d
+     )
+)
+assert_that(all.equal(unname(sort(coefs.v1)), unname(sort(coefs.v2))))
+
+
+
+##########################
+## hierarchical testing ##
+##########################
+
+## initialize top layer p-values:
+##   does treatment have any effect on any outcome in family
+families <- c('platform',
+              'gunpolicy',
+              'media',
+              'affpol'
+              )
+layer1.pvals <- rep(NA_real_, length(families))
+layer1.notes <- rep('', length(families))
+names(layer1.pvals) <- families
+
+## initialize 2nd layer p-values:
+##   which treatment has detectable effect?
+contrast.pvals <- rep(NA_real_, nrow(contrasts))
+names(contrast.pvals) <- paste(contrasts[, 'treat'],
+                               contrasts[, 'ctrl'],
+                               sep = '.vs.'
+                               )
+layer2.pvals <- list(platform = contrast.pvals,
+                     gunpolicy = contrast.pvals,
+                     media = contrast.pvals,
+                     affpol = contrast.pvals
+                     )
+rm(contrast.pvals)
+
+## initialize 3rd layer p-values:
+##   on which specific outcome in family?
+layer3.pvals <- list()
+layer3.ests <- list()
+layer3.ses <- list()
+layer3.notes <- list()
+for (i in 1:length(families)){
+  family <- families[i]
+  layer3.pvals[[family]] <- list()
+  layer3.ests[[family]] <- list()
+  layer3.ses[[family]] <- list()
+  layer3.notes[[family]] <- list()
+  outcomes <- get(family %.% '.outcomes')
+  for (j in 1:nrow(contrasts)){
+    contrast <- paste(contrasts[j, 'treat'],
+                      contrasts[j, 'ctrl'],
+                      sep = '.vs.'
+                      )
+    layer3.pvals[[family]][[contrast]] <- numeric(0)
+    layer3.ests[[family]][[contrast]] <- numeric(0)
+    layer3.ses[[family]][[contrast]] <- numeric(0)
+    for (k in 1:length(outcomes)){
+      outcome <- outcomes[k]
+      layer3.pvals[[family]][[contrast]][outcome] <- NA_real_
+      layer3.ests[[family]][[contrast]][outcome] <- NA_real_
+      layer3.ses[[family]][[contrast]][outcome] <- NA_real_
+      layer3.notes[[family]][outcome] <- ''
+    }
+  }
+}
+
+
+
+### begin nested analyses ###
+
+for (i in 1:length(families)){
+
+  family <- families[i]
+  family.outcomes <- get(family %.% '.outcomes')
+  family.controls <- get(family %.% '.controls')
+  family.controls.interactions <- as.character(
+    outer(treatments,
+          family.controls,
+          FUN = function(x, y) x %.% ':' %.% y
+          )
+  )
+
+  family.formula <-
+    'cbind(' %.%                # outcomes
+    paste(family.outcomes,
+          collapse = ', '
+          ) %.%  ') ~\n0 +\n' %.%
+    paste(treatments,           # treatments (base terms)
+          collapse = ' +\n'
+          ) %.% ' +\n' %.%
+    paste(family.controls,      # controls (base terms)
+          collapse = ' +\n'
+          )
+
+  cat(rep('=', 80),
+      '\n\nHYPOTHESIS FAMILY: ',
+      family,
+      '\n\nrunning mlm:\n\n',
+      family.formula,
+      '\n\n',
+      sep = ''
+      )
+
+  ## run model
+  family.mod <- lm(family.formula, d)
+  ## hack to eliminate NA coefs
+  if (any(is.na(coef(family.mod)))){
+    if ('mlm' %in% class(family.mod)){
+      drop <- rownames(coef(family.mod))[is.na(coef(family.mod))[, 1]]
+    } else {
+      drop <- names(coef(family.mod))[is.na(coef(family.mod))]
+    }
+    drop <- convert.interaction.names(drop,
+                                      c(family.controls,
+                                        family.controls.interactions
+                                        )
+                                      )
+    layer1.notes[[i]] <-
+      layer1.notes[[i]] %.%
+      'dropped the following coefs: ' %.%
+      paste(drop, sep = ', ') %.%
+      '\n\n'
+    family.formula <- gsub(
+      '\\s+\\+\\s+(' %.% paste(drop, collapse = '|') %.% ')',
+      '',
+      family.formula
+    )
+    family.mod <- lm(family.formula, d)
+  }
+
+  family.vcov <- vcovHC(family.mod)
+  if (is.null(dim(coef(family.mod)))){
+    coef.names <- names(coef(family.mod))
+  } else {
+    coef.names <- rownames(coef(family.mod))
+  }
+
+  ### top layer: test overall significance of all contrasts on all outcomes ###
+  ## convert interaction terms to whatever mlm() named it
+  treats <- convert.interaction.names(contrasts[, 'treat'], coef.names)
+  ctrls <- convert.interaction.names(contrasts[, 'ctrl'], coef.names)
+  ## test jointly
+  lht.attempt <- tryCatch({
+    if ('mlm' %in% class(family.mod)){
+      contrast.lht <- linearHypothesis(
+        family.mod,
+        vcov. = family.vcov,
+        hypothesis.matrix = sprintf('%s - %s', treats, ctrls),
+        rhs = matrix(0, nrow = nrow(contrasts), ncol = length(family.outcomes)),
+        test = 'Pillai'
+      )
+      layer1.pvals[[i]] <- extract.lht(contrast.lht)[, 'Pr(>F)']
+    } else {
+      contrast.lht <- linearHypothesis(
+        family.mod,
+        vcov. = family.vcov,
+        hypothesis.matrix = sprintf('%s - %s', treats, ctrls),
+        rhs = matrix(0, nrow = nrow(contrasts), ncol = length(family.outcomes)),
+        test = 'F'
+      )
+      layer1.pvals[[i]] <- contrast.lht[['Pr(>F)']][2]
+    }
+  },
+  error = function(e){
+    warning(sprintf('caught error in %s family:', family), e)
+    ## return error as string for inclusion in notes
+    'caught error: ' %.%
+      e %.%
+      '\n\n'
+  })
+  if (lht.attempt %like% 'caught error'){
+    layer1.notes[[i]] <-
+      layer1.notes[[i]] %.% lht.attempt
+  }
+
+
+
+  ### layer 2: test each contrast individually on all outcomes ###
+
+  for (j in 1:nrow(contrasts)){
+    ## test group equality on all outcomes
+    if ('mlm' %in% class(family.mod)){
+      contrast.lht <-
+        linearHypothesis(
+        family.mod,
+        vcov. = family.vcov,
+        hypothesis.matrix = sprintf('%s - %s', treats[j], ctrls[j]),
+        rhs = matrix(0, nrow = 1, ncol = length(family.outcomes)),
+        test = 'Pillai'
+      )
+      layer2.pvals[[i]][j] <- extract.lht(contrast.lht)[, 'Pr(>F)']
+    } else {
+      contrast.lht <- linearHypothesis(
+        family.mod,
+        vcov. = family.vcov,
+        hypothesis.matrix = sprintf('%s - %s', treats[j], ctrls[j]),
+        rhs = matrix(0, nrow = 1, ncol = length(family.outcomes)),
+        test = 'F'
+      )
+      layer2.pvals[[i]][j] <- contrast.lht[['Pr(>F)']][2]
+    }
+  }
+
+  ### layer 3: test each contrast on each outcome individually ###
+
+  for (k in 1:length(family.outcomes)){
+
+    outcome <- family.outcomes[k]
+
+    outcome.formula <-
+      outcome %.% ' ~\n0 +\n' %.%
+      paste(treatments,         # treatments (base terms)
+            collapse = ' +\n'
+            ) %.% ' +\n' %.%
+    paste(family.controls,      # controls (base terms)
+          collapse = ' +\n'
+          )
+
+    cat(rep('-', 40), '\n\nrunning lm:\n\n', outcome.formula, '\n\n', sep = '')
+
+    outcome.mod <- lm(outcome.formula, d)
+    ## hack to eliminate NA coefs
+    if (any(is.na(coef(outcome.mod)))){
+      drop <- names(coef(outcome.mod))[is.na(coef(outcome.mod))]
+      drop <- convert.interaction.names(drop,
+                                        c(family.controls,
+                                          family.controls.interactions
+                                          )
+                                        )
+      layer3.notes[[i]][k] <-
+        layer3.notes[[i]][k] %.%
+        'dropped the following coefs: ' %.%
+        paste(drop, sep = ', ') %.%
+        '\n\n'
+      outcome.formula <- gsub(
+        '\\s+\\+\\s+(' %.% paste(drop, collapse = '|') %.% ')',
+        '',
+        outcome.formula
+      )
+      outcome.mod <- lm(outcome.formula, d)
+    }
+
+    outcome.vcov <- vcovHC(outcome.mod)
+    if (any(!is.finite(outcome.vcov))){
+      outcome.vcov <- vcov(outcome.mod)
+      layer3.notes[[i]][k] <-
+        layer3.notes[[i]][k] %.%
+        'falling back to non-robust vcov\n\n'
+    }
+    coef.names <- names(coef(outcome.mod))
+
+    for (j in 1:nrow(contrasts)){
+
+      ## convert this interaction term to whatever lm() named it
+      treat <- convert.interaction.names(contrasts[j, 'treat'], coef.names)
+      ctrl <- convert.interaction.names(contrasts[j, 'ctrl'], coef.names)
+      ## test group equality on this outcome
+      contrast.lht <- linearHypothesis(
+        outcome.mod,
+        vcov. = outcome.vcov,
+        hypothesis.matrix = sprintf('%s - %s', treat, ctrl),
+        test = 'F'
+      )
+      layer3.pvals[[i]][[j]][k] <- contrast.lht[['Pr(>F)']][2]
+      layer3.ests[[i]][[j]][k] <- (
+        coef(outcome.mod)[treat] - coef(outcome.mod)[ctrl]
+      ) ## * attr(d[[outcome]], 'scaled:scale')  # note: uncomment if rescaling
+      layer3.ses[[i]][[j]][k] <- sqrt(
+        outcome.vcov[treat, treat] +
+          outcome.vcov[ctrl, ctrl] -
+          2 * outcome.vcov[treat, ctrl]
+      )
+
+      ## ## confirm
+      ## linearHypothesis(
+      ##   outcome.mod,
+      ##   vcov. = outcome.vcov,
+      ##   hypothesis.matrix = sprintf('%s - %s', treat, ctrl),
+      ##   test = 'F'
+      ## )
+      ## (coef(outcome.mod)[treat] - coef(outcome.mod)[ctrl])^2 /
+      ##   (
+      ##     outcome.vcov[treat, treat] +
+      ##       outcome.vcov[ctrl, ctrl] -
+      ##       2 * outcome.vcov[treat, ctrl]
+      ##   )
+      ## linearHypothesis(
+      ##   outcome.mod,
+      ##   vcov. = outcome.vcov,
+      ##   hypothesis.matrix = sprintf('%s - %s', treat, ctrl),
+      ##   test = 'Chisq'
+      ## )
+      ## 2 - 2 * pnorm(abs(
+      ## (coef(outcome.mod)[treat] - coef(outcome.mod)[ctrl]) /
+      ##   sqrt(
+      ##     outcome.vcov[treat, treat] +
+      ##       outcome.vcov[ctrl, ctrl] -
+      ##       2 * outcome.vcov[treat, ctrl]
+      ##   )
+      ## ))
+
+    }
+
+  }
+
+}
+
+
+
+#################################
+## multiple testing correction ##
+#################################
+
+thresh <- .05
+
+## if layer-1 f-test is infeasible for a family due to collinearity,
+##   obtain layer-1 p-values for that family by simes
+for (i in which(is.na(layer1.pvals))){
+  layer1.pvals[i] <- simes(layer2.pvals[[i]])
+}
+
+## multiple testing adjustment for layer 1
+layer1.pvals.adj <- p.adjust(layer1.pvals, 'BH')
+layer1.nonnull.prop <- mean(layer1.pvals.adj < thresh)
+
+## test layer-2 hypotheses only if layer 1 passes
+layer2.pvals.adj <- layer2.pvals  # start by copying unadjusted layer-2 p-values
+layer2.nonnull.prop <- rep(NA, length(layer1.pvals.adj))
+names(layer2.nonnull.prop) <- names(layer1.pvals.adj)
+for (i in 1:length(layer1.pvals)){
+  if (layer1.pvals.adj[i] < thresh){  # if layer 1 passes
+    ## adjust for multiplicity within layer 2...
+    layer2.pvals.adj[[i]] <- p.adjust(layer2.pvals[[i]], 'BH')
+    ## ... and inflate to account for selection at layer 1
+    layer2.pvals.adj[[i]] <-
+      pmin(layer2.pvals.adj[[i]] / layer1.nonnull.prop, 1)
+    ## keep track of selection at layer 2 for use in layer 3
+    layer2.nonnull.prop[i] <- mean(layer2.pvals.adj[[i]] < thresh)
+  } else {  # if layer 1 fails
+    layer2.pvals.adj[[i]] <- rep(NA_real_, length(layer2.pvals[[i]]))
+    names(layer2.pvals.adj[[i]]) <- names(layer2.pvals[[i]])
+  }
+}
+
+## test layer-3 hypotheses only if layers 1 & 2 pass
+layer3.pvals.adj <- layer3.pvals  # start by copying unadjusted layer-3 p-values
+for (i in 1:length(layer1.pvals.adj)){
+  for (j in 1:length(layer2.pvals.adj[[i]])){
+    ##
+    if (layer1.pvals.adj[i] < thresh &&      # if layer 1 passes...
+          layer2.pvals.adj[[i]][j] < thresh  # ... and if layer 2 passes
+        ){
+      ## adjust for multiplicity within layer 3...
+      layer3.pvals.adj[[i]][[j]] <- p.adjust(layer3.pvals[[i]][[j]], 'BH')
+      ## ... and inflate to account for selection at layer 1
+      layer3.pvals.adj[[i]][[j]] <- pmin(
+        layer3.pvals.adj[[i]][[j]] / layer1.nonnull.prop / layer2.nonnull.prop[i],
+        1
+      )
+    } else {
+      layer3.pvals.adj[[i]][[j]] <- rep(NA_real_, length(layer3.pvals[[i]][[j]]))
+      names(layer3.pvals.adj[[i]][[j]]) <- names(layer3.pvals[[i]][[j]])
+    }
+  }
+}
+
+pvals.adj <- data.table(layer1 = character(0),
+                        layer2 = character(0),
+                        layer3 = character(0),
+                        p.adj = numeric(0),
+                        est = numeric(0),
+                        se = numeric(0)
+                        )
+for (i in 1:length(layer1.pvals.adj)){
+  pvals.adj <- rbind(pvals.adj,
+                     data.table(layer1 = names(layer1.pvals.adj)[i],
+                                layer2 = 'overall',
+                                layer3 = 'overall',
+                                p.adj = layer1.pvals.adj[i],
+                                est = NA_real_,
+                                se = NA_real_
+                                )
+                     )
+  for (j in 1:length(layer2.pvals.adj[[i]])){
+    pvals.adj <- rbind(pvals.adj,
+                       data.table(layer1 = names(layer1.pvals.adj)[i],
+                                  layer2 = names(layer2.pvals.adj[[i]])[j],
+                                  layer3 = 'overall',
+                                  p.adj = layer2.pvals.adj[[i]][j],
+                                  est = NA_real_,
+                                  se = NA_real_
+                                  )
+                       )
+    for (k in 1:length(layer3.pvals.adj[[i]][[j]])){
+      pvals.adj <- rbind(pvals.adj,
+                         data.table(layer1 = names(layer1.pvals.adj)[i],
+                                    layer2 = names(layer2.pvals.adj[[i]])[j],
+                                    layer3 = names(layer3.pvals.adj[[i]][[j]])[k],
+                                    p.adj = layer3.pvals.adj[[i]][[j]][k],
+                                    est = layer3.ests[[i]][[j]][k],
+                                    se = layer3.ses[[i]][[j]][k]
+                                    )
+                         )
+    }
+  }
+}
+
+## write out
+fwrite(pvals.adj, '../results/intermediate data/guncontrol_padj_basecontrol.csv')
+
+## prettify for reading
+pvals.adj.pretty <- pvals.adj
+colnames(pvals.adj.pretty) <- gsub('layer1',
+                                   'layer1_hypothesisfamily',
+                                   colnames(pvals.adj.pretty)
+                                   )
+colnames(pvals.adj.pretty) <- gsub('layer2',
+                                   'layer2_treatmentcontrast',
+                                   colnames(pvals.adj.pretty)
+                                   )
+colnames(pvals.adj.pretty) <- gsub('layer3',
+                                   'layer3_specificoutcome',
+                                   colnames(pvals.adj.pretty)
+                                   )
+pvals.adj.pretty[, layer2_treatmentcontrast := gsub(
+  'attitude\\.(pro|anti|neutral)(:seed\\.(pro|anti))?:recsys.(31|22)',
+  '\\1 \\3 \\4',
+  layer2_treatmentcontrast
+)]
+pvals.adj.pretty[, layer2_treatmentcontrast := gsub(
+  '.vs.',
+  ' - ',
+  layer2_treatmentcontrast,
+  fixed = TRUE
+)]
+pvals.adj.pretty[, layer2_treatmentcontrast := gsub(
+  ' +',
+  ' ',
+  layer2_treatmentcontrast
+)]
+fwrite(pvals.adj.pretty,
+       '../results/intermediate data/gun control (issue 1)/guncontrol_padj_basecontrol_pretty.csv'
+       )
+
+print('preregistered results:')
+pvals.adj.pretty[p.adj < .05 & layer3_specificoutcome != 'overall',]
+
+
+
+##############################################
+## added 4 jun 2024 at request of reviewers ##
+##############################################
+
+## analyze components of main policy outcome index that relate to
+## post-experiment w2 "understanding" of an issue, using w1 version
+## of that same outcome as the only control (analogous to outcome index
+## regression, which uses w2 index as outcome and w1 index as control)
+
+## initialize results table
+understanding.results <- data.table(layer2_treatmentcontrast = character(0),
+                                    layer3_specificoutcome = character(0),
+                                    est = numeric(0),
+                                    se = numeric(0),
+                                    p = numeric(0)
+                                    )
+
+## loop over outcomes
+for (k in 1:length(gunpolicy.outcomes.understanding)){
+
+  outcome <- gunpolicy.outcomes.understanding[k]
+
+  outcome.formula <-
+    outcome %.% ' ~\n0 +\n' %.%
+    paste(treatments,         # treatments (base terms)
+          collapse = ' +\n'
+          ) %.% ' +\n' %.%
+    paste(gsub('_w2', '', outcome),      # controls (w1 outcome)
+          collapse = ' +\n'
+          )
+
+  cat(rep('-', 40), '\n\nrunning lm:\n\n', outcome.formula, '\n\n', sep = '')
+
+  outcome.mod <- lm(outcome.formula, d)
+  ## hack to eliminate NA coefs
+  if (any(is.na(coef(outcome.mod)))){
+    drop <- names(coef(outcome.mod))[is.na(coef(outcome.mod))]
+    drop <- convert.interaction.names(drop,
+                                      c(family.controls,
+                                        family.controls.interactions
+                                        )
+                                      )
+    layer3.notes[[i]][k] <-
+      layer3.notes[[i]][k] %.%
+      'dropped the following coefs: ' %.%
+      paste(drop, sep = ', ') %.%
+      '\n\n'
+    outcome.formula <- gsub(
+      '\\s+\\+\\s+(' %.% paste(drop, collapse = '|') %.% ')',
+      '',
+      outcome.formula
+    )
+    outcome.mod <- lm(outcome.formula, d)
+  }
+
+  outcome.vcov <- vcovHC(outcome.mod)
+  if (any(!is.finite(outcome.vcov))){
+    outcome.vcov <- vcov(outcome.mod)
+    layer3.notes[[i]][k] <-
+      layer3.notes[[i]][k] %.%
+      'falling back to non-robust vcov\n\n'
+  }
+  coef.names <- names(coef(outcome.mod))
+
+  ## loop over treatment contrasts
+  for (j in 1:nrow(contrasts)){
+
+    ## convert this interaction term to whatever llm() named it
+    treat <- convert.interaction.names(contrasts[j, 'treat'], coef.names)
+    ctrl <- convert.interaction.names(contrasts[j, 'ctrl'], coef.names)
+    ## test group equality on this outcome
+    contrast.lht <- linearHypothesis(
+      outcome.mod,
+      vcov. = outcome.vcov,
+      hypothesis.matrix = sprintf('%s - %s', treat, ctrl),
+      test = 'F'
+    )
+
+    ## prettify name of contrast for readability
+
+    contrast <- treat %.% ' - ' %.% ctrl
+    contrast <- gsub('attitude\\.(pro|anti|neutral)', '\\1', contrast)
+    contrast <- gsub('seed\\.(pro|anti)', '\\1', contrast)
+    contrast <- gsub('recsys.(31|22)', '\\1', contrast)
+    contrast <- gsub(':', ' ', contrast)
+    contrast <- gsub(' +', ' ', contrast)
+
+    p <- contrast.lht[['Pr(>F)']][2]
+    est <- (
+      coef(outcome.mod)[treat] - coef(outcome.mod)[ctrl]
+    ) ## * attr(d[[outcome]], 'scaled:scale')  # note: uncomment if rescaling
+    se <- sqrt(
+      outcome.vcov[treat, treat] +
+        outcome.vcov[ctrl, ctrl] -
+        2 * outcome.vcov[treat, ctrl]
+    )
+
+    understanding.results <- rbind(
+      understanding.results,
+      data.table(
+        layer2_treatmentcontrast = contrast,
+        layer3_specificoutcome = outcome,
+        p,
+        est,
+        se
+      )
+    )
+
+  }
+
+}
+
+## conduct multiple testing adjustment within newly exploratory results
+understanding.results[, p.adj := p.adjust(p, 'BH')]
+print('exploratory results on understanding-related questions:')
+understanding.results[p.adj < .05,]
+
+fwrite(understanding.results,
+       '../results/intermediate data/gun control (issue 1)/guncontrol_understanding_basecontrol_pretty.csv'
+       )
+
+
+
+#############################################################
+## preregistered exploratory heterogeneous effect analysis ##
+#############################################################
+
+# outcome is gun_index_w2
+# construct moderators by cutting demographics & pre-treatment vars at midpoint
+
+d[,
+  pol_interest_hi := as.numeric(
+    pol_interest > median(pol_interest, na.rm = TRUE)
+  )]
+d[,
+  age_hi := as.numeric(
+    age > median(age, na.rm = TRUE)
+  )]
+d[,
+  freq_youtube_hi := as.numeric(
+    freq_youtube > median(freq_youtube, na.rm = TRUE)
+  )]
+
+moderator_variables <- c('pol_interest_hi',
+                         'age_hi',
+                         'male',
+                         'freq_youtube_hi'
+                         )
+## added 4 jun 2024 at request of reviewer
+moderator_variables_revision <- 'college'
+
+interaction_results <- data.table()
+for (moderator_variable in c(moderator_variables, moderator_variables_revision)){
+
+  d[, moderator := get(moderator_variable)]
+
+  mod.attitude.anti <- lm(
+    gun_index_w2 ~
+      recsys.31 * moderator +
+      gun_index,   # only control is pre-treatment outcome, as in primary analysis
+    data = d[attitude.anti == 1]
+  )
+  vcov.attitude.anti <- vcovHC(mod.attitude.anti)
+  test.attitude.anti <- coeftest(mod.attitude.anti, vcov.attitude.anti)
+  interaction_results <- rbind(
+    interaction_results,
+    data.table(subset = 'attitude.anti',
+               interaction = 'recsys.31:' %.% moderator_variable,
+               test.attitude.anti['recsys.31:moderator', , drop = FALSE]
+               ),
+    fill = TRUE
+  )
+
+  mod.attitude.pro <- lm(
+    gun_index_w2 ~
+      recsys.31 * moderator +
+      gun_index,   # only control is pre-treatment outcome, as in primary analysis
+    data = d[attitude.pro == 1]
+  )
+  vcov.attitude.pro <- vcovHC(mod.attitude.pro)
+  test.attitude.pro <- coeftest(mod.attitude.pro, vcov.attitude.pro)
+  interaction_results <- rbind(
+    interaction_results,
+    data.table(subset = 'attitude.pro',
+               interaction = 'recsys.31:' %.% moderator_variable,
+               test.attitude.pro['recsys.31:moderator', , drop = FALSE]
+               ),
+    fill = TRUE
+  )
+
+  mod.attitude.neutral.seed.anti <- lm(
+    gun_index_w2 ~
+      recsys.31 * moderator +
+      gun_index,   # only control is pre-treatment outcome, as in primary analysis
+    data = d[attitude.neutral == 1 & seed.anti == 1]
+  )
+  vcov.attitude.neutral.seed.anti <- vcovHC(mod.attitude.neutral.seed.anti)
+  test.attitude.neutral.seed.anti <- coeftest(mod.attitude.neutral.seed.anti,
+                                             vcov.attitude.neutral.seed.anti
+                                             )
+  interaction_results <- rbind(
+    interaction_results,
+    data.table(subset = 'attitude.neutral.seed.anti',
+               interaction = 'recsys.31:' %.% moderator_variable,
+               test.attitude.neutral.seed.anti[
+                 'recsys.31:moderator', , drop = FALSE
+               ]
+               ),
+    fill = TRUE
+  )
+
+  mod.attitude.neutral.seed.pro <- lm(
+    gun_index_w2 ~
+      recsys.31 * moderator +
+      gun_index,   # only control is pre-treatment outcome, as in primary analysis
+    data = d[attitude.neutral == 1 & seed.pro == 1]
+  )
+  vcov.attitude.neutral.seed.pro <- vcovHC(mod.attitude.neutral.seed.pro)
+  test.attitude.neutral.seed.pro <- coeftest(mod.attitude.neutral.seed.pro,
+                                             vcov.attitude.neutral.seed.pro                                             )
+  interaction_results <- rbind(
+    interaction_results,
+    data.table(subset = 'attitude.neutral.seed.pro',
+               interaction = 'recsys.31:' %.% moderator_variable,
+               test.attitude.neutral.seed.pro[
+                 'recsys.31:moderator', , drop = FALSE
+               ]
+               ),
+    fill = TRUE
+  )
+
+}
+
+# no significant heterogeneity even before multiple testing correction
+print('heterogeneity results before multiple correction:')
+interaction_results[`Pr(>|t|)` < .05,]
+# none survives a BH correction
+interaction_results[, p.adj := p.adjust(`Pr(>|t|)`, 'BH')]
+print('heterogeneity p-values after multiple correction:')
+interaction_results[, p.adj]
+
+## updated 4 jun 2024 at request of reviewer
+colnames(interaction_results) <- c(
+  subset = 'subset',
+  interaction = 'interaction',
+  Estimate = 'est',
+  `Std. Error` = 'se',
+  `t value` = 't',
+  `Pr(>|t|)` = 'p',
+  p.adj = 'p.adj'
+)[colnames(interaction_results)]
+fwrite(interaction_results,
+        '../results/intermediate data/gun control (issue 1)/guncontrol_heterogeneity_basecontrol.csv'
+        )
+
+
+
+###############################################
+## added 30 sep 2024 at request of reviewers ##
+###############################################
+
+## what are minimum detectable effects, given multiple testing correction?
+
+n_sims <- 1000
+params_sims <- expand.grid(seed = 19104 + 0:(n_sims - 1),
+                           effect = seq(from = .01, to = .05, by = .001)
+                           )
+
+## step 1: identify largest p-value s.t. we would have rejected layer-1 null
+## (that at least one treatment contrast has effect on policy index)
+## to do this, we hold fixed p-values for all other layer-1 hypothesis families
+layer1.pvals.mde <- layer1.pvals
+layer1.pvals.mde['gunpolicy'] <- 0
+while (p.adjust(layer1.pvals.mde, 'BH')['gunpolicy'] <= .05){
+  layer1.pvals.mde['gunpolicy'] <- layer1.pvals.mde['gunpolicy'] + .001
+}
+pval.cutoff <- layer1.pvals.mde['gunpolicy']
+print('to achieve significance of policy attitude family at layer 1 (pooled test of any effect on policy index from any contrast) when correcting for multiple layer-1 hypothesis families, this is the minimum cutoff value after conducting simes correction of layer 2 pvals:')
+pval.cutoff
+
+## if layer-1 null was rejected for the policy outcome, then we would use this
+## correction factor when interpreting layer-2 p-values (for specific contrasts)
+layer1.nonnull.prop.if.gt.cutoff <- mean(c(
+  p.adjust(layer1.pvals.mde, 'BH')[c('platform', 'media', 'affpol')] < .05,
+  TRUE
+))
+
+## the sims below will only examine 3/1 vs 2/2 treatment contrasts, so we will
+## hold fixed the layer-2 p-values that relate to seed contrasts
+pvals.for.seed.contrasts.on.policyindex <- layer2.pvals$mwpolicy[
+  c('attitude.neutral:seed.pro:recsys.31.vs.attitude.neutral:seed.anti:recsys.31',
+    'attitude.neutral:seed.pro:recsys.22.vs.attitude.neutral:seed.anti:recsys.22'
+    )
+]
+
+
+
+## step 2: prepare simulations based on real data ------------------------------
+
+mod.attitude.anti <- lm(
+  gun_index_w2 ~ recsys.31 + gun_index,
+  data = d[attitude.anti == 1]
+)
+X.attitude.anti <- model.matrix(mod.attitude.anti)
+residual.sd.attitude.anti <- sd(resid(mod.attitude.anti))
+## confirm that this recovers fitted values
+##   model.matrix(mod.attitude.anti) %*% coef(mod.attitude.anti)
+assert_that(all(
+  predict(mod.attitude.anti) ==
+    X.attitude.anti %*% coef(mod.attitude.anti)
+))
+## we will create simulated outcomes, given hypothesized treatment effect
+## == intercept +                                              <-- part A
+##    real coef * real pretreatment attitude +                 <-- part A
+##    hypothesized treatment effect * real treatment status +  <-- part B
+##    rnorm(mean = 0, sd = real residual outcome sd)           <-- part C
+## A: generate fitted values under hypothesized effect size
+coef.attitude.anti.baseline <- coef(mod.attitude.anti)
+coef.attitude.anti.baseline['recsys.31'] <- 0
+Y.attitude.anti.baseline <-
+  as.numeric(X.attitude.anti %*% coef.attitude.anti.baseline)
+## C: will be added below with hypothesized effect * treatment
+## B: will be drawn below with rnorm(mean=0, sd=residual_sd)
+
+## repeat above for respondents with pro attitude
+mod.attitude.pro <- lm(
+  gun_index_w2 ~ recsys.31 + gun_index,
+  data = d[attitude.pro == 1]
+)
+X.attitude.pro <- model.matrix(mod.attitude.pro)
+residual.sd.attitude.pro <- sd(resid(mod.attitude.pro))
+coef.attitude.pro.baseline <- coef(mod.attitude.pro)
+coef.attitude.pro.baseline['recsys.31'] <- 0
+Y.attitude.pro.baseline <-
+  as.numeric(X.attitude.pro %*% coef.attitude.pro.baseline)
+
+## repeat above for respondents with neutral attitude assigned to pro seed
+mod.attitude.neutral.seed.pro <- lm(
+  gun_index_w2 ~ recsys.31 + gun_index,
+  data = d[attitude.neutral == 1 & seed.pro == 1]
+)
+X.attitude.neutral.seed.pro <- model.matrix(mod.attitude.neutral.seed.pro)
+residual.sd.attitude.neutral.seed.pro <- sd(resid(mod.attitude.neutral.seed.pro))
+coef.attitude.neutral.seed.pro.baseline <- coef(mod.attitude.neutral.seed.pro)
+coef.attitude.neutral.seed.pro.baseline['recsys.31'] <- 0
+Y.attitude.neutral.seed.pro.baseline <-
+  as.numeric(X.attitude.neutral.seed.pro %*% coef.attitude.neutral.seed.pro.baseline)
+
+## repeat above for respondents with neutral attitude assigned to anti seed
+mod.attitude.neutral.seed.anti <- lm(
+  gun_index_w2 ~ recsys.31 + gun_index,
+  data = d[attitude.neutral == 1 & seed.anti == 1]
+)
+X.attitude.neutral.seed.anti <- model.matrix(mod.attitude.neutral.seed.anti)
+residual.sd.attitude.neutral.seed.anti <- sd(resid(mod.attitude.neutral.seed.anti))
+coef.attitude.neutral.seed.anti.baseline <- coef(mod.attitude.neutral.seed.anti)
+coef.attitude.neutral.seed.anti.baseline['recsys.31'] <- 0
+Y.attitude.neutral.seed.anti.baseline <-
+  as.numeric(X.attitude.neutral.seed.anti %*% coef.attitude.neutral.seed.anti.baseline)
+
+
+
+## step 3: conduct sims --------------------------------------------------------
+
+sims.attitude.anti <- foreach(seed = params_sims$seed,
+                              effect = params_sims$effect,
+                              .combine = rbind
+                              ) %dopar%
+  {
+    set.seed(seed)
+    Y <-
+      Y.attitude.anti.baseline +
+      effect * X.attitude.anti[, 'recsys.31'] +
+      rnorm(
+        n = nrow(X.attitude.anti),
+        mean = 0,
+        sd = residual.sd.attitude.anti
+      )
+    mod <- lm(Y ~ 0 + X.attitude.anti)
+    smry <- coeftest(mod, vcovHC(mod))
+    cbind(
+      seed,
+      effect,
+      data.table(smry['X.attitude.antirecsys.31', , drop = FALSE])
+    )
+  }
+
+sims.attitude.pro <- foreach(seed = params_sims$seed,
+                              effect = params_sims$effect,
+                              .combine = rbind
+                              ) %dopar%
+  {
+    set.seed(seed)
+    Y <-
+      Y.attitude.pro.baseline +
+      effect * X.attitude.pro[, 'recsys.31'] +
+      rnorm(
+        n = nrow(X.attitude.pro),
+        mean = 0,
+        sd = residual.sd.attitude.pro
+      )
+    mod <- lm(Y ~ 0 + X.attitude.pro)
+    smry <- coeftest(mod, vcovHC(mod))
+    cbind(
+      seed,
+      effect,
+      data.table(smry['X.attitude.prorecsys.31', , drop = FALSE])
+    )
+  }
+
+sims.attitude.neutral.seed.anti <- foreach(seed = params_sims$seed,
+                                           effect = params_sims$effect,
+                                           .combine = rbind
+                                           ) %dopar%
+  {
+    set.seed(seed)
+    Y <-
+      Y.attitude.neutral.seed.anti.baseline +
+      effect * X.attitude.neutral.seed.anti[, 'recsys.31'] +
+      rnorm(
+        n = nrow(X.attitude.neutral.seed.anti),
+        mean = 0,
+        sd = residual.sd.attitude.neutral.seed.anti
+      )
+    mod <- lm(Y ~ 0 + X.attitude.neutral.seed.anti)
+    smry <- coeftest(mod, vcovHC(mod))
+    cbind(
+      seed,
+      effect,
+      data.table(smry['X.attitude.neutral.seed.antirecsys.31', , drop = FALSE])
+    )
+  }
+
+sims.attitude.neutral.seed.pro <- foreach(seed = params_sims$seed,
+                                           effect = params_sims$effect,
+                                           .combine = rbind
+                                           ) %dopar%
+  {
+    set.seed(seed)
+    Y <-
+      Y.attitude.neutral.seed.pro.baseline +
+      effect * X.attitude.neutral.seed.pro[, 'recsys.31'] +
+      rnorm(
+        n = nrow(X.attitude.neutral.seed.pro),
+        mean = 0,
+        sd = residual.sd.attitude.neutral.seed.pro
+      )
+    mod <- lm(Y ~ 0 + X.attitude.neutral.seed.pro)
+    smry <- coeftest(mod, vcovHC(mod))
+    cbind(
+      seed,
+      effect,
+      data.table(smry['X.attitude.neutral.seed.prorecsys.31', , drop = FALSE])
+    )
+  }
+
+
+
+## step 4: analyze power results -----------------------------------------------
+
+## without multiple-testing corrections
+
+print('mde for respondents with anti attitude (conventional analysis w/o correction):')
+sims.attitude.anti[,
+                   .(p.reject = mean(`Pr(>|t|)` < .05)),
+                   by = effect
+                   ][p.reject >= .8, min(effect)]
+
+print('mde for respondents with pro attitude (conventional analysis w/o correction):')
+sims.attitude.pro[,
+                  .(p.reject = mean(`Pr(>|t|)` < .05)),
+                  by = effect
+                  ][p.reject >= .8, min(effect)]
+
+print('mde for respondents with neutral attitude assigned to pro seed (conventional analysis w/o correction):')
+sims.attitude.neutral.seed.anti[,
+                                .(p.reject = mean(`Pr(>|t|)` < .05)),
+                                by = effect
+                                ][p.reject >= .8, min(effect)]
+
+print('mde for respondents with neutral attitude assigned to anti seed (conventional analysis w/o correction):')
+sims.attitude.neutral.seed.pro[,
+                               .(p.reject = mean(`Pr(>|t|)` < .05)),
+                               by = effect
+                               ][p.reject >= .8, min(effect)]
+
+
+
+## with multiple testing correction
+
+sims <- rbind(
+  sims.attitude.anti,
+  sims.attitude.pro,
+  sims.attitude.neutral.seed.anti,
+  sims.attitude.neutral.seed.pro
+)
+
+sims.layer1 <- sims[
+,
+  .(pval.pooled = ifelse(
+    ## if these results would lead us to reject layer-1 pooled null of no effect
+    ## on policy attitudes from any treatment contrast
+    simes(c(
+      `Pr(>|t|)`,
+      pvals.for.seed.contrasts.on.policyindex
+    )) <= pval.cutoff,
+    ## disaggregate layer-2 results report with procedure from above
+    ## (BH correction, then inflate by 1/prop of layer-1 sig results)
+    ## then subset to only those p-values relating to 3/1 vs 2/2 contrast
+    ## to see if any are <.05 after full correction procedure
+    yes = min(
+      p.adjust(c(`Pr(>|t|)`, pvals.for.seed.contrasts.on.policyindex),
+               'BH'
+               )[1:4] / layer1.nonnull.prop.if.gt.cutoff
+    ),
+    no = Inf
+  )
+  ),
+  by = .(seed, effect)
+]
+print('with multiple testing correction:')
+sims.layer1[, .(p.reject = mean(pval.pooled <= pval.cutoff)), by = effect]
+print('mde:')
+sims.layer1[,
+            .(p.reject = mean(pval.pooled <= pval.cutoff)),
+            by = effect
+            ][p.reject >= .8, min(effect)]
+

code/minimum wage (issue 2)/01_trt_assign.R

@@ -0,0 +1,239 @@
+cat(rep('=', 80),
+    '\n\n',
+    'OUTPUT FROM: minimum wage (issue 2)/01_trt_assign.R',
+    '\n\n',
+    sep = ''
+    )
+    
+library(tidyverse)
+library(janitor)
+library(lubridate)
+library(randomizr)
+library(haven)
+
+# create directory to hold cached intermediate files
+dir.create("../results/intermediate data/minimum wage (issue 2)",
+           recursive = TRUE, showWarnings = FALSE)
+
+
+w1 <- read_csv("../data/minimum wage (issue 2)/YouTube+Min+Wage+-+Apr+2022+presurvey_May+24,+2022_02.57.csv")[-c(1,2),] %>% clean_names() %>%
+  filter(finished == "True", q62 == "I agree to participate\u2028")
+
+# Recodes ======================================================================
+
+w1 <- w1 %>% mutate(start_date = as_datetime(start_date),
+                    end_date = as_datetime(end_date),
+                    survey_time = as.numeric(end_date-start_date))
+
+print('wave 1 survey time:')
+summary(w1$survey_time)
+
+w1 <- w1 %>%
+  mutate(man = ifelse(q26 == "Man", 1, 0),
+         black = ifelse(str_detect(q29, "Black"), 1, 0),
+         white = ifelse(str_detect(q29, "White"), 1, 0),
+         college = ifelse(str_detect(q30, "college ") | str_detect(q30, "Post"), 1, 0),
+         income_gt50k = ifelse(q31 %in% names(table(w1$q31))[c(2,3,5,10:13)], 1, 0)
+  )
+
+# PID:
+w1$pid <- ifelse(w1$pid1=="Democrat",-1,NA)
+w1$pid <- ifelse(w1$pid1=="Republican",1,w1$pid)
+w1$pid <- ifelse(w1$pid4=="Closer to the Republican Party",1,w1$pid)
+w1$pid <- ifelse(w1$pid4=="Closer to the Democratic Party",-1,w1$pid)
+w1$pid <- ifelse(w1$pid4=="Neither",0,w1$pid)
+
+print('wave 1 party id:')
+round(table(w1$pid) / sum(table(w1$pid)), digits=2)
+
+w1$ideo <- ifelse(w1$ideo1=="Liberal",-1,NA)
+w1$ideo <- ifelse(w1$ideo1=="Conservative",1,w1$ideo)
+w1$ideo <- ifelse(w1$ideo4=="Closer to liberals",-1,w1$ideo)
+w1$ideo <- ifelse(w1$ideo4=="Closer to conservatives",1,w1$ideo)
+w1$ideo <- ifelse(w1$ideo4=="Neither",0,w1$ideo)
+
+print('wave 1 ideology:')
+round(table(w1$ideo) / sum(table(w1$ideo)), digits=2)
+
+w1$age <- 2022 - as.numeric(w1$q27)
+
+
+
+# A/V check ====================================================================
+
+print("audio check:")
+length(which(w1$q87 == "Quick and easy")) / length(w1$q87)
+
+print("video check:")
+length(which(w1$q89 == "wikiHow")) / length(w1$q89)
+
+w1$audio_ok <- 1*(w1$q87 == "Quick and easy")
+w1$video_ok <- 1*(w1$q89 == "wikiHow")
+
+# Convert pre-treatment DV to numeric unit scale -------------------------------
+
+w1 <- w1 %>%
+  mutate( # higher = more conservative
+    minwage15 = recode(minwage15,"Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4,
+    rtwa_v1 = recode(rtwa_v1, "Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4,
+    #minwage_inflation = recode(minwage_inflation,"Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4,
+    rtwa_v2 = recode(rtwa_v2, "Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4,
+    mw_support = recode(mw_support, "Strongly oppose raising the minimum wage"=4,"Somewhat oppose raising the minimum wage"=3,"Neither support nor oppose raising the minimum wage"=2,"Somewhat support raising the minimum wage"=1,"Strongly support raising the minimum wage"=0)/4,
+    minwage_howhigh = recode(minwage_howhigh, "Much lower than the current level"=4,"Somewhat lower than the current level"=3,"About the current level"=2,"Somewhat higher than the current level"=1,"Much higher than the current level"=0)/4,
+    mw_help = recode(mw_help, "Would hurt low-income workers\n10\n"=9,"9"=8,"8"=7,"7"=6,"6"=5,"5"=4,"4"=3,"3"=2,"2"=1,"Would help low-income workers\n1"=0)/9,
+    mw_restrict = recode(mw_restrict, "Would restrict businesses' freedom\n1\n"=9,"2"=8,"3"=7,"4"=6,"5"=5,"6"=4,"7"=3,"8"=2,"9"=1,"Would protect workers from exploitation\n10\n"=0)/9,
+    minwage_text = (25-as.numeric(minwage_text))/25
+    )
+
+w1 <- w1 %>%
+  rowwise() %>%
+  mutate(mw_index = mean(c(minwage15, rtwa_v1, rtwa_v2, mw_support, minwage_howhigh, mw_help, mw_restrict, minwage_text),
+                        na.rm=T))
+
+
+
+# trim sample -------------------------------------------------------------
+
+# We exclude respondents who took less than 120 seconds to complete the Wave 1 survey, failed either
+# an audio check or a video check, as well as those whose gun policy opinions fall within the most
+# extreme 5% of the gun policy index outcome (i.e. < 0.25 or > 4.75 on the 0-5 scale, to guard
+# against eventual ceiling/floor effects; in our pilot study this was 15% of the sample).
+
+w1 <- w1 %>% filter(audio_ok == 1, video_ok == 1)
+w1 <- w1 %>% filter(survey_time >= 2)
+w1 <- w1 %>% filter(mw_index >= 0.025, mw_index <= 0.975)
+w1 <- w1 %>% filter(!is.na(worker_id))
+w1 <- w1 %>% distinct(worker_id, .keep_all = TRUE)
+
+print('mw index:')
+summary(w1$mw_index)
+
+
+
+# Block random assignment ======================================================
+
+# We randomly assign respondents to both a seed video type (pro-gun vs. anti-gun) and a recommendation system (3/1 vs. 2/2)
+# blocking on Wave 1 gun policy opinions. In the sample of respondents
+# who will be invited to Wave 2, we form terciles of the Wave 1 gun policy opinion index, referring
+# to the lower, middle and upper terciles as anti-gun, moderate and pro-gun respectively
+
+w1$tercile <- cut(w1$mw_index, breaks = quantile(w1$mw_index, c(0, 1/3, 2/3, 1)), include.lowest = TRUE, labels = 1:3)
+tapply(w1$mw_index, w1$tercile, mean)
+table(w1$tercile)
+
+# pure control (with 1/5 probability), anti-MW 2/2 (with 2/5 probability), or anti-MW 3/1 (with 2/5 probability).
+# seed position (pro-MW or anti-MW), recommendation system (2/2 or 3/1), or a
+# pure control group (i.e. one of five possible conditions) with equal probability
+
+# For MTurk --------------------------------------------------------------------
+
+set.seed(2022)
+
+w1$trt_system <- block_ra(blocks = w1$tercile, prob_each = c(2/5, 2/5, 1/5), conditions = c("2/2", "3/1", "pure control"))
+
+w1$seed <- rep("", nrow(w1))
+w1[w1$tercile == 1,]$seed <- "pro-minwage seed"
+w1[w1$tercile == 3,]$seed <- "anti-minwage seed"
+w1[w1$tercile == 2,]$seed <- complete_ra(N = length(which(w1$tercile == 2)), prob = 0.5, conditions = c("pro-minwage seed",
+                                                                                                        "anti-minwage seed"))
+with(w1[w1$tercile == 1,], round(prop.table(table(seed, trt_system)), digits = 3))
+with(w1[w1$tercile == 2,], round(prop.table(table(seed, trt_system)), digits = 3))
+with(w1[w1$tercile == 3,], round(prop.table(table(seed, trt_system)), digits = 3))
+
+w1 <- w1 %>% mutate(trt_assign = case_when(seed == "anti-minwage seed" & trt_system == "2/2" ~ 1,
+                                           seed == "anti-minwage seed" & trt_system == "3/1" ~ 2,
+                                           seed == "pro-minwage seed" & trt_system == "2/2" ~ 3,
+                                           seed == "pro-minwage seed" & trt_system == "3/1" ~ 4,
+                                           trt_system == "pure control" ~ 5))
+
+print('treatment assignment:')
+table(w1$trt_assign)
+print('seed assignment:')
+table(w1$seed)
+print('system assignment:')
+table(w1$trt_system)
+print('seed & system assignment:')
+table(w1$trt_system, w1$seed)
+
+# w1$batch <- sample(c(rep(1:floor(nrow(w1)/500), 500), rep(6, nrow(w1)-500*5)))
+# sent to Qualtrics
+# write_csv(data.frame(trt = w1$trt_assign, id = w1$worker_id), "mw_mturk_wave1_assignments.csv")
+
+
+
+# YouGov -----------------------------------------------------------------------
+
+w1 <- read_sav("../data/minimum wage (issue 2)/PRIN0016_W1_OUTPUT.sav") %>% filter(consent == 22)
+w1$caseid <- as.character(w1$caseid)
+
+# Convert pre-treatment DV to numeric unit scale
+w1 <- w1 %>%
+  mutate( # higher = more conservative
+    minwage15 = (minwage15-1)/4,
+    rtwa_v1 = (RTWA_v1-1)/4,
+    rtwa_v2 = (RTWA_v2-1)/4,
+    mw_support = (mw_support-1)/4,
+    minwage_howhigh = (minwage_howhigh-1)/4,
+    mw_help = (mw_help_a-1)/9,
+    mw_restrict = (10-mw_restrict_1)/9,
+    minwage_text = (25-minwage_text)/25
+  )
+
+
+w1 <- w1 %>%
+  rowwise() %>%
+  mutate(mw_index = mean(c(minwage15, rtwa_v1, rtwa_v2, mw_support, minwage_howhigh, mw_help, mw_restrict, minwage_text),
+                         na.rm=T))
+
+w1 <- w1 %>% mutate(start_date = as_datetime(starttime),
+                    end_date = as_datetime(endtime),
+                    survey_time = as.numeric(end_date-start_date))
+
+print('wave 1 survey time:')
+summary(w1$survey_time)
+
+w1 <- w1 %>% filter(survey_time >= 2)
+w1 <- w1 %>% filter(mw_index >= 0.025, mw_index <= 0.975)
+
+print('mw index:')
+summary(w1$mw_index)
+
+w1$tercile <- cut(w1$mw_index, breaks = quantile(w1$mw_index, c(0, 1/3, 2/3, 1)), include.lowest = TRUE, labels = 1:3)
+
+write_csv(select(w1, caseid, tercile, mw_index), "../results/intermediate data/minimum wage (issue 2)/yougov_terciles.csv")
+
+
+# pure control (with 1/5 probability), anti-MW 2/2 (with 2/5 probability), or anti-MW 3/1 (with 2/5 probability).
+# seed position (pro-MW or anti-MW), recommendation system (2/2 or 3/1), or a
+# pure control group (i.e. one of five possible conditions) with equal probability
+
+set.seed(22022)
+
+# For YouGov
+w1$trt_system <- block_ra(blocks = w1$tercile, prob_each = c(.5, .5), conditions = c("2/2", "3/1"))
+
+w1$seed <- rep("", nrow(w1))
+w1[w1$tercile == 1,]$seed <- "pro-minwage seed"
+w1[w1$tercile == 3,]$seed <- "anti-minwage seed"
+w1[w1$tercile == 2,]$seed <- complete_ra(N = length(which(w1$tercile == 2)), prob = 0.5, conditions = c("pro-minwage seed",
+                                                                                                        "anti-minwage seed"))
+with(w1[w1$tercile == 1,], round(prop.table(table(seed, trt_system)), digits = 3))
+with(w1[w1$tercile == 2,], round(prop.table(table(seed, trt_system)), digits = 3))
+with(w1[w1$tercile == 3,], round(prop.table(table(seed, trt_system)), digits = 3))
+
+w1 <- w1 %>% mutate(trt_assign = case_when(seed == "anti-minwage seed" & trt_system == "2/2" ~ 1,
+                                           seed == "anti-minwage seed" & trt_system == "3/1" ~ 2,
+                                           seed == "pro-minwage seed" & trt_system == "2/2" ~ 3,
+                                           seed == "pro-minwage seed" & trt_system == "3/1" ~ 4))
+
+print('treatment assignment:')
+table(w1$trt_assign)
+print('seed assignment:')
+table(w1$seed)
+print('system assignment:')
+table(w1$trt_system)
+print('seed & system assignment:')
+table(w1$trt_system, w1$seed)
+
+# sent to YouGov
+# write_csv(select(w1, caseid, trt_system, seed, trt_assign), "mw_yg_wave1_assignments.csv")

code/minimum wage (issue 2)/02_clean_merge.R

@@ -0,0 +1,488 @@
+## YouTube Algorithms and Minimum Wage Opinions
+## Data collected April 2022 via MTurk/CloudResearch
+
+cat(rep('=', 80),
+    '\n\n',
+    'OUTPUT FROM: minimum wage (issue 2)/02_clean_merge.R',
+    '\n\n',
+    sep = ''
+    )
+
+## Preamble ----------------------------
+library(tidyverse)
+library(janitor)
+library(lubridate)
+library(stargazer)
+library(broom)
+
+a <- read_csv("../data/minimum wage (issue 2)/YouTube+Min+Wage+-+Apr+2022+presurvey_July+5,+2022_12.50.csv")[-c(1,2),] %>%
+  clean_names()
+
+
+# Wave 1 =======================================================================
+
+## Recodes:
+a <- a %>% mutate(start_date = as_datetime(start_date),
+                  end_date = as_datetime(end_date),
+                  survey_time = as.numeric(end_date-start_date))
+
+print('wave 1 survey time')
+summary(a$survey_time)
+
+# Demographics -----------------------------------------------------------------
+
+a <- a %>%
+  mutate(female = ifelse(q26 == "Woman", 1, 0),
+         male = ifelse(q26 == "Man", 1, 0),
+         black = ifelse(str_detect(q29, "Black"), 1, 0),
+         white = ifelse(str_detect(q29, "White"), 1, 0),
+         college = ifelse(str_detect(q30, "college ") | str_detect(q30, "Post"), 1, 0),
+         income_gt50k = ifelse(q31 %in% names(table(a$q31))[c(2,3,5,10:13)], 1, 0)
+  )
+a$income_gt50k[is.na(a$q31)] <- NA
+
+# PID:
+
+a <- a %>%
+  mutate(pid = case_when(pid1=="Democrat" ~ -1,
+                         pid1=="Republican" ~ 1,
+                         pid4=="Closer to the Republican Party" ~ 1,
+                         pid4=="Closer to the Democratic Party" ~ -1,
+                         pid4=="Neither" ~ 0))
+
+tabyl(a,pid)
+
+
+a <- a %>%
+  mutate(ideo = case_when(ideo1=="Liberal" ~ -1,
+                          ideo1=="Conservative" ~ 1,
+                          ideo4=="Closer to conservatives" ~ 1,
+                          ideo4=="Closer to liberals" ~ -1,
+                          ideo4=="Neither" ~ 0))
+
+tabyl(a,ideo)
+
+a$age <- 2022-as.numeric(a$q27)
+
+# age categories: 18-29;  30-44;  45-64;  65+
+a <- a %>%
+  mutate(age_cat = case_when(age>=18 & age<=29 ~ "18-29",
+                             age>=30 & age<=44 ~ "30-44",
+                             age>=45 & age<=64 ~ "45-64",
+                             age>=65 ~ "65+"
+  ))
+a <- a %>%
+  fastDummies::dummy_cols(select_columns = "age_cat",remove_selected_columns = F)
+
+## Need:
+# political  interest  (5-point  scale:   1=Not  atall interested, 5=Extremely interested),
+# self-reported YouTube usage frequency (7-pointscale: 0=None, 6=More than 3 hours per day),
+# number of self-reported favorite YouTubechannels (count coded from open-ended question: “Who/what are your favorite YouTubebroadcasters or channels?”; 0 if blank),
+# indicator for having watched videos from popularchannels (1 if any selected:  “In the past week, have you watched videos from any of thefollowing  YouTube  broadcasters  or  channels?”),
+# video  vs.   text  preference  (1=Alwaysprefer  videos,  10=Always  prefer  text),
+# gun  enthusiasm  (additive  index  of  “Do  you  ordoes anyone in your household own a gun?”  with yes=1 and “How often, if ever, do youvisit websites about guns, hunting or other shooting sports?”  from 0=Never or Hardlyever to 1=Sometimes or Often),
+# gun policy issue importance (4-point scale:  1=Not atall important, 4=Very important)
+
+a <- a %>%
+  mutate(pol_interest = dplyr::recode(q91,"Extremely interested"=5,"Very interested"=4,"Somewhat interested"=3,"Not very interested"=2,"Not at all interested"=1),
+         freq_youtube = dplyr::recode(q77,"More than 3 hours per day"=6,"2–3 hours per day"=5,"1–2 hours per day"=4,"31–59 minutes per day"=3,"10–30 minutes per day"=2,"Less than 10 minutes per day"=1,"None"=0),
+  )
+
+descr_data <- as.data.frame(select(a,
+                     female,
+                     white,
+                     black,
+                     age,
+                     college,
+                     income_gt50k))
+descr_data <- descr_data %>% filter(rowSums(is.na(.)) != ncol(.))
+descriptive_tab <- stargazer(descr_data,
+                             summary = T, digits=2,
+                             summary.stat=c("mean","sd","median","min","max","n"),
+                             covariate.labels = c("Female",
+                                                  "White",
+                                                  "Black",
+                                                  "Age",
+                                                  "College educated",
+                                                  "Income \\textgreater 50k"),
+                             float = F,
+                             out = "../results/minwage_descriptive_tab.tex")
+
+summary_tab <- a %>%
+  dplyr::summarize(female = mean(female,na.rm=T),
+                   white = mean(white,na.rm=T),
+                   black = mean(black,na.rm=T),
+                   age1829 = mean(`age_cat_18-29`,na.rm=T),
+                   age3044 = mean(`age_cat_30-44`,na.rm=T),
+                   age4564 = mean(`age_cat_45-64`,na.rm=T),
+                   age65p = mean(`age_cat_65+`,na.rm=T),
+                   college = mean(college,na.rm=T),
+                   income_gt50k = mean(income_gt50k,na.rm=T),
+                   democrat = mean(pid==-1,na.rm=T),
+                   republican = mean(pid==1,na.rm=T))
+
+summary_tab <- pivot_longer(summary_tab,
+                            cols=c(female,
+                                   white,
+                                   black,
+                                   age1829,
+                                   age3044,
+                                   age4564,
+                                   age65p,
+                                   college,
+                                   income_gt50k,
+                                   democrat,
+                                   republican),
+                            names_to = "outcome",values_to = "survey_avg")
+outcome_labels <- data.frame(outcome_pretty = c("Female",
+                                                "White",
+                                                "Black",
+                                                "Age 18-29",
+                                                "Age 30-44",
+                                                "Age 45-64",
+                                                "Age 65+",
+                                                "College educated",
+                                                "Income >$50k",
+                                                "Democrat",
+                                                "Republican"),
+                             outcome = c("female",
+                                         "white",
+                                         "black",
+                                         "age1829",
+                                         "age3044",
+                                         "age4564",
+                                         "age65p",
+                                         "college",
+                                         "income_gt50k",
+                                         "democrat",
+                                         "republican"))
+summary_tab$outcome_pretty <-  outcome_labels$outcome_pretty[match(summary_tab$outcome,outcome_labels$outcome)]
+summary_tab <- summary_tab %>%
+  mutate(outcome_pretty = factor(outcome_pretty,levels = c("Republican",
+                                                           "Democrat",
+                                                           "Income >$50k",
+                                                           "College educated",
+                                                           "Age 65+",
+                                                           "Age 45-64",
+                                                           "Age 30-44",
+                                                           "Age 18-29",
+                                                           "Female",
+                                                           "Black",
+                                                           "White"
+  ),ordered=T))
+
+(descrip_fig <- ggplot(summary_tab) +
+    geom_point(aes(y=outcome_pretty,x=survey_avg)) +
+    geom_text(aes(y=outcome_pretty,x=survey_avg,label=paste0(round(100*survey_avg,0),"%")),nudge_x = 0.1) +
+    scale_y_discrete("") +
+    scale_x_continuous("",labels=scales::percent_format(),limits=c(0,1)) +
+    theme_bw()
+)
+ggsave(descrip_fig,filename = "../results/minwage_demographics.pdf",height=5,width=4)
+
+
+
+# A/V check
+print('audio ok:')
+length(which(a$q87 == "Quick and easy"))/length(a$q87)
+print('video ok:')
+length(which(a$q89 == "wikiHow"))/length(a$q89)#dk
+
+#### Outcomes ####
+
+##### policy opinions #####
+# convert to numeric unit scale:
+a <- a %>%
+  mutate( # higher = more conservative or anti-min wage
+    minwage15 = dplyr::recode(minwage15,"Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4,
+    rtwa_v1 = dplyr::recode(rtwa_v1, "Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4,
+    # minwage_inflation = recode(minwage_inflation,"Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4,
+    rtwa_v2 = dplyr::recode(rtwa_v2, "Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4,
+    mw_support = dplyr::recode(mw_support, "Strongly oppose raising the minimum wage"=4,"Somewhat oppose raising the minimum wage"=3,"Neither support nor oppose raising the minimum wage"=2,"Somewhat support raising the minimum wage"=1,"Strongly support raising the minimum wage"=0)/4,
+    minwage_howhigh = dplyr::recode(minwage_howhigh, "Much lower than the current level"=4,"Somewhat lower than the current level"=3,"About the current level"=2,"Somewhat higher than the current level"=1,"Much higher than the current level"=0)/4,
+    mw_help = dplyr::recode(mw_help, "Would hurt low-income workers\n10\n"=9,"9"=8,"8"=7,"7"=6,"6"=5,"5"=4,"4"=3,"3"=2,"2"=1,"Would help low-income workers\n1"=0)/9,
+    mw_restrict = dplyr::recode(mw_restrict, "Would restrict businesses' freedom\n1\n"=9,"2"=8,"3"=7,"4"=6,"5"=5,"6"=4,"7"=3,"8"=2,"9"=1,"Would protect workers from exploitation\n10\n"=0)/9,
+    minwage_text_r = (25-as.numeric(minwage_text))/25,
+  )
+a$minwage_text_r[as.numeric(a$minwage_text)>25] <- NA
+
+a <- a %>%
+  rowwise() %>%
+  mutate(mw_index = mean(c(minwage15, rtwa_v1, rtwa_v2, mw_support, minwage_howhigh, mw_help, mw_restrict, minwage_text_r), na.rm=T)) %>%
+  ungroup()
+
+# Cronbach's alpha
+index_fa <- psych::alpha(select(a, minwage15, rtwa_v1, rtwa_v2, mw_support, minwage_howhigh, mw_help, mw_restrict, minwage_text_r), check.keys = TRUE)
+write.csv(data.frame(cor(select(a, minwage15, rtwa_v1, rtwa_v2, mw_support, minwage_howhigh, mw_help, mw_restrict, minwage_text_r), use = "complete.obs")),row.names = T,
+          file = "../results/minwage_cormat_mwindex_w1.csv")
+
+pdf("../results/corrplot_mwindex_w1.pdf")
+w1_corrplot <- corrplot::corrplot(cor(select(a, minwage15, rtwa_v1, rtwa_v2, mw_support, minwage_howhigh, mw_help, mw_restrict, minwage_text_r), use = "complete.obs"),method = "shade")
+dev.off()
+
+alpha <- index_fa$total["raw_alpha"]
+writeLines(as.character(round(alpha,2)),con = "../results/minwage_outcomes_alpha_w1_mturk.tex",sep = "%")
+
+# FACTOR ANALYSIS WITH VARIMAX ROTATION (PRE)
+pca2 <- psych::principal(select(a, minwage15, rtwa_v1, rtwa_v2, mw_support, minwage_howhigh, mw_help, mw_restrict, minwage_text_r),
+                         rotate="varimax",
+                         nfactors=1
+)
+pc2 <- pca2$Vaccounted[2]
+writeLines(as.character(round(pc2, 2)),con = "../results/outcomes_pc2_study2_pre.tex",sep = "%")
+
+##### media trust #####
+a <- a %>%
+  mutate( # higher = more trusting
+    trust_majornews = dplyr::recode(q58_1,"A lot"=3,"Some"=2,"Not too much"=1,"Not at all"=0)/3,
+    trust_localnews = dplyr::recode(q58_2,"A lot"=3,"Some"=2,"Not too much"=1,"Not at all"=0)/3,
+    trust_social = dplyr::recode(q58_3,"A lot"=3,"Some"=2,"Not too much"=1,"Not at all"=0)/3,
+    trust_youtube = dplyr::recode(q58_4,"A lot"=3,"Some"=2,"Not too much"=1,"Not at all"=0)/3,
+    fabricate_majornews = dplyr::recode(q89_1,"Never"=4,"Once in a while"=3,"About half the time"=2,"Most of the time"=1,"All the time"=0)/4,
+    fabricate_youtube = dplyr::recode(q90,"Never"=4,"Once in a while"=3,"About half the time"=2,"Most of the time"=1,"All the time"=0)/4
+  ) %>%
+  rowwise() %>%
+  mutate(media_trust = mean(trust_majornews,trust_localnews,fabricate_majornews,na.rm=T)) %>%
+  ungroup()
+
+media_trust_fa <- psych::alpha(select(a, trust_majornews,trust_localnews,fabricate_majornews), check.keys = TRUE)
+print('media trust alpha:')
+media_trust_fa$total["raw_alpha"]
+
+
+##### affective polarization #####
+# check FTs:
+a %>%
+  group_by(pid) %>%
+  summarize(mean_2=mean(as.numeric(q5_2),na.rm=T), # Trump
+            mean_5=mean(as.numeric(q5_5),na.rm=T), # Biden
+            mean_11=mean(as.numeric(q5_11),na.rm=T), # dems
+            mean_12=mean(as.numeric(q5_12),na.rm=T)) # reps
+
+a <- a %>%
+  mutate( # higher = more trusting
+    smart_dems = dplyr::recode(q61, "Extremely"=4,"Very"=3,"Somewhat"=2,"A little"=1,"Not at all"=0)/4,
+    smart_reps = dplyr::recode(q62_1, "Extremely"=4,"Very"=3,"Somewhat"=2,"A little"=1,"Not at all"=0)/4,
+    comfort_dems = dplyr::recode(q87_1,"Extremely comfortable"=3,"Somewhat comfortable"=2,"Not too comfortable"=1,"Not at all comfortable"=0)/3,
+    comfort_reps = dplyr::recode(q88,"Extremely comfortable"=3,"Somewhat comfortable"=2,"Not too comfortable"=1,"Not at all comfortable"=0)/3,
+    ft_dems = as.numeric(q5_11),
+    ft_reps = as.numeric(q5_12),
+    affpol_smart = case_when(
+      pid==-1 ~ smart_dems-smart_reps,
+      pid==1 ~ smart_reps-smart_dems
+    ),
+    affpol_comfort = case_when(
+      pid==-1 ~ comfort_dems-comfort_reps,
+      pid==1 ~ comfort_reps-comfort_dems
+    ),
+    affpol_ft = case_when(
+      pid==-1 ~ ft_dems-ft_reps,
+      pid==1 ~ ft_reps-ft_dems
+    )
+  )
+
+
+
+## for reinvitations:
+w1_reinvited <- a %>% filter(q87 == "Quick and easy", q89 == "wikiHow") # AV checks
+w1_reinvited <- w1_reinvited %>% filter(mw_index >= 0.025, mw_index <= 0.975)
+
+
+w1_reinvited$thirds <- cut(w1_reinvited$mw_index, breaks = quantile(w1_reinvited$mw_index, c(0, 1/3, 2/3, 1)), include.lowest = TRUE, labels = 1:3)
+a$thirds <- w1_reinvited$thirds[match(a$worker_id,w1_reinvited$worker_id)]
+
+write_csv(a, "../results/intermediate data/minimum wage (issue 2)/qualtrics_w1_clean.csv")
+
+
+# Wave 2 (main survey) =========================================================
+
+
+w2 <- read_csv("../data/minimum wage (issue 2)/YouTube+Min+Wage+-+Apr+2022+main+survey_July+5,+2022_12.47.csv")[-c(1,2),] %>%
+  clean_names() %>%
+  select(-thirds) # remove all-NA column
+
+w2 <- w2 %>% mutate(start_date_w2 = as_datetime(start_date),
+                    end_date_w2 = as_datetime(end_date),
+                    survey_time_w2 = as.numeric(end_date_w2-start_date_w2))
+
+print('wave 2 survey time:')
+summary(w2$survey_time_w2)
+
+print('audio ok:')
+length(which(w2$q81 == "Quick and easy"))/length(w2$q81)
+print('video ok:')
+length(which(w2$q82 == "wikiHow"))/length(w2$q82)
+
+
+#### Outcomes ####
+
+##### policy opinions ######
+# convert to numeric unit scale:
+w2 <- w2 %>%
+  mutate( # higher = more pro-gun
+    minwage15 = dplyr::recode(minwage15,"Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4,
+    rtwa_v1 = dplyr::recode(rtwa_v1, "Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4,
+    rtwa_v2 = dplyr::recode(rtwa_v2, "Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4,
+    mw_support = dplyr::recode(mw_support, "Strongly oppose raising the minimum wage"=4,"Somewhat oppose raising the minimum wage"=3,"Neither support nor oppose raising the minimum wage"=2,"Somewhat support raising the minimum wage"=1,"Strongly support raising the minimum wage"=0)/4,
+    minwage_howhigh = dplyr::recode(minwage_howhigh, "Much lower than the current level"=4,"Somewhat lower than the current level"=3,"About the current level"=2,"Somewhat higher than the current level"=1,"Much higher than the current level"=0)/4,
+    mw_help = dplyr::recode(mw_help, "Would hurt low-income workers\n10\n"=9,"9"=8,"8"=7,"7"=6,"6"=5,"5"=4,"4"=3,"3"=2,"2"=1,"Would help low-income workers\n1"=0)/9,
+    mw_restrict = dplyr::recode(mw_restrict, "Would restrict businesses' freedom\n1\n"=9,"2"=8,"3"=7,"4"=6,"5"=5,"6"=4,"7"=3,"8"=2,"9"=1,"Would protect workers from exploitation\n10\n"=0)/9,
+    minwage_text_r = (25-as.numeric(minwage_text))/25,
+  )
+w2$minwage_text_r[as.numeric(w2$minwage_text)>25] <- NA
+
+w2 <- w2 %>%
+  rowwise() %>%
+  mutate(mw_index = mean(c(minwage15, rtwa_v1, rtwa_v2, mw_support, minwage_howhigh, mw_help, mw_restrict, minwage_text_r), na.rm=T)) %>%
+  ungroup()
+
+
+# Cronbach's alpha
+index_fa <- psych::alpha(select(w2, minwage15, rtwa_v1, rtwa_v2, mw_support, minwage_howhigh, mw_help, mw_restrict, minwage_text_r), check.keys = T)
+write.csv(data.frame(cor(select(w2, minwage15, rtwa_v1, rtwa_v2, mw_support, minwage_howhigh, mw_help, mw_restrict, minwage_text_r), use = "complete.obs")),row.names = T,
+          file = "../results/minwage_cormat_mw_index_w2.csv")
+
+pdf("../results/minwage_corrplot_mwindex_w2.pdf")
+w2_corrplot <- corrplot::corrplot(cor(select(w2, minwage15, rtwa_v1, rtwa_v2, mw_support, minwage_howhigh, mw_help, mw_restrict, minwage_text_r), use = "complete.obs"),method = "shade")
+dev.off()
+
+print('wave 2 policy opinion alpha:')
+(alpha <- index_fa$total["raw_alpha"])
+writeLines(as.character(round(alpha,2)),con = "../results/minwage_outcomes_alpha_w2_mturk.tex",sep = "%")
+
+# FACTOR ANALYSIS WITH VARIMAX ROTATION (POST)
+pca2 <- psych::principal(select(w2, minwage15, rtwa_v1, rtwa_v2, mw_support, minwage_howhigh, mw_help, mw_restrict, minwage_text_r),
+                         rotate="varimax",
+                         nfactors=1
+)
+pc2 <- pca2$Vaccounted[2]
+writeLines(as.character(round(pc2, 2)),con = "../results/outcomes_pc2_study2_post.tex",sep = "%")
+
+##### media trust #####
+w2 <- w2 %>%
+  mutate( # higher = more trusting
+    trust_majornews = dplyr::recode(q96_1,"A lot"=3,"Some"=2,"Not too much"=1,"Not at all"=0)/3,
+    trust_localnews = dplyr::recode(q96_2,"A lot"=3,"Some"=2,"Not too much"=1,"Not at all"=0)/3,
+    trust_social = dplyr::recode(q96_3,"A lot"=3,"Some"=2,"Not too much"=1,"Not at all"=0)/3,
+    trust_youtube = dplyr::recode(q96_4,"A lot"=3,"Some"=2,"Not too much"=1,"Not at all"=0)/3,
+    fabricate_majornews = dplyr::recode(q98,"Never"=4,"Once in a while"=3,"About half the time"=2,"Most of the time"=1,"All the time"=0)/4,
+    fabricate_youtube = dplyr::recode(q100_1,"Never"=4,"Once in a while"=3,"About half the time"=2,"Most of the time"=1,"All the time"=0)/4
+  ) %>%
+  rowwise() %>%
+  mutate(media_trust = mean(trust_majornews,trust_localnews,fabricate_majornews,na.rm=T)) %>%
+  ungroup()
+
+##### affective polarization #####
+print('check affpol feeling thermometers:')
+w2 <- w2 %>%
+  mutate(
+    smart_dems = dplyr::recode(q61, "Extremely"=4,"Very"=3,"Somewhat"=2,"A little"=1,"Not at all"=0)/4,
+    smart_reps = dplyr::recode(q62_1, "Extremely"=4,"Very"=3,"Somewhat"=2,"A little"=1,"Not at all"=0)/4,
+    comfort_dems = dplyr::recode(q92,"Extremely comfortable"=3,"Somewhat comfortable"=2,"Not too comfortable"=1,"Not at all comfortable"=0)/3,
+    comfort_reps = dplyr::recode(q94,"Extremely comfortable"=3,"Somewhat comfortable"=2,"Not too comfortable"=1,"Not at all comfortable"=0)/3,
+    ft_dems = as.numeric(q90_11),
+    ft_reps = as.numeric(q90_12)
+  )
+
+write_csv(w2, "../results/intermediate data/minimum wage (issue 2)/qualtrics_w2_clean.csv")
+
+
+# join to W1 by MT worker ID:
+w12 <- left_join(a, filter(w2,!is.na(worker_id)), by = "worker_id",suffix=c("_w1","_w2"))
+names(w12)
+
+w12 <- w12 %>%
+  mutate(
+         affpol_smart_w2 = case_when(
+           pid==-1 ~ smart_dems_w2-smart_reps_w2,
+           pid==1 ~ smart_reps_w2-smart_dems_w2
+         ),
+         affpol_comfort_w2 = case_when(
+           pid==-1 ~ comfort_dems_w2-comfort_reps_w2,
+           pid==1 ~ comfort_reps_w2-comfort_dems_w2
+         ),
+         affpol_ft_w2 = case_when(
+           pid==-1 ~ ft_dems_w2-ft_reps_w2,
+           pid==1 ~ ft_reps_w2-ft_dems_w2
+         ))
+
+write_csv(w12, "../results/intermediate data/minimum wage (issue 2)/qualtrics_w12_clean.csv")
+
+
+## YTRecs session data: -------------------------------------------------------
+
+ytrecs <- read_rds("../data/minimum wage (issue 2)/min_wage_data.rds") %>%
+  clean_names() %>%
+  as_tibble()
+
+ytrecs <- ytrecs %>%
+  mutate(duration = end_time2 - start_time2) %>%
+  select(topic_id,urlid,pro,anti,duration,pro_up,pro_down,anti_up,anti_down,pro_save,anti_save,start_time2, end_time2) %>%
+  filter(str_detect(urlid,"mt_") & !is.na(pro))
+
+ytrecs <- ytrecs %>%
+  group_by(topic_id,urlid) %>%
+  mutate(dupes = n(),
+         max_duration = ifelse(duration==max(duration),1,0)
+  ) %>%
+  filter(max_duration==1) # using longest session as valid one
+
+ytrecs <- ytrecs %>%
+  mutate(
+    pro_up = replace_na(pro_up,0),
+    pro_down = replace_na(pro_down,0),
+    anti_up = replace_na(anti_up,0),
+    anti_down = replace_na(anti_down,0),
+    pro_save = replace_na(pro_save,0),
+    anti_save = replace_na(anti_save,0)) %>%
+  rowwise() %>%
+  mutate(total_likes = sum(pro_up,anti_up,na.rm=T),
+         total_dislikes = sum(pro_down,anti_down,na.rm=T),
+         total_thumbs = sum(pro_up,pro_down,anti_up,anti_down,na.rm=T),
+         total_saved = sum(pro_save,anti_save,na.rm=T),
+         total_interactions = sum(pro_up,pro_down,anti_up,anti_down,pro_save,anti_save,na.rm=T),
+         positive_interactions = total_likes + total_saved - total_dislikes
+  )
+
+ytrecs <- ytrecs %>%
+  mutate(seed = str_replace(topic_id,".*_(\\w+)$","\\1")) %>%
+  mutate(pro = as.numeric(pro),
+         anti = as.numeric(anti)) %>%
+  mutate(pro_fraction_chosen = case_when(
+    seed=="anti" ~ pro/(pro+anti-1),
+    seed=="pro" ~ (pro-1)/(pro+anti-1)
+  ))
+# adjust for zeros:
+ytrecs$pro_fraction_chosen[ytrecs$pro==0 & ytrecs$anti==0] <- NA
+
+
+w12 <- w12 %>%
+  ungroup() %>%
+  mutate(topic_id = str_replace(video_link_w2,".*&topicid=(.*)&allowDupe=1&id=(.*)$","\\1"),
+         urlid = str_replace(video_link_w2,".*&topicid=(.*?)&allowDupe=1&id=(.*)$","\\2"),
+  )
+
+w12 <- left_join(w12,ytrecs,by=c("topic_id","urlid"))
+
+w12 <- w12 %>%
+  arrange(worker_id, start_time2) %>%
+  group_by(worker_id) %>%
+  slice(1) %>% # Keep first resp
+  ungroup()
+print("ISSUE 2 NUMBERS:")
+print(paste('count w/ valid ytrecs data:', sum(!is.na(w12$pro))))
+print(paste('count w/ valid ytrecs interactions:', sum(!is.na(w12$total_thumbs))))
+print('interactions:')
+summary(w12$total_interactions)
+
+
+# create numeric dosage version of treatment:
+w12 <- w12 %>%
+  mutate(treatment_dose = dplyr::recode(treatment_arm,
+                                 "anti_31"= 1, "anti_22" = 0,
+                                 "pro_31"= 1, "pro_22" = 0,
+                                 "control"=NA_real_),
+         treatment_seed = str_replace(treatment_arm,"(.*)\\_\\d{2}","\\1")
+  )
+
+write_csv(w12, "../results/intermediate data/minimum wage (issue 2)/qualtrics_w12_clean.csv")

code/minimum wage (issue 2)/02b_clean_merge_yg.R

@@ -0,0 +1,436 @@
+cat(rep('=', 80),
+    '\n\n',
+    'OUTPUT FROM: minimum wage (issue 2)/02b_clean_merge_yg.R',
+    '\n\n',
+    sep = ''
+    )
+
+library(tidyverse)
+library(lubridate)
+library(stargazer)
+library(haven)
+library(janitor)
+
+yg <- read_sav("../data/minimum wage (issue 2)/PRIN0016_MERGED_OUTPUT.sav")
+
+## Recodes:
+yg <- yg %>% mutate(start_date = as_datetime(starttime),
+                    end_date = as_datetime(endtime),
+                    start_date_w2 = as_datetime(starttime_W2),
+                    end_date_w2 = as_datetime(endtime_W2),
+                    survey_time = as.numeric(end_date-start_date),
+                    survey_time_w2 = as.numeric(end_date_w2-start_date_w2),
+)
+
+print('wave 1 survey time')
+summary(yg$survey_time)
+
+print('wave 2 survey time')
+summary(yg$survey_time_w2)
+
+#### Demographics ####
+yg <- yg %>%
+  mutate(female = ifelse(gender4 == 2, 1, 0),
+         male = ifelse(gender4 == 1, 1, 0),
+         black = ifelse(race == 2, 1, 0),
+         white = ifelse(race == 1, 1, 0),
+         college = ifelse(educ == 5 | educ == 6, 1, 0),
+         income_gt50k = ifelse(faminc_new >= 6 & faminc_new <= 16, 1, 0)
+  )
+
+# PID:
+yg <- yg %>%
+  mutate(pid = case_when(pid3==1 ~ -1,
+                         pid3==2 ~ 1,
+                         pid7>4 & pid7<8 ~ 1,
+                         pid7<4 ~ -1,
+                         pid7==4 ~ 0))
+
+yg <- yg %>%
+  mutate(ideo = case_when(ideo5<3 ~ -1,
+                          ideo5>3 & ideo5<6 ~ 1,
+                          ideo5==3 ~ 0))
+
+yg$age <- 2022 - yg$birthyr
+
+# age categories: 18-29;  30-44;  45-64;  65+
+yg <- yg %>%
+  mutate(age_cat = case_when(age>=18 & age<=29 ~ "18-29",
+                             age>=30 & age<=44 ~ "30-44",
+                             age>=45 & age<=64 ~ "45-64",
+                             age>=65 ~ "65+"
+  ))
+
+yg <- yg %>%
+  fastDummies::dummy_cols(select_columns = "age_cat",remove_selected_columns = F)
+
+yg <- yg %>%
+  mutate(pol_interest = ifelse(newsint>4,NA_real_,newsint),
+         pol_interest = (4-pol_interest)/3,
+         youtube_freq_v2 = ifelse(youtube_freq>10,NA_real_,youtube_freq),
+         freq_youtube_v2 = 10-youtube_freq_v2,
+         freq_youtube = (Q77-1)
+  )
+
+
+# Descriptives ------------------------------------------------------------
+
+descr_data <- as.data.frame(select(yg,
+                     female,
+                     white,
+                     black,
+                     age,
+                     college,
+                     income_gt50k))
+descr_data <- descr_data %>% filter(rowSums(is.na(.)) != ncol(.))
+descriptive_tab <- stargazer(descr_data,
+                             summary = T, digits=2,
+                             summary.stat=c("mean","sd","median","min","max","n"),
+                             covariate.labels = c("Female",
+                                                  "White",
+                                                  "Black",
+                                                  "Age",
+                                                  "College educated",
+                                                  "Income \\textgreater 50k"),
+                             float = F,
+                             out = "../results/minwage_descriptive_tab_yg.tex")
+
+summary_tab <- yg %>%
+  dplyr::summarize(female = mean(female,na.rm=T),
+                   white = mean(white,na.rm=T),
+                   black = mean(black,na.rm=T),
+                   age1829 = mean(`age_cat_18-29`,na.rm=T),
+                   age3044 = mean(`age_cat_30-44`,na.rm=T),
+                   age4564 = mean(`age_cat_45-64`,na.rm=T),
+                   age65p = mean(`age_cat_65+`,na.rm=T),
+                   college = mean(college,na.rm=T),
+                   income_gt50k = mean(income_gt50k,na.rm=T),
+                   democrat = mean(pid==-1,na.rm=T),
+                   republican = mean(pid==1,na.rm=T))
+
+summary_tab <- pivot_longer(summary_tab,
+                            cols=c(female,
+                                   white,
+                                   black,
+                                   age1829,
+                                   age3044,
+                                   age4564,
+                                   age65p,
+                                   college,
+                                   income_gt50k,
+                                   democrat,
+                                   republican),
+                            names_to = "outcome",values_to = "survey_avg")
+outcome_labels <- data.frame(outcome_pretty = c("Female",
+                                                "White",
+                                                "Black",
+                                                "Age 18-29",
+                                                "Age 30-44",
+                                                "Age 45-64",
+                                                "Age 65+",
+                                                "College educated",
+                                                "Income >$50k",
+                                                "Democrat",
+                                                "Republican"),
+                             outcome = c("female",
+                                         "white",
+                                         "black",
+                                         "age1829",
+                                         "age3044",
+                                         "age4564",
+                                         "age65p",
+                                         "college",
+                                         "income_gt50k",
+                                         "democrat",
+                                         "republican"))
+summary_tab$outcome_pretty <-  outcome_labels$outcome_pretty[match(summary_tab$outcome,outcome_labels$outcome)]
+summary_tab <- summary_tab %>%
+  mutate(outcome_pretty = factor(outcome_pretty,levels = c("Republican",
+                                                           "Democrat",
+                                                           "Income >$50k",
+                                                           "College educated",
+                                                           "Age 65+",
+                                                           "Age 45-64",
+                                                           "Age 30-44",
+                                                           "Age 18-29",
+                                                           "Female",
+                                                           "Black",
+                                                           "White"
+  ),ordered=T))
+
+(descrip_fig <- ggplot(summary_tab) +
+    geom_point(aes(y=outcome_pretty,x=survey_avg)) +
+    geom_text(aes(y=outcome_pretty,x=survey_avg,label=paste0(round(100*survey_avg,0),"%")),nudge_x = 0.1) +
+    scale_y_discrete("") +
+    scale_x_continuous("",labels=scales::percent_format(),limits=c(0,1)) +
+    theme_bw()
+)
+
+ggsave(descrip_fig,filename = "../results/minwage_demographics_yg.pdf",height=5,width=4)
+
+
+
+#### A/V check
+print('audio ok:')
+length(which(yg$Q81_W2 == 1))/length(which(!is.na(yg$Q81_W2)))
+print('video ok:')
+length(which(yg$Q82_W2 == 1))/length(which(!is.na(yg$Q82_W2)))
+
+
+
+#### Outcomes ####
+
+##### policy opinions #####
+# convert to numeric unit scale:
+yg <- yg %>%
+  mutate( # higher = more conservative or anti-min wage
+    minwage15_w1 = (minwage15-1)/4,
+    rtwa_v1_w1 = (RTWA_v1-1)/4,
+    rtwa_v2_w1 = (RTWA_v2-1)/4,
+    mw_support_w1 = (mw_support-1)/4,
+    minwage_howhigh_w1 = (minwage_howhigh-1)/4,
+    mw_help_w1 = (mw_help_a-1)/9,
+    mw_restrict_w1 = (10-mw_restrict_1)/9,
+    minwage_text_r_w1 = (25-as.numeric(minwage_text))/25,
+  )
+
+yg <- yg %>%
+  rowwise() %>%
+  mutate(mw_index_w1 = mean(c(minwage15_w1, rtwa_v1_w1, rtwa_v2_w1, mw_support_w1, minwage_howhigh_w1, mw_help_w1, mw_restrict_w1, minwage_text_r_w1), na.rm=T)) %>%
+  ungroup()
+
+# Cronbach's alpha
+index_fa <- psych::alpha(select(yg, minwage15_w1, rtwa_v1_w1, rtwa_v2_w1, mw_support_w1, minwage_howhigh_w1, mw_help_w1, mw_restrict_w1, minwage_text_r_w1), check.keys = TRUE)
+write.csv(data.frame(cor(select(yg, minwage15_w1, rtwa_v1_w1, rtwa_v2_w1, mw_support_w1, minwage_howhigh_w1, mw_help_w1, mw_restrict_w1, minwage_text_r_w1), use = "complete.obs")),row.names = T,
+          file = "../results/cormat_mwindex_w1_yg.csv")
+
+pdf("../results/corrplot_mwindex_w1_yg.pdf")
+w1_corrplot <- corrplot::corrplot(cor(select(yg, minwage15_w1, rtwa_v1_w1, rtwa_v2_w1, mw_support_w1, minwage_howhigh_w1, mw_help_w1, mw_restrict_w1, minwage_text_r_w1), use = "complete.obs"),method = "shade")
+dev.off()
+
+alpha <- index_fa$total["raw_alpha"]
+writeLines(as.character(round(alpha,2)),con = "../results/minwage_outcomes_alpha_w1_yg.tex",sep = "%")
+
+# FACTOR ANALYSIS WITH VARIMAX ROTATION (PRE)
+pca2 <- psych::principal(select(yg, minwage15_w1, rtwa_v1_w1, rtwa_v2_w1, mw_support_w1, minwage_howhigh_w1, mw_help_w1, mw_restrict_w1, minwage_text_r_w1),
+                         rotate="varimax",
+                         nfactors=1
+)
+pc2 <- pca2$Vaccounted[2]
+writeLines(as.character(round(pc2, 2)),con = "../results/outcomes_pc2_study3_pre.tex",sep = "%")
+
+##### media trust #####
+yg <- yg %>%
+  mutate( # higher = more trusting
+    trust_majornews_w1 = (4-Q58_a)/3,
+    trust_localnews_w1 = (4-Q58_b)/3,
+    trust_social_w1 = (4-Q58_c)/3,
+    trust_youtube_w1 = (4-Q58_d)/3,
+    fabricate_majornews_w1 = (5-Q89b)/4,
+    fabricate_youtube_w1 = (5-Q90)/4
+  ) %>%
+  rowwise() %>%
+  mutate(media_trust_w1 = mean(trust_majornews_w1,trust_localnews_w1,fabricate_majornews_w1,na.rm=T)) %>%
+  ungroup()
+
+media_trust_fa <- psych::alpha(select(yg, trust_majornews_w1,trust_localnews_w1,fabricate_majornews_w1), check.keys = TRUE)
+print('media trust alpha:')
+media_trust_fa$total["raw_alpha"]
+
+
+##### affective polarization #####
+# check FTs:
+yg %>%
+  group_by(pid) %>%
+  summarize(mean_2=mean(as.numeric(Q5_a),na.rm=T), # Trump
+            mean_5=mean(as.numeric(Q5_b),na.rm=T), # Biden
+            mean_11=mean(as.numeric(Q5_c),na.rm=T), # dems
+            mean_12=mean(as.numeric(Q5_d),na.rm=T)) # reps
+
+yg <- yg %>%
+  mutate(
+    smart_dems = (5-Q61)/4,
+    smart_reps = (5-Q62)/4,
+    comfort_dems = (Q87b-1)/3,
+    comfort_reps = (Q88-1)/3,
+    ft_dems = as.numeric(Q5_c),
+    ft_reps = as.numeric(Q5_d),
+    affpol_smart = case_when(
+      pid==-1 ~ smart_dems-smart_reps,
+      pid==1 ~ smart_reps-smart_dems
+    ),
+    affpol_comfort = case_when(
+      pid==-1 ~ comfort_dems-comfort_reps,
+      pid==1 ~ comfort_reps-comfort_dems
+    ),
+    affpol_ft = case_when(
+      pid==-1 ~ ft_dems-ft_reps,
+      pid==1 ~ ft_reps-ft_dems
+    )
+  )
+
+
+
+# W2 ----------------------------------------------------------------------
+
+##### policy opinions #####
+# convert to numeric unit scale:
+yg <- yg %>%
+  mutate( # higher = more conservative or anti-min wage
+    minwage15_w2 = (minwage15_W2-1)/4,
+    rtwa_v1_w2 = (RTWA_v1_W2-1)/4,
+    rtwa_v2_w2 = (RTWA_v2_W2-1)/4,
+    mw_support_w2 = (mw_support_W2-1)/4,
+    minwage_howhigh_w2 = (minwage_howhigh_W2-1)/4,
+    mw_help_w2 = (mw_help_a_W2-1)/9,
+    mw_restrict_w2 = (10-mw_restrict_1_W2)/9,
+    minwage_text_r_w2 = (25-as.numeric(minwage_text_W2))/25,
+  )
+
+yg <- yg %>%
+  rowwise() %>%
+  mutate(mw_index_w2 = mean(c(minwage15_w2, rtwa_v1_w2, rtwa_v2_w2, mw_support_w2, minwage_howhigh_w2, mw_help_w2, mw_restrict_w2, minwage_text_r_w2), na.rm=T)) %>%
+  ungroup()
+
+# Cronbach's alpha
+index_fa <- psych::alpha(select(yg, minwage15_w2, rtwa_v1_w2, rtwa_v2_w2, mw_support_w2, minwage_howhigh_w2, mw_help_w2, mw_restrict_w2, minwage_text_r_w2), check.keys = TRUE)
+write.csv(data.frame(cor(select(yg, minwage15_w2, rtwa_v1_w2, rtwa_v2_w2, mw_support_w2, minwage_howhigh_w2, mw_help_w2, mw_restrict_w2, minwage_text_r_w2), use = "complete.obs")),row.names = T,
+          file = "../results/cormat_mwindex_w2_yg.csv")
+
+pdf("../results/corrplot_mwindex_w2_yg.pdf")
+w2_corrplot <- corrplot::corrplot(cor(select(yg, minwage15_w2, rtwa_v1_w2, rtwa_v2_w2, mw_support_w2, minwage_howhigh_w2, mw_help_w2, mw_restrict_w2, minwage_text_r_w2), use = "complete.obs"),method = "shade")
+dev.off()
+
+print('wave 2 policy opinion alpha:')
+(alpha <- index_fa$total["raw_alpha"])
+writeLines(as.character(round(alpha,2)),con = "../results/minwage_outcomes_alpha_w2_mturk.tex",sep = "%")
+
+# FACTOR ANALYSIS WITH VARIMAX ROTATION (POST)
+pca2 <- psych::principal(select(yg, minwage15_w2, rtwa_v1_w2, rtwa_v2_w2, mw_support_w2, minwage_howhigh_w2, mw_help_w2, mw_restrict_w2, minwage_text_r_w2),
+                         rotate="varimax",
+                         nfactors=1
+)
+pc2 <- pca2$Vaccounted[2]
+writeLines(as.character(round(pc2, 2)),con = "../results/outcomes_pc2_study3_post.tex",sep = "%")
+
+##### media trust #####
+yg <- yg %>%
+  mutate( # higher = more trusting
+    trust_majornews_w2 = (4-Q58_a_W2)/3,
+    trust_localnews_w2 = (4-Q58_b_W2)/3,
+    trust_social_w2 = (4-Q58_c_W2)/3,
+    trust_youtube_w2 = (4-Q58_d_W2)/3,
+    fabricate_majornews_w2 = (5-Q89b_W2)/4,
+    fabricate_youtube_w2 = (5-Q90_W2)/4
+  ) %>%
+  rowwise() %>%
+  mutate(media_trust_w2 = mean(c(trust_majornews_w2,trust_localnews_w2,fabricate_majornews_w2),na.rm=T)) %>%
+  ungroup()
+
+##### affective polarization #####
+print('check affpol feeling thermometers:')
+yg %>%
+  group_by(pid) %>%
+  summarize(mean_2=mean(as.numeric(Q5_a_W2),na.rm=T), # Trump
+            mean_5=mean(as.numeric(Q5_b_W2),na.rm=T), # Biden
+            mean_11=mean(as.numeric(Q5_c_W2),na.rm=T), # dems
+            mean_12=mean(as.numeric(Q5_d_W2),na.rm=T)) # reps
+
+yg <- yg %>%
+  mutate( # higher = more trusting
+    smart_dems_w2 = (5-Q61_W2)/4,
+    smart_reps_w2 = (5-Q62_W2)/4,
+    comfort_dems_w2 = (Q92_W2-1)/3,
+    comfort_reps_w2 = (Q94_W2-1)/3,
+    ft_dems_w2 = as.numeric(Q5_c_W2),
+    ft_reps_w2 = as.numeric(Q5_d_W2),
+    affpol_smart_w2 = case_when(
+      pid==-1 ~ smart_dems_w2-smart_reps_w2,
+      pid==1 ~ smart_reps_w2-smart_dems_w2
+    ),
+    affpol_comfort_w2 = case_when(
+      pid==-1 ~ comfort_dems_w2-comfort_reps_w2,
+      pid==1 ~ comfort_reps_w2-comfort_dems_w2
+    ),
+    affpol_ft_w2 = case_when(
+      pid==-1 ~ ft_dems_w2-ft_reps_w2,
+      pid==1 ~ ft_reps_w2-ft_dems_w2
+    )
+  )
+
+
+## YTRecs session data: -------------------------------------------------------
+
+ytrecs <- read_rds("../data/minimum wage (issue 2)/min_wage_data.rds") %>%
+  clean_names() %>%
+  as_tibble()
+
+ytrecs <- ytrecs %>%
+  mutate(duration = end_time2 - start_time2) %>%
+  select(topic_id,urlid,pro,anti,duration,pro_up,pro_down,anti_up,anti_down,pro_save,anti_save,start_time2, end_time2) %>%
+  filter(str_detect(urlid,"mt_",negate = T) & !is.na(pro))
+
+ytrecs <- ytrecs %>%
+  group_by(topic_id,urlid) %>%
+  mutate(dupes = n(),
+         max_duration = ifelse(duration==max(duration),1,0)
+  ) %>%
+  filter(max_duration==1) # using longest session as valid one
+
+ytrecs <- ytrecs %>%
+  mutate(
+    pro_up = replace_na(pro_up,0),
+    pro_down = replace_na(pro_down,0),
+    anti_up = replace_na(anti_up,0),
+    anti_down = replace_na(anti_down,0),
+    pro_save = replace_na(pro_save,0),
+    anti_save = replace_na(anti_save,0)) %>%
+  rowwise() %>%
+  mutate(total_likes = sum(pro_up,anti_up,na.rm=T),
+         total_dislikes = sum(pro_down,anti_down,na.rm=T),
+         total_thumbs = sum(pro_up,pro_down,anti_up,anti_down,na.rm=T),
+         total_saved = sum(pro_save,anti_save,na.rm=T),
+         total_interactions = sum(pro_up,pro_down,anti_up,anti_down,pro_save,anti_save,na.rm=T),
+         positive_interactions = total_likes + total_saved - total_dislikes
+  )
+
+ytrecs <- ytrecs %>%
+  mutate(seed = str_replace(topic_id,".*_(\\w+)$","\\1")) %>%
+  mutate(pro = as.numeric(pro),
+         anti = as.numeric(anti)) %>%
+  mutate(pro_fraction_chosen = case_when(
+    seed=="anti" ~ pro/(pro+anti-1),
+    seed=="pro" ~ (pro-1)/(pro+anti-1)
+  ))
+# adjust for zeros:
+ytrecs$pro_fraction_chosen[ytrecs$pro==0 & ytrecs$anti==0] <- NA
+
+
+yg <- yg %>%
+  ungroup() %>%
+  mutate(
+    urlid = session_visa_W2
+  )
+
+yg <- left_join(yg,ytrecs,by=c("urlid"))
+
+print("ISSUE 2 NUMBERS:")
+print(paste('count w/ valid ytrecs data:', sum(!is.na(yg$pro))))
+print(paste('count w/ valid ytrecs interactions:', sum(!is.na(yg$total_thumbs))))
+print('interactions:')
+summary(yg$total_interactions)
+
+# create numeric dosage version of treatment:
+yg <- yg %>%
+  mutate(treatment_arm = haven::as_factor(treatment_arm_W2),
+         treatment_dose = dplyr::recode(treatment_arm,
+                                 "anti_31"= 1, "anti_22" = 0,
+                                 "pro_31"= 1, "pro_22" = 0,
+                                 "control"=NA_real_),
+         treatment_seed = str_replace(treatment_arm,"(.*)\\_\\d{2}","\\1")
+  )
+
+terciles <- read_csv("../results/intermediate data/minimum wage (issue 2)/yougov_terciles.csv")
+yg <- left_join(yg,select(terciles,caseid,thirds=tercile),by="caseid")
+
+write_csv(yg, "../results/intermediate data/minimum wage (issue 2)/yg_w12_clean.csv")

code/minimum wage (issue 2)/03_analysis_multipletesting.R

@@ -0,0 +1,1293 @@
+cat(rep('=', 80),
+    '\n\n',
+    'OUTPUT FROM: minimum wage (issue 2)/03_analysis_multipletesting.R',
+    '\n\n',
+    sep = ''
+    )
+
+library(data.table)
+library(car)
+library(sandwich)
+library(lmtest)
+library(ggplot2)
+library(assertthat)
+library(foreach)
+library(doParallel)
+registerDoParallel(cores = detectCores() - 1)
+
+
+
+###############
+## functions ##
+###############
+
+`%.%` <- paste0
+
+simes <- function(ps){
+  min(sort(length(ps) * ps / rank(ps)))
+}
+
+### functions to handle inconsistent interaction ordering of mlm() ###
+
+## convert interaction terms of form 'b#:a#' to 'a#:b#'
+reorder.interaction.names <- function(x, prefix = ''){
+  x <- gsub('^' %.% prefix, '', x)
+  sapply(strsplit(x, ':'),
+         function(y){
+           paste(sort(y), collapse = ':')
+         })
+}
+
+## take term of form 'a1:b1', look up in vector of form 'b#:a#, return 'b1:a1'
+convert.interaction.names <- function(x, y, prefix.y = ''){
+  ind <- match(reorder.interaction.names(x),
+               reorder.interaction.names(y, prefix = prefix.y)
+               )
+  return(y[ind])
+}
+
+## modified from print.linearHypothesis.mlm to use alternate df & return pvals
+##   (print method is responsible for doing the actual computation of pvals)
+extract.lht <- function(x,
+                        SSP = TRUE,
+                        SSPE = SSP,
+                        digits = getOption('digits'),
+                        df.residual = x$df.residual
+                        ){
+  test <- x$test
+  if (!is.null(x$P) && SSP) {
+    P <- x$P
+    cat("\n Response transformation matrix:\n")
+    attr(P, "assign") <- NULL
+    attr(P, "contrasts") <- NULL
+    print(P, digits = digits)
+  }
+  if (SSP) {
+    cat("\nSum of squares and products for the hypothesis:\n")
+    print(x$SSPH, digits = digits)
+  }
+  if (SSPE) {
+    cat("\nSum of squares and products for error:\n")
+    print(x$SSPE, digits = digits)
+  }
+  if ((!is.null(x$singular)) && x$singular) {
+    warning("the error SSP matrix is singular; multivariate tests are unavailable")
+    return(invisible(x))
+  }
+  SSPE.qr <- qr(x$SSPE)
+  eigs <- Re(eigen(qr.coef(SSPE.qr, x$SSPH), symmetric = FALSE)$values)
+  tests <- matrix(NA, 4, 4)
+  rownames(tests) <- c("Pillai", "Wilks", "Hotelling-Lawley",
+                       "Roy")
+  if ("Pillai" %in% test)
+    tests[1, 1:4] <- car:::Pillai(eigs, x$df, df.residual)
+  if ("Wilks" %in% test)
+    tests[2, 1:4] <- car:::Wilks(eigs, x$df, df.residual)
+  if ("Hotelling-Lawley" %in% test)
+    tests[3, 1:4] <- car:::HL(eigs, x$df, df.residual)
+  if ("Roy" %in% test)
+    tests[4, 1:4] <- car:::Roy(eigs, x$df, df.residual)
+  tests <- na.omit(tests)
+  ok <- tests[, 2] >= 0 & tests[, 3] > 0 & tests[, 4] > 0
+  ok <- !is.na(ok) & ok
+  tests <- cbind(x$df, tests, pf(tests[ok, 2], tests[ok, 3],
+                                 tests[ok, 4], lower.tail = FALSE))
+  colnames(tests) <- c("Df", "test stat", "approx F", "num Df",
+                       "den Df", "Pr(>F)")
+  tests <- structure(as.data.frame(tests),
+                     heading = paste("\nMultivariate Test",
+                                     if (nrow(tests) > 1)
+                                       "s", ": ", x$title, sep = ""),
+                     class = c("anova",
+                               "data.frame"
+                               )
+                     )
+  return(tests)
+}
+
+
+
+###############
+## load data ##
+###############
+
+d <- fread('../results/intermediate data/minimum wage (issue 2)/qualtrics_w12_clean.csv')
+
+## drop pure control
+d <- d[treatment_arm != 'control',]
+
+## drop NA video counts
+d <- d[!is.na(pro) & !is.na(anti),]
+
+
+
+##############
+## controls ##
+##############
+
+platform.controls <- c('age_cat',
+                       'male',
+                       'pol_interest',
+                       'freq_youtube'
+                       )
+
+mwpolicy.controls <- 'mw_index_w1'
+
+media.controls <- c('trust_majornews_w1',
+                    'trust_youtube_w1',
+                    'fabricate_majornews_w1',
+                    'fabricate_youtube_w1'
+                    )
+
+affpol.controls <- c('affpol_ft',
+                     'affpol_smart',
+                     'affpol_comfort'
+                     )
+
+controls.raw <- unique(c(platform.controls,
+                         mwpolicy.controls,
+                         media.controls,
+                         affpol.controls
+                         )
+                       )
+
+## transform control variables by creating dummies and demeaning
+controls.trans <- list()
+for (j in controls.raw){
+  ## convert to dummies if needed
+  controls.j <- model.matrix(as.formula('~ 0 + ' %.% j),
+                             model.frame(as.formula('~ 0 + ' %.% j),
+                                         data = d,
+                                         na.action = 'na.pass'
+                                         )
+                             )
+  ## demean by column
+  controls.j <- sweep(controls.j,
+                      MARGIN = 2,
+                      STATS = colMeans(controls.j, na.rm = TRUE),
+                      FUN = `-`,
+                      )
+  colnames(controls.j) <- make.names(colnames(controls.j))
+  ## remove control from original data
+  d[[j]] <- NULL
+  ## reinsert transformed control
+  d <- cbind(d, controls.j)
+  ## keep track of which original controls map to which transformed controls
+  controls.trans[[j]] <- colnames(controls.j)
+}
+
+## map original control variables to transformed versions
+platform.controls <- unlist(controls.trans[platform.controls])
+mwpolicy.controls <- unlist(controls.trans[mwpolicy.controls])
+media.controls <- unlist(controls.trans[media.controls])
+affpol.controls <- unlist(controls.trans[affpol.controls])
+
+
+
+##############
+## outcomes ##
+##############
+
+### hypothesis family 1: platform interactions ###
+
+## platform interaction time: compute windorized usage time
+warning('diverges from pap, 95% windsorized due to extreme outliers')
+d[, platform_duration := duration]
+d[platform_duration <= quantile(d$duration, .025),
+  platform_duration := quantile(d$duration, .025)
+  ]
+d[platform_duration >= quantile(d$duration, .975),
+  platform_duration := quantile(d$duration, .975)
+  ]
+## all platform interaction outcomes
+platform.outcomes <- c('pro_fraction_chosen',
+                       'positive_interactions',  # positive - negative (dislike)
+                       'platform_duration'
+                       )
+
+
+
+### hypothesis family 2: MW policy attitudes ###
+
+## only one preregistered outcome in this family
+mwpolicy.outcomes <- 'mw_index_w2'
+## added 4 jun 2024 at request of reviewers
+mwpolicy.outcomes.understanding <- c('mw_restrict_w2',
+                                     'mw_help_w2'
+                                     )
+
+
+
+### hypothesis family 3: media trust ###
+media.outcomes <- c('trust_majornews_w2',
+                    'trust_youtube_w2',
+                    'fabricate_majornews_w2',
+                    'fabricate_youtube_w2'
+                    )
+
+
+
+### hypothesis family 4: affective polarization ###
+affpol.outcomes <- c('affpol_ft_w2',
+                     'affpol_smart_w2',
+                     'affpol_comfort_w2'
+                     )
+
+outcomes <- unique(c(
+  platform.outcomes,
+  mwpolicy.outcomes,
+  media.outcomes,
+  affpol.outcomes
+)
+)
+
+
+
+################
+## treatments ##
+################
+
+## create attitude dummies
+## (pro/anti stance on issue has opposite lib/con meaning from study 1)
+d[, attitude := c('pro', 'neutral', 'anti')[thirds]]
+d[, attitude.anti := as.numeric(attitude == 'anti')]
+d[, attitude.neutral := as.numeric(attitude == 'neutral')]
+d[, attitude.pro := as.numeric(attitude == 'pro')]
+
+## create seed dummies
+d[, seed.anti := as.numeric(treatment_seed == 'anti')]
+d[, seed.pro := as.numeric(treatment_seed == 'pro')]
+
+## create recsys dummies
+d[, recsys.22 := as.numeric(treatment_arm %like% '22')]
+d[, recsys.31 := as.numeric(treatment_arm %like% '31')]
+
+## manually define coefficients to estimate
+treatments <- c('attitude.anti:recsys.22',
+                'attitude.anti:recsys.31',
+                'attitude.neutral:seed.anti:recsys.22',
+                'attitude.neutral:seed.pro:recsys.22',
+                'attitude.neutral:seed.anti:recsys.31',
+                'attitude.neutral:seed.pro:recsys.31',
+                'attitude.pro:recsys.22',
+                'attitude.pro:recsys.31'
+                )
+
+contrasts <- rbind(
+  i = c(treat = 'attitude.pro:recsys.31',
+        ctrl = 'attitude.pro:recsys.22'
+        ),
+  ii = c(treat = 'attitude.anti:recsys.31',
+         ctrl = 'attitude.anti:recsys.22'
+         ),
+  iii = c(treat = 'attitude.neutral:seed.pro:recsys.31',
+          ctrl = 'attitude.neutral:seed.pro:recsys.22'
+          ),
+  iv = c(treat = 'attitude.neutral:seed.anti:recsys.31',
+         ctrl = 'attitude.neutral:seed.anti:recsys.22'
+         ),
+  # in (v-vi), pro/anti order is reversed from study 1 to ensure that
+  # - 1st condition (treatment) is always the conservative video
+  # - 2nd condition (control) is always the liberal video
+  v = c(treat = 'attitude.neutral:seed.anti:recsys.31',
+        ctrl = 'attitude.neutral:seed.pro:recsys.31'
+  ),
+  vi = c(treat = 'attitude.neutral:seed.anti:recsys.22',
+         ctrl = 'attitude.neutral:seed.pro:recsys.22'
+  )
+)
+
+## check that contrasts are valid
+assert_that(all(unlist(contrasts) %in% treatments))
+
+## check that specifications are equivalent
+coefs.v1 <- coef(lm(mw_index_w2 ~ 0 + attitude:treatment_arm, d))
+coefs.v2 <- coef(
+  lm(mw_index_w2 ~
+       0 +
+       attitude.anti:recsys.22 +
+       attitude.anti:recsys.31 +
+       attitude.neutral:seed.anti:recsys.22 +
+       attitude.neutral:seed.pro:recsys.22 +
+       attitude.neutral:seed.anti:recsys.31 +
+       attitude.neutral:seed.pro:recsys.31 +
+       attitude.pro:recsys.22 +
+       attitude.pro:recsys.31,
+     d
+     )
+)
+assert_that(all.equal(unname(sort(coefs.v1)), unname(sort(coefs.v2))))
+
+##########################
+## hierarchical testing ##
+##########################
+
+## initialize top layer p-values:
+##   does treatment have any effect on any outcome in family
+families <- c(
+  'platform',
+  'mwpolicy',
+  'media',
+  'affpol'
+)
+layer1.pvals <- rep(NA_real_, length(families))
+layer1.notes <- rep('', length(families))
+names(layer1.pvals) <- families
+
+## initialize 2nd layer p-values:
+##   which treatment has detectable effect?
+contrast.pvals <- rep(NA_real_, nrow(contrasts))
+names(contrast.pvals) <- paste(contrasts[, 'treat'],
+                               contrasts[, 'ctrl'],
+                               sep = '.vs.'
+                               )
+layer2.pvals <- list(
+  platform = contrast.pvals,
+  mwpolicy = contrast.pvals,
+  media = contrast.pvals,
+  affpol = contrast.pvals
+)
+rm(contrast.pvals)
+
+## initialize 3rd layer p-values:
+##   on which specific outcome in family?
+layer3.pvals <- list()
+layer3.ests <- list()
+layer3.ses <- list()
+layer3.notes <- list()
+for (i in 1:length(families)){
+  family <- families[i]
+  layer3.pvals[[family]] <- list()
+  layer3.ests[[family]] <- list()
+  layer3.ses[[family]] <- list()
+  layer3.notes[[family]] <- list()
+  outcomes <- get(family %.% '.outcomes')
+  for (j in 1:nrow(contrasts)){
+    contrast <- paste(contrasts[j, 'treat'],
+                      contrasts[j, 'ctrl'],
+                      sep = '.vs.'
+                      )
+    layer3.pvals[[family]][[contrast]] <- numeric(0)
+    layer3.ests[[family]][[contrast]] <- numeric(0)
+    layer3.ses[[family]][[contrast]] <- numeric(0)
+    for (k in 1:length(outcomes)){
+      outcome <- outcomes[k]
+      layer3.pvals[[family]][[contrast]][outcome] <- NA_real_
+      layer3.ests[[family]][[contrast]][outcome] <- NA_real_
+      layer3.ses[[family]][[contrast]][outcome] <- NA_real_
+      layer3.notes[[family]][outcome] <- ''
+    }
+  }
+}
+
+
+
+### begin nested analyses ###
+
+for (i in 1:length(families)){
+
+  family <- families[i]
+  family.outcomes <- get(family %.% '.outcomes')
+  family.controls <- get(family %.% '.controls')
+  family.controls.interactions <- as.character(
+    outer(treatments,
+          family.controls,
+          FUN = function(x, y) x %.% ':' %.% y
+          )
+  )
+
+  family.formula <-
+    'cbind(' %.%                # outcomes
+    paste(family.outcomes,
+          collapse = ', '
+          ) %.%  ') ~\n0 +\n' %.%
+    paste(treatments,           # treatments (base terms)
+          collapse = ' +\n'
+          ) %.% ' +\n' %.%
+    paste(family.controls,      # controls (base terms)
+          collapse = ' +\n'
+          )
+
+  cat(rep('=', 80),
+      '\n\nHYPOTHESIS FAMILY: ',
+      family,
+      '\n\nrunning mlm:\n\n',
+      family.formula,
+      '\n\n',
+      sep = ''
+      )
+
+  ## run model
+  family.mod <- lm(family.formula, d)
+  ## hack to eliminate NA coefs
+  if (any(is.na(coef(family.mod)))){
+    if ('mlm' %in% class(family.mod)){
+      drop <- rownames(coef(family.mod))[is.na(coef(family.mod))[, 1]]
+    } else {
+      drop <- names(coef(family.mod))[is.na(coef(family.mod))]
+    }
+    drop <- convert.interaction.names(drop,
+                                      c(family.controls,
+                                        family.controls.interactions
+                                        )
+                                      )
+    layer1.notes[[i]] <-
+      layer1.notes[[i]] %.%
+      'dropped the following coefs: ' %.%
+      paste(drop, sep = ', ') %.%
+      '\n\n'
+    family.formula <- gsub(
+      '\\s+\\+\\s+(' %.% paste(drop, collapse = '|') %.% ')',
+      '',
+      family.formula
+    )
+    family.mod <- lm(family.formula, d)
+  }
+
+  family.vcov <- vcovHC(family.mod)
+  if (is.null(dim(coef(family.mod)))){
+    coef.names <- names(coef(family.mod))
+  } else {
+    coef.names <- rownames(coef(family.mod))
+  }
+
+  ### top layer: test overall significance of all contrasts on all outcomes ###
+  ## convert interaction terms to whatever mlm() named it
+  treats <- convert.interaction.names(contrasts[, 'treat'], coef.names)
+  ctrls <- convert.interaction.names(contrasts[, 'ctrl'], coef.names)
+  ## test jointly
+  lht.attempt <- tryCatch({
+    if ('mlm' %in% class(family.mod)){
+      contrast.lht <- linearHypothesis(
+        family.mod,
+        vcov. = family.vcov,
+        hypothesis.matrix = sprintf('%s - %s', treats, ctrls),
+        rhs = matrix(0, nrow = nrow(contrasts), ncol = length(family.outcomes)),
+        test = 'Pillai'
+      )
+      layer1.pvals[[i]] <- extract.lht(contrast.lht)[, 'Pr(>F)']
+    } else {
+      contrast.lht <- linearHypothesis(
+        family.mod,
+        vcov. = family.vcov,
+        hypothesis.matrix = sprintf('%s - %s', treats, ctrls),
+        rhs = matrix(0, nrow = nrow(contrasts), ncol = length(family.outcomes)),
+        test = 'F'
+      )
+      layer1.pvals[[i]] <- contrast.lht[['Pr(>F)']][2]
+    }
+  },
+  error = function(e){
+    warning(sprintf('caught error in %s family:', family), e)
+    ## return error as string for inclusion in notes
+    'caught error: ' %.%
+      e %.%
+      '\n\n'
+  })
+  if (lht.attempt %like% 'caught error'){
+    layer1.notes[[i]] <-
+      layer1.notes[[i]] %.% lht.attempt
+  }
+
+
+
+  ### layer 2: test each contrast individually on all outcomes ###
+
+  for (j in 1:nrow(contrasts)){
+    ## test group equality on all outcomes
+    if ('mlm' %in% class(family.mod)){
+      contrast.lht <-
+        linearHypothesis(
+        family.mod,
+        vcov. = family.vcov,
+        hypothesis.matrix = sprintf('%s - %s', treats[j], ctrls[j]),
+        rhs = matrix(0, nrow = 1, ncol = length(family.outcomes)),
+        test = 'Pillai'
+      )
+      layer2.pvals[[i]][j] <- extract.lht(contrast.lht)[, 'Pr(>F)']
+    } else {
+      contrast.lht <- linearHypothesis(
+        family.mod,
+        vcov. = family.vcov,
+        hypothesis.matrix = sprintf('%s - %s', treats[j], ctrls[j]),
+        rhs = matrix(0, nrow = 1, ncol = length(family.outcomes)),
+        test = 'F'
+      )
+      layer2.pvals[[i]][j] <- contrast.lht[['Pr(>F)']][2]
+    }
+  }
+
+  ### layer 3: test each contrast on each outcome individually ###
+
+  for (k in 1:length(family.outcomes)){
+
+    outcome <- family.outcomes[k]
+
+    outcome.formula <-
+      outcome %.% ' ~\n0 +\n' %.%
+      paste(treatments,         # treatments (base terms)
+            collapse = ' +\n'
+            ) %.% ' +\n' %.%
+    paste(family.controls,      # controls (base terms)
+          collapse = ' +\n'
+          )
+
+    cat(rep('-', 40), '\n\nrunning lm:\n\n', outcome.formula, '\n\n', sep = '')
+
+    outcome.mod <- lm(outcome.formula, d)
+    ## hack to eliminate NA coefs
+    if (any(is.na(coef(outcome.mod)))){
+      drop <- names(coef(outcome.mod))[is.na(coef(outcome.mod))]
+      drop <- convert.interaction.names(drop,
+                                        c(family.controls,
+                                          family.controls.interactions
+                                          )
+                                        )
+      layer3.notes[[i]][k] <-
+        layer3.notes[[i]][k] %.%
+        'dropped the following coefs: ' %.%
+        paste(drop, sep = ', ') %.%
+        '\n\n'
+      outcome.formula <- gsub(
+        '\\s+\\+\\s+(' %.% paste(drop, collapse = '|') %.% ')',
+        '',
+        outcome.formula
+      )
+      outcome.mod <- lm(outcome.formula, d)
+    }
+
+    outcome.vcov <- vcovHC(outcome.mod)
+    if (any(!is.finite(outcome.vcov))){
+      outcome.vcov <- vcov(outcome.mod)
+      layer3.notes[[i]][k] <-
+        layer3.notes[[i]][k] %.%
+        'falling back to non-robust vcov\n\n'
+    }
+    coef.names <- names(coef(outcome.mod))
+
+    for (j in 1:nrow(contrasts)){
+
+      ## convert this interaction term to whatever llm() named it
+      treat <- convert.interaction.names(contrasts[j, 'treat'], coef.names)
+      ctrl <- convert.interaction.names(contrasts[j, 'ctrl'], coef.names)
+      ## test group equality on this outcome
+      contrast.lht <- linearHypothesis(
+        outcome.mod,
+        vcov. = outcome.vcov,
+        hypothesis.matrix = sprintf('%s - %s', treat, ctrl),
+        test = 'F'
+      )
+      layer3.pvals[[i]][[j]][k] <- contrast.lht[['Pr(>F)']][2]
+      layer3.ests[[i]][[j]][k] <- (
+        coef(outcome.mod)[treat] - coef(outcome.mod)[ctrl]
+      ) ## * attr(d[[outcome]], 'scaled:scale')  # note: uncomment if rescaling
+      layer3.ses[[i]][[j]][k] <- sqrt(
+        outcome.vcov[treat, treat] +
+          outcome.vcov[ctrl, ctrl] -
+          2 * outcome.vcov[treat, ctrl]
+      )
+
+      ## ## confirm
+      ## linearHypothesis(
+      ##   outcome.mod,
+      ##   vcov. = outcome.vcov,
+      ##   hypothesis.matrix = sprintf('%s - %s', treat, ctrl),
+      ##   test = 'F'
+      ## )
+      ## (coef(outcome.mod)[treat] - coef(outcome.mod)[ctrl])^2 /
+      ##   (
+      ##     outcome.vcov[treat, treat] +
+      ##       outcome.vcov[ctrl, ctrl] -
+      ##       2 * outcome.vcov[treat, ctrl]
+      ##   )
+      ## linearHypothesis(
+      ##   outcome.mod,
+      ##   vcov. = outcome.vcov,
+      ##   hypothesis.matrix = sprintf('%s - %s', treat, ctrl),
+      ##   test = 'Chisq'
+      ## )
+      ## 2 - 2 * pnorm(abs(
+      ## (coef(outcome.mod)[treat] - coef(outcome.mod)[ctrl]) /
+      ##   sqrt(
+      ##     outcome.vcov[treat, treat] +
+      ##       outcome.vcov[ctrl, ctrl] -
+      ##       2 * outcome.vcov[treat, ctrl]
+      ##   )
+      ## ))
+
+    }
+
+  }
+
+}
+
+
+
+#################################
+## multiple testing correction ##
+#################################
+
+thresh <- .05
+
+## if layer-1 f-test is infeasible for a family due to collinearity,
+##   obtain layer-1 p-values for that family by simes
+for (i in which(is.na(layer1.pvals))){
+  layer1.pvals[i] <- simes(layer2.pvals[[i]])
+}
+
+## multiple testing adjustment for layer 1
+layer1.pvals.adj <- p.adjust(layer1.pvals, 'BH')
+layer1.nonnull.prop <- mean(layer1.pvals.adj < thresh)
+
+## test layer-2 hypotheses only if layer 1 passes
+layer2.pvals.adj <- layer2.pvals  # start by copying unadjusted layer-2 p-values
+layer2.nonnull.prop <- rep(NA, length(layer1.pvals.adj))
+names(layer2.nonnull.prop) <- names(layer1.pvals.adj)
+for (i in 1:length(layer1.pvals)){
+  if (layer1.pvals.adj[i] < thresh){  # if layer 1 passes
+    ## adjust for multiplicity within layer 2...
+    layer2.pvals.adj[[i]] <- p.adjust(layer2.pvals[[i]], 'BH')
+    ## ... and inflate to account for selection at layer 1
+    layer2.pvals.adj[[i]] <-
+      pmin(layer2.pvals.adj[[i]] / layer1.nonnull.prop, 1)
+    ## keep track of selection at layer 2 for use in layer 3
+    layer2.nonnull.prop[i] <- mean(layer2.pvals.adj[[i]] < thresh)
+  } else {  # if layer 1 fails
+    layer2.pvals.adj[[i]] <- rep(NA_real_, length(layer2.pvals[[i]]))
+    names(layer2.pvals.adj[[i]]) <- names(layer2.pvals[[i]])
+  }
+}
+
+## test layer-3 hypotheses only if layers 1 & 2 pass
+layer3.pvals.adj <- layer3.pvals  # start by copying unadjusted layer-3 p-values
+for (i in 1:length(layer1.pvals.adj)){
+  for (j in 1:length(layer2.pvals.adj[[i]])){
+    ##
+    if (layer1.pvals.adj[i] < thresh &&      # if layer 1 passes...
+          layer2.pvals.adj[[i]][j] < thresh  # ... and if layer 2 passes
+        ){
+      ## adjust for multiplicity within layer 3...
+      layer3.pvals.adj[[i]][[j]] <- p.adjust(layer3.pvals[[i]][[j]], 'BH')
+      ## ... and inflate to account for selection at layer 1
+      layer3.pvals.adj[[i]][[j]] <- pmin(
+        layer3.pvals.adj[[i]][[j]] / layer1.nonnull.prop / layer2.nonnull.prop[i],
+        1
+      )
+    } else {
+      layer3.pvals.adj[[i]][[j]] <- rep(NA_real_, length(layer3.pvals[[i]][[j]]))
+      names(layer3.pvals.adj[[i]][[j]]) <- names(layer3.pvals[[i]][[j]])
+    }
+  }
+}
+
+pvals.adj <- data.table(layer1 = character(0),
+                        layer2 = character(0),
+                        layer3 = character(0),
+                        p.adj = numeric(0),
+                        est = numeric(0),
+                        se = numeric(0)
+                        )
+for (i in 1:length(layer1.pvals.adj)){
+  pvals.adj <- rbind(pvals.adj,
+                     data.table(layer1 = names(layer1.pvals.adj)[i],
+                                layer2 = 'overall',
+                                layer3 = 'overall',
+                                p.adj = layer1.pvals.adj[i],
+                                est = NA_real_,
+                                se = NA_real_
+                                )
+                     )
+  for (j in 1:length(layer2.pvals.adj[[i]])){
+    pvals.adj <- rbind(pvals.adj,
+                       data.table(layer1 = names(layer1.pvals.adj)[i],
+                                  layer2 = names(layer2.pvals.adj[[i]])[j],
+                                  layer3 = 'overall',
+                                  p.adj = layer2.pvals.adj[[i]][j],
+                                  est = NA_real_,
+                                  se = NA_real_
+                                  )
+                       )
+    for (k in 1:length(layer3.pvals.adj[[i]][[j]])){
+      pvals.adj <- rbind(pvals.adj,
+                         data.table(layer1 = names(layer1.pvals.adj)[i],
+                                    layer2 = names(layer2.pvals.adj[[i]])[j],
+                                    layer3 = names(layer3.pvals.adj[[i]][[j]])[k],
+                                    p.adj = layer3.pvals.adj[[i]][[j]][k],
+                                    est = layer3.ests[[i]][[j]][k],
+                                    se = layer3.ses[[i]][[j]][k]
+                                    )
+                         )
+    }
+  }
+}
+
+## write out
+fwrite(pvals.adj, '../results/intermediate data/minimum wage (issue 2)/padj_basecontrol.csv')
+
+## prettify for reading
+pvals.adj.pretty <- pvals.adj
+colnames(pvals.adj.pretty) <- gsub('layer1',
+                                   'layer1_hypothesisfamily',
+                                   colnames(pvals.adj.pretty)
+                                   )
+colnames(pvals.adj.pretty) <- gsub('layer2',
+                                   'layer2_treatmentcontrast',
+                                   colnames(pvals.adj.pretty)
+                                   )
+colnames(pvals.adj.pretty) <- gsub('layer3',
+                                   'layer3_specificoutcome',
+                                   colnames(pvals.adj.pretty)
+                                   )
+pvals.adj.pretty[, layer2_treatmentcontrast := gsub(
+  'attitude\\.(pro|anti|neutral)(:seed\\.(pro|anti))?:recsys.(31|22)',
+  '\\1 \\3 \\4',
+  layer2_treatmentcontrast
+)]
+pvals.adj.pretty[, layer2_treatmentcontrast := gsub(
+  '.vs.',
+  ' - ',
+  layer2_treatmentcontrast,
+  fixed = TRUE
+)]
+pvals.adj.pretty[, layer2_treatmentcontrast := gsub(
+  ' +',
+  ' ',
+  layer2_treatmentcontrast
+)]
+fwrite(pvals.adj.pretty,
+       '../results/intermediate data/minimum wage (issue 2)/padj_basecontrol_pretty.csv'
+       )
+
+print('preregistered results:')
+pvals.adj.pretty[p.adj < .05 & layer3_specificoutcome != 'overall',]
+
+
+
+##############################################
+## added 4 jun 2024 at request of reviewers ##
+##############################################
+
+## analyze components of main policy outcome index that relate to
+## post-experiment w2 "understanding" of an issue, using w1 version
+## of that same outcome as the only control (analogous to outcome index
+## regression, which uses w2 index as outcome and w1 index as control)
+
+## initialize results table
+understanding.results <- data.table(layer2_treatmentcontrast = character(0),
+                                    layer3_specificoutcome = character(0),
+                                    est = numeric(0),
+                                    se = numeric(0),
+                                    p = numeric(0)
+                                    )
+
+## loop over outcomes
+for (k in 1:length(mwpolicy.outcomes.understanding)){
+
+  outcome <- mwpolicy.outcomes.understanding[k]
+
+  outcome.formula <-
+    outcome %.% ' ~\n0 +\n' %.%
+    paste(treatments,         # treatments (base terms)
+          collapse = ' +\n'
+          ) %.% ' +\n' %.%
+    paste(gsub('w2', 'w1', outcome),      # controls (w1 outcome)
+          collapse = ' +\n'
+          )
+
+  cat(rep('-', 40), '\n\nrunning lm:\n\n', outcome.formula, '\n\n', sep = '')
+
+  outcome.mod <- lm(outcome.formula, d)
+  ## hack to eliminate NA coefs
+  if (any(is.na(coef(outcome.mod)))){
+    drop <- names(coef(outcome.mod))[is.na(coef(outcome.mod))]
+    drop <- convert.interaction.names(drop,
+                                      c(family.controls,
+                                        family.controls.interactions
+                                        )
+                                      )
+    layer3.notes[[i]][k] <-
+      layer3.notes[[i]][k] %.%
+      'dropped the following coefs: ' %.%
+      paste(drop, sep = ', ') %.%
+      '\n\n'
+    outcome.formula <- gsub(
+      '\\s+\\+\\s+(' %.% paste(drop, collapse = '|') %.% ')',
+      '',
+      outcome.formula
+    )
+    outcome.mod <- lm(outcome.formula, d)
+  }
+
+  outcome.vcov <- vcovHC(outcome.mod)
+  if (any(!is.finite(outcome.vcov))){
+    outcome.vcov <- vcov(outcome.mod)
+    layer3.notes[[i]][k] <-
+      layer3.notes[[i]][k] %.%
+      'falling back to non-robust vcov\n\n'
+  }
+  coef.names <- names(coef(outcome.mod))
+
+  ## loop over treatment contrasts
+  for (j in 1:nrow(contrasts)){
+
+    ## convert this interaction term to whatever llm() named it
+    treat <- convert.interaction.names(contrasts[j, 'treat'], coef.names)
+    ctrl <- convert.interaction.names(contrasts[j, 'ctrl'], coef.names)
+    ## test group equality on this outcome
+    contrast.lht <- linearHypothesis(
+      outcome.mod,
+      vcov. = outcome.vcov,
+      hypothesis.matrix = sprintf('%s - %s', treat, ctrl),
+      test = 'F'
+    )
+
+    ## prettify name of contrast for readability
+
+    contrast <- treat %.% ' - ' %.% ctrl
+    contrast <- gsub('attitude\\.(pro|anti|neutral)', '\\1', contrast)
+    contrast <- gsub('seed\\.(pro|anti)', '\\1', contrast)
+    contrast <- gsub('recsys.(31|22)', '\\1', contrast)
+    contrast <- gsub(':', ' ', contrast)
+    contrast <- gsub(' +', ' ', contrast)
+
+    p <- contrast.lht[['Pr(>F)']][2]
+    est <- (
+      coef(outcome.mod)[treat] - coef(outcome.mod)[ctrl]
+    ) ## * attr(d[[outcome]], 'scaled:scale')  # note: uncomment if rescaling
+    se <- sqrt(
+      outcome.vcov[treat, treat] +
+        outcome.vcov[ctrl, ctrl] -
+        2 * outcome.vcov[treat, ctrl]
+    )
+
+    understanding.results <- rbind(
+      understanding.results,
+      data.table(
+        layer2_treatmentcontrast = contrast,
+        layer3_specificoutcome = outcome,
+        p,
+        est,
+        se
+      )
+    )
+
+  }
+
+}
+
+## conduct multiple testing adjustment within newly exploratory results
+understanding.results[, p.adj := p.adjust(p, 'BH')]
+print('exploratory results on understanding-related questions:')
+understanding.results[p.adj < .05,]
+fwrite(understanding.results,
+       '../results/intermediate data/minimum wage (issue 2)/understanding_basecontrol_pretty.csv'
+       )
+
+
+
+#############################################################
+## preregistered exploratory heterogeneous effect analysis ##
+#############################################################
+
+# outcome is mw_index_w2
+# construct moderators by cutting demographics & pre-treatment vars at midpoint
+
+d[,
+  pol_interest_hi := as.numeric(
+    pol_interest > median(pol_interest, na.rm = TRUE)
+  )]
+d[,
+  age_hi := as.numeric(
+    age > median(age, na.rm = TRUE)
+  )]
+d[,
+  freq_youtube_hi := as.numeric(
+    freq_youtube > median(freq_youtube, na.rm = TRUE)
+  )]
+
+moderator_variables <- c('pol_interest_hi',
+                         'age_hi',
+                         'male',
+                         'freq_youtube_hi'
+                         )
+## added 4 jun 2024 at request of reviewer
+moderator_variables_revision <- 'college'
+
+interaction_results <- data.table()
+for (moderator_variable in c(moderator_variables, moderator_variables_revision)){
+
+  d[, moderator := get(moderator_variable)]
+
+  mod.attitude.anti <- lm(
+    mw_index_w2 ~
+      recsys.31 * moderator +
+      mw_index_w1,   # only control is pre-treatment outcome, as in primary analysis
+    data = d[attitude.anti == 1]
+  )
+  vcov.attitude.anti <- vcovHC(mod.attitude.anti)
+  test.attitude.anti <- coeftest(mod.attitude.anti, vcov.attitude.anti)
+  interaction_results <- rbind(
+    interaction_results,
+    data.table(subset = 'attitude.anti',
+               interaction = 'recsys.31:' %.% moderator_variable,
+               test.attitude.anti['recsys.31:moderator', , drop = FALSE]
+               ),
+    fill = TRUE
+  )
+
+  mod.attitude.pro <- lm(
+    mw_index_w2 ~
+      recsys.31 * moderator +
+      mw_index_w1,   # only control is pre-treatment outcome, as in primary analysis
+    data = d[attitude.pro == 1]
+  )
+  vcov.attitude.pro <- vcovHC(mod.attitude.pro)
+  test.attitude.pro <- coeftest(mod.attitude.pro, vcov.attitude.pro)
+  interaction_results <- rbind(
+    interaction_results,
+    data.table(subset = 'attitude.pro',
+               interaction = 'recsys.31:' %.% moderator_variable,
+               test.attitude.pro['recsys.31:moderator', , drop = FALSE]
+               ),
+    fill = TRUE
+  )
+
+  mod.attitude.neutral.seed.anti <- lm(
+    mw_index_w2 ~
+      recsys.31 * moderator +
+      mw_index_w1,   # only control is pre-treatment outcome, as in primary analysis
+    data = d[attitude.neutral == 1 & seed.anti == 1]
+  )
+  vcov.attitude.neutral.seed.anti <- vcovHC(mod.attitude.neutral.seed.anti)
+  test.attitude.neutral.seed.anti <- coeftest(mod.attitude.neutral.seed.anti,
+                                             vcov.attitude.neutral.seed.anti
+                                             )
+  interaction_results <- rbind(
+    interaction_results,
+    data.table(subset = 'attitude.neutral.seed.anti',
+               interaction = 'recsys.31:' %.% moderator_variable,
+               test.attitude.neutral.seed.anti[
+                 'recsys.31:moderator', , drop = FALSE
+               ]
+               ),
+    fill = TRUE
+  )
+
+  mod.attitude.neutral.seed.pro <- lm(
+    mw_index_w2 ~
+      recsys.31 * moderator +
+      mw_index_w1,   # only control is pre-treatment outcome, as in primary analysis
+    data = d[attitude.neutral == 1 & seed.pro == 1]
+  )
+  vcov.attitude.neutral.seed.pro <- vcovHC(mod.attitude.neutral.seed.pro)
+  test.attitude.neutral.seed.pro <- coeftest(mod.attitude.neutral.seed.pro,
+                                             vcov.attitude.neutral.seed.pro                                             )
+  interaction_results <- rbind(
+    interaction_results,
+    data.table(subset = 'attitude.neutral.seed.pro',
+               interaction = 'recsys.31:' %.% moderator_variable,
+               test.attitude.neutral.seed.pro[
+                 'recsys.31:moderator', , drop = FALSE
+               ]
+               ),
+    fill = TRUE
+  )
+
+}
+
+# very little significant heterogeneity even before multiple testing correction
+# out of 16 tests, 2 have p values of .043 and .032
+print('heterogeneity results before multiple correction:')
+interaction_results[`Pr(>|t|)` < .05,]
+# none survives a BH correction
+interaction_results[, p.adj := p.adjust(`Pr(>|t|)`, 'BH')]
+print('heterogeneity p-values after multiple correction:')
+interaction_results[, p.adj]
+
+## added 4 jun 2024 at request of reviewers
+colnames(interaction_results) <- c(
+  subset = 'subset',
+  interaction = 'interaction',
+  Estimate = 'est',
+  `Std. Error` = 'se',
+  `t value` = 't',
+  `Pr(>|t|)` = 'p',
+  p.adj = 'p.adj'
+)[colnames(interaction_results)]
+fwrite(interaction_results,
+        '../results/intermediate data/minimum wage (issue 2)/heterogeneity_basecontrol.csv'
+        )
+
+
+
+###############################################
+## added 30 sep 2024 at request of reviewers ##
+###############################################
+
+## what are minimum detectable effects, given multiple testing correction?
+
+n_sims <- 1000
+params_sims <- expand.grid(seed = 19104 + 0:(n_sims - 1),
+                           effect = seq(from = .01, to = .05, by = .001)
+                           )
+
+## step 1: identify largest p-value s.t. we would have rejected layer-1 null
+## (that at least one treatment contrast has effect on policy index)
+## to do this, we hold fixed p-values for all other layer-1 hypothesis families
+layer1.pvals.mde <- layer1.pvals
+layer1.pvals.mde['mwpolicy'] <- 0
+while (p.adjust(layer1.pvals.mde, 'BH')['mwpolicy'] <= .05){
+  layer1.pvals.mde['mwpolicy'] <- layer1.pvals.mde['mwpolicy'] + .001
+}
+pval.cutoff <- layer1.pvals.mde['mwpolicy']
+print('to achieve significance of policy attitude family at layer 1 (pooled test of any effect on policy index from any contrast) when correcting for multiple layer-1 hypothesis families, this is the minimum cutoff value after conducting simes correction of layer 2 pvals:')
+pval.cutoff
+
+## if layer-1 null was rejected for the policy outcome, then we would use this
+## correction factor when interpreting layer-2 p-values (for specific contrasts)
+layer1.nonnull.prop.if.gt.cutoff <- mean(c(
+  p.adjust(layer1.pvals.mde, 'BH')[c('platform', 'media', 'affpol')] < .05,
+  TRUE
+))
+
+## the sims below will only examine 3/1 vs 2/2 treatment contrasts, so we will
+## hold fixed the layer-2 p-values that relate to seed contrasts
+pvals.for.seed.contrasts.on.policyindex <- layer2.pvals$mwpolicy[
+  c('attitude.neutral:seed.pro:recsys.31.vs.attitude.neutral:seed.anti:recsys.31',
+    'attitude.neutral:seed.pro:recsys.22.vs.attitude.neutral:seed.anti:recsys.22'
+    )
+]
+
+
+
+## step 2: prepare simulations based on real data ------------------------------
+
+mod.attitude.anti <- lm(
+  mw_index_w2 ~ recsys.31 + mw_index_w1,
+  data = d[attitude.anti == 1]
+)
+X.attitude.anti <- model.matrix(mod.attitude.anti)
+residual.sd.attitude.anti <- sd(resid(mod.attitude.anti))
+## confirm that this recovers fitted values
+##   model.matrix(mod.attitude.anti) %*% coef(mod.attitude.anti)
+assert_that(all(
+  predict(mod.attitude.anti) ==
+    X.attitude.anti %*% coef(mod.attitude.anti)
+))
+## we will create simulated outcomes, given hypothesized treatment effect
+## == intercept +                                              <-- part A
+##    real coef * real pretreatment attitude +                 <-- part A
+##    hypothesized treatment effect * real treatment status +  <-- part B
+##    rnorm(mean = 0, sd = real residual outcome sd)           <-- part C
+## A: generate fitted values under hypothesized effect size
+coef.attitude.anti.baseline <- coef(mod.attitude.anti)
+coef.attitude.anti.baseline['recsys.31'] <- 0
+Y.attitude.anti.baseline <-
+  as.numeric(X.attitude.anti %*% coef.attitude.anti.baseline)
+## C: will be added below with hypothesized effect * treatment
+## B: will be drawn below with rnorm(mean=0, sd=residual_sd)
+
+## repeat above for respondents with pro attitude
+mod.attitude.pro <- lm(
+  mw_index_w2 ~ recsys.31 + mw_index_w1,
+  data = d[attitude.pro == 1]
+)
+X.attitude.pro <- model.matrix(mod.attitude.pro)
+residual.sd.attitude.pro <- sd(resid(mod.attitude.pro))
+coef.attitude.pro.baseline <- coef(mod.attitude.pro)
+coef.attitude.pro.baseline['recsys.31'] <- 0
+Y.attitude.pro.baseline <-
+  as.numeric(X.attitude.pro %*% coef.attitude.pro.baseline)
+
+## repeat above for respondents with neutral attitude assigned to pro seed
+mod.attitude.neutral.seed.pro <- lm(
+  mw_index_w2 ~ recsys.31 + mw_index_w1,
+  data = d[attitude.neutral == 1 & seed.pro == 1]
+)
+X.attitude.neutral.seed.pro <- model.matrix(mod.attitude.neutral.seed.pro)
+residual.sd.attitude.neutral.seed.pro <- sd(resid(mod.attitude.neutral.seed.pro))
+coef.attitude.neutral.seed.pro.baseline <- coef(mod.attitude.neutral.seed.pro)
+coef.attitude.neutral.seed.pro.baseline['recsys.31'] <- 0
+Y.attitude.neutral.seed.pro.baseline <-
+  as.numeric(X.attitude.neutral.seed.pro %*% coef.attitude.neutral.seed.pro.baseline)
+
+## repeat above for respondents with neutral attitude assigned to anti seed
+mod.attitude.neutral.seed.anti <- lm(
+  mw_index_w2 ~ recsys.31 + mw_index_w1,
+  data = d[attitude.neutral == 1 & seed.anti == 1]
+)
+X.attitude.neutral.seed.anti <- model.matrix(mod.attitude.neutral.seed.anti)
+residual.sd.attitude.neutral.seed.anti <- sd(resid(mod.attitude.neutral.seed.anti))
+coef.attitude.neutral.seed.anti.baseline <- coef(mod.attitude.neutral.seed.anti)
+coef.attitude.neutral.seed.anti.baseline['recsys.31'] <- 0
+Y.attitude.neutral.seed.anti.baseline <-
+  as.numeric(X.attitude.neutral.seed.anti %*% coef.attitude.neutral.seed.anti.baseline)
+
+
+
+## step 3: conduct sims --------------------------------------------------------
+
+sims.attitude.anti <- foreach(seed = params_sims$seed,
+                              effect = params_sims$effect,
+                              .combine = rbind
+                              ) %dopar%
+  {
+    set.seed(seed)
+    Y <-
+      Y.attitude.anti.baseline +
+      effect * X.attitude.anti[, 'recsys.31'] +
+      rnorm(
+        n = nrow(X.attitude.anti),
+        mean = 0,
+        sd = residual.sd.attitude.anti
+      )
+    mod <- lm(Y ~ 0 + X.attitude.anti)
+    smry <- coeftest(mod, vcovHC(mod))
+    cbind(
+      seed,
+      effect,
+      data.table(smry['X.attitude.antirecsys.31', , drop = FALSE])
+    )
+  }
+
+sims.attitude.pro <- foreach(seed = params_sims$seed,
+                              effect = params_sims$effect,
+                              .combine = rbind
+                              ) %dopar%
+  {
+    set.seed(seed)
+    Y <-
+      Y.attitude.pro.baseline +
+      effect * X.attitude.pro[, 'recsys.31'] +
+      rnorm(
+        n = nrow(X.attitude.pro),
+        mean = 0,
+        sd = residual.sd.attitude.pro
+      )
+    mod <- lm(Y ~ 0 + X.attitude.pro)
+    smry <- coeftest(mod, vcovHC(mod))
+    cbind(
+      seed,
+      effect,
+      data.table(smry['X.attitude.prorecsys.31', , drop = FALSE])
+    )
+  }
+
+sims.attitude.neutral.seed.anti <- foreach(seed = params_sims$seed,
+                                           effect = params_sims$effect,
+                                           .combine = rbind
+                                           ) %dopar%
+  {
+    set.seed(seed)
+    Y <-
+      Y.attitude.neutral.seed.anti.baseline +
+      effect * X.attitude.neutral.seed.anti[, 'recsys.31'] +
+      rnorm(
+        n = nrow(X.attitude.neutral.seed.anti),
+        mean = 0,
+        sd = residual.sd.attitude.neutral.seed.anti
+      )
+    mod <- lm(Y ~ 0 + X.attitude.neutral.seed.anti)
+    smry <- coeftest(mod, vcovHC(mod))
+    cbind(
+      seed,
+      effect,
+      data.table(smry['X.attitude.neutral.seed.antirecsys.31', , drop = FALSE])
+    )
+  }
+
+sims.attitude.neutral.seed.pro <- foreach(seed = params_sims$seed,
+                                           effect = params_sims$effect,
+                                           .combine = rbind
+                                           ) %dopar%
+  {
+    set.seed(seed)
+    Y <-
+      Y.attitude.neutral.seed.pro.baseline +
+      effect * X.attitude.neutral.seed.pro[, 'recsys.31'] +
+      rnorm(
+        n = nrow(X.attitude.neutral.seed.pro),
+        mean = 0,
+        sd = residual.sd.attitude.neutral.seed.pro
+      )
+    mod <- lm(Y ~ 0 + X.attitude.neutral.seed.pro)
+    smry <- coeftest(mod, vcovHC(mod))
+    cbind(
+      seed,
+      effect,
+      data.table(smry['X.attitude.neutral.seed.prorecsys.31', , drop = FALSE])
+    )
+  }
+
+
+
+## step 4: analyze power results -----------------------------------------------
+
+## without multiple-testing corrections
+
+print('mde for respondents with anti attitude (conventional analysis w/o correction):')
+sims.attitude.anti[,
+                   .(p.reject = mean(`Pr(>|t|)` < .05)),
+                   by = effect
+                   ][p.reject >= .8, min(effect)]
+
+print('mde for respondents with pro attitude (conventional analysis w/o correction):')
+sims.attitude.pro[,
+                  .(p.reject = mean(`Pr(>|t|)` < .05)),
+                  by = effect
+                  ][p.reject >= .8, min(effect)]
+
+print('mde for respondents with neutral attitude assigned to pro seed (conventional analysis w/o correction):')
+sims.attitude.neutral.seed.anti[,
+                                .(p.reject = mean(`Pr(>|t|)` < .05)),
+                                by = effect
+                                ][p.reject >= .8, min(effect)]
+
+## respondents with neutral attitude assigned to anti seed
+sims.attitude.neutral.seed.pro[,
+                               .(p.reject = mean(`Pr(>|t|)` < .05)),
+                               by = effect
+                               ][p.reject >= .8, min(effect)]
+
+
+
+## with multiple testing correction
+
+sims <- rbind(
+  sims.attitude.anti,
+  sims.attitude.pro,
+  sims.attitude.neutral.seed.anti,
+  sims.attitude.neutral.seed.pro
+)
+
+sims.layer1 <- sims[
+,
+  .(pval.pooled = ifelse(
+    ## if these results would lead us to reject layer-1 pooled null of no effect
+    ## on policy attitudes from any treatment contrast
+    simes(c(
+      `Pr(>|t|)`,
+      pvals.for.seed.contrasts.on.policyindex
+    )) <= pval.cutoff,
+    ## disaggregate layer-2 results report with procedure from above
+    ## (BH correction, then inflate by 1/prop of layer-1 sig results)
+    ## then subset to only those p-values relating to 3/1 vs 2/2 contrast
+    ## to see if any are <.05 after full correction procedure
+    yes = min(
+      p.adjust(c(`Pr(>|t|)`, pvals.for.seed.contrasts.on.policyindex),
+               'BH'
+               )[1:4] / layer1.nonnull.prop.if.gt.cutoff
+    ),
+    no = Inf
+  )
+  ),
+  by = .(seed, effect)
+]
+print('with multiple testing correction:')
+sims.layer1[, .(p.reject = mean(pval.pooled <= pval.cutoff)), by = effect]
+print('mde:')
+sims.layer1[,
+            .(p.reject = mean(pval.pooled <= pval.cutoff)),
+            by = effect
+            ][p.reject >= .8, min(effect)]
+

code/minimum wage (issue 2)/03b_analysis_multipletesting_yg.R

@@ -0,0 +1,1295 @@
+cat(rep('=', 80),
+    '\n\n',
+    'OUTPUT FROM: minimum wage (issue 2)/03b_analysis_multipletesting_yg.R',
+    '\n\n',
+    sep = ''
+    )
+
+library(data.table)
+library(car)
+library(sandwich)
+library(lmtest)
+library(ggplot2)
+library(assertthat)
+library(foreach)
+library(doParallel)
+registerDoParallel(cores = detectCores() - 1)
+
+
+
+###############
+## functions ##
+###############
+
+`%.%` <- paste0
+
+simes <- function(ps){
+  min(sort(length(ps) * ps / rank(ps)))
+}
+
+### functions to handle inconsistent interaction ordering of mlm() ###
+
+## convert interaction terms of form 'b#:a#' to 'a#:b#'
+reorder.interaction.names <- function(x, prefix = ''){
+  x <- gsub('^' %.% prefix, '', x)
+  sapply(strsplit(x, ':'),
+         function(y){
+           paste(sort(y), collapse = ':')
+         })
+}
+
+## take term of form 'a1:b1', look up in vector of form 'b#:a#, return 'b1:a1'
+convert.interaction.names <- function(x, y, prefix.y = ''){
+  ind <- match(reorder.interaction.names(x),
+               reorder.interaction.names(y, prefix = prefix.y)
+               )
+  return(y[ind])
+}
+
+## modified from print.linearHypothesis.mlm to use alternate df & return pvals
+##   (print method is responsible for doing the actual computation of pvals)
+extract.lht <- function(x,
+                        SSP = TRUE,
+                        SSPE = SSP,
+                        digits = getOption('digits'),
+                        df.residual = x$df.residual
+                        ){
+  test <- x$test
+  if (!is.null(x$P) && SSP) {
+    P <- x$P
+    cat("\n Response transformation matrix:\n")
+    attr(P, "assign") <- NULL
+    attr(P, "contrasts") <- NULL
+    print(P, digits = digits)
+  }
+  if (SSP) {
+    cat("\nSum of squares and products for the hypothesis:\n")
+    print(x$SSPH, digits = digits)
+  }
+  if (SSPE) {
+    cat("\nSum of squares and products for error:\n")
+    print(x$SSPE, digits = digits)
+  }
+  if ((!is.null(x$singular)) && x$singular) {
+    warning("the error SSP matrix is singular; multivariate tests are unavailable")
+    return(invisible(x))
+  }
+  SSPE.qr <- qr(x$SSPE)
+  eigs <- Re(eigen(qr.coef(SSPE.qr, x$SSPH), symmetric = FALSE)$values)
+  tests <- matrix(NA, 4, 4)
+  rownames(tests) <- c("Pillai", "Wilks", "Hotelling-Lawley",
+                       "Roy")
+  if ("Pillai" %in% test)
+    tests[1, 1:4] <- car:::Pillai(eigs, x$df, df.residual)
+  if ("Wilks" %in% test)
+    tests[2, 1:4] <- car:::Wilks(eigs, x$df, df.residual)
+  if ("Hotelling-Lawley" %in% test)
+    tests[3, 1:4] <- car:::HL(eigs, x$df, df.residual)
+  if ("Roy" %in% test)
+    tests[4, 1:4] <- car:::Roy(eigs, x$df, df.residual)
+  tests <- na.omit(tests)
+  ok <- tests[, 2] >= 0 & tests[, 3] > 0 & tests[, 4] > 0
+  ok <- !is.na(ok) & ok
+  tests <- cbind(x$df, tests, pf(tests[ok, 2], tests[ok, 3],
+                                 tests[ok, 4], lower.tail = FALSE))
+  colnames(tests) <- c("Df", "test stat", "approx F", "num Df",
+                       "den Df", "Pr(>F)")
+  tests <- structure(as.data.frame(tests),
+                     heading = paste("\nMultivariate Test",
+                                     if (nrow(tests) > 1)
+                                       "s", ": ", x$title, sep = ""),
+                     class = c("anova",
+                               "data.frame"
+                               )
+                     )
+  return(tests)
+}
+
+
+
+###############
+## load data ##
+###############
+
+d <- fread('../results/intermediate data/minimum wage (issue 2)/yg_w12_clean.csv')
+
+## drop pure control
+d <- d[treatment_arm != 'control',]
+
+## drop NA video counts
+d <- d[!is.na(pro) & !is.na(anti),]
+
+
+
+##############
+## controls ##
+##############
+
+platform.controls <- c('age_cat',
+                       'male',
+                       'pol_interest',
+                       'freq_youtube'
+                       )
+
+mwpolicy.controls <- 'mw_index_w1'
+
+media.controls <- c('trust_majornews_w1',
+                    'trust_youtube_w1',
+                    'fabricate_majornews_w1',
+                    'fabricate_youtube_w1'
+                    )
+
+affpol.controls <- c('affpol_ft',
+                     'affpol_smart',
+                     'affpol_comfort'
+                     )
+
+controls.raw <- unique(c(platform.controls,
+                         mwpolicy.controls,
+                         media.controls,
+                         affpol.controls
+                         )
+                       )
+
+## transform control variables by creating dummies and demeaning
+controls.trans <- list()
+for (j in controls.raw){
+  ## convert to dummies if needed
+  controls.j <- model.matrix(as.formula('~ 0 + ' %.% j),
+                             model.frame(as.formula('~ 0 + ' %.% j),
+                                         data = d,
+                                         na.action = 'na.pass'
+                                         )
+                             )
+  ## demean by column
+  controls.j <- sweep(controls.j,
+                      MARGIN = 2,
+                      STATS = colMeans(controls.j, na.rm = TRUE),
+                      FUN = `-`,
+                      )
+  colnames(controls.j) <- make.names(colnames(controls.j))
+  ## remove control from original data
+  d[[j]] <- NULL
+  ## reinsert transformed control
+  d <- cbind(d, controls.j)
+  ## keep track of which original controls map to which transformed controls
+  controls.trans[[j]] <- colnames(controls.j)
+}
+
+## map original control variables to transformed versions
+platform.controls <- unlist(controls.trans[platform.controls])
+mwpolicy.controls <- unlist(controls.trans[mwpolicy.controls])
+media.controls <- unlist(controls.trans[media.controls])
+affpol.controls <- unlist(controls.trans[affpol.controls])
+
+
+
+##############
+## outcomes ##
+##############
+
+### hypothesis family 1: platform interactions ###
+
+## platform interaction time: compute windorized usage time
+warning('diverges from pap, 95% windsorized due to extreme outliers')
+d[, platform_duration := duration]
+d[platform_duration <= quantile(d$duration, .025),
+  platform_duration := quantile(d$duration, .025)
+  ]
+d[platform_duration >= quantile(d$duration, .975),
+  platform_duration := quantile(d$duration, .975)
+  ]
+## all platform interaction outcomes
+platform.outcomes <- c('pro_fraction_chosen',
+                       'positive_interactions',  # positive - negative (dislike)
+                       'platform_duration'
+                       )
+
+
+
+### hypothesis family 2: MW policy attitudes ###
+
+## only one preregistered outcome in this family
+mwpolicy.outcomes <- 'mw_index_w2'
+## added 4 jun 2024 at request of reviewers
+mwpolicy.outcomes.understanding <- c('mw_restrict_w2',
+                                     'mw_help_w2'
+                                     )
+
+
+
+### hypothesis family 3: media trust ###
+media.outcomes <- c('trust_majornews_w2',
+                    'trust_youtube_w2',
+                    'fabricate_majornews_w2',
+                    'fabricate_youtube_w2'
+                    )
+
+
+
+### hypothesis family 4: affective polarization ###
+affpol.outcomes <- c('affpol_ft_w2',
+                     'affpol_smart_w2',
+                     'affpol_comfort_w2'
+                     )
+
+outcomes <- unique(c(
+  platform.outcomes,
+  mwpolicy.outcomes,
+  media.outcomes,
+  affpol.outcomes
+)
+)
+
+
+
+################
+## treatments ##
+################
+
+## create attitude dummies
+## (pro/anti stance on issue has opposite lib/con meaning from study 1)
+d[, attitude := c('pro', 'neutral', 'anti')[thirds]]
+d[, attitude.anti := as.numeric(attitude == 'anti')]
+d[, attitude.neutral := as.numeric(attitude == 'neutral')]
+d[, attitude.pro := as.numeric(attitude == 'pro')]
+
+## create seed dummies
+d[, seed.anti := as.numeric(treatment_seed == 'anti')]
+d[, seed.pro := as.numeric(treatment_seed == 'pro')]
+
+## create recsys dummies
+d[, recsys.22 := as.numeric(treatment_arm %like% '22')]
+d[, recsys.31 := as.numeric(treatment_arm %like% '31')]
+
+## manually define coefficients to estimate
+treatments <- c('attitude.anti:recsys.22',
+                'attitude.anti:recsys.31',
+                'attitude.neutral:seed.anti:recsys.22',
+                'attitude.neutral:seed.pro:recsys.22',
+                'attitude.neutral:seed.anti:recsys.31',
+                'attitude.neutral:seed.pro:recsys.31',
+                'attitude.pro:recsys.22',
+                'attitude.pro:recsys.31'
+                )
+
+contrasts <- rbind(
+  i = c(treat = 'attitude.pro:recsys.31',
+        ctrl = 'attitude.pro:recsys.22'
+        ),
+  ii = c(treat = 'attitude.anti:recsys.31',
+         ctrl = 'attitude.anti:recsys.22'
+         ),
+  iii = c(treat = 'attitude.neutral:seed.pro:recsys.31',
+          ctrl = 'attitude.neutral:seed.pro:recsys.22'
+          ),
+  iv = c(treat = 'attitude.neutral:seed.anti:recsys.31',
+         ctrl = 'attitude.neutral:seed.anti:recsys.22'
+         ),
+  # in (v-vi), pro/anti order is reversed from study 1 to ensure that
+  # - 1st condition (treatment) is always the conservative video
+  # - 2nd condition (control) is always the liberal video
+  v = c(treat = 'attitude.neutral:seed.anti:recsys.31',
+        ctrl = 'attitude.neutral:seed.pro:recsys.31'
+  ),
+  vi = c(treat = 'attitude.neutral:seed.anti:recsys.22',
+         ctrl = 'attitude.neutral:seed.pro:recsys.22'
+  )
+)
+
+## check that contrasts are valid
+assert_that(all(unlist(contrasts) %in% treatments))
+
+## check that specifications are equivalent
+coefs.v1 <- coef(lm(mw_index_w2 ~ 0 + attitude:treatment_arm, d))
+coefs.v2 <- coef(
+  lm(mw_index_w2 ~
+       0 +
+       attitude.anti:recsys.22 +
+       attitude.anti:recsys.31 +
+       attitude.neutral:seed.anti:recsys.22 +
+       attitude.neutral:seed.pro:recsys.22 +
+       attitude.neutral:seed.anti:recsys.31 +
+       attitude.neutral:seed.pro:recsys.31 +
+       attitude.pro:recsys.22 +
+       attitude.pro:recsys.31,
+     d
+     )
+)
+assert_that(all.equal(unname(sort(coefs.v1)), unname(sort(coefs.v2))))
+
+
+
+##########################
+## hierarchical testing ##
+##########################
+
+## initialize top layer p-values:
+##   does treatment have any effect on any outcome in family
+families <- c(
+  'platform',
+  'mwpolicy',
+  'media',
+  'affpol'
+)
+layer1.pvals <- rep(NA_real_, length(families))
+layer1.notes <- rep('', length(families))
+names(layer1.pvals) <- families
+
+## initialize 2nd layer p-values:
+##   which treatment has detectable effect?
+contrast.pvals <- rep(NA_real_, nrow(contrasts))
+names(contrast.pvals) <- paste(contrasts[, 'treat'],
+                               contrasts[, 'ctrl'],
+                               sep = '.vs.'
+                               )
+layer2.pvals <- list(
+  platform = contrast.pvals,
+  mwpolicy = contrast.pvals,
+  media = contrast.pvals,
+  affpol = contrast.pvals
+)
+rm(contrast.pvals)
+
+## initialize 3rd layer p-values:
+##   on which specific outcome in family?
+layer3.pvals <- list()
+layer3.ests <- list()
+layer3.ses <- list()
+layer3.notes <- list()
+for (i in 1:length(families)){
+  family <- families[i]
+  layer3.pvals[[family]] <- list()
+  layer3.ests[[family]] <- list()
+  layer3.ses[[family]] <- list()
+  layer3.notes[[family]] <- list()
+  outcomes <- get(family %.% '.outcomes')
+  for (j in 1:nrow(contrasts)){
+    contrast <- paste(contrasts[j, 'treat'],
+                      contrasts[j, 'ctrl'],
+                      sep = '.vs.'
+                      )
+    layer3.pvals[[family]][[contrast]] <- numeric(0)
+    layer3.ests[[family]][[contrast]] <- numeric(0)
+    layer3.ses[[family]][[contrast]] <- numeric(0)
+    for (k in 1:length(outcomes)){
+      outcome <- outcomes[k]
+      layer3.pvals[[family]][[contrast]][outcome] <- NA_real_
+      layer3.ests[[family]][[contrast]][outcome] <- NA_real_
+      layer3.ses[[family]][[contrast]][outcome] <- NA_real_
+      layer3.notes[[family]][outcome] <- ''
+    }
+  }
+}
+
+
+
+### begin nested analyses ###
+
+for (i in 1:length(families)){
+
+  family <- families[i]
+  family.outcomes <- get(family %.% '.outcomes')
+  family.controls <- get(family %.% '.controls')
+  family.controls.interactions <- as.character(
+    outer(treatments,
+          family.controls,
+          FUN = function(x, y) x %.% ':' %.% y
+          )
+  )
+
+  family.formula <-
+    'cbind(' %.%                # outcomes
+    paste(family.outcomes,
+          collapse = ', '
+          ) %.%  ') ~\n0 +\n' %.%
+    paste(treatments,           # treatments (base terms)
+          collapse = ' +\n'
+          ) %.% ' +\n' %.%
+    paste(family.controls,      # controls (base terms)
+          collapse = ' +\n'
+          )
+
+  cat(rep('=', 80),
+      '\n\nHYPOTHESIS FAMILY: ',
+      family,
+      '\n\nrunning mlm:\n\n',
+      family.formula,
+      '\n\n',
+      sep = ''
+      )
+
+  ## run model
+  family.mod <- lm(family.formula, d)
+
+  ## hack to eliminate NA coefs
+  if (any(is.na(coef(family.mod)))){
+    if ('mlm' %in% class(family.mod)){
+      drop <- rownames(coef(family.mod))[is.na(coef(family.mod))[, 1]]
+    } else {
+      drop <- names(coef(family.mod))[is.na(coef(family.mod))]
+    }
+    drop <- convert.interaction.names(drop,
+                                      c(family.controls,
+                                        family.controls.interactions
+                                        )
+                                      )
+    layer1.notes[[i]] <-
+      layer1.notes[[i]] %.%
+      'dropped the following coefs: ' %.%
+      paste(drop, sep = ', ') %.%
+      '\n\n'
+    family.formula <- gsub(
+      '\\s+\\+\\s+(' %.% paste(drop, collapse = '|') %.% ')',
+      '',
+      family.formula
+    )
+    family.mod <- lm(family.formula, d)
+  }
+
+  family.vcov <- vcovHC(family.mod)
+  if (is.null(dim(coef(family.mod)))){
+    coef.names <- names(coef(family.mod))
+  } else {
+    coef.names <- rownames(coef(family.mod))
+  }
+
+  ### top layer: test overall significance of all contrasts on all outcomes ###
+  ## convert interaction terms to whatever mlm() named it
+  treats <- convert.interaction.names(contrasts[, 'treat'], coef.names)
+  ctrls <- convert.interaction.names(contrasts[, 'ctrl'], coef.names)
+  ## test jointly
+  lht.attempt <- tryCatch({
+    if ('mlm' %in% class(family.mod)){
+      contrast.lht <- linearHypothesis(
+        family.mod,
+        vcov. = family.vcov,
+        hypothesis.matrix = sprintf('%s - %s', treats, ctrls),
+        rhs = matrix(0, nrow = nrow(contrasts), ncol = length(family.outcomes)),
+        test = 'Pillai'
+      )
+      layer1.pvals[[i]] <- extract.lht(contrast.lht)[, 'Pr(>F)']
+    } else {
+      contrast.lht <- linearHypothesis(
+        family.mod,
+        vcov. = family.vcov,
+        hypothesis.matrix = sprintf('%s - %s', treats, ctrls),
+        rhs = matrix(0, nrow = nrow(contrasts), ncol = length(family.outcomes)),
+        test = 'F'
+      )
+      layer1.pvals[[i]] <- contrast.lht[['Pr(>F)']][2]
+    }
+  },
+  error = function(e){
+    warning(sprintf('caught error in %s family:', family), e)
+    ## return error as string for inclusion in notes
+    'caught error: ' %.%
+      e %.%
+      '\n\n'
+  })
+  if (lht.attempt %like% 'caught error'){
+    layer1.notes[[i]] <-
+      layer1.notes[[i]] %.% lht.attempt
+  }
+
+
+
+  ### layer 2: test each contrast individually on all outcomes ###
+
+  for (j in 1:nrow(contrasts)){
+    ## test group equality on all outcomes
+    if ('mlm' %in% class(family.mod)){
+      contrast.lht <-
+        linearHypothesis(
+        family.mod,
+        vcov. = family.vcov,
+        hypothesis.matrix = sprintf('%s - %s', treats[j], ctrls[j]),
+        rhs = matrix(0, nrow = 1, ncol = length(family.outcomes)),
+        test = 'Pillai'
+      )
+      layer2.pvals[[i]][j] <- extract.lht(contrast.lht)[, 'Pr(>F)']
+    } else {
+      contrast.lht <- linearHypothesis(
+        family.mod,
+        vcov. = family.vcov,
+        hypothesis.matrix = sprintf('%s - %s', treats[j], ctrls[j]),
+        rhs = matrix(0, nrow = 1, ncol = length(family.outcomes)),
+        test = 'F'
+      )
+      layer2.pvals[[i]][j] <- contrast.lht[['Pr(>F)']][2]
+    }
+  }
+
+  ### layer 3: test each contrast on each outcome individually ###
+
+  for (k in 1:length(family.outcomes)){
+
+    outcome <- family.outcomes[k]
+
+    outcome.formula <-
+      outcome %.% ' ~\n0 +\n' %.%
+      paste(treatments,         # treatments (base terms)
+            collapse = ' +\n'
+            ) %.% ' +\n' %.%
+    paste(family.controls,      # controls (base terms)
+          collapse = ' +\n'
+          )
+
+    cat(rep('-', 40), '\n\nrunning lm:\n\n', outcome.formula, '\n\n', sep = '')
+
+    outcome.mod <- lm(outcome.formula, d)
+    ## hack to eliminate NA coefs
+    if (any(is.na(coef(outcome.mod)))){
+      drop <- names(coef(outcome.mod))[is.na(coef(outcome.mod))]
+      drop <- convert.interaction.names(drop,
+                                        c(family.controls,
+                                          family.controls.interactions
+                                          )
+                                        )
+      layer3.notes[[i]][k] <-
+        layer3.notes[[i]][k] %.%
+        'dropped the following coefs: ' %.%
+        paste(drop, sep = ', ') %.%
+        '\n\n'
+      outcome.formula <- gsub(
+        '\\s+\\+\\s+(' %.% paste(drop, collapse = '|') %.% ')',
+        '',
+        outcome.formula
+      )
+      outcome.mod <- lm(outcome.formula, d)
+    }
+
+    outcome.vcov <- vcovHC(outcome.mod)
+    if (any(!is.finite(outcome.vcov))){
+      outcome.vcov <- vcov(outcome.mod)
+      layer3.notes[[i]][k] <-
+        layer3.notes[[i]][k] %.%
+        'falling back to non-robust vcov\n\n'
+    }
+    coef.names <- names(coef(outcome.mod))
+
+    for (j in 1:nrow(contrasts)){
+
+      ## convert this interaction term to whatever llm() named it
+      treat <- convert.interaction.names(contrasts[j, 'treat'], coef.names)
+      ctrl <- convert.interaction.names(contrasts[j, 'ctrl'], coef.names)
+      ## test group equality on this outcome
+      contrast.lht <- linearHypothesis(
+        outcome.mod,
+        vcov. = outcome.vcov,
+        hypothesis.matrix = sprintf('%s - %s', treat, ctrl),
+        test = 'F'
+      )
+      layer3.pvals[[i]][[j]][k] <- contrast.lht[['Pr(>F)']][2]
+      layer3.ests[[i]][[j]][k] <- (
+        coef(outcome.mod)[treat] - coef(outcome.mod)[ctrl]
+      ) ## * attr(d[[outcome]], 'scaled:scale')  # note: uncomment if rescaling
+      layer3.ses[[i]][[j]][k] <- sqrt(
+        outcome.vcov[treat, treat] +
+          outcome.vcov[ctrl, ctrl] -
+          2 * outcome.vcov[treat, ctrl]
+      )
+
+      ## ## confirm
+      ## linearHypothesis(
+      ##   outcome.mod,
+      ##   vcov. = outcome.vcov,
+      ##   hypothesis.matrix = sprintf('%s - %s', treat, ctrl),
+      ##   test = 'F'
+      ## )
+      ## (coef(outcome.mod)[treat] - coef(outcome.mod)[ctrl])^2 /
+      ##   (
+      ##     outcome.vcov[treat, treat] +
+      ##       outcome.vcov[ctrl, ctrl] -
+      ##       2 * outcome.vcov[treat, ctrl]
+      ##   )
+      ## linearHypothesis(
+      ##   outcome.mod,
+      ##   vcov. = outcome.vcov,
+      ##   hypothesis.matrix = sprintf('%s - %s', treat, ctrl),
+      ##   test = 'Chisq'
+      ## )
+      ## 2 - 2 * pnorm(abs(
+      ## (coef(outcome.mod)[treat] - coef(outcome.mod)[ctrl]) /
+      ##   sqrt(
+      ##     outcome.vcov[treat, treat] +
+      ##       outcome.vcov[ctrl, ctrl] -
+      ##       2 * outcome.vcov[treat, ctrl]
+      ##   )
+      ## ))
+
+    }
+
+  }
+
+}
+
+
+
+#################################
+## multiple testing correction ##
+#################################
+
+thresh <- .05
+
+## if layer-1 f-test is infeasible for a family due to collinearity,
+##   obtain layer-1 p-values for that family by simes
+for (i in which(is.na(layer1.pvals))){
+  layer1.pvals[i] <- simes(layer2.pvals[[i]])
+}
+
+## multiple testing adjustment for layer 1
+layer1.pvals.adj <- p.adjust(layer1.pvals, 'BH')
+layer1.nonnull.prop <- mean(layer1.pvals.adj < thresh)
+
+## test layer-2 hypotheses only if layer 1 passes
+layer2.pvals.adj <- layer2.pvals  # start by copying unadjusted layer-2 p-values
+layer2.nonnull.prop <- rep(NA, length(layer1.pvals.adj))
+names(layer2.nonnull.prop) <- names(layer1.pvals.adj)
+for (i in 1:length(layer1.pvals)){
+  if (layer1.pvals.adj[i] < thresh){  # if layer 1 passes
+    ## adjust for multiplicity within layer 2...
+    layer2.pvals.adj[[i]] <- p.adjust(layer2.pvals[[i]], 'BH')
+    ## ... and inflate to account for selection at layer 1
+    layer2.pvals.adj[[i]] <-
+      pmin(layer2.pvals.adj[[i]] / layer1.nonnull.prop, 1)
+    ## keep track of selection at layer 2 for use in layer 3
+    layer2.nonnull.prop[i] <- mean(layer2.pvals.adj[[i]] < thresh)
+  } else {  # if layer 1 fails
+    layer2.pvals.adj[[i]] <- rep(NA_real_, length(layer2.pvals[[i]]))
+    names(layer2.pvals.adj[[i]]) <- names(layer2.pvals[[i]])
+  }
+}
+
+## test layer-3 hypotheses only if layers 1 & 2 pass
+layer3.pvals.adj <- layer3.pvals  # start by copying unadjusted layer-3 p-values
+for (i in 1:length(layer1.pvals.adj)){
+  for (j in 1:length(layer2.pvals.adj[[i]])){
+    ##
+    if (layer1.pvals.adj[i] < thresh &&      # if layer 1 passes...
+          layer2.pvals.adj[[i]][j] < thresh  # ... and if layer 2 passes
+        ){
+      ## adjust for multiplicity within layer 3...
+      layer3.pvals.adj[[i]][[j]] <- p.adjust(layer3.pvals[[i]][[j]], 'BH')
+      ## ... and inflate to account for selection at layer 1
+      layer3.pvals.adj[[i]][[j]] <- pmin(
+        layer3.pvals.adj[[i]][[j]] / layer1.nonnull.prop / layer2.nonnull.prop[i],
+        1
+      )
+    } else {
+      layer3.pvals.adj[[i]][[j]] <- rep(NA_real_, length(layer3.pvals[[i]][[j]]))
+      names(layer3.pvals.adj[[i]][[j]]) <- names(layer3.pvals[[i]][[j]])
+    }
+  }
+}
+
+pvals.adj <- data.table(layer1 = character(0),
+                        layer2 = character(0),
+                        layer3 = character(0),
+                        p.adj = numeric(0),
+                        est = numeric(0),
+                        se = numeric(0)
+                        )
+for (i in 1:length(layer1.pvals.adj)){
+  pvals.adj <- rbind(pvals.adj,
+                     data.table(layer1 = names(layer1.pvals.adj)[i],
+                                layer2 = 'overall',
+                                layer3 = 'overall',
+                                p.adj = layer1.pvals.adj[i],
+                                est = NA_real_,
+                                se = NA_real_
+                                )
+                     )
+  for (j in 1:length(layer2.pvals.adj[[i]])){
+    pvals.adj <- rbind(pvals.adj,
+                       data.table(layer1 = names(layer1.pvals.adj)[i],
+                                  layer2 = names(layer2.pvals.adj[[i]])[j],
+                                  layer3 = 'overall',
+                                  p.adj = layer2.pvals.adj[[i]][j],
+                                  est = NA_real_,
+                                  se = NA_real_
+                                  )
+                       )
+    for (k in 1:length(layer3.pvals.adj[[i]][[j]])){
+      pvals.adj <- rbind(pvals.adj,
+                         data.table(layer1 = names(layer1.pvals.adj)[i],
+                                    layer2 = names(layer2.pvals.adj[[i]])[j],
+                                    layer3 = names(layer3.pvals.adj[[i]][[j]])[k],
+                                    p.adj = layer3.pvals.adj[[i]][[j]][k],
+                                    est = layer3.ests[[i]][[j]][k],
+                                    se = layer3.ses[[i]][[j]][k]
+                                    )
+                         )
+    }
+  }
+}
+
+## write out
+dir.create('../results/intermediate data/minimum wage (issue 2)')
+fwrite(pvals.adj, '../results/intermediate data/minimum wage (issue 2)/padj_basecontrol_yg.csv')
+
+## prettify for reading
+pvals.adj.pretty <- pvals.adj
+colnames(pvals.adj.pretty) <- gsub('layer1',
+                                   'layer1_hypothesisfamily',
+                                   colnames(pvals.adj.pretty)
+                                   )
+colnames(pvals.adj.pretty) <- gsub('layer2',
+                                   'layer2_treatmentcontrast',
+                                   colnames(pvals.adj.pretty)
+                                   )
+colnames(pvals.adj.pretty) <- gsub('layer3',
+                                   'layer3_specificoutcome',
+                                   colnames(pvals.adj.pretty)
+                                   )
+pvals.adj.pretty[, layer2_treatmentcontrast := gsub(
+  'attitude\\.(pro|anti|neutral)(:seed\\.(pro|anti))?:recsys.(31|22)',
+  '\\1 \\3 \\4',
+  layer2_treatmentcontrast
+)]
+pvals.adj.pretty[, layer2_treatmentcontrast := gsub(
+  '.vs.',
+  ' - ',
+  layer2_treatmentcontrast,
+  fixed = TRUE
+)]
+pvals.adj.pretty[, layer2_treatmentcontrast := gsub(
+  ' +',
+  ' ',
+  layer2_treatmentcontrast
+)]
+fwrite(pvals.adj.pretty,
+       '../results/intermediate data/minimum wage (issue 2)/padj_basecontrol_pretty_yg.csv'
+       )
+
+print('preregistered results:')
+pvals.adj.pretty[p.adj < .05 & layer3_specificoutcome != 'overall',]
+
+
+
+##############################################
+## added 4 jun 2024 at request of reviewers ##
+##############################################
+
+## analyze components of main policy outcome index that relate to
+## post-experiment w2 "understanding" of an issue, using w1 version
+## of that same outcome as the only control (analogous to outcome index
+## regression, which uses w2 index as outcome and w1 index as control)
+
+## initialize results table
+understanding.results <- data.table(layer2_treatmentcontrast = character(0),
+                                    layer3_specificoutcome = character(0),
+                                    est = numeric(0),
+                                    se = numeric(0),
+                                    p = numeric(0)
+                                    )
+
+## loop over outcomes
+for (k in 1:length(mwpolicy.outcomes.understanding)){
+
+  outcome <- mwpolicy.outcomes.understanding[k]
+
+  outcome.formula <-
+    outcome %.% ' ~\n0 +\n' %.%
+    paste(treatments,         # treatments (base terms)
+          collapse = ' +\n'
+          ) %.% ' +\n' %.%
+    paste(gsub('w2', 'w1', outcome),      # controls (w1 outcome)
+          collapse = ' +\n'
+          )
+
+  cat(rep('-', 40), '\n\nrunning lm:\n\n', outcome.formula, '\n\n', sep = '')
+
+  outcome.mod <- lm(outcome.formula, d)
+  ## hack to eliminate NA coefs
+  if (any(is.na(coef(outcome.mod)))){
+    drop <- names(coef(outcome.mod))[is.na(coef(outcome.mod))]
+    drop <- convert.interaction.names(drop,
+                                      c(family.controls,
+                                        family.controls.interactions
+                                        )
+                                      )
+    layer3.notes[[i]][k] <-
+      layer3.notes[[i]][k] %.%
+      'dropped the following coefs: ' %.%
+      paste(drop, sep = ', ') %.%
+      '\n\n'
+    outcome.formula <- gsub(
+      '\\s+\\+\\s+(' %.% paste(drop, collapse = '|') %.% ')',
+      '',
+      outcome.formula
+    )
+    outcome.mod <- lm(outcome.formula, d)
+  }
+
+  outcome.vcov <- vcovHC(outcome.mod)
+  if (any(!is.finite(outcome.vcov))){
+    outcome.vcov <- vcov(outcome.mod)
+    layer3.notes[[i]][k] <-
+      layer3.notes[[i]][k] %.%
+      'falling back to non-robust vcov\n\n'
+  }
+  coef.names <- names(coef(outcome.mod))
+
+  ## loop over treatment contrasts
+  for (j in 1:nrow(contrasts)){
+
+    ## convert this interaction term to whatever llm() named it
+    treat <- convert.interaction.names(contrasts[j, 'treat'], coef.names)
+    ctrl <- convert.interaction.names(contrasts[j, 'ctrl'], coef.names)
+    ## test group equality on this outcome
+    contrast.lht <- linearHypothesis(
+      outcome.mod,
+      vcov. = outcome.vcov,
+      hypothesis.matrix = sprintf('%s - %s', treat, ctrl),
+      test = 'F'
+    )
+
+    ## prettify name of contrast for readability
+
+    contrast <- treat %.% ' - ' %.% ctrl
+    contrast <- gsub('attitude\\.(pro|anti|neutral)', '\\1', contrast)
+    contrast <- gsub('seed\\.(pro|anti)', '\\1', contrast)
+    contrast <- gsub('recsys.(31|22)', '\\1', contrast)
+    contrast <- gsub(':', ' ', contrast)
+    contrast <- gsub(' +', ' ', contrast)
+
+    p <- contrast.lht[['Pr(>F)']][2]
+    est <- (
+      coef(outcome.mod)[treat] - coef(outcome.mod)[ctrl]
+    ) ## * attr(d[[outcome]], 'scaled:scale')  # note: uncomment if rescaling
+    se <- sqrt(
+      outcome.vcov[treat, treat] +
+        outcome.vcov[ctrl, ctrl] -
+        2 * outcome.vcov[treat, ctrl]
+    )
+
+    understanding.results <- rbind(
+      understanding.results,
+      data.table(
+        layer2_treatmentcontrast = contrast,
+        layer3_specificoutcome = outcome,
+        p,
+        est,
+        se
+      )
+    )
+
+  }
+
+}
+
+## conduct multiple testing adjustment within newly exploratory results
+understanding.results[, p.adj := p.adjust(p, 'BH')]
+print('exploratory results on understanding-related questions:')
+understanding.results[p.adj < .05,]
+fwrite(understanding.results,
+       '../results/intermediate data/minimum wage (issue 2)/understanding_basecontrol_pretty_yg.csv'
+       )
+
+
+
+#############################################################
+## preregistered exploratory heterogeneous effect analysis ##
+#############################################################
+
+# outcome is mw_index_w2
+# construct moderators by cutting demographics & pre-treatment vars at midpoint
+
+d[,
+  pol_interest_hi := as.numeric(
+    pol_interest >= median(pol_interest, na.rm = TRUE)
+  )]
+d[,
+  age_hi := as.numeric(
+    age > median(age, na.rm = TRUE)
+  )]
+d[,
+  freq_youtube_hi := as.numeric(
+    freq_youtube > median(freq_youtube, na.rm = TRUE)
+  )]
+
+moderator_variables <- c('pol_interest_hi',
+                         'age_hi',
+                         'male',
+                         'freq_youtube_hi'
+                         )
+## added 4 jun 2024 at request of reviewer
+moderator_variables_revision <- 'college'
+
+interaction_results <- data.table()
+for (moderator_variable in c(moderator_variables, moderator_variables_revision)){
+
+  d[, moderator := get(moderator_variable)]
+
+  mod.attitude.anti <- lm(
+    mw_index_w2 ~
+      recsys.31 * moderator +
+      mw_index_w1,   # only control is pre-treatment outcome, as in primary analysis
+    data = d[attitude.anti == 1]
+  )
+  vcov.attitude.anti <- vcovHC(mod.attitude.anti)
+  test.attitude.anti <- coeftest(mod.attitude.anti, vcov.attitude.anti)
+  interaction_results <- rbind(
+    interaction_results,
+    data.table(subset = 'attitude.anti',
+               interaction = 'recsys.31:' %.% moderator_variable,
+               test.attitude.anti['recsys.31:moderator', , drop = FALSE]
+               ),
+    fill = TRUE
+  )
+
+  mod.attitude.pro <- lm(
+    mw_index_w2 ~
+      recsys.31 * moderator +
+      mw_index_w1,   # only control is pre-treatment outcome, as in primary analysis
+    data = d[attitude.pro == 1]
+  )
+  vcov.attitude.pro <- vcovHC(mod.attitude.pro)
+  test.attitude.pro <- coeftest(mod.attitude.pro, vcov.attitude.pro)
+  interaction_results <- rbind(
+    interaction_results,
+    data.table(subset = 'attitude.pro',
+               interaction = 'recsys.31:' %.% moderator_variable,
+               test.attitude.pro['recsys.31:moderator', , drop = FALSE]
+               ),
+    fill = TRUE
+  )
+
+  mod.attitude.neutral.seed.anti <- lm(
+    mw_index_w2 ~
+      recsys.31 * moderator +
+      mw_index_w1,   # only control is pre-treatment outcome, as in primary analysis
+    data = d[attitude.neutral == 1 & seed.anti == 1]
+  )
+  vcov.attitude.neutral.seed.anti <- vcovHC(mod.attitude.neutral.seed.anti)
+  test.attitude.neutral.seed.anti <- coeftest(mod.attitude.neutral.seed.anti,
+                                             vcov.attitude.neutral.seed.anti
+                                             )
+  interaction_results <- rbind(
+    interaction_results,
+    data.table(subset = 'attitude.neutral.seed.anti',
+               interaction = 'recsys.31:' %.% moderator_variable,
+               test.attitude.neutral.seed.anti[
+                 'recsys.31:moderator', , drop = FALSE
+               ]
+               ),
+    fill = TRUE
+  )
+
+  mod.attitude.neutral.seed.pro <- lm(
+    mw_index_w2 ~
+      recsys.31 * moderator +
+      mw_index_w1,   # only control is pre-treatment outcome, as in primary analysis
+    data = d[attitude.neutral == 1 & seed.pro == 1]
+  )
+  vcov.attitude.neutral.seed.pro <- vcovHC(mod.attitude.neutral.seed.pro)
+  test.attitude.neutral.seed.pro <- coeftest(mod.attitude.neutral.seed.pro,
+                                             vcov.attitude.neutral.seed.pro                                             )
+  interaction_results <- rbind(
+    interaction_results,
+    data.table(subset = 'attitude.neutral.seed.pro',
+               interaction = 'recsys.31:' %.% moderator_variable,
+               test.attitude.neutral.seed.pro[
+                 'recsys.31:moderator', , drop = FALSE
+               ]
+               ),
+    fill = TRUE
+  )
+
+}
+
+# very little significant heterogeneity even before multiple testing correction
+# out of 16 tests, 1 has p values of .013
+print('heterogeneity results before multiple correction:')
+interaction_results[`Pr(>|t|)` < .05,]
+# does not survive a BH correction
+interaction_results[, p.adj := p.adjust(`Pr(>|t|)`, 'BH')]
+print('heterogeneity p-values after multiple correction:')
+interaction_results[, p.adj]
+
+## updated 4 jun 2024 at request of reviewer
+colnames(interaction_results) <- c(
+  subset = 'subset',
+  interaction = 'interaction',
+  Estimate = 'est',
+  `Std. Error` = 'se',
+  `t value` = 't',
+  `Pr(>|t|)` = 'p',
+  p.adj = 'p.adj'
+)[colnames(interaction_results)]
+fwrite(interaction_results,
+        '../results/intermediate data/minimum wage (issue 2)/heterogeneity_basecontrol_yg.csv'
+        )
+
+
+
+###############################################
+## added 30 sep 2024 at request of reviewers ##
+###############################################
+
+## what are minimum detectable effects, given multiple testing correction?
+
+n_sims <- 1000
+params_sims <- expand.grid(seed = 19104 + 0:(n_sims - 1),
+                           effect = seq(from = .01, to = .05, by = .001)
+                           )
+
+## step 1: identify largest p-value s.t. we would have rejected layer-1 null
+## (that at least one treatment contrast has effect on policy index)
+layer1.pvals.mde <- layer1.pvals
+layer1.pvals.mde['mwpolicy'] <- 0
+while (p.adjust(layer1.pvals.mde, 'BH')['mwpolicy'] <= .05){
+  layer1.pvals.mde['mwpolicy'] <- layer1.pvals.mde['mwpolicy'] + .001
+}
+pval.cutoff <- layer1.pvals.mde['mwpolicy']
+print('to achieve significance of policy attitude family at layer 1 (pooled test of any effect on policy index from any contrast) when correcting for multiple layer-1 hypothesis families, this is the minimum cutoff value after conducting simes correction of layer 2 pvals:')
+pval.cutoff
+
+## if layer-1 null was rejected for the policy outcome, then we would use this
+## correction factor when interpreting layer-2 p-values (for specific contrasts)
+layer1.nonnull.prop.if.gt.cutoff <- mean(c(
+  p.adjust(layer1.pvals.mde, 'BH')[c('platform', 'media', 'affpol')] < .05,
+  TRUE
+))
+
+## the sims below will only examine 3/1 vs 2/2 treatment contrasts, so we will
+## hold fixed the layer-2 p-values that relate to seed contrasts
+pvals.for.seed.contrasts.on.policyindex <- layer2.pvals$mwpolicy[
+  c('attitude.neutral:seed.pro:recsys.31.vs.attitude.neutral:seed.anti:recsys.31',
+    'attitude.neutral:seed.pro:recsys.22.vs.attitude.neutral:seed.anti:recsys.22'
+    )
+]
+
+
+
+## step 2: prepare simulations based on real data ------------------------------
+
+mod.attitude.anti <- lm(
+  mw_index_w2 ~ recsys.31 + mw_index_w1,
+  data = d[attitude.anti == 1]
+)
+X.attitude.anti <- model.matrix(mod.attitude.anti)
+residual.sd.attitude.anti <- sd(resid(mod.attitude.anti))
+## confirm that this recovers fitted values
+##   model.matrix(mod.attitude.anti) %*% coef(mod.attitude.anti)
+assert_that(all(
+  predict(mod.attitude.anti) ==
+    X.attitude.anti %*% coef(mod.attitude.anti)
+))
+## we will create simulated outcomes, given hypothesized treatment effect
+## == intercept +                                              <-- part A
+##    real coef * real pretreatment attitude +                 <-- part A
+##    hypothesized treatment effect * real treatment status +  <-- part B
+##    rnorm(mean = 0, sd = real residual outcome sd)           <-- part C
+## A: generate fitted values under hypothesized effect size
+coef.attitude.anti.baseline <- coef(mod.attitude.anti)
+coef.attitude.anti.baseline['recsys.31'] <- 0
+Y.attitude.anti.baseline <-
+  as.numeric(X.attitude.anti %*% coef.attitude.anti.baseline)
+## C: will be added below with hypothesized effect * treatment
+## B: will be drawn below with rnorm(mean=0, sd=residual_sd)
+
+## repeat above for respondents with pro attitude
+mod.attitude.pro <- lm(
+  mw_index_w2 ~ recsys.31 + mw_index_w1,
+  data = d[attitude.pro == 1]
+)
+X.attitude.pro <- model.matrix(mod.attitude.pro)
+residual.sd.attitude.pro <- sd(resid(mod.attitude.pro))
+coef.attitude.pro.baseline <- coef(mod.attitude.pro)
+coef.attitude.pro.baseline['recsys.31'] <- 0
+Y.attitude.pro.baseline <-
+  as.numeric(X.attitude.pro %*% coef.attitude.pro.baseline)
+
+## repeat above for respondents with neutral attitude assigned to pro seed
+mod.attitude.neutral.seed.pro <- lm(
+  mw_index_w2 ~ recsys.31 + mw_index_w1,
+  data = d[attitude.neutral == 1 & seed.pro == 1]
+)
+X.attitude.neutral.seed.pro <- model.matrix(mod.attitude.neutral.seed.pro)
+residual.sd.attitude.neutral.seed.pro <- sd(resid(mod.attitude.neutral.seed.pro))
+coef.attitude.neutral.seed.pro.baseline <- coef(mod.attitude.neutral.seed.pro)
+coef.attitude.neutral.seed.pro.baseline['recsys.31'] <- 0
+Y.attitude.neutral.seed.pro.baseline <-
+  as.numeric(X.attitude.neutral.seed.pro %*% coef.attitude.neutral.seed.pro.baseline)
+
+## repeat above for respondents with neutral attitude assigned to anti seed
+mod.attitude.neutral.seed.anti <- lm(
+  mw_index_w2 ~ recsys.31 + mw_index_w1,
+  data = d[attitude.neutral == 1 & seed.anti == 1]
+)
+X.attitude.neutral.seed.anti <- model.matrix(mod.attitude.neutral.seed.anti)
+residual.sd.attitude.neutral.seed.anti <- sd(resid(mod.attitude.neutral.seed.anti))
+coef.attitude.neutral.seed.anti.baseline <- coef(mod.attitude.neutral.seed.anti)
+coef.attitude.neutral.seed.anti.baseline['recsys.31'] <- 0
+Y.attitude.neutral.seed.anti.baseline <-
+  as.numeric(X.attitude.neutral.seed.anti %*% coef.attitude.neutral.seed.anti.baseline)
+
+
+
+## step 3: conduct sims --------------------------------------------------------
+
+sims.attitude.anti <- foreach(seed = params_sims$seed,
+                              effect = params_sims$effect,
+                              .combine = rbind
+                              ) %dopar%
+  {
+    set.seed(seed)
+    Y <-
+      Y.attitude.anti.baseline +
+      effect * X.attitude.anti[, 'recsys.31'] +
+      rnorm(
+        n = nrow(X.attitude.anti),
+        mean = 0,
+        sd = residual.sd.attitude.anti
+      )
+    mod <- lm(Y ~ 0 + X.attitude.anti)
+    smry <- coeftest(mod, vcovHC(mod))
+    cbind(
+      seed,
+      effect,
+      data.table(smry['X.attitude.antirecsys.31', , drop = FALSE])
+    )
+  }
+
+sims.attitude.pro <- foreach(seed = params_sims$seed,
+                              effect = params_sims$effect,
+                              .combine = rbind
+                              ) %dopar%
+  {
+    set.seed(seed)
+    Y <-
+      Y.attitude.pro.baseline +
+      effect * X.attitude.pro[, 'recsys.31'] +
+      rnorm(
+        n = nrow(X.attitude.pro),
+        mean = 0,
+        sd = residual.sd.attitude.pro
+      )
+    mod <- lm(Y ~ 0 + X.attitude.pro)
+    smry <- coeftest(mod, vcovHC(mod))
+    cbind(
+      seed,
+      effect,
+      data.table(smry['X.attitude.prorecsys.31', , drop = FALSE])
+    )
+  }
+
+sims.attitude.neutral.seed.anti <- foreach(seed = params_sims$seed,
+                                           effect = params_sims$effect,
+                                           .combine = rbind
+                                           ) %dopar%
+  {
+    set.seed(seed)
+    Y <-
+      Y.attitude.neutral.seed.anti.baseline +
+      effect * X.attitude.neutral.seed.anti[, 'recsys.31'] +
+      rnorm(
+        n = nrow(X.attitude.neutral.seed.anti),
+        mean = 0,
+        sd = residual.sd.attitude.neutral.seed.anti
+      )
+    mod <- lm(Y ~ 0 + X.attitude.neutral.seed.anti)
+    smry <- coeftest(mod, vcovHC(mod))
+    cbind(
+      seed,
+      effect,
+      data.table(smry['X.attitude.neutral.seed.antirecsys.31', , drop = FALSE])
+    )
+  }
+
+sims.attitude.neutral.seed.pro <- foreach(seed = params_sims$seed,
+                                           effect = params_sims$effect,
+                                           .combine = rbind
+                                           ) %dopar%
+  {
+    set.seed(seed)
+    Y <-
+      Y.attitude.neutral.seed.pro.baseline +
+      effect * X.attitude.neutral.seed.pro[, 'recsys.31'] +
+      rnorm(
+        n = nrow(X.attitude.neutral.seed.pro),
+        mean = 0,
+        sd = residual.sd.attitude.neutral.seed.pro
+      )
+    mod <- lm(Y ~ 0 + X.attitude.neutral.seed.pro)
+    smry <- coeftest(mod, vcovHC(mod))
+    cbind(
+      seed,
+      effect,
+      data.table(smry['X.attitude.neutral.seed.prorecsys.31', , drop = FALSE])
+    )
+  }
+
+
+
+## step 4: analyze power results -----------------------------------------------
+
+## without multiple-testing corrections
+
+print('mde for respondents with anti attitude (conventional analysis w/o correction):')
+sims.attitude.anti[,
+                   .(p.reject = mean(`Pr(>|t|)` < .05)),
+                   by = effect
+                   ][p.reject >= .8, min(effect)]
+
+print('mde for respondents with pro attitude (conventional analysis w/o correction):')
+sims.attitude.pro[,
+                  .(p.reject = mean(`Pr(>|t|)` < .05)),
+                  by = effect
+                  ][p.reject >= .8, min(effect)]
+
+print('mde for respondents with neutral attitude assigned to pro seed (conventional analysis w/o correction):')
+sims.attitude.neutral.seed.anti[,
+                                .(p.reject = mean(`Pr(>|t|)` < .05)),
+                                by = effect
+                                ][p.reject >= .8, min(effect)]
+
+print('mde for respondents with neutral attitude assigned to anti seed (conventional analysis w/o correction):')
+sims.attitude.neutral.seed.pro[,
+                               .(p.reject = mean(`Pr(>|t|)` < .05)),
+                               by = effect
+                               ][p.reject >= .8, min(effect)]
+
+
+
+## with multiple testing correction
+
+sims <- rbind(
+  sims.attitude.anti,
+  sims.attitude.pro,
+  sims.attitude.neutral.seed.anti,
+  sims.attitude.neutral.seed.pro
+)
+
+sims.layer1 <- sims[
+,
+  .(pval.pooled = ifelse(
+    ## if these results would lead us to reject layer-1 pooled null of no effect
+    ## on policy attitudes from any treatment contrast
+    simes(c(
+      `Pr(>|t|)`,
+      pvals.for.seed.contrasts.on.policyindex
+    )) <= pval.cutoff,
+    ## disaggregate layer-2 results report with procedure from above
+    ## (BH correction, then inflate by 1/prop of layer-1 sig results)
+    ## then subset to only those p-values relating to 3/1 vs 2/2 contrast
+    ## to see if any are <.05 after full correction procedure
+    yes = min(
+      p.adjust(c(`Pr(>|t|)`, pvals.for.seed.contrasts.on.policyindex),
+               'BH'
+               )[1:4] / layer1.nonnull.prop.if.gt.cutoff
+    ),
+    no = Inf
+  )
+  ),
+  by = .(seed, effect)
+]
+print('with multiple testing correction:')
+sims.layer1[, .(p.reject = mean(pval.pooled <= pval.cutoff)), by = effect]
+print('mde:')
+sims.layer1[,
+            .(p.reject = mean(pval.pooled <= pval.cutoff)),
+            by = effect
+            ][p.reject >= .8, min(effect)]

code/run

@@ -0,0 +1,45 @@
+#!/usr/bin/env bash
+set -ex
+
+## Study 1
+Rscript 'gun control (issue 1)/01_trt_assign.R'
+Rscript 'gun control (issue 1)/02_clean_merge.R'
+Rscript 'gun control (issue 1)/03_analysis_multipletesting.R'
+
+## Study 2
+Rscript 'minimum wage (issue 2)/01_trt_assign.R'  # also covers study 3
+Rscript 'minimum wage (issue 2)/02_clean_merge.R'
+Rscript 'minimum wage (issue 2)/03_analysis_multipletesting.R'
+
+## Study 3 (YouGov)
+Rscript 'minimum wage (issue 2)/02b_clean_merge_yg.R'
+Rscript 'minimum wage (issue 2)/03b_analysis_multipletesting_yg.R'
+
+## Combined results from Studies 1-3
+Rscript '04_postprocessing_exploration_issues12.R'
+
+## Study 4
+Rscript 'shorts/05_clean_shorts_data.R' # returns demographics for the shorts (SI Fig1)
+Rscript 'shorts/06_analysis_multipletesting.R'
+Rscript 'shorts/07_postprocessing_exploration.R' # -- may not need the figure plotting section
+Rscript 'shorts/08_plot_shorts_figure.R' # returns the main figure for the Shorts exp
+
+## Supplementary analyses
+
+# (1) Experiment durations
+# SI Fig S2 comes from this code.
+python3 'supplemental/experiment durations/09_experiment_times.py' 
+
+# (2) Increasingly extreme recommendations 
+# SI Table S11, Fig S15 comes from this code
+python3 'supplemental/increasingly extreme recommendations/10_partisanship_increase.py'
+
+# SI Fig S12-13 comes from this code
+python3 'supplemental/increasingly extreme recommendations/11_gpt_rating_plots.py'
+
+# (3) Thumbnail ("First Impressions") analysis
+python3 'supplemental/thumbnails (first impressions)/12_thumbnail_analysis.py'
+python3 'supplemental/thumbnails (first impressions)/13_thumbnail_null_comparison.py'
+
+# SI Fig S3 comes from this code.
+Rscript 'supplemental/14_api_browser_comparison.R'

code/shorts/05_clean_shorts_data.R

@@ -0,0 +1,354 @@
+cat(rep('=', 80),
+    '\n\n',
+    'OUTPUT FROM: shorts/05_clean_shorts_data.R',
+    '\n\n',
+    sep = ''
+    )
+    
+## Extremizing Sequences and Minimum Wage Opinions
+## Data collected May 2024 via MTurk/CloudResearch
+## Analysis for the Extremizing Sequences Experiment
+
+## Preamble ----------------------------
+library(tidyverse)
+library(janitor)
+library(lubridate)
+library(stargazer)
+library(broom)
+
+# create a folder for the shorts intermediate data
+dir.create("../results/intermediate data/shorts/", recursive = TRUE, showWarnings = FALSE)
+
+# SURVEY DATA (FROM QUALTRICS)
+a <- read_csv("../data/shorts/ytrecs_surveys_may2024.csv")[-c(1,2),] %>%
+  clean_names() # 1315 obs.
+
+# DATE FILTER
+a <- a %>% filter(start_date >= '2024-05-28') # 1032 obs.
+
+# ATTENTION CHECK -- 932 obs.
+a <- a %>% filter(a$q81 == "Quick and easy")
+a <- a %>% filter(a$q82 == "wikiHow")
+a <- a %>% filter(is.na(video_link) == FALSE) ## failed respondents don't have a valid link
+
+# SURVEY TIME (ALL)
+a <- a %>% mutate(start_date = as_datetime(start_date),
+                  end_date = as_datetime(end_date),
+                  survey_time = as.numeric(end_date-start_date))
+
+summary(a$survey_time) # 5.5 mins to 74 mins (median 34 mins)
+
+# DEMOGRAPHICS -------------------------------------------------
+
+# GENDER, EDUCATION, INCOME
+a <- a %>%
+  mutate(female = ifelse(gender == "Woman", 1, 0),
+         male = ifelse(gender == "Man", 1, 0),
+         black = ifelse(str_detect(race_ethnicity, "Black"), 1, 0),
+         white = ifelse(str_detect(race_ethnicity, "White"), 1, 0),
+         college = ifelse(str_detect(highest_education, "college ") | str_detect(highest_education, "Post"), 1, 0),
+         income_gt50k = ifelse(income %in% names(table(a$income))[c(2,3,5,10,11,12,13)], 1, 0)
+  )
+a$income_gt50k[is.na(a$income)] <- NA
+
+# PID
+a <- a %>%
+  mutate(pid = case_when(pid1=="Democrat" ~ -1,
+                         pid1=="Republican" ~ 1,
+                         pid4=="Closer to the Republican Party" ~ 1,
+                         pid4=="Closer to the Democratic Party" ~ -1,
+                         pid4=="Neither" ~ 0))
+
+tabyl(a,pid)
+
+# IDEO
+a <- a %>%
+  mutate(ideo = case_when(ideo1=="Liberal" ~ -1,
+                          ideo1=="Conservative" ~ 1,
+                          ideo4=="Closer to conservatives" ~ 1,
+                          ideo4=="Closer to liberals" ~ -1,
+                          ideo4=="Neither" ~ 0))
+
+tabyl(a,ideo)
+
+# AGE
+a$age <- 2024-as.numeric(a$year_born)
+
+# AGE CATEGORIES: 18-29;  30-44;  45-64;  65+
+a <- a %>%
+  mutate(age_cat = case_when(age>=18 & age<=29 ~ "18-29",
+                             age>=30 & age<=44 ~ "30-44",
+                             age>=45 & age<=64 ~ "45-64",
+                             age>=65 ~ "65+"
+  ))
+a <- a %>%
+  fastDummies::dummy_cols(select_columns = "age_cat",remove_selected_columns = F)
+
+# POLITICAL INTEREST AND YOUTUBE FREQUENCY RECODING
+a <- a %>%
+  mutate(pol_interest = dplyr::recode(political_interest,"Extremely interested"=5,"Very interested"=4,"Somewhat interested"=3,"Not very interested"=2,"Not at all interested"=1),
+         freq_youtube = dplyr::recode(youtube_time,"More than 3 hours per day"=6,"2–3 hours per day"=5,"1–2 hours per day"=4,"31–59 minutes per day"=3,"10–30 minutes per day"=2,"Less than 10 minutes per day"=1,"None"=0)
+  )
+
+# SUMMARY TABLE FOR DEMOGRAPHICS
+summary_tab <- a %>%
+  dplyr::summarize(female = mean(female,na.rm=T),
+                   white = mean(white,na.rm=T),
+                   black = mean(black,na.rm=T),
+                   age1829 = mean(`age_cat_18-29`,na.rm=T),
+                   age3044 = mean(`age_cat_30-44`,na.rm=T),
+                   age4564 = mean(`age_cat_45-64`,na.rm=T),
+                   age65p = mean(`age_cat_65+`,na.rm=T),
+                   college = mean(college,na.rm=T),
+                   income_gt50k = mean(income_gt50k,na.rm=T),
+                   democrat = mean(pid==-1,na.rm=T),
+                   republican = mean(pid==1,na.rm=T))
+
+summary_tab <- pivot_longer(summary_tab,
+                            cols=c(female,
+                                   white,
+                                   black,
+                                   age1829,
+                                   age3044,
+                                   age4564,
+                                   age65p,
+                                   college,
+                                   income_gt50k,
+                                   democrat,
+                                   republican),
+                            names_to = "outcome",values_to = "survey_avg")
+outcome_labels <- data.frame(outcome_pretty = c("Female",
+                                                "White",
+                                                "Black",
+                                                "Age 18-29",
+                                                "Age 30-44",
+                                                "Age 45-64",
+                                                "Age 65+",
+                                                "College educated",
+                                                "Income >$50k",
+                                                "Democrat",
+                                                "Republican"),
+                             outcome = c("female",
+                                         "white",
+                                         "black",
+                                         "age1829",
+                                         "age3044",
+                                         "age4564",
+                                         "age65p",
+                                         "college",
+                                         "income_gt50k",
+                                         "democrat",
+                                         "republican"))
+
+summary_tab$outcome_pretty <-  outcome_labels$outcome_pretty[match(summary_tab$outcome,outcome_labels$outcome)]
+summary_tab <- summary_tab %>%
+  mutate(outcome_pretty = factor(outcome_pretty,levels = c("Republican",
+                                                           "Democrat",
+                                                           "Income >$50k",
+                                                           "College educated",
+                                                           "Age 65+",
+                                                           "Age 45-64",
+                                                           "Age 30-44",
+                                                           "Age 18-29",
+                                                           "Female",
+                                                           "Black",
+                                                           "White"),ordered=T))
+
+# DEMOGRAPHICS DESCRIPTIVE FIGURE
+(descrip_fig <- ggplot(summary_tab) + 
+    geom_point(aes(y=outcome_pretty,x=survey_avg)) + 
+    geom_text(aes(y=outcome_pretty,x=survey_avg,label=paste0(round(100*survey_avg,0),"%")),nudge_x = 0.1) + 
+    scale_y_discrete("") + 
+    scale_x_continuous("",labels=scales::percent_format(),limits=c(0,1)) + 
+    theme_bw()
+)
+ggsave(descrip_fig,filename = "../results/shorts_demographics.pdf",height=5,width=4)
+
+
+### DEMOGRAPHICS DONE ###
+
+#### OUTCOMES ####
+
+##### POLICY OPINIONS #####
+
+# convert to numeric unit scale:
+a <- a %>%
+  mutate( # higher = more conservative or anti-min wage
+    minwage15_pre = dplyr::recode(minwage15_pre,"Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4,
+    rtwa_v1_pre = dplyr::recode(rtwa_v1_pre, "Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4,
+    rtwa_v2_pre = dplyr::recode(rtwa_v2_pre, "Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4,
+    mw_support_pre = dplyr::recode(mw_support_pre, "Strongly oppose raising the minimum wage"=4,"Somewhat oppose raising the minimum wage"=3,"Neither support nor oppose raising the minimum wage"=2,"Somewhat support raising the minimum wage"=1,"Strongly support raising the minimum wage"=0)/4,
+    minwage_howhigh_pre = dplyr::recode(minwage_howhigh_pre, "Much lower than the current level"=4,"Somewhat lower than the current level"=3,"About the current level"=2,"Somewhat higher than the current level"=1,"Much higher than the current level"=0)/4,
+    mw_help_pre_1 = dplyr::recode(mw_help_pre_1, "10"=9,"9"=8,"8"=7,"7"=6,"6"=5,"5"=4,"4"=3,"3"=2,"2"=1,"1"=0)/9,
+    mw_restrict_pre_1 = dplyr::recode(mw_restrict_pre_1, "1"=9,"2"=8,"3"=7,"4"=6,"5"=5,"6"=4,"7"=3,"8"=2,"9"=1,"10"=0)/9,
+    minwage_text_r_pre = (25-as.numeric(minwage_text_pre))/25,
+  )
+a$minwage_text_r_pre[as.numeric(a$minwage_text_pre)>25] <- NA
+
+
+a <- a %>% 
+  rowwise() %>%
+  mutate(mw_index_pre = mean(c(minwage15_pre, rtwa_v1_pre,
+                               rtwa_v2_pre, mw_support_pre, 
+                               minwage_howhigh_pre, mw_help_pre_1,
+                               mw_restrict_pre_1, minwage_text_r_pre), na.rm=T)) %>%
+  ungroup()
+
+
+# CRONBACH'S ALPHA
+index_fa <- psych::alpha(select(a, minwage15_pre, rtwa_v1_pre, 
+                                rtwa_v2_pre, mw_support_pre, minwage_howhigh_pre,
+                                mw_help_pre_1, mw_restrict_pre_1, minwage_text_r_pre), check.keys = TRUE)
+
+write.csv(data.frame(cor(select(a, minwage15_pre, rtwa_v1_pre, rtwa_v2_pre, 
+                                mw_support_pre, minwage_howhigh_pre, mw_help_pre_1, 
+                                mw_restrict_pre_1, minwage_text_r_pre), use = "complete.obs")),
+row.names = T,file = "../results/cormat_mwindex_w1.csv")
+
+# CORRELATION PLOT PRE-MINIMUM WAGE OPINION
+pdf("corrplot_mwindex_w1.pdf")
+w1_corrplot <- corrplot::corrplot(cor(select(a, minwage15_pre, rtwa_v1_pre, rtwa_v2_pre,
+                                             mw_support_pre, minwage_howhigh_pre, mw_help_pre_1, 
+                                             mw_restrict_pre_1, minwage_text_r_pre), 
+                                      use = "complete.obs"),method = "shade")
+dev.off()
+
+(alpha <- index_fa$total["raw_alpha"]) # 0.9407615
+writeLines(as.character(round(alpha,2)),con = "../results/outcomes_alpha_w1_mturk.tex",sep = "%")
+
+tabyl(a,mw_index_pre)
+
+##### MEDIA TRUST #####
+a <- a %>%
+  mutate( # higher = more trusting
+    trust_majornews = dplyr::recode(info_trust_1,"A lot"=3,"Some"=2,"Not too much"=1,"Not at all"=0)/3,
+    trust_localnews = dplyr::recode(info_trust_2,"A lot"=3,"Some"=2,"Not too much"=1,"Not at all"=0)/3,
+    trust_social = dplyr::recode(info_trust_3,"A lot"=3,"Some"=2,"Not too much"=1,"Not at all"=0)/3,
+    trust_youtube = dplyr::recode(info_trust_4,"A lot"=3,"Some"=2,"Not too much"=1,"Not at all"=0)/3,
+    fabricate_majornews = dplyr::recode(mainstream_fakenews,"Never"=4,"Once in a while"=3,"About half the time"=2,"Most of the time"=1,"All the time"=0)/4,
+    fabricate_youtube = dplyr::recode(youtube_fakenews,"Never"=4,"Once in a while"=3,"About half the time"=2,"Most of the time"=1,"All the time"=0)/4
+  ) %>%
+  rowwise() %>%
+  mutate(media_trust = mean(trust_majornews,trust_localnews,fabricate_majornews,na.rm=T)) %>%
+  ungroup()
+
+media_trust_fa <- psych::alpha(select(a, trust_majornews,trust_localnews,fabricate_majornews), 
+                               check.keys = TRUE)
+(alpha <- media_trust_fa$total["raw_alpha"]) #. 0.7698292
+
+##### AFFECTIVE POLARIZATION #####
+a %>% 
+  group_by(pid) %>% 
+  summarize(mean_2=mean(as.numeric(political_lead_feels_2),na.rm=T), # Trump
+            mean_5=mean(as.numeric(political_lead_feels_5),na.rm=T), # Biden
+            mean_11=mean(as.numeric(political_lead_feels_11),na.rm=T), # dems
+            mean_12=mean(as.numeric(political_lead_feels_12),na.rm=T)) # reps
+
+a <- a %>%
+  mutate( # higher = more trusting
+    smart_dems = dplyr::recode(democrat_smart, "Extremely"=4,"Very"=3,"Somewhat"=2,"A little"=1,"Not at all"=0)/4,
+    smart_reps = dplyr::recode(republican_smart, "Extremely"=4,"Very"=3,"Somewhat"=2,"A little"=1,"Not at all"=0)/4,
+    comfort_dems = dplyr::recode(democrat_friends,"Extremely comfortable"=3,"Somewhat comfortable"=2,"Not too comfortable"=1,"Not at all comfortable"=0)/3,
+    comfort_reps = dplyr::recode(republican_friends,"Extremely comfortable"=3,"Somewhat comfortable"=2,"Not too comfortable"=1,"Not at all comfortable"=0)/3,
+    affpol_smart = case_when(
+      pid==-1 ~ smart_dems-smart_reps,
+      pid==1 ~ smart_reps-smart_dems
+    ),
+    affpol_comfort = case_when(
+      pid==-1 ~ comfort_dems-comfort_reps,
+      pid==1 ~ comfort_reps-comfort_dems
+    )
+  )
+
+# Create a new variable 'thirds' based on attributes
+a$thirds <- ifelse(!is.na(a$liberals_do) & is.na(a$moderates_do) & is.na(a$conservatives_do), 1, 
+                   ifelse(is.na(a$liberals_do) & !is.na(a$moderates_do) & is.na(a$conservatives_do), 2, 
+                          ifelse(is.na(a$liberals_do) & is.na(a$moderates_do) & !is.na(a$conservatives_do), 3, NA)))
+
+tabyl(a$thirds)
+
+#### OUTCOMES ####
+
+##### POLICY OPINIONS ######
+# convert to numeric unit scale:
+a <- a %>%
+  mutate( # higher = more pro-gun
+    minwage15 = dplyr::recode(minwage15,"Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4,
+    rtwa_v1 = dplyr::recode(rtwa_v1_updated, "Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4,
+    rtwa_v2 = dplyr::recode(rtwa_v2_updated, "Strongly oppose"=4,"Somewhat oppose"=3,"Neither support nor oppose"=2,"Somewhat support"=1,"Strongly support"=0)/4,
+    mw_support = dplyr::recode(mw_support, "Strongly oppose raising the minimum wage"=4,"Somewhat oppose raising the minimum wage"=3,"Neither support nor oppose raising the minimum wage"=2,"Somewhat support raising the minimum wage"=1,"Strongly support raising the minimum wage"=0)/4,
+    minwage_howhigh = dplyr::recode(minwage_howhigh, "Much lower than the current level"=4,"Somewhat lower than the current level"=3,"About the current level"=2,"Somewhat higher than the current level"=1,"Much higher than the current level"=0)/4,
+    mw_help_1 = dplyr::recode(mw_help_1, "10"=9,"9"=8,"8"=7,"7"=6,"6"=5,"5"=4,"4"=3,"3"=2,"2"=1,"1"=0)/9,
+    mw_restrict_1 = dplyr::recode(mw_restrict_1, "1"=9,"2"=8,"3"=7,"4"=6,"5"=5,"6"=4,"7"=3,"8"=2,"9"=1,"10"=0)/9,
+    minwage_text_r = (25-as.numeric(minwage_text))/25,
+  )
+a$minwage_text_r[as.numeric(a$minwage_text)>25] <- NA
+
+a <- a %>% 
+  rowwise() %>%
+  mutate(mw_index = mean(c(minwage15, rtwa_v1, rtwa_v2, mw_support, minwage_howhigh, 
+                           mw_help_1, mw_restrict_1, minwage_text_r), na.rm=T)) %>%
+  ungroup()
+
+# CRONBACH-S ALPHA
+index_fa <- psych::alpha(select(a, minwage15, rtwa_v1, rtwa_v2, mw_support, minwage_howhigh, 
+                                mw_help_1, mw_restrict_1, minwage_text_r), check.keys = T)
+
+write.csv(data.frame(cor(select(a, minwage15, rtwa_v1, rtwa_v2, mw_support, minwage_howhigh, 
+                                mw_help_1, mw_restrict_1, minwage_text_r), use = "complete.obs")),
+          row.names = T,file = "../results/cormat_mw_index_w2.csv")
+
+pdf("corrplot_mwindex_w2.pdf")
+a_corrplot <- corrplot::corrplot(cor(select(a, minwage15, rtwa_v1, rtwa_v2, mw_support, 
+                                            minwage_howhigh, mw_help_1, mw_restrict_1, minwage_text_r), 
+                                     use = "complete.obs"),method = "shade")
+dev.off()
+
+(alpha <- index_fa$total["raw_alpha"]) # 0.9582061
+
+### SURVEY PREPROCESSING DONE ###
+
+## YTRECS SESSION DATA -------------------------------------------------------
+ytrecs <- read_rds("../data/shorts/ytrecs_sessions_may2024.rds") %>%
+  clean_names() %>%
+  as_tibble()
+
+## EXTRACTING TOPICID AND URLID
+a <- a %>%
+  ungroup() %>%
+  mutate(
+    topic_id = str_extract(video_link, "topicid=([a-z]{2}[1-6])") %>% str_replace("topicid=", ""),
+    urlid = str_extract(video_link, "id=(mt_\\d+)") %>% str_replace("id=", "")
+  )
+
+## USING THE FIRST SESSION AS THE VALID ONE IF A PERSON HAS MULTIPLE ATTEMPTS
+ytrecs <- ytrecs %>%
+  group_by(topic_id, urlid) %>% 
+  mutate(dupes = n(),
+         first_session = ifelse(row_number() == 1, 1, 0)
+  ) %>%
+  filter(first_session == 1) # using the first session as valid one
+
+a <- left_join(a, ytrecs,by=c("topic_id","urlid"))
+
+## EXTRACTING TREATMENT ARM
+extract_treatmentarm <- function(url) {
+  pattern <- "topicid=([a-z]{2})" #[a-z]{2}[1-6]
+  match <- str_match(url, pattern)
+  if (!is.na(match[2])) {
+    return(match[2])
+  } else {
+    return(NA)
+  }
+}
+
+# APPLY THE FUNCTION TO THE VIDEO_LINK COLUMN
+a <- a %>%
+  rowwise() %>%
+  mutate(treatment_arm = extract_treatmentarm(video_link)) %>%
+  ungroup()
+
+write_csv(a, "../results/intermediate data/shorts/qualtrics_w12_clean_ytrecs_may2024.csv")
+rm(list = ls())
+
+### PREPROCESSING DONE ----------------------

code/shorts/06_analysis_multipletesting.R

@@ -0,0 +1,667 @@
+cat(rep('=', 80),
+    '\n\n',
+    'OUTPUT FROM: shorts/06_analysis_multipletesting.R',
+    '\n\n',
+    sep = ''
+    )
+    
+library(data.table)
+library(car)
+library(sandwich)
+library(lmtest)
+library(ggplot2)
+library(tidyverse)
+
+###############
+## functions ##
+###############
+
+`%.%` <- paste0
+
+simes <- function(ps){
+  min(sort(length(ps) * ps / rank(ps)))
+}
+
+### functions to handle inconsistent interaction ordering of mlm() ###
+
+## convert interaction terms of form 'b#:a#' to 'a#:b#'
+reorder.interaction.names <- function(x, prefix = ''){
+  x <- gsub('^' %.% prefix, '', x)
+  sapply(strsplit(x, ':'),
+         function(y){
+           paste(sort(y), collapse = ':')
+         })
+}
+
+## take term of form 'a1:b1', look up in vector of form 'b#:a#, return 'b1:a1'
+convert.interaction.names <- function(x, y, prefix.y = ''){
+  ind <- match(reorder.interaction.names(x),
+               reorder.interaction.names(y, prefix = prefix.y)
+  )
+  return(y[ind])
+}
+
+## modified from print.linearHypothesis.mlm to use alternate df & return pvals
+##   (print method is responsible for doing the actual computation of pvals)
+extract.lht <- function(x,
+                        SSP = TRUE,
+                        SSPE = SSP,
+                        digits = getOption('digits'),
+                        df.residual = x$df.residual
+){
+  test <- x$test
+  if (!is.null(x$P) && SSP) {
+    P <- x$P
+    cat("\n Response transformation matrix:\n")
+    attr(P, "assign") <- NULL
+    attr(P, "contrasts") <- NULL
+    print(P, digits = digits)
+  }
+  if (SSP) {
+    cat("\nSum of squares and products for the hypothesis:\n")
+    print(x$SSPH, digits = digits)
+  }
+  if (SSPE) {
+    cat("\nSum of squares and products for error:\n")
+    print(x$SSPE, digits = digits)
+  }
+  if ((!is.null(x$singular)) && x$singular) {
+    warning("the error SSP matrix is singular; multivariate tests are unavailable")
+    return(invisible(x))
+  }
+  SSPE.qr <- qr(x$SSPE)
+  eigs <- Re(eigen(qr.coef(SSPE.qr, x$SSPH), symmetric = FALSE)$values)
+  tests <- matrix(NA, 4, 4)
+  rownames(tests) <- c("Pillai", "Wilks", "Hotelling-Lawley",
+                       "Roy")
+  if ("Pillai" %in% test)
+    tests[1, 1:4] <- car:::Pillai(eigs, x$df, df.residual)
+  if ("Wilks" %in% test)
+    tests[2, 1:4] <- car:::Wilks(eigs, x$df, df.residual)
+  if ("Hotelling-Lawley" %in% test)
+    tests[3, 1:4] <- car:::HL(eigs, x$df, df.residual)
+  if ("Roy" %in% test)
+    tests[4, 1:4] <- car:::Roy(eigs, x$df, df.residual)
+  tests <- na.omit(tests)
+  ok <- tests[, 2] >= 0 & tests[, 3] > 0 & tests[, 4] > 0
+  ok <- !is.na(ok) & ok
+  tests <- cbind(x$df, tests, pf(tests[ok, 2], tests[ok, 3],
+                                 tests[ok, 4], lower.tail = FALSE))
+  colnames(tests) <- c("Df", "test stat", "approx F", "num Df",
+                       "den Df", "Pr(>F)")
+  tests <- structure(as.data.frame(tests),
+                     heading = paste("\nMultivariate Test",
+                                     if (nrow(tests) > 1)
+                                       "s", ": ", x$title, sep = ""),
+                     class = c("anova",
+                               "data.frame"
+                     )
+  )
+  return(tests)
+}
+
+###############
+## load data ##
+###############
+
+d <- fread('../results/intermediate data/shorts/qualtrics_w12_clean_ytrecs_may2024.csv')
+
+##############
+## controls ##
+##############
+
+platform.controls <- c('age_cat',
+                       'male',
+                       'pol_interest',
+                       'freq_youtube')
+
+mwpolicy.controls <- 'mw_index_pre'
+
+media.controls <- c('trust_majornews',
+                    'trust_youtube',
+                    'fabricate_majornews',
+                    'fabricate_youtube')
+
+affpol.controls <- c('affpol_smart',
+                     'affpol_comfort')
+
+controls.raw <- unique(c(platform.controls,
+                         mwpolicy.controls,
+                         media.controls,
+                         affpol.controls))
+
+## transform control variables by creating dummies and demeaning
+controls.trans <- list()
+for (j in controls.raw){
+  ## convert to dummies if needed
+  controls.j <- model.matrix(as.formula('~ 0 + ' %.% j),
+                             model.frame(as.formula('~ 0 + ' %.% j),
+                                         data = d,
+                                         na.action = 'na.pass'
+                             )
+  )
+  ## demean by column
+  controls.j <- sweep(controls.j,
+                      MARGIN = 2,
+                      STATS = colMeans(controls.j, na.rm = TRUE),
+                      FUN = `-`,
+  )
+  colnames(controls.j) <- make.names(colnames(controls.j))
+  ## remove control from original data
+  d[[j]] <- NULL
+  ## reinsert transformed control
+  d <- cbind(d, controls.j)
+  ## keep track of which original controls map to which transformed controls
+  controls.trans[[j]] <- colnames(controls.j)
+}
+
+## map original control variables to transformed versions
+platform.controls <- unlist(controls.trans[platform.controls])
+mwpolicy.controls <- unlist(controls.trans[mwpolicy.controls])
+media.controls <- unlist(controls.trans[media.controls])
+affpol.controls <- unlist(controls.trans[affpol.controls])
+
+### Platform interactions ###
+d <- d %>% filter(!is.na(interface_duration)) # -- 929 observations
+
+##############
+## outcomes ##
+##############
+
+### HYPOTHESIS FAMILY: MIN WAGE POLICY ATTITUDES ###
+
+## ONLY HAVE ONE OUTCOME
+mwpolicy.outcomes <- 'mw_index'
+
+outcomes <- unique(c(mwpolicy.outcomes))
+
+################
+## treatments ##
+################
+
+## CREATE ATTITUDE DUMMIES
+# 1-LIBERALS, 2-MODERATES, 3-CONSERVATIVES
+d[, attitude := c('pro', 'neutral', 'anti')[thirds]]
+d[, attitude.pro := as.numeric(attitude == 'pro')]
+d[, attitude.neutral := as.numeric(attitude == 'neutral')]
+d[, attitude.anti := as.numeric(attitude == 'anti')]
+
+## CREATE SEQUENCE DUMMIES -- AC, PC, AI, PI
+d[, recsys.ac := as.numeric(treatment_arm %like% 'ac')]
+d[, recsys.pc := as.numeric(treatment_arm %like% 'pc')]
+d[, recsys.ai := as.numeric(treatment_arm %like% 'ai')]
+d[, recsys.pi := as.numeric(treatment_arm %like% 'pi')]
+
+# (a)  Increasing vs.  Constant assignment among Pro participants;
+# (b)  Increasing vs.  Constant assignment among Anti participants;
+# (c)  Increasing  vs.   Constant  assignment  among  Moderate  participants  assigned  to  a  Prosequence;
+# (d)  Increasing vs.  Constant assignment among moderate participants assigned to an Antisequence;
+# (e)  Pro vs.  Anti sequence assignment among moderate participants with Increasing assignment;
+# (f)  Pro vs.  Anti seed among moderate participants with Constant assignment.
+
+# Treatments:
+treatments <- c('attitude.pro:recsys.pi', # (a)
+                'attitude.pro:recsys.pc', # (a)
+                'attitude.anti:recsys.ai', # (b)
+                'attitude.anti:recsys.ac', # (b)
+                'attitude.neutral:recsys.ai', # (d-e)
+                'attitude.neutral:recsys.pi', # (c-e)
+                'attitude.neutral:recsys.ac', # (d-f)
+                'attitude.neutral:recsys.pc') # (c-f)
+
+# Contrasts:
+contrasts <- rbind(
+  # Increasing vs. Constant assignment among Pro participants
+  i = c(treat = 'attitude.pro:recsys.pi',
+        ctrl = 'attitude.pro:recsys.pc'
+  ),
+  # Increasing vs. Constant assignment among Anti participants
+  ii = c(treat = 'attitude.anti:recsys.ai',
+         ctrl = 'attitude.anti:recsys.ac'
+  ),
+  # Increasing vs. Constant assignment among Moderate participants assigned to a Pro sequence
+  iii = c(treat = 'attitude.neutral:recsys.pi',
+          ctrl = 'attitude.neutral:recsys.pc'
+  ),
+  # Increasing vs. Constant assignment among moderate participants assigned to an Anti sequence
+  iv = c(treat = 'attitude.neutral:recsys.ai',
+         ctrl = 'attitude.neutral:recsys.ac'
+  ),
+  # Pro vs. Anti sequence assignment among moderate participants with Increasing assignment
+  v = c(treat = 'attitude.neutral:recsys.ai',
+        ctrl = 'attitude.neutral:recsys.pi'
+  ),
+  # Pro vs. Anti sequence assignment among moderate participants with Constant assignment
+  vi = c(treat = 'attitude.neutral:recsys.ac',
+         ctrl = 'attitude.neutral:recsys.pc'
+  )
+)
+
+##########################
+## hierarchical testing ##
+##########################
+
+## initialize top layer p-values:
+## does treatment have any effect on any outcome in family
+families <- c('mwpolicy')
+layer1.pvals <- rep(NA_real_, length(families))
+layer1.notes <- rep('', length(families))
+names(layer1.pvals) <- families
+
+## initialize 2nd layer p-values:
+##   which treatment has detectable effect?
+contrast.pvals <- rep(NA_real_, nrow(contrasts))
+names(contrast.pvals) <- paste(contrasts[, 'treat'],
+                               contrasts[, 'ctrl'],
+                               sep = '.vs.'
+)
+layer2.pvals <- list( mwpolicy = contrast.pvals)
+rm(contrast.pvals)
+
+## initialize 3rd layer p-values:
+##   on which specific outcome in family?
+layer3.pvals <- list()
+layer3.ests <- list()
+layer3.ses <- list()
+layer3.notes <- list()
+for (i in 1:length(families)){
+  family <- families[i]
+  layer3.pvals[[family]] <- list()
+  layer3.ests[[family]] <- list()
+  layer3.ses[[family]] <- list()
+  layer3.notes[[family]] <- list()
+  outcomes <- get(family %.% '.outcomes')
+  for (j in 1:nrow(contrasts)){
+    contrast <- paste(contrasts[j, 'treat'],
+                      contrasts[j, 'ctrl'],
+                      sep = '.vs.'
+    )
+    layer3.pvals[[family]][[contrast]] <- numeric(0)
+    layer3.ests[[family]][[contrast]] <- numeric(0)
+    layer3.ses[[family]][[contrast]] <- numeric(0)
+    for (k in 1:length(outcomes)){
+      outcome <- outcomes[k]
+      layer3.pvals[[family]][[contrast]][outcome] <- NA_real_
+      layer3.ests[[family]][[contrast]][outcome] <- NA_real_
+      layer3.ses[[family]][[contrast]][outcome] <- NA_real_
+      layer3.notes[[family]][outcome] <- ''
+    }
+  }
+}
+
+### begin nested analyses ###
+for (i in 1:length(families)){
+  
+  family <- families[i]
+  family.outcomes <- get(family %.% '.outcomes')
+  family.controls <- get(family %.% '.controls')
+
+  
+  family.controls.interactions <- as.character(
+    outer(treatments,
+          family.controls,
+          FUN = function(x, y) x %.% ':' %.% y
+    )
+  )
+  
+  family.formula <-
+    'cbind(' %.%                # outcomes
+    paste(family.outcomes,
+          collapse = ', '
+    ) %.%  ') ~\n0 +\n' %.%
+    paste(treatments,           # treatments (base terms)
+          collapse = ' +\n'
+    ) %.% ' +\n' %.%
+    paste(family.controls,      # controls (base terms)
+          collapse = ' +\n'
+    )##  %.% ' +\n' %.%
+  ## paste(                      # treat-ctrl interactions
+  ##   family.controls.interactions,
+  ##   collapse = ' +\n'
+  ## )
+  
+  cat(rep('=', 80),
+      '\n\nHYPOTHESIS FAMILY: ',
+      family,
+      '\n\nrunning mlm:\n\n',
+      family.formula,
+      '\n\n',
+      sep = ''
+  )
+  
+  ## run model
+  family.mod <- lm(family.formula, d)
+  
+  ## hack to eliminate NA coefs
+  if (any(is.na(coef(family.mod)))){
+    if ('mlm' %in% class(family.mod)){
+      drop <- rownames(coef(family.mod))[is.na(coef(family.mod))[, 1]]
+    } else {
+      drop <- names(coef(family.mod))[is.na(coef(family.mod))]
+    }
+    drop <- convert.interaction.names(drop,
+                                      c(family.controls,
+                                        family.controls.interactions
+                                      )
+    )
+    layer1.notes[[i]] <-
+      layer1.notes[[i]] %.%
+      'dropped the following coefs: ' %.%
+      paste(drop, sep = ', ') %.%
+      '\n\n'
+    family.formula <- gsub(
+      '\\s+\\+\\s+(' %.% paste(drop, collapse = '|') %.% ')',
+      '',
+      family.formula
+    )
+    family.mod <- lm(family.formula, d)
+  }
+  
+  family.vcov <- vcovHC(family.mod)
+  if (is.null(dim(coef(family.mod)))){
+    coef.names <- names(coef(family.mod))
+  } else {
+    coef.names <- rownames(coef(family.mod))
+  }
+  
+  ### top layer: test overall significance of all contrasts on all outcomes ###
+  ## convert interaction terms to whatever mlm() named it
+  treats <- convert.interaction.names(contrasts[, 'treat'], coef.names)
+  ctrls <- convert.interaction.names(contrasts[, 'ctrl'], coef.names)
+  
+  ## test jointly
+  lht.attempt <- tryCatch({
+    if ('mlm' %in% class(family.mod)){
+      contrast.lht <- linearHypothesis(
+        family.mod,
+        vcov. = family.vcov,
+        hypothesis.matrix = sprintf('%s - %s', treats, ctrls),
+        rhs = matrix(0, nrow = nrow(contrasts), ncol = length(family.outcomes)),
+        test = 'Pillai'
+      )
+      layer1.pvals[[i]] <- extract.lht(contrast.lht)[, 'Pr(>F)']
+    } else {
+      contrast.lht <- linearHypothesis(
+        family.mod,
+        vcov. = family.vcov,
+        hypothesis.matrix = sprintf('%s - %s', treats, ctrls),
+        rhs = matrix(0, nrow = nrow(contrasts), ncol = length(family.outcomes)),
+        test = 'F'
+      )
+      layer1.pvals[[i]] <- contrast.lht[['Pr(>F)']][2]
+    }
+  },
+  error = function(e){
+    warning(sprintf('caught error in %s family:', family), e)
+    ## return error as string for inclusion in notes
+    'caught error: ' %.%
+      e %.%
+      '\n\n'
+  })
+  if (lht.attempt %like% 'caught error'){
+    layer1.notes[[i]] <-
+      layer1.notes[[i]] %.% lht.attempt
+  }
+
+  ### layer 2: test each contrast individually on all outcomes ###
+  
+  for (j in 1:nrow(contrasts)){
+    ## test group equality on all outcomes
+    if ('mlm' %in% class(family.mod)){
+      contrast.lht <-
+        linearHypothesis(
+          family.mod,
+          vcov. = family.vcov,
+          hypothesis.matrix = sprintf('%s - %s', treats[j], ctrls[j]),
+          rhs = matrix(0, nrow = 1, ncol = length(family.outcomes)),
+          test = 'Pillai'
+        )
+      layer2.pvals[[i]][j] <- extract.lht(contrast.lht)[, 'Pr(>F)']
+    } else {
+      contrast.lht <- linearHypothesis(
+        family.mod,
+        vcov. = family.vcov,
+        hypothesis.matrix = sprintf('%s - %s', treats[j], ctrls[j]),
+        rhs = matrix(0, nrow = 1, ncol = length(family.outcomes)),
+        test = 'F'
+      )
+      layer2.pvals[[i]][j] <- contrast.lht[['Pr(>F)']][2]
+    }
+  }
+  
+  ### layer 3: test each contrast on each outcome individually ###
+  
+  for (k in 1:length(family.outcomes)){
+    
+    outcome <- family.outcomes[k]
+    
+    outcome.formula <-
+      outcome %.% ' ~\n0 +\n' %.%
+      paste(treatments,         # treatments (base terms)
+            collapse = ' +\n'
+      ) %.% ' +\n' %.%
+      paste(family.controls,      # controls (base terms)
+            collapse = ' +\n'
+      )##  %.% ' +\n' %.%
+    ## paste(                      # treat-ctrl interactions
+    ##   family.controls.interactions,
+    ##   collapse = ' +\n'
+    ## )
+    
+    cat(rep('-', 40), '\n\nrunning lm:\n\n', outcome.formula, '\n\n', sep = '')
+    
+    outcome.mod <- lm(outcome.formula, d)
+    ## hack to eliminate NA coefs
+    if (any(is.na(coef(outcome.mod)))){
+      drop <- names(coef(outcome.mod))[is.na(coef(outcome.mod))]
+      drop <- convert.interaction.names(drop,
+                                        c(family.controls,
+                                          family.controls.interactions
+                                        )
+      )
+      layer3.notes[[i]][k] <-
+        layer3.notes[[i]][k] %.%
+        'dropped the following coefs: ' %.%
+        paste(drop, sep = ', ') %.%
+        '\n\n'
+      outcome.formula <- gsub(
+        '\\s+\\+\\s+(' %.% paste(drop, collapse = '|') %.% ')',
+        '',
+        outcome.formula
+      )
+      outcome.mod <- lm(outcome.formula, d)
+    }
+    
+    outcome.vcov <- vcovHC(outcome.mod)
+    if (any(!is.finite(outcome.vcov))){
+      outcome.vcov <- vcov(outcome.mod)
+      layer3.notes[[i]][k] <-
+        layer3.notes[[i]][k] %.%
+        'falling back to non-robust vcov\n\n'
+    }
+    coef.names <- names(coef(outcome.mod))
+    
+    for (j in 1:nrow(contrasts)){
+      
+      ## convert this interaction term to whatever llm() named it
+      treat <- convert.interaction.names(contrasts[j, 'treat'], coef.names)
+      ctrl <- convert.interaction.names(contrasts[j, 'ctrl'], coef.names)
+      ## test group equality on this outcome
+  
+      
+      contrast.lht <- linearHypothesis(
+        outcome.mod,
+        vcov. = outcome.vcov,
+        hypothesis.matrix = sprintf('%s - %s', treat, ctrl),
+        test = 'F'
+      )
+      
+      layer3.pvals[[i]][[j]][k] <- contrast.lht[['Pr(>F)']][2]
+      layer3.ests[[i]][[j]][k] <- (
+        coef(outcome.mod)[treat] - coef(outcome.mod)[ctrl]
+      ) ## * attr(d[[outcome]], 'scaled:scale')  # note: uncomment if rescaling
+      layer3.ses[[i]][[j]][k] <- sqrt(
+        outcome.vcov[treat, treat] +
+          outcome.vcov[ctrl, ctrl] -
+          2 * outcome.vcov[treat, ctrl]
+      )
+    }
+  }
+}
+
+#################################
+## multiple testing correction ##
+#################################
+
+thresh <- .05
+
+## if layer-1 f-test is infeasible for a family due to collinearity,
+##   obtain layer-1 p-values for that family by simes
+for (i in which(is.na(layer1.pvals))){
+  layer1.pvals[i] <- simes(layer2.pvals[[i]])
+}
+
+## multiple testing adjustment for layer 1
+layer1.pvals.adj <- p.adjust(layer1.pvals, 'BH')
+layer1.nonnull.prop <- mean(layer1.pvals.adj < thresh)
+
+## test layer-2 hypotheses only if layer 1 passes
+layer2.pvals.adj <- layer2.pvals  # start by copying unadjusted layer-2 p-values
+layer2.nonnull.prop <- rep(NA, length(layer1.pvals.adj))
+names(layer2.nonnull.prop) <- names(layer1.pvals.adj)
+for (i in 1:length(layer1.pvals)){
+  if (layer1.pvals.adj[i] < thresh){  # if layer 1 passes
+    ## adjust for multiplicity within layer 2...
+    layer2.pvals.adj[[i]] <- p.adjust(layer2.pvals[[i]], 'BH')
+    ## ... and inflate to account for selection at layer 1
+    layer2.pvals.adj[[i]] <-
+      pmin(layer2.pvals.adj[[i]] / layer1.nonnull.prop, 1)
+    ## keep track of selection at layer 2 for use in layer 3
+    layer2.nonnull.prop[i] <- mean(layer2.pvals.adj[[i]] < thresh)
+  } else {  # if layer 1 fails
+    layer2.pvals.adj[[i]] <- rep(NA_real_, length(layer2.pvals[[i]]))
+    names(layer2.pvals.adj[[i]]) <- names(layer2.pvals[[i]])
+  }
+}
+
+## test layer-3 hypotheses only if layers 1 & 2 pass
+layer3.pvals.adj <- layer3.pvals  # start by copying unadjusted layer-3 p-values
+for (i in 1:length(layer1.pvals.adj)){
+  for (j in 1:length(layer2.pvals.adj[[i]])){
+    ##
+    if (layer1.pvals.adj[i] < thresh &&      # if layer 1 passes...
+        layer2.pvals.adj[[i]][j] < thresh  # ... and if layer 2 passes
+    ){
+      ## adjust for multiplicity within layer 3...
+      layer3.pvals.adj[[i]][[j]] <- p.adjust(layer3.pvals[[i]][[j]], 'BH')
+      ## ... and inflate to account for selection at layer 1
+      layer3.pvals.adj[[i]][[j]] <- pmin(
+        layer3.pvals.adj[[i]][[j]] / layer1.nonnull.prop / layer2.nonnull.prop[i],
+        1
+      )
+    } else {
+      layer3.pvals.adj[[i]][[j]] <- rep(NA_real_, length(layer3.pvals[[i]][[j]]))
+      names(layer3.pvals.adj[[i]][[j]]) <- names(layer3.pvals[[i]][[j]])
+    }
+  }
+}
+
+pvals.adj <- data.table(layer1 = character(0),
+                        layer2 = character(0),
+                        layer3 = character(0),
+                        p.adj = numeric(0),
+                        est = numeric(0),
+                        se = numeric(0)
+)
+for (i in 1:length(layer1.pvals.adj)){
+  pvals.adj <- rbind(pvals.adj,
+                     data.table(layer1 = names(layer1.pvals.adj)[i],
+                                layer2 = 'overall',
+                                layer3 = 'overall',
+                                p.adj = layer1.pvals.adj[i],
+                                est = NA_real_,
+                                se = NA_real_
+                     )
+  )
+  for (j in 1:length(layer2.pvals.adj[[i]])){
+    pvals.adj <- rbind(pvals.adj,
+                       data.table(layer1 = names(layer1.pvals.adj)[i],
+                                  layer2 = names(layer2.pvals.adj[[i]])[j],
+                                  layer3 = 'overall',
+                                  p.adj = layer2.pvals.adj[[i]][j],
+                                  est = NA_real_,
+                                  se = NA_real_
+                       )
+    )
+    for (k in 1:length(layer3.pvals.adj[[i]][[j]])){
+      pvals.adj <- rbind(pvals.adj,
+                         data.table(layer1 = names(layer1.pvals.adj)[i],
+                                    layer2 = names(layer2.pvals.adj[[i]])[j],
+                                    layer3 = names(layer3.pvals.adj[[i]][[j]])[k],
+                                    p.adj = layer3.pvals.adj[[i]][[j]][k],
+                                    est = layer3.ests[[i]][[j]][k],
+                                    se = layer3.ses[[i]][[j]][k]
+                         )
+      )
+    }
+  }
+}
+
+## write out
+fwrite(pvals.adj, '../results/padj_basecontrol_may2024.csv')
+
+
+## prettify for reading
+pvals.adj.pretty <- pvals.adj
+colnames(pvals.adj.pretty) <- gsub('layer1',
+                                   'layer1_hypothesisfamily',
+                                   colnames(pvals.adj.pretty)
+)
+colnames(pvals.adj.pretty) <- gsub('layer2',
+                                   'layer2_treatmentcontrast',
+                                   colnames(pvals.adj.pretty)
+)
+colnames(pvals.adj.pretty) <- gsub('layer3',
+                                   'layer3_specificoutcome',
+                                   colnames(pvals.adj.pretty)
+)
+
+pvals.adj.pretty[, layer2_treatmentcontrast := gsub(
+  'attitude\\.(pro|anti|neutral)(:assg\\.(inc|cons))?:recsys.(ca|cp|ip|ia)',
+  '\\1 \\3 \\4',
+  layer2_treatmentcontrast
+)]
+pvals.adj.pretty[, layer2_treatmentcontrast := gsub(
+  '.vs.',
+  ' - ',
+  layer2_treatmentcontrast,
+  fixed = TRUE
+)]
+pvals.adj.pretty[, layer2_treatmentcontrast := gsub(
+  ' +',
+  ' ',
+  layer2_treatmentcontrast
+)]
+fwrite(pvals.adj.pretty,
+       '../results/padj_basecontrol_pretty_ytrecs_may2024.csv'
+)
+
+# pvals.adj.pretty[p.adj < .05 & layer3_specificoutcome != 'overall',]
+
+################################
+######### OMNIBUS TEST #########
+################################
+
+# Step 1: Create a binary variable indicating increasing condition
+d$is_increasing <- ifelse(d$treatment_arm == "pi" | d$treatment_arm == "ai", 1, 0)
+
+# Step 2: Reverse values for individuals in the Pro condition
+d$mw_index_pre[d$treatment_arm %like% "pi|pc"] <- 1 - d$mw_index_pre[d$treatment_arm %like% "pi|pc"]
+d$mw_index[d$treatment_arm %like% "pi|pc"] <- 1 - d$mw_index[d$treatment_arm %like% "pi|pc"]
+
+# Step 3: Perform the linear regression (omnibus test)
+model <- lm(I(mw_index - mw_index_pre) ~ is_increasing, data = d)
+
+# View the summary of the model
+summary(model)
+rm(list = ls())

code/shorts/07_postprocessing_exploration.R

@@ -0,0 +1,622 @@
+cat(rep('=', 80),
+    '\n\n',
+    'OUTPUT FROM: shorts/07_postprocessing_exploration.R',
+    '\n\n',
+    sep = ''
+    )
+
+## Extremizing Sequences and Minimum Wage Opinions
+## Data collected May 2024 via MTurk/CloudResearch
+## Analysis for the Extremizing Sequences Experiment
+
+## Preamble ----------------------------
+library(tidyverse)
+library(janitor)
+library(lubridate)
+library(stargazer)
+library(broom)
+library(psych)
+
+w12 <- read_csv("../results/intermediate data/shorts/qualtrics_w12_clean_ytrecs_may2024.csv")
+
+## SAMPLE SIZE AND CRONBACH'S ALPHA ------------------
+
+# SAMPLE SIZE
+w12 %>%
+  filter(!is.na(treatment_arm)) %>%
+  count() %>%
+  as.integer() %>%
+  format(big.mark = ',')
+
+# CRONBACH'S ALPHA ON POLICY INDEX
+w12 %>%
+  select(minwage15_pre,
+         rtwa_v1_pre,
+         rtwa_v2_pre,
+         mw_support_pre,
+         minwage_howhigh_pre,
+         mw_help_pre_1,
+         mw_restrict_pre_1,
+         minwage_text_r_pre
+  ) %>%
+  alpha() %>%
+  `[[`('total') %>%
+  `[`('raw_alpha') %>%
+  as.numeric() %>%
+  format(digits = 2, nsmall = 2) %>%
+  paste0('%') %>%  # trailing comment char to prevent latex import issue
+  writeLines('../results/alpha_study4.txt')
+
+
+# FACTOR ANALYSIS WITH VARIMAX ROTATION (PRE)
+pca2 <- psych::principal(select(w12, minwage15_pre, rtwa_v1_pre, 
+                                rtwa_v2_pre, mw_support_pre, minwage_howhigh_pre, 
+                                mw_help_pre_1, mw_restrict_pre_1, minwage_text_r_pre),
+                         rotate="varimax",
+                         nfactors=1
+)
+pc2 <- pca2$Vaccounted[2]
+writeLines(as.character(round(pc2, 2)),con = "../results/outcomes_pc2_study4_pre.tex",sep = "%")
+
+
+# FACTOR ANALYSIS WITH VARIMAX ROTATION (POST)
+pca2 <- psych::principal(
+  select(w12, minwage15, rtwa_v1, rtwa_v2, mw_support, minwage_howhigh, mw_help_1, 
+         mw_restrict_1, minwage_text_r),
+  rotate="varimax",
+  nfactors=1
+)
+pc2 <- pca2$Vaccounted[2]
+writeLines(as.character(round(pc2, 2)),con = "../results/outcomes_pc2_study4_post.tex",sep = "%")
+
+
+## BASIC DESCRIPTIVE FIGURES ------------------
+
+## TIME SPENT DURING THE SURVEY
+(surveytime_plot <- ggplot(w12) +
+   geom_histogram(aes(x=survey_time,y=..density../sum(..density..))) +
+   scale_x_continuous("Overall survey time taken (minutes)",
+                       breaks=seq(0,100,10),
+                       limits=c(-1,100)
+                      ) +
+   scale_y_continuous("Density") +
+   geom_vline(xintercept = mean(w12$survey_time,na.rm=T),lty=3,col="red") +
+   annotate(x=mean(w12$survey_time+1,na.rm=T),y=0.13,geom = "text",
+            label=paste0("Average: ",round(mean(w12$survey_time,na.rm=T),0)," minutes"),hjust=0) +
+   geom_vline(xintercept = median(w12$survey_time,na.rm=T),lty=2,col="red") +
+   annotate(x=median(w12$survey_time+1,na.rm=T),y=0.16,geom = "text",
+            label=paste0("Median: ",round(median(w12$survey_time,na.rm=T),0)," minutes"),hjust=0) +
+   theme_minimal()
+)
+
+## TIME SPENT ON THE INTERFACE
+(ytrecstime_plot <- ggplot(w12) +
+    geom_histogram(aes(x=interface_duration/60,y=..density../sum(..density..))) +
+    scale_x_continuous("Interface Time Taken (minutes)",
+                       breaks=seq(0,80,10),
+                       limits=c(-1,70)) +
+    scale_y_continuous("Density") +
+    geom_vline(xintercept = mean(w12$interface_duration/60,na.rm=T),lty=3,col="red") +
+    annotate(x=mean(w12$interface_duration/60+1,na.rm=T),y=0.1,geom = "text",
+             label=paste0("Average: ",round(mean(w12$interface_duration/60,na.rm=T),0)," minutes"),hjust=0) +
+    geom_vline(xintercept = median(w12$interface_duration/60,na.rm=T),lty=2,col="red") +
+    annotate(x=median(w12$interface_duration/60+1,na.rm=T),y=0.13,geom = "text",
+             label=paste0("Median: ",round(median(w12$interface_duration/60,na.rm=T),0)," minutes"),hjust=0) +
+    theme_minimal()
+)
+
+## PRE OPINIONS OVERALL
+(hist_mwindex <- ggplot(w12) +
+    geom_histogram(aes(x=mw_index_pre)) +
+    scale_x_continuous("Minimum Wage Opinions Index, Pre") +
+    scale_y_continuous("Count",limits=c(-5,200)) +
+    annotate(x = 0.92,y=-3,geom = "text",label="More conservative\nopinions",col="red",hjust=1,size=3,lineheight=0.75) +
+    annotate(x = 0.98,xend=1,y=-3,yend=-3,geom = "segment",arrow=arrow(type = "closed",angle = 20),col="red") +
+    annotate(x = 0.08,y=-3,geom = "text",label="More liberal\nopinions",col="blue",hjust=0,size=3,lineheight=0.75) +
+    annotate(x = 0.02,xend=0.00,y=-3,yend=-3,geom = "segment",arrow=arrow(type = "closed",angle = 20),col="blue") +
+    theme_minimal()
+)
+
+## PRE OPINION BY TERCILE
+(hist_mwindex_thirds <- ggplot(w12,aes(x=mw_index_pre)) +
+    geom_histogram(data=filter(w12,thirds==1),aes(x=mw_index_pre),fill="blue") +
+    geom_histogram(data=filter(w12,thirds==2),aes(x=mw_index_pre),fill="grey") +
+    geom_histogram(data=filter(w12,thirds==3),aes(x=mw_index_pre),fill="red") +
+    scale_x_continuous("Minimum Wage Opinions Index, Pre") +
+    scale_y_continuous("Count",limits=c(-5,200)) +
+    annotate(x = 0.92,y=-5,geom = "text",label="More conservative\nopinions",col="red",hjust=1,size=3,lineheight=0.75) +
+    annotate(x = 0.98,xend=1,y=-5,yend=-5,geom = "segment",arrow=arrow(type = "closed",angle = 20),col="red") +
+    annotate(x = 0.08,y=-5,geom = "text",label="More liberal\nopinions",col="blue",hjust=0,size=3,lineheight=0.75) +
+    annotate(x = 0.02,xend=0.00,y=-5,yend=-5,geom = "segment",arrow=arrow(type = "closed",angle = 20),col="blue") +
+    theme_minimal()
+)
+
+(hist_mwindex_thirds_nocolor <- ggplot(w12,aes(x=mw_index_pre)) +
+    geom_histogram(data=filter(w12,thirds==1),aes(x=mw_index_pre),fill="grey") +
+    geom_histogram(data=filter(w12,thirds==2),aes(x=mw_index_pre),fill="grey") +
+    geom_histogram(data=filter(w12,thirds==3),aes(x=mw_index_pre),fill="grey") +
+    scale_x_continuous("Minimum Wage Opinions Index, W1") +
+    scale_y_continuous("Count",limits=c(-5,200)) +
+    annotate(x = 0.92,y=-5,geom = "text",label="More conservative\nopinions",col="red",hjust=1,size=3,lineheight=0.75) +
+    annotate(x = 0.98,xend=1,y=-5,yend=-5,geom = "segment",arrow=arrow(type = "closed",angle = 20),col="red") +
+    annotate(x = 0.08,y=-5,geom = "text",label="More liberal\nopinions",col="blue",hjust=0,size=3,lineheight=0.75) +
+    annotate(x = 0.02,xend=0.00,y=-5,yend=-5,geom = "segment",arrow=arrow(type = "closed",angle = 20),col="blue") +
+    theme_minimal()
+)
+
+# SUMMARY PRE OPINIONS FOR EACH CONDITION
+groupsumm_bythirds <- w12 %>%
+  group_by(treatment_arm,thirds) %>%
+  summarize(n = n()) %>%
+  na.omit() %>%
+  mutate(treatment_arm = factor(treatment_arm,levels=c("pc", "pi","ac" , "ai"),
+                                labels = c("Liberal\nconstant",
+                                           "Liberal\nincreasing",
+                                           "Conservative\nconstant",
+                                           "Conservative\nincreasing"),ordered=T),
+         thirds = factor(thirds,levels=c(1,2,3),ordered=T))
+
+groupsumm <- w12 %>%
+  group_by(treatment_arm) %>%
+  summarize(
+    minwage15 = mean(minwage15_pre,na.rm=T),
+    rtwa_v1 =  mean(rtwa_v1_pre, na.rm = T),
+    rtwa_v2 = mean(rtwa_v2_pre, na.rm = T),
+    mw_support = mean(mw_support_pre,na.rm = T),
+    minwage_howhigh = mean(minwage_howhigh_pre, na.rm = T),
+    mw_help_1 = mean(mw_help_pre_1, na.rm = T),
+    mw_restrict_1 = mean(mw_restrict_pre_1,na.rm = T),
+    minwage_text_r = mean(minwage_text_r_pre,na.rm = T),
+    mw_index_pre = mean(mw_index_pre,na.rm = T),
+    n = n()) %>%
+  na.omit() %>%
+  mutate(treatment_arm = factor(treatment_arm,levels=c("pc", 
+                                                       "pi",
+                                                       "ac" , 
+                                                       "ai"),
+                                labels = c("Liberal\nconstant",
+                                           "Liberal\nincreasing",
+                                           "Conservative\nconstant",
+                                           "Conservative\nincreasing"),ordered=T))
+
+# N IN EACH TREATMENT CONDITION
+(plot_hist_n <- ggplot(groupsumm) +
+    geom_bar(aes(x=treatment_arm,y=n),stat="identity") +
+    geom_text(aes(x=treatment_arm,y=n+15,label=n),stat="identity") +
+    scale_x_discrete("Treatment Condition") +
+    scale_y_continuous("N") +
+    theme_minimal()
+)
+
+## N IN EACH TREATMENT CONDITION COLORED BY THIRDS
+(plot_hist_n_bythirds <- ggplot(groupsumm_bythirds) +
+    geom_bar(aes(x=treatment_arm,y=n,fill=thirds),stat="identity") +
+    geom_text(data=groupsumm,aes(x=treatment_arm,y=n+15,label=n),stat="identity") +
+    scale_x_discrete("Treatment Condition") +
+    scale_y_continuous("N") +
+    scale_fill_manual("Tercile of\nPre-Opinion",breaks=c(1,2,3),values=c("blue","grey","red")) +
+    theme_minimal()
+)
+
+## AVERAGE PRE-OPINION ON MINIMUM WAGE INDEX
+(plot_hist_mwindex <- ggplot(groupsumm) +
+    geom_bar(aes(x=treatment_arm,y=mw_index_pre),stat="identity") +
+    scale_x_discrete("Treatment Condition") +
+    scale_y_continuous("Average Pre-Opinion\non Minimum Wage Index",
+                       limits=c(0,0.6),
+                       breaks = seq(0,0.6,0.2),
+                       labels=c("\n0.0\nMore\nliberal\nopinions","0.2","0.4","More\nconservative\nopinions\n0.6\n\n\n")) +
+    theme_minimal() +
+    theme(plot.margin = unit(c(1.75,0.5,0.5,0.5),"lines"))
+)
+
+# SUMMARY FOR EACH CONDITION
+groupsumm <- w12 %>%
+  group_by(treatment_arm) %>%
+  summarize(
+    minwage15 = mean(minwage15,na.rm=T),
+    rtwa_v1 =  mean(rtwa_v1, na.rm = T),
+    rtwa_v2 = mean(rtwa_v2, na.rm = T),
+    mw_support = mean(mw_support,na.rm = T),
+    minwage_howhigh = mean(minwage_howhigh, na.rm = T),
+    mw_help_1 = mean(mw_help_1, na.rm = T),
+    mw_restrict_1 = mean(mw_restrict_1,na.rm = T),
+    minwage_text_r = mean(minwage_text_r,na.rm = T),
+    mw_index = mean(mw_index,na.rm = T),
+    n = n()) %>%
+  na.omit() %>%
+  mutate(treatment_arm = factor(treatment_arm,levels=c("pc", 
+                                                       "pi",
+                                                       "ac" , 
+                                                       "ai"),
+                                labels = c("Liberal\nconstant",
+                                           "Liberal\nincreasing",
+                                           "Conservative\nconstant",
+                                           "Conservative\nincreasing"),
+                                ordered=T))
+
+(plot_hist_mwindex <- ggplot(groupsumm) +
+    geom_bar(aes(x=treatment_arm,y=mw_index),stat="identity") +
+    scale_x_discrete("Treatment Condition") +
+    scale_y_continuous("Average Post-Opinion\non Minimum Wage Index",
+                       limits=c(0,0.6),
+                       breaks = seq(0,0.6,0.2),
+                       labels=c("\n0.0\nMore\nliberal\nopinions","0.2","0.4","More\nconservative\nopinions\n0.6\n\n\n")) +
+    theme_minimal() +
+    theme(plot.margin = unit(c(1.75,0.5,0.5,0.5),"lines"))
+)
+
+## CHANGES IN OPINION BETWEEN WAVES
+treatsumm <- w12 %>%
+  group_by(treatment_arm) %>%
+  summarize(minwage15 = mean(minwage15-minwage15_pre,na.rm=T),
+            rtwa_v1 =  mean(rtwa_v1-rtwa_v1_pre, na.rm = T),
+            rtwa_v2 = mean(rtwa_v2-rtwa_v2_pre, na.rm = T),
+            mw_support = mean(mw_support-mw_support_pre,na.rm = T),
+            minwage_howhigh = mean(minwage_howhigh-minwage_howhigh_pre, na.rm = T),
+            mw_help_1 = mean(mw_help_1-mw_help_pre_1, na.rm = T),
+            mw_restrict_1 = mean(mw_restrict_1-mw_restrict_pre_1,na.rm = T),
+            minwage_text_r = mean(minwage_text_r-minwage_text_r_pre,na.rm = T),
+            mw_index_change = mean(mw_index - mw_index_pre,na.rm = T),
+            n = n()) %>%
+  na.omit() %>%
+  mutate(treatment_arm = factor(treatment_arm,levels=c("pc", 
+                                                       "pi",
+                                                       "ac" , 
+                                                       "ai"),
+                                labels = c("Liberal\nconstant",
+                                           "Liberal\nincreasing",
+                                           "Conservative\nconstant",
+                                           "Conservative\nincreasing"),
+                                ordered=T))
+
+w1w2_corrplot <- corrplot::corrplot(cor(select(w12,
+                                               minwage15_pre, rtwa_v1_pre, rtwa_v2_pre, mw_support_pre, 
+                                               minwage_howhigh_pre, mw_help_pre_1, mw_restrict_pre_1, minwage_text_r_pre,
+                                               minwage15, rtwa_v1, rtwa_v2, mw_support, minwage_howhigh, 
+                                               mw_help_1, mw_restrict_1, minwage_text_r), use = "complete.obs")[1:8,9:16],method = "shade")
+dev.off()
+
+## AVERAGE OPINION CHANGE POST-PRE ON MIN WAGE POLICY INDEX
+(plot_hist_mwindex <- ggplot(treatsumm) +
+    geom_bar(aes(x=treatment_arm,y=mw_index_change),stat="identity") +
+    scale_x_discrete("Treatment Condition") +
+    scale_y_continuous("Average Opinion Change Post-Pre\non Min. Wage Policy Index",
+                       limits=c(-0.2,0.2),
+                       breaks = seq(-0.2,0.2,0.1),
+                       labels=c("\n\n\n-0.2\nLiberal\nopinion\nchange","-0.1","0.00","0.1","Conservative\nopinion\nchange\n0.2\n\n\n")
+                       ) +
+    theme_minimal() +
+    theme(plot.margin = unit(c(1.75,0.5,0.5,0.5),"lines"))
+)
+
+### CHANGE FOR MODERATES
+treatsumm_thirds <- w12 %>%
+  group_by(thirds, treatment_arm) %>%
+  summarize(minwage15 = mean(minwage15-minwage15_pre,na.rm=T),
+            rtwa_v1 =  mean(rtwa_v1-rtwa_v1_pre, na.rm = T),
+            rtwa_v2 = mean(rtwa_v2-rtwa_v2_pre, na.rm = T),
+            mw_support = mean(mw_support-mw_support_pre,na.rm = T),
+            minwage_howhigh = mean(minwage_howhigh-minwage_howhigh_pre, na.rm = T),
+            mw_help_1 = mean(mw_help_1-mw_help_pre_1, na.rm = T),
+            mw_restrict_1 = mean(mw_restrict_1-mw_restrict_pre_1,na.rm = T),
+            minwage_text_r = mean(minwage_text_r-minwage_text_r_pre,na.rm = T),
+            mw_index_change = mean(mw_index - mw_index_pre,na.rm = T),
+            n = n()) %>%
+  na.omit() %>%
+  mutate(treatment_arm = factor(treatment_arm,levels=c("pc", 
+                                                       "pi",
+                                                       "ac" , 
+                                                       "ai"),
+                                labels = c("Liberal\nconstant",
+                                           "Liberal\nincreasing",
+                                           "Conservative\nconstant",
+                                           "Conservative\nincreasing"),
+                                ordered=T))
+
+(plot_hist_mwindex_thirds <- ggplot(treatsumm_thirds %>% filter(thirds == 2)) +
+    geom_bar(aes(x=treatment_arm,y=mw_index_change),stat="identity") +
+    scale_x_discrete("Treatment Condition") +
+    scale_y_continuous("Average Opinion Change Post-Pre\non Min. Wage Policy Index\nfor Moderates",
+                       limits=c(-0.2,0.2),
+                       breaks = seq(-0.2,0.2,0.1),
+                       labels=c("\n\n\n-0.2\nLiberal\nopinion\nchange","-0.1","0.00","0.1","Conservative\nopinion\nchange\n0.2\n\n\n")
+    ) +
+    theme_minimal() +
+    theme(plot.margin = unit(c(1.75,0.5,0.5,0.5),"lines"))
+)
+
+
+## BASE CONTROL FIGURES --------------------------------------
+
+##
+## RUN 04_analysis_multipletesting_basecontrol_may2024.R, THEN READ IN ADJUSTED P-VALUES
+##
+
+coefs_basecontrol <- read_csv("../results/padj_basecontrol_pretty_ytrecs_may2024.csv")
+
+outcome_labels <- data.frame(outcome = c("Minimum wage\nindex"),
+                             specificoutcome = c("mw_index"),
+                             family = c(rep("Policy Attitudes\n(unit scale, + is more conservative)",1)))
+
+
+#### THE effect of INCREASING vs. CONSTANT assignment among LIBERAL participants ####
+coefs_third1_basecontrol <- coefs_basecontrol %>%
+  filter(layer2_treatmentcontrast == "attitude.pro:recsys.pi - attitude.pro:recsys.pc" &
+           layer3_specificoutcome != "overall")
+
+
+coefs_third1_basecontrol$outcome = outcome_labels$outcome[match(coefs_third1_basecontrol$layer3_specificoutcome,
+                                                                outcome_labels$specificoutcome)]
+
+coefs_third1_basecontrol$family = outcome_labels$family[match(coefs_third1_basecontrol$layer3_specificoutcome,outcome_labels$specificoutcome)]
+
+coefs_third1_basecontrol <- mutate(coefs_third1_basecontrol,
+                                   family = factor(family,
+                                                   levels = c("Policy Attitudes\n(unit scale, + is more conservative)"
+                                                     ),ordered = T))
+
+coefs_third1_basecontrol <- coefs_third1_basecontrol %>%
+  mutate(ci_lo_99 = est + qnorm(0.001)*se,
+         ci_hi_99 = est + qnorm(0.995)*se,
+         ci_lo_95 = est + qnorm(0.025)*se,
+         ci_hi_95 = est + qnorm(0.975)*se,
+         ci_lo_90 = est + qnorm(0.05)*se,
+         ci_hi_90 = est + qnorm(0.95)*se,
+         plotorder = nrow(coefs_third1_basecontrol):1
+  )
+
+writeLines(as.character(round(100*abs(filter(coefs_third1_basecontrol,layer3_specificoutcome=="pro_fraction_chosen")$est),0)),
+           con = "../results/beta_recsys_pro_fraction_chosen_third1.tex",sep="%")
+
+
+#### THE effect of INCREASING vs. CONSTANT assignment among LIBERAL participants ####
+(coefplot_third1_basecontrol <- ggplot(filter(coefs_third1_basecontrol),aes(y=plotorder)) +
+    geom_errorbarh(aes(xmin=ci_lo_95,xmax=ci_hi_95),height=0,lwd=0.5) +
+    geom_errorbarh(aes(xmin=ci_lo_90,xmax=ci_hi_90),height=0,lwd=1) +
+    geom_point(aes(x=est),size=1.5) +
+    geom_vline(xintercept = 0,lty=2) +
+    facet_wrap(~family,ncol=1,scales="free") +
+    scale_y_continuous("",
+                       breaks = coefs_third1_basecontrol$plotorder,
+                       labels = coefs_third1_basecontrol$outcome) +
+    scale_x_continuous("Increasing Liberal seed vs. Constant Liberal seed assignment \namong Liberal participants \n(95% and 90% CIs)") +
+    coord_cartesian(xlim=c(-0.2,0.2)) +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"))
+)
+
+#### THE effect of INCREASING vs. CONSTANT assignment among CONSERVATIVE participants ####
+coefs_third3_basecontrol <- coefs_basecontrol %>%
+  filter(layer2_treatmentcontrast == "attitude.anti:recsys.ai - attitude.anti:recsys.ac" &
+           layer3_specificoutcome != "overall")
+
+coefs_third3_basecontrol$outcome = outcome_labels$outcome[match(coefs_third3_basecontrol$layer3_specificoutcome,
+                                                                outcome_labels$specificoutcome)]
+
+coefs_third3_basecontrol$family = outcome_labels$family[match(coefs_third3_basecontrol$layer3_specificoutcome,
+                                                              outcome_labels$specificoutcome)]
+
+coefs_third3_basecontrol <- mutate(coefs_third3_basecontrol,
+                                   family = factor(family,levels = c("Policy Attitudes\n(unit scale, + is more conservative)"
+                                                                     ),ordered = T))
+
+
+coefs_third3_basecontrol <- coefs_third3_basecontrol %>%
+  mutate(ci_lo_99 = est + qnorm(0.001)*se,
+         ci_hi_99 = est + qnorm(0.995)*se,
+         ci_lo_95 = est + qnorm(0.025)*se,
+         ci_hi_95 = est + qnorm(0.975)*se,
+         ci_lo_90 = est + qnorm(0.05)*se,
+         ci_hi_90 = est + qnorm(0.95)*se,
+         plotorder = nrow(coefs_third3_basecontrol):1
+  )
+
+writeLines(as.character(round(100*abs(filter(coefs_third3_basecontrol,layer3_specificoutcome=="pro_fraction_chosen")$est),0)),con = "../results/beta_recsys_pro_fraction_chosen_third3.tex",sep="%")
+writeLines(as.character(round(abs(filter(coefs_third3_basecontrol,layer3_specificoutcome=="mw_index_w2")$est),2)),con = "../results/beta_recsys_mwindex_third3.tex",sep="%")
+writeLines(as.character(round(abs(filter(coefs_third3_basecontrol,layer3_specificoutcome=="mw_index_w2")$ci_hi_95),2)),con = "../results/cihi_recsys_mwindex_third3.tex",sep="%")
+
+
+#### THE effect of INCREASING vs. CONSTANT assignment among CONSERVATIVE participants ####
+(coefplot_third3_basecontrol <- ggplot(filter(coefs_third3_basecontrol),aes(y=plotorder)) +
+    geom_errorbarh(aes(xmin=ci_lo_95,xmax=ci_hi_95),height=0,lwd=0.5) +
+    geom_errorbarh(aes(xmin=ci_lo_90,xmax=ci_hi_90),height=0,lwd=1) +
+    geom_point(aes(x=est),size=1.5) +
+    geom_vline(xintercept = 0,lty=2) +
+    facet_wrap(~family,ncol=1,scales="free") +
+    scale_y_continuous("",
+                       breaks = coefs_third3_basecontrol$plotorder,labels = coefs_third3_basecontrol$outcome) +
+    scale_x_continuous("Increasing Conservative vs. Constant Conservative \n seed among Conservative participants \n(95% and 90% CIs)") +
+    coord_cartesian(xlim=c(-0.2,0.2)) +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"))
+)
+
+#### THE effect of INCREASING vs. CONSTANT assignment among MODERATE participants assigned to a LIBERAL sequence ####
+coefs_third2_pro_basecontrol <- coefs_basecontrol %>%
+  filter(layer2_treatmentcontrast == "attitude.neutral:recsys.pi - attitude.neutral:recsys.pc" &
+         layer3_specificoutcome != "overall")
+
+
+coefs_third2_pro_basecontrol$outcome = outcome_labels$outcome[match(coefs_third2_pro_basecontrol$layer3_specificoutcome,
+                                                                    outcome_labels$specificoutcome)]
+
+coefs_third2_pro_basecontrol$family = outcome_labels$family[match(coefs_third2_pro_basecontrol$layer3_specificoutcome,
+                                                                  outcome_labels$specificoutcome)]
+
+coefs_third2_pro_basecontrol <- mutate(coefs_third2_pro_basecontrol,
+                                       family = factor(family,levels = c("Policy Attitudes\n(unit scale, + is more conservative)"
+                                                                         ),ordered = T))
+
+coefs_third2_pro_basecontrol <- coefs_third2_pro_basecontrol %>%
+  mutate(ci_lo_99 = est + qnorm(0.001)*se,
+         ci_hi_99 = est + qnorm(0.995)*se,
+         ci_lo_95 = est + qnorm(0.025)*se,
+         ci_hi_95 = est + qnorm(0.975)*se,
+         ci_lo_90 = est + qnorm(0.05)*se,
+         ci_hi_90 = est + qnorm(0.95)*se,
+         plotorder = nrow(coefs_third2_pro_basecontrol):1
+  )
+writeLines(as.character(round(100*abs(filter(coefs_third2_pro_basecontrol,layer3_specificoutcome=="pro_fraction_chosen")$est),0)),con = "../results/beta_recsys_pro_fraction_chosen_third2_proseed.tex",sep="%")
+writeLines(as.character(abs(round(filter(coefs_third2_pro_basecontrol,layer3_specificoutcome=="platform_duration")$est,2))),con = "../results/beta_recsys_duration_third2_proseed.tex",sep="%")
+writeLines(as.character(abs(round(filter(coefs_third2_pro_basecontrol,layer3_specificoutcome=="platform_duration")$est*60,1))),con = "../results/beta_minutes_recsys_duration_third2_proseed.tex",sep="%")
+
+#### THE effect of INCREASING vs. CONSTANT assignment among MODERATE participants assigned to a LIBERAL sequence ####
+(coefplot_third2_pro_basecontrol <- ggplot(filter(coefs_third2_pro_basecontrol),aes(y=plotorder)) +
+    geom_errorbarh(aes(xmin=ci_lo_95,xmax=ci_hi_95),height=0,lwd=0.5) +
+    geom_errorbarh(aes(xmin=ci_lo_90,xmax=ci_hi_90),height=0,lwd=1) +
+    geom_point(aes(x=est),size=1.5) +
+    geom_vline(xintercept = 0,lty=2) +
+    facet_wrap(~family,ncol=1,scales="free") +
+    scale_y_continuous("",
+                       breaks = coefs_third2_pro_basecontrol$plotorder,labels = coefs_third2_pro_basecontrol$outcome) +
+    scale_x_continuous("Increasing Liberal vs. Constant Liberal seed among Moderates \n(95% and 90% CIs)") +
+    coord_cartesian(xlim=c(-0.2,0.2)) +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"))
+)
+ggsave(coefplot_third2_pro_basecontrol,
+       filename = "../results/coefplot_third2_pro_basecontrol.png",width=5,height=8)
+
+#### THE effect of INCREASING vs. CONSTANT assignment among MODERATE participants assigned to a CONSERVATIVE sequence ####
+coefs_third2_anti_basecontrol <- coefs_basecontrol %>%
+  filter(layer2_treatmentcontrast == "attitude.neutral:recsys.ai - attitude.neutral:recsys.ac" &
+           layer3_specificoutcome != "overall")
+
+
+coefs_third2_anti_basecontrol$outcome = outcome_labels$outcome[match(coefs_third2_anti_basecontrol$layer3_specificoutcome,
+                                                                     outcome_labels$specificoutcome)]
+
+coefs_third2_anti_basecontrol$family = outcome_labels$family[match(coefs_third2_anti_basecontrol$layer3_specificoutcome,
+                                                                   outcome_labels$specificoutcome)]
+
+coefs_third2_anti_basecontrol <- mutate(coefs_third2_anti_basecontrol,
+                                        family = factor(family,levels = c("Policy Attitudes\n(unit scale, + is more conservative)"
+                                                                          ),ordered = T))
+
+
+coefs_third2_anti_basecontrol <- coefs_third2_anti_basecontrol %>%
+  mutate(ci_lo_99 = est + qnorm(0.001)*se,
+         ci_hi_99 = est + qnorm(0.995)*se,
+         ci_lo_95 = est + qnorm(0.025)*se,
+         ci_hi_95 = est + qnorm(0.975)*se,
+         ci_lo_90 = est + qnorm(0.05)*se,
+         ci_hi_90 = est + qnorm(0.95)*se,
+         plotorder = nrow(coefs_third2_anti_basecontrol):1
+  )
+
+writeLines(as.character(round(100*abs(filter(coefs_third2_anti_basecontrol,layer3_specificoutcome=="pro_fraction_chosen")$est),0)),con = "../results/beta_recsys_pro_fraction_chosen_third2_antiseed.tex",sep="%")
+writeLines(as.character(round(filter(coefs_third2_anti_basecontrol,layer1_hypothesisfamily=="gunpolicy")$est,2)),con = "../results/beta_recsys_mwindex_third2_antiseed.tex",sep="%")
+writeLines(as.character(round(filter(coefs_third2_anti_basecontrol,layer1_hypothesisfamily=="gunpolicy")$est + qnorm(0.975)*filter(coefs_third2_anti_basecontrol,layer1_hypothesisfamily=="gunpolicy")$se,2)),con = "../results/cihi_recsys_mwindex_third2_antiseed.tex",sep="%")
+writeLines(as.character(round(filter(coefs_third2_anti_basecontrol,layer1_hypothesisfamily=="gunpolicy")$est + qnorm(0.025)*filter(coefs_third2_anti_basecontrol,layer1_hypothesisfamily=="gunpolicy")$se,2)),con = "../results/cilo_recsys_mwindex_third2_antiseed.tex",sep="%")
+
+#### THE effect of INCREASING vs. CONSTANT assignment among MODERATE participants assigned to a CONSERVATIVE sequence ####
+(coefplot_third2_anti_basecontrol <- ggplot(filter(coefs_third2_anti_basecontrol),aes(y=plotorder)) +
+    geom_errorbarh(aes(xmin=ci_lo_95,xmax=ci_hi_95),height=0,lwd=0.5) +
+    geom_errorbarh(aes(xmin=ci_lo_90,xmax=ci_hi_90),height=0,lwd=1) +
+    geom_point(aes(x=est),size=1.5) +
+    geom_vline(xintercept = 0,lty=2) +
+    facet_wrap(~family,ncol=1,scales="free") +
+    scale_y_continuous("",
+                       breaks = coefs_third2_anti_basecontrol$plotorder,labels = coefs_third2_anti_basecontrol$outcome) +
+    scale_x_continuous("Increasing Conservative vs. Constant Conservative seed \namong Moderates \n(95% and 90% CIs)") +
+    coord_cartesian(xlim=c(-0.2,0.2)) +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"))
+)
+ggsave(coefplot_third2_anti_basecontrol,
+       filename = "../results/coefplot_third2_anti_basecontrol.png",width=5,height=8)
+
+
+#### THE effect of CONSERVATIVE vs. LIBERAL assignment among MODERATE participants assigned to an INCREASING sequence ####
+coefs_third2_31_basecontrol <- coefs_basecontrol %>%
+  filter(layer2_treatmentcontrast == "attitude.neutral:recsys.ai - attitude.neutral:recsys.pi" &
+           layer3_specificoutcome != "overall")
+
+
+coefs_third2_31_basecontrol$outcome = outcome_labels$outcome[match(coefs_third2_31_basecontrol$layer3_specificoutcome,
+                                                                   outcome_labels$specificoutcome)]
+
+coefs_third2_31_basecontrol$family = outcome_labels$family[match(coefs_third2_31_basecontrol$layer3_specificoutcome,
+                                                                 outcome_labels$specificoutcome)]
+
+coefs_third2_31_basecontrol <- mutate(coefs_third2_31_basecontrol,
+                                      family = factor(family,levels = c("Policy Attitudes\n(unit scale, + is more conservative)"
+                                                                        ),ordered = T))
+
+
+coefs_third2_31_basecontrol <- coefs_third2_31_basecontrol %>%
+  mutate(ci_lo_99 = est + qnorm(0.001)*se,
+         ci_hi_99 = est + qnorm(0.995)*se,
+         ci_lo_95 = est + qnorm(0.025)*se,
+         ci_hi_95 = est + qnorm(0.975)*se,
+         ci_lo_90 = est + qnorm(0.05)*se,
+         ci_hi_90 = est + qnorm(0.95)*se,
+         plotorder = nrow(coefs_third2_31_basecontrol):1
+  )
+writeLines(as.character(round(100*abs(filter(coefs_third2_31_basecontrol,layer3_specificoutcome=="pro_fraction_chosen")$est),0)),con = "../results/beta_seed_pro_fraction_chosen_third2_31.tex",sep="%")
+
+
+#### THE effect of CONSERVATIVE vs. LIBERAL assignment among MODERATE participants assigned to an INCREASING sequence ####
+(coefplot_third2_31_basecontrol <- ggplot(filter(coefs_third2_31_basecontrol),aes(y=plotorder)) +
+    geom_errorbarh(aes(xmin=ci_lo_95,xmax=ci_hi_95),height=0,lwd=0.5) +
+    geom_errorbarh(aes(xmin=ci_lo_90,xmax=ci_hi_90),height=0,lwd=1) +
+    geom_point(aes(x=est),size=1.5) +
+    geom_vline(xintercept = 0,lty=2) +
+    facet_wrap(~family,ncol=1,scales="free") +
+    scale_y_continuous("",
+                       breaks = coefs_third2_31_basecontrol$plotorder,labels = coefs_third2_31_basecontrol$outcome) +
+    scale_x_continuous("Conservative vs. Liberal seed assignment among Moderates\n with Increasing assignment\n(95% and 90% CIs)") +
+    coord_cartesian(xlim=c(-0.2,0.2)) +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"))
+)
+ggsave(coefplot_third2_31_basecontrol,
+       filename = "../results/coefplot_third2_31_basecontrol.png",width=5,height=8)
+
+#### THE effect of CONSERVATIVE vs. LIBERAL assignment among MODERATE participants assigned to an CONSTANT sequence ####
+coefs_third2_22_basecontrol <- coefs_basecontrol %>%
+  filter(layer2_treatmentcontrast == "attitude.neutral:recsys.ac - attitude.neutral:recsys.pc" &
+           layer3_specificoutcome != "overall")
+
+coefs_third2_22_basecontrol$outcome = outcome_labels$outcome[match(coefs_third2_22_basecontrol$layer3_specificoutcome,
+                                                                   outcome_labels$specificoutcome)]
+
+coefs_third2_22_basecontrol$family = outcome_labels$family[match(coefs_third2_22_basecontrol$layer3_specificoutcome,
+                                                                 outcome_labels$specificoutcome)]
+
+coefs_third2_22_basecontrol <- mutate(coefs_third2_22_basecontrol,
+                                      family = factor(family,levels = c(#"Platform Interaction",
+                                                                        "Policy Attitudes\n(unit scale, + is more conservative)"
+                                                                        #"Media Trust\n(unit scale, + is more trusting)",
+                                                                        #"Affective Polarization\n(unit scale, + is greater polarization)"
+                                                                        ),ordered = T))
+
+#### THE effect of CONSERVATIVE vs. LIBERAL assignment among MODERATE participants assigned to an CONSTANT sequence ####
+coefs_third2_22_basecontrol <- coefs_third2_22_basecontrol %>%
+  mutate(ci_lo_99 = est + qnorm(0.001)*se,
+         ci_hi_99 = est + qnorm(0.995)*se,
+         ci_lo_95 = est + qnorm(0.025)*se,
+         ci_hi_95 = est + qnorm(0.975)*se,
+         ci_lo_90 = est + qnorm(0.05)*se,
+         ci_hi_90 = est + qnorm(0.95)*se,
+         plotorder = nrow(coefs_third2_22_basecontrol):1
+  )
+writeLines(as.character(round(100*abs(filter(coefs_third2_22_basecontrol,layer3_specificoutcome=="pro_fraction_chosen")$est),0)),con = "../results/beta_seed_pro_fraction_chosen_third2_22.tex",sep="%")
+
+
+(coefplot_third2_22_basecontrol <- ggplot(filter(coefs_third2_22_basecontrol),aes(y=plotorder)) +
+    geom_errorbarh(aes(xmin=ci_lo_95,xmax=ci_hi_95),height=0,lwd=0.5) +
+    geom_errorbarh(aes(xmin=ci_lo_90,xmax=ci_hi_90),height=0,lwd=1) +
+    geom_point(aes(x=est),size=1.5) +
+    geom_vline(xintercept = 0,lty=2) +
+    facet_wrap(~family,ncol=1,scales="free") +
+    scale_y_continuous("",
+                       breaks = coefs_third2_22_basecontrol$plotorder,labels = coefs_third2_22_basecontrol$outcome) +
+    scale_x_continuous("Conservative vs. Liberal seed assignment among Moderates\n with Constant assignment\n(95% and 90% CIs)") +
+    coord_cartesian(xlim=c(-0.2,0.2)) +
+    theme_bw(base_family = "sans") +
+    theme(strip.background = element_rect(fill="white"))
+)
+
+rm(list = ls())

code/shorts/08_plot_shorts_figure.R

@@ -0,0 +1,326 @@
+cat(rep('=', 80),
+    '\n\n',
+    'OUTPUT FROM: shorts/08_plot_shorts_figure.R',
+    '\n\n',
+    sep = ''
+    )
+
+library(tidyverse)
+library(janitor)
+library(lubridate)
+library(stargazer)
+library(broom)
+library(psych)
+library(ggtext)
+library(ggplot2)
+
+# plotting w/ custom colors (optional)
+red_mit = '#A31F34'
+red_light = '#A9606C'
+blue_mit = '#315485'
+grey_light= '#C2C0BF'
+grey_dark = '#8A8B8C'
+black = '#353132'
+vpurple = "#440154FF"
+vyellow = "#FDE725FF"
+vgreen = "#21908CFF"
+
+
+## MODEL RESULTS
+coefs_basecontrol <- read_csv("../results/padj_basecontrol_pretty_ytrecs_may2024.csv")
+
+outcome_labels <- data.frame(outcome = c("Minimum wage<br>index"),
+                             specificoutcome = c("mw_index"),
+                             family = c(rep("Policy Attitudes<br>(unit scale, + is more conservative)",1)))
+
+
+# HYP 1
+#### THE effect of INCREASING vs. CONSTANT assignment among LIBERAL participants ####
+coefs_hyp1 <- coefs_basecontrol %>%
+  filter(layer2_treatmentcontrast == "attitude.pro:recsys.pi - attitude.pro:recsys.pc" &
+           layer3_specificoutcome != "overall")
+
+
+coefs_hyp1$outcome = outcome_labels$outcome[match(coefs_hyp1$layer3_specificoutcome,
+                                                                outcome_labels$specificoutcome)]
+
+coefs_hyp1$family = outcome_labels$family[match(coefs_hyp1$layer3_specificoutcome,outcome_labels$specificoutcome)]
+
+coefs_hyp1 <- mutate(coefs_hyp1,
+                     family = factor(family,
+                                     levels = c("Policy Attitudes<br>(unit scale, + is more conservative)"
+                                                ),ordered = T))
+
+coefs_hyp1 <- coefs_hyp1 %>%
+  mutate(ci_lo_99 = est + qnorm(0.001)*se,
+         ci_hi_99 = est + qnorm(0.995)*se,
+         ci_lo_95 = est + qnorm(0.025)*se,
+         ci_hi_95 = est + qnorm(0.975)*se,
+         ci_lo_90 = est + qnorm(0.05)*se,
+         ci_hi_90 = est + qnorm(0.95)*se,
+         plotorder = nrow(coefs_hyp1):1
+  )
+
+## HYP 2
+#### THE effect of INCREASING vs. CONSTANT assignment among CONSERVATIVE participants ####
+coefs_hyp2 <- coefs_basecontrol %>%
+  filter(layer2_treatmentcontrast == "attitude.anti:recsys.ai - attitude.anti:recsys.ac" &
+           layer3_specificoutcome != "overall")
+
+coefs_hyp2$outcome = outcome_labels$outcome[match(coefs_hyp2$layer3_specificoutcome,
+                                                                outcome_labels$specificoutcome)]
+
+coefs_hyp2$family = outcome_labels$family[match(coefs_hyp2$layer3_specificoutcome,
+                                                              outcome_labels$specificoutcome)]
+
+coefs_hyp2 <- mutate(coefs_hyp2,
+                                   family = factor(family,levels = c("Policy Attitudes<br>(unit scale, + is more conservative)"
+                                   ),ordered = T))
+
+
+coefs_hyp2 <- coefs_hyp2 %>%
+  mutate(ci_lo_99 = est + qnorm(0.001)*se,
+         ci_hi_99 = est + qnorm(0.995)*se,
+         ci_lo_95 = est + qnorm(0.025)*se,
+         ci_hi_95 = est + qnorm(0.975)*se,
+         ci_lo_90 = est + qnorm(0.05)*se,
+         ci_hi_90 = est + qnorm(0.95)*se,
+         plotorder = nrow(coefs_hyp2):1
+  )
+
+# HYP 3
+#### THE effect of INCREASING vs. CONSTANT assignment among MODERATE participants assigned to a LIBERAL sequence ####
+coefs_hyp3 <- coefs_basecontrol %>%
+  filter(layer2_treatmentcontrast == "attitude.neutral:recsys.pi - attitude.neutral:recsys.pc" &
+           layer3_specificoutcome != "overall")
+
+
+coefs_hyp3$outcome = outcome_labels$outcome[match(coefs_hyp3$layer3_specificoutcome,
+                                                                    outcome_labels$specificoutcome)]
+
+coefs_hyp3$family = outcome_labels$family[match(coefs_hyp3$layer3_specificoutcome,
+                                                                  outcome_labels$specificoutcome)]
+
+coefs_hyp3 <- mutate(coefs_hyp3,
+                                       family = factor(family,levels = c("Policy Attitudes<br>(unit scale, + is more conservative)"
+                                       ),ordered = T))
+
+coefs_hyp3 <- coefs_hyp3 %>%
+  mutate(ci_lo_99 = est + qnorm(0.001)*se,
+         ci_hi_99 = est + qnorm(0.995)*se,
+         ci_lo_95 = est + qnorm(0.025)*se,
+         ci_hi_95 = est + qnorm(0.975)*se,
+         ci_lo_90 = est + qnorm(0.05)*se,
+         ci_hi_90 = est + qnorm(0.95)*se,
+         plotorder = nrow(coefs_hyp3):1
+  )
+
+# HYP 4
+#### THE effect of INCREASING vs. CONSTANT assignment among MODERATE participants assigned to a CONSERVATIVE sequence ####
+coefs_hyp4 <- coefs_basecontrol %>%
+  filter(layer2_treatmentcontrast == "attitude.neutral:recsys.ai - attitude.neutral:recsys.ac" &
+           layer3_specificoutcome != "overall")
+
+
+coefs_hyp4$outcome = outcome_labels$outcome[match(coefs_hyp4$layer3_specificoutcome,
+                                                                     outcome_labels$specificoutcome)]
+
+coefs_hyp4$family = outcome_labels$family[match(coefs_hyp4$layer3_specificoutcome,
+                                                                   outcome_labels$specificoutcome)]
+
+coefs_hyp4 <- mutate(coefs_hyp4,
+                                        family = factor(family,levels = c("Policy Attitudes<br>(unit scale, + is more conservative)"
+                                        ),ordered = T))
+
+
+coefs_hyp4 <- coefs_hyp4 %>%
+  mutate(ci_lo_99 = est + qnorm(0.001)*se,
+         ci_hi_99 = est + qnorm(0.995)*se,
+         ci_lo_95 = est + qnorm(0.025)*se,
+         ci_hi_95 = est + qnorm(0.975)*se,
+         ci_lo_90 = est + qnorm(0.05)*se,
+         ci_hi_90 = est + qnorm(0.95)*se,
+         plotorder = nrow(coefs_hyp4):1
+  )
+
+# HYP 5
+#### THE effect of CONSERVATIVE vs. LIBERAL assignment among MODERATE participants assigned to an INCREASING sequence ####
+coefs_hyp5 <- coefs_basecontrol %>%
+  filter(layer2_treatmentcontrast == "attitude.neutral:recsys.ai - attitude.neutral:recsys.pi" &
+           layer3_specificoutcome != "overall")
+
+
+coefs_hyp5$outcome = outcome_labels$outcome[match(coefs_hyp5$layer3_specificoutcome,
+                                                                   outcome_labels$specificoutcome)]
+
+coefs_hyp5$family = outcome_labels$family[match(coefs_hyp5$layer3_specificoutcome,
+                                                                 outcome_labels$specificoutcome)]
+
+coefs_hyp5 <- mutate(coefs_hyp5,
+                     family = factor(family,levels = c("Policy Attitudes<br>(unit scale, + is more conservative)"
+                                      ),ordered = T))
+
+
+coefs_hyp5 <- coefs_hyp5 %>%
+  mutate(ci_lo_99 = est + qnorm(0.001)*se,
+         ci_hi_99 = est + qnorm(0.995)*se,
+         ci_lo_95 = est + qnorm(0.025)*se,
+         ci_hi_95 = est + qnorm(0.975)*se,
+         ci_lo_90 = est + qnorm(0.05)*se,
+         ci_hi_90 = est + qnorm(0.95)*se,
+         plotorder = nrow(coefs_hyp5):1
+  )
+
+# HYP 6
+#### THE effect of CONSERVATIVE vs. LIBERAL assignment among MODERATE participants assigned to an CONSTANT sequence ####
+coefs_hyp6 <- coefs_basecontrol %>%
+  filter(layer2_treatmentcontrast == "attitude.neutral:recsys.ac - attitude.neutral:recsys.pc" &
+           layer3_specificoutcome != "overall")
+
+coefs_hyp6$outcome = outcome_labels$outcome[match(coefs_hyp6$layer3_specificoutcome,
+                                                                   outcome_labels$specificoutcome)]
+
+coefs_hyp6$family = outcome_labels$family[match(coefs_hyp6$layer3_specificoutcome,
+                                                                 outcome_labels$specificoutcome)]
+
+coefs_hyp6 <- mutate(coefs_hyp6,
+                     family = factor(family,levels = c("Policy Attitudes<br>(unit scale, + is more conservative)"
+                                       ),ordered = T))
+
+#### THE effect of CONSERVATIVE vs. LIBERAL assignment among MODERATE participants assigned to an CONSTANT sequence ####
+coefs_hyp6 <- coefs_hyp6 %>%
+  mutate(ci_lo_99 = est + qnorm(0.001)*se,
+         ci_hi_99 = est + qnorm(0.995)*se,
+         ci_lo_95 = est + qnorm(0.025)*se,
+         ci_hi_95 = est + qnorm(0.975)*se,
+         ci_lo_90 = est + qnorm(0.05)*se,
+         ci_hi_90 = est + qnorm(0.95)*se,
+         plotorder = nrow(coefs_hyp6):1,
+  )
+
+# Combine all data frames into one
+all_coefs <- bind_rows(
+  mutate(coefs_hyp1, hypothesis = "**Increasing vs. Constant**<br>Liberal Seed<br>Liberal Ideologues", Sample="**Increasing vs. Constant**<br>Liberal Seed"),
+  mutate(coefs_hyp2, hypothesis = "**Increasing vs. Constant**<br>Conservative Seed<br>Conservative Ideologues", Sample="**Increasing vs. Constant**<br>Conservative Seed"),
+  mutate(coefs_hyp3, hypothesis = "**Increasing vs. Constant**<br>Liberal Seed<br>Moderates", Sample="**Increasing vs. Constant**<br>Liberal Seed"),
+  mutate(coefs_hyp4, hypothesis = "**Increasing vs. Constant**<br>Conservative Seed<br>Moderates", Sample="**Increasing vs. Constant**<br>Conservative Seed"),
+  mutate(coefs_hyp5, hypothesis = "**Conservative vs. Liberal**<br>Increasing Extremity<br>Moderates", Sample="**Conservative vs. Liberal**<br>Increasing Extremity"),
+  mutate(coefs_hyp6, hypothesis = "**Conservative vs. Liberal**<br>Constant Extremity<br>Moderates", Sample="**Conservative vs. Liberal**<br>Constant Extremity")
+)
+
+# Define the order of hypotheses
+hypothesis_order <- c("**Increasing vs. Constant**<br>Liberal Seed<br>Liberal Ideologues", 
+                      "**Increasing vs. Constant**<br>Conservative Seed<br>Conservative Ideologues", 
+                      "**Increasing vs. Constant**<br>Liberal Seed<br>Moderates", 
+                      "**Increasing vs. Constant**<br>Conservative Seed<br>Moderates", 
+                      "**Conservative vs. Liberal**<br>Increasing Extremity<br>Moderates", 
+                      "**Conservative vs. Liberal**<br>Constant Extremity<br>Moderates")
+
+# Reorder the factor levels
+all_coefs$hypothesis <- factor(all_coefs$hypothesis, levels = hypothesis_order)
+
+all_coefs <- all_coefs %>%
+  mutate(
+    attitude = case_when(
+      row_number() == 1 ~ "Liberal Ideologues",
+      row_number() == 2 ~ "Conservative Ideologues",
+      TRUE ~ "Moderates"
+    ),
+    alpha = ifelse(p.adj<0.05, T, F),
+    alpha = as.logical(alpha),
+    alpha = replace_na(alpha,F),
+    Sample_color = as.character(Sample),
+    Sample_color = replace(Sample_color,alpha==F,"insig")
+  )
+
+all_coefs <- all_coefs %>%
+  mutate(
+    sign_color = case_when(
+      ci_lo_95 < 0 & ci_hi_95 > 0 ~ grey_dark,  # black color code
+      TRUE ~ "darkgreen"  # blue color code (or replace with your desired color code)
+    )
+  )
+
+all_coefs <- all_coefs %>%
+  mutate(
+    attitude_color = case_when(
+      attitude == "Liberal Ideologues" ~ blue_mit,
+      attitude == "Conservative Ideologues" ~ red_mit,
+      attitude == "Moderates" ~ "darkgreen"
+    )
+  )
+
+
+all_coefs <- all_coefs %>%
+  mutate(Sample = factor(Sample,levels=c("**Increasing vs. Constant**<br>Liberal Seed",
+                                         "**Increasing vs. Constant**<br>Conservative Seed",
+                                         "**Conservative vs. Liberal**<br>Increasing Extremity",
+                                         "**Conservative vs. Liberal**<br>Constant Extremity"),
+                         ordered=T)) #%>%
+  #mutate(layer1_hypothesisfamily = recode(layer1_hypothesisfamily,
+  #                                        "mwpolicy"="policy"),
+  #       layer3_specificoutcome = recode(layer3_specificoutcome,
+  #                                       "mw_index"="policyindex"))
+
+
+# Create a data frame for attitude shapes
+attitude_shapes <- data.frame(attitude = c("Liberal Ideologues", "Conservative Ideologues", "Moderates"))
+
+# Plot the attitude shapes
+attitude_bar <- ggplot(attitude_shapes, aes(x = attitude)) +
+  geom_point(aes(shape = attitude), size = 3) +
+  scale_shape_manual(values = c("Liberal Ideologues" = 16, "Conservative Ideologues" = 17, "Moderates" = 15)) +
+  theme_void() +
+  theme(legend.position = "none")
+
+# Create a data frame for attitude shapes
+attitude_shapes <- data.frame(attitude = c("Liberal Ideologues", "Conservative Ideologues", "Moderates"))
+
+# Plot the attitude shapes
+attitude_bar <- ggplot(attitude_shapes, aes(x = attitude)) +
+  geom_point(aes(shape = attitude), size = 5) +
+  scale_shape_manual(values = c("Liberal Ideologues" = 16, "Conservative Ideologues" = 17, "Moderates" = 15)) +
+  theme_void() +
+  theme(legend.position = "none")
+
+# Plot
+combined_plot <- ggplot(all_coefs, aes(x = est, y = Sample, group = attitude, shape = attitude)) +
+  # 95% CI: Adjust alpha based on significance
+  geom_errorbarh(aes(xmin = ci_lo_95, xmax = ci_hi_95, color = sign_color, alpha = 0.8), 
+                 height = 0, lwd = 1, position = position_dodge(width = 0.8)) +
+  
+  # 90% CI: Adjust alpha based on significance
+  geom_errorbarh(aes(xmin = ci_lo_90, xmax = ci_hi_90, color = sign_color, alpha = 0.8), 
+                 height = 0, lwd = 1.5, position = position_dodge(width = 0.8)) +
+  
+  # Points: Adjust alpha directly for better visibility of insignificant shapes
+  geom_point(aes(color = sign_color), 
+             size = 4, position = position_dodge(width = 0.8), 
+             alpha = ifelse(all_coefs$alpha, 1, 0.7)) +  # Make insignificant points more visible with 0.7 alpha
+  
+  # Labels: Adjust alpha based on significance
+  geom_text(data = all_coefs, 
+            aes(x = est, label = attitude, color = attitude_color), 
+            alpha = 1, size = 6, 
+            position = position_dodge(width = 0.8), vjust = -0.6) + 
+  
+  geom_vline(xintercept = 0, lty = 2) +
+  facet_wrap(~ family, ncol = 1, scales = "free") +
+  coord_cartesian(xlim = c(-0.06, 0.18), clip="off") +
+  scale_x_continuous(" Minimum Wage Policy Effect Size\n(95% and 90% CIs)") +
+  scale_color_identity() +  # Ensure that the color column is used directly
+  labs(y = NULL) +  # Remove y-axis title
+  theme_bw(base_family = "sans") +
+  theme(strip.background = element_rect(fill = "white"),
+        legend.position = "none",
+        axis.text.y = element_markdown(color = "black", size=16),
+        axis.title.x = element_markdown(color = "black", size=16),
+        strip.text = element_markdown(size = 18)
+  )
+combined_plot
+ggsave(combined_plot, filename = "../results/shorts_combined_intervals.pdf", width = 8.5, height = 5)
+rm(list = ls())
+
+
+
+

code/supplemental/14_api_browser_comparison.R

@@ -0,0 +1,90 @@
+cat(rep('=', 80),
+    '\n\n',
+    'OUTPUT FROM: supplemental/14_api_browser_comparison.R',
+    '\n\n',
+    sep = ''
+    )
+
+# Load the API tree and the natural tree datasets
+
+api.tree <- read.csv('../data/supplemental/api_tree.csv')
+natural.tree <- readRDS('../data/supplemental/natural_tree.rds')
+
+#############
+# Figure S3 #
+#############
+
+natural.tree$w <- NA
+natural.tree$w[natural.tree$step == 1] <- 1 # All recs in the first step come from a common origin video
+natural.tree$w[natural.tree$step == 2] <- 1 # All recs in the second step come from a unique origin (one of the 20 recs from the first video)
+
+# If a recommendation appears in a step multiple times, we only get recs for that video once, but upweight those videos according to freq
+for(i in 1:nrow(natural.tree)){
+    if(natural.tree$step[i] == 1) next
+    if(natural.tree$step[i] == 2) next
+    paths.to.rec <- (natural.tree$originID[i] == natural.tree$recID) & (natural.tree$step == natural.tree$step[i] - 1)
+    sum(natural.tree$w[paths.to.rec])    
+    natural.tree$w[i] <- sum(natural.tree$w[paths.to.rec])
+}
+
+weighted.means <- c(weighted.mean(natural.tree$in.api.tree[natural.tree$step == 1], w = natural.tree$w[natural.tree$step == 1]),
+                    weighted.mean(natural.tree$in.api.tree[natural.tree$step == 2], w = natural.tree$w[natural.tree$step == 2]),
+                    weighted.mean(natural.tree$in.api.tree[natural.tree$step == 3], w = natural.tree$w[natural.tree$step == 3]),
+                    weighted.mean(natural.tree$in.api.tree[natural.tree$step == 4], w = natural.tree$w[natural.tree$step == 4]),
+                    weighted.mean(natural.tree$in.api.tree[natural.tree$step == 5], w = natural.tree$w[natural.tree$step == 5])
+                    )
+
+pdf('../results/proportion_by_step_in_tree_weighted.pdf')
+barplot(weighted.means,
+        main = 'Weighted Proportion of Natural Recs in API Tree (In Step)',
+        names.arg = c('Step 1', 'Step 2', 'Step 3', 'Step 4', 'Step 5'),
+        ylab = 'Proportion of Naturalistic Recs in API Tree',
+        ylim = c(0,1)
+        )
+dev.off()
+
+###############################
+# Table in Appendix Section 4 #
+###############################
+
+set.seed(63110)
+to.label <- natural.tree[sample(which(natural.tree$in.api.tree == 0 & natural.tree$step == 5), 10),]
+step.five.recs <- to.label$recID
+
+for(i in 1:nrow(to.label)){    
+    step.six.origin <- to.label$recID[i]
+    step.six.origin.str <- paste0('\\verb|', step.six.origin, '|')
+    if(!(step.six.origin %in% api.tree$RecID)) step.six.origin.str <- paste0(step.six.origin.str, '***')
+        
+    step.five.origin <- to.label$originID[i]
+    step.five.origin.str <- paste0('\\verb|', step.five.origin, '|')
+    if(!(step.five.origin %in% api.tree$RecID)) step.five.origin.str <- paste0(step.five.origin.str, '***')
+
+    step.four.origin <- natural.tree$originID[natural.tree$step == 4 & natural.tree$recID == step.five.origin][1]
+    step.four.origin.str <- paste0('\\verb|', step.four.origin, '|')
+    if(!(step.four.origin %in% api.tree$RecID)) step.four.origin.str <- paste0(step.four.origin.str, '***')
+    
+    step.three.origin <- natural.tree$originID[natural.tree$step == 3 & natural.tree$recID == step.four.origin][1]
+    step.three.origin.str <- paste0('\\verb|', step.three.origin, '|')
+    if(!(step.three.origin %in% api.tree$RecID)) step.three.origin.str <- paste0(step.three.origin.str, '***')
+
+    step.two.origin <- natural.tree$originID[natural.tree$step == 2 & natural.tree$recID == step.three.origin][1]
+    step.two.origin.str <- paste0('\\verb|', step.two.origin, '|')
+    if(!(step.two.origin %in% api.tree$RecID)) step.two.origin.str <- paste0(step.two.origin.str, '***')
+
+    step.one.origin <- natural.tree$originID[natural.tree$step == 1 & natural.tree$recID == step.two.origin][1]
+    step.one.origin.str <- paste0('\\verb|', step.one.origin, '|')
+    if(!(step.one.origin %in% api.tree$RecID)) step.one.origin.str <- paste0(step.one.origin.str, '***')
+
+    row <- paste(step.six.origin.str,
+                 step.five.origin.str,
+                 step.four.origin.str,
+                 step.three.origin.str,
+                 step.two.origin.str,
+                 step.one.origin.str,
+                 sep = ' & ')
+    row <- paste0(row, ' \\\\')
+    
+    cat(row)
+    cat('\n')
+}

code/supplemental/experiment durations/09_experiment_times.py

@@ -0,0 +1,307 @@
+### Average Time Spent in Experiments
+
+import numpy as np
+import pandas as pd
+import seaborn as sns
+import matplotlib.pyplot as plt
+import rpy2.robjects as robjects
+from rpy2.robjects import pandas2ri
+from IPython.display import Image
+
+print('=' * 80 + '\n\n' + 'OUTPUT FROM: supplemental/experiment durations/09_experiment_times.py' + '\n\n')
+
+study1 = pd.read_csv('../results/intermediate data/gun control (issue 1)/guncontrol_qualtrics_w123_clean.csv')
+study2 = pd.read_csv('../results/intermediate data/minimum wage (issue 2)/qualtrics_w12_clean.csv')
+study3 = pd.read_csv('../results/intermediate data/minimum wage (issue 2)/yg_w12_clean.csv')
+study4 = pd.read_csv('../results/intermediate data/shorts/qualtrics_w12_clean_ytrecs_may2024.csv')
+
+# ## Outlier Elimination
+# Manually patching the guaranteed incorrect values 
+# by taking the interface end time to be `pmin(survey end time, raw interface end time)`
+# Create a new column 'interface_end_time' by taking the minimum of the two columns
+
+# Convert 'end_date_w2' and 'end_time2' to datetime format
+study1['end_date_w2'] = pd.to_datetime(study1['end_date_w2'], format='%Y-%m-%d %H:%M:%S', errors='coerce')
+study1['end_time2'] = pd.to_datetime(study1['end_time2'], format='%Y-%m-%dT%H:%M:%SZ',errors='coerce').dt.tz_localize('UTC').dt.tz_convert('America/New_York').dt.tz_localize(None)
+study1['start_time2'] = pd.to_datetime(study1['start_time2'], format='%Y-%m-%dT%H:%M:%SZ',errors='coerce').dt.tz_localize('UTC').dt.tz_convert('America/New_York').dt.tz_localize(None)
+
+# Create a new column 'interface_end_time_fixed' by taking the minimum of both dates
+study1['interface_end_time_fixed'] = study1['end_date_w2'].combine(study1['end_time2'], 
+    lambda x, y: x if pd.notna(y) and (pd.isna(x) or x < y) else y)
+
+study1['interface_end_time_fixed'] = study1['interface_end_time_fixed'].where(
+    pd.notna(study1['interface_end_time_fixed']), np.nan
+)
+
+# Convert 'end_date_w2' and 'end_time2' to datetime format
+study2['end_date_w2'] = pd.to_datetime(study2['end_date_w2'], format='%Y-%m-%d %H:%M:%S', errors='coerce')
+study2['end_time2'] = pd.to_datetime(study2['end_time2'], format='%Y-%m-%dT%H:%M:%SZ',errors='coerce').dt.tz_localize('UTC').dt.tz_convert('America/New_York').dt.tz_localize(None)
+study2['start_time2'] = pd.to_datetime(study2['start_time2'], format='%Y-%m-%dT%H:%M:%SZ',errors='coerce').dt.tz_localize('UTC').dt.tz_convert('America/New_York').dt.tz_localize(None)
+
+# Create a new column 'interface_end_time_fixed' by taking the minimum of both dates
+study2['interface_end_time_fixed'] = study2['end_date_w2'].combine(study2['end_time2'], 
+    lambda x, y: x if pd.notna(y) and (pd.isna(x) or x < y) else y)
+
+study2['interface_end_time_fixed'] = study2['interface_end_time_fixed'].where(
+    pd.notna(study2['interface_end_time_fixed']), np.nan
+)
+
+# Convert 'end_date_w2' and 'end_time2' to datetime format
+study3['start_date_w2'] = pd.to_datetime(study3['start_date_w2'], format='%Y-%m-%dT%H:%M:%SZ').dt.tz_localize(None)
+study3['end_date_w2'] = pd.to_datetime(study3['end_date_w2'], format='%Y-%m-%dT%H:%M:%SZ').dt.tz_localize(None)
+study3['end_time2'] = pd.to_datetime(study3['end_time2'], format='%Y-%m-%dT%H:%M:%SZ').dt.tz_localize(None)
+study3['start_time2'] = pd.to_datetime(study3['start_time2'], format='%Y-%m-%dT%H:%M:%SZ').dt.tz_localize(None)
+# Create a new column 'interface_time_fixed' with the minimum of both dates
+study3['interface_end_time_fixed'] = np.minimum(study3['end_date_w2'], study3['end_time2'])
+
+# Fixed duration values for study 4 is done.
+study4['interface_time_fixed'] = np.minimum(study4['survey_time'], study4['interface_duration']/60)
+
+# Ensure both columns are in datetime format, and make them timezone-naive
+study1['interface_end_time_fixed'] = pd.to_datetime(study1['interface_end_time_fixed'], errors='coerce').dt.tz_localize(None)
+study1['start_time2'] = pd.to_datetime(study1['start_time2'], errors='coerce').dt.tz_localize(None)
+# Subtract the times while handling nulls
+study1['interface_time_fixed'] = study1['interface_end_time_fixed'] - study1['start_time2']
+
+# Ensure both columns are in datetime format, and make them timezone-naive
+study2['interface_end_time_fixed'] = pd.to_datetime(study2['interface_end_time_fixed'], errors='coerce').dt.tz_localize(None)
+study2['start_time2'] = pd.to_datetime(study2['start_time2'], errors='coerce').dt.tz_localize(None)
+# Subtract the times while handling nulls
+study2['interface_time_fixed'] = study2['interface_end_time_fixed'] - study2['start_time2']
+
+# Ensure both columns are in datetime format, and make them timezone-naive
+study3['interface_end_time_fixed'] = pd.to_datetime(study3['interface_end_time_fixed'], errors='coerce').dt.tz_localize(None)
+study3['start_time2'] = pd.to_datetime(study3['start_time2'], errors='coerce').dt.tz_localize(None)
+# Subtract the times while handling nulls
+study3['interface_time_fixed'] = study3['interface_end_time_fixed'] - study3['start_time2']
+
+# Convert the timedelta to minutes
+study1['interface_time_fixed_minutes'] = study1['interface_time_fixed'].dt.total_seconds() / 60
+study2['interface_time_fixed_minutes'] = study2['interface_time_fixed'].dt.total_seconds() / 60
+study3['interface_time_fixed_minutes'] = study3['interface_time_fixed'].dt.total_seconds() / 60
+
+
+#### Windsorization
+
+# Copy the 'duration' column to a new 'platform_duration' column
+study1['platform_duration'] = study1['duration']
+study2['platform_duration'] = study2['duration']
+study3['platform_duration'] = study3['duration']
+study4['platform_duration'] = study4['interface_duration']
+
+# Calculate the 2.5% and 97.5% quantiles
+lower_quantile_study1 = study1['duration'].quantile(0.025)
+upper_quantile_study1 = study1['duration'].quantile(0.975)
+
+lower_quantile_study2 = study2['duration'].quantile(0.025)
+upper_quantile_study2 = study2['duration'].quantile(0.975)
+
+lower_quantile_study3 = study3['duration'].quantile(0.025)
+upper_quantile_study3 = study3['duration'].quantile(0.975)
+
+lower_quantile_study4 = study4['interface_duration'].quantile(0.025)
+upper_quantile_study4 = study4['interface_duration'].quantile(0.975)
+
+# Apply Windsorization: cap the values at 2.5% and 97.5%
+study1['platform_duration'] = study1['platform_duration'].apply(
+    lambda x: lower_quantile_study1 if x <= lower_quantile_study1 else upper_quantile_study1 if x >= upper_quantile_study1 else x
+)
+
+study2['platform_duration'] = study2['platform_duration'].apply(
+    lambda x: lower_quantile_study2 if x <= lower_quantile_study2 else upper_quantile_study2 if x >= upper_quantile_study2 else x
+)
+
+study3['platform_duration'] = study3['platform_duration'].apply(
+    lambda x: lower_quantile_study3 if x <= lower_quantile_study3 else upper_quantile_study3 if x >= upper_quantile_study3 else x
+)
+
+study4['platform_duration'] = study4['platform_duration'].apply(
+    lambda x: lower_quantile_study4 if x <= lower_quantile_study4 else upper_quantile_study4 if x >= upper_quantile_study4 else x
+)
+
+# Copy the 'duration' column to a new 'platform_duration' column
+study1['platform_duration'] = study1['interface_time_fixed_minutes']
+study2['platform_duration'] = study2['interface_time_fixed_minutes']
+study3['platform_duration'] = study3['interface_time_fixed_minutes']
+study4['platform_duration'] = study4['interface_time_fixed']
+
+# Calculate the 2.5% and 97.5% quantiles
+lower_quantile_study1 = study1['interface_time_fixed_minutes'].quantile(0.025)
+upper_quantile_study1 = study1['interface_time_fixed_minutes'].quantile(0.975)
+
+lower_quantile_study2 = study2['interface_time_fixed_minutes'].quantile(0.025)
+upper_quantile_study2 = study2['interface_time_fixed_minutes'].quantile(0.975)
+
+lower_quantile_study3 = study3['interface_time_fixed_minutes'].quantile(0.025)
+upper_quantile_study3 = study3['interface_time_fixed_minutes'].quantile(0.975)
+
+lower_quantile_study4 = study4['interface_time_fixed'].quantile(0.025)
+upper_quantile_study4 = study4['interface_time_fixed'].quantile(0.975)
+
+# Apply Windsorization: cap the values at 2.5% and 97.5%
+study1['platform_duration'] = study1['platform_duration'].apply(
+    lambda x: lower_quantile_study1 if x <= lower_quantile_study1 else upper_quantile_study1 if x >= upper_quantile_study1 else x
+)
+
+study2['platform_duration'] = study2['platform_duration'].apply(
+    lambda x: lower_quantile_study2 if x <= lower_quantile_study2 else upper_quantile_study2 if x >= upper_quantile_study2 else x
+)
+
+study3['platform_duration'] = study3['platform_duration'].apply(
+    lambda x: lower_quantile_study3 if x <= lower_quantile_study3 else upper_quantile_study3 if x >= upper_quantile_study3 else x
+)
+
+study4['platform_duration'] = study4['platform_duration'].apply(
+    lambda x: lower_quantile_study4 if x <= lower_quantile_study4 else upper_quantile_study4 if x >= upper_quantile_study4 else x
+)
+
+# Overall interface time spent (mean, Studies 1-3)
+print('Mean Interface Time for Studies 1-3:', pd.concat([study1[study1.treatment_arm != 'control']['platform_duration'], 
+                                                      study2[study2.treatment_arm != 'control']['platform_duration'], 
+                                                      study3[study3.treatment_arm != 'control']['platform_duration']], 
+                                                     ignore_index=True).mean())
+
+print('******')
+# Overall interface time spent (mean, Studies 1-4)
+print('Mean Interface Time for Studies 1-4:', pd.concat([study1[study1.treatment_arm != 'control']['platform_duration'], 
+                                                      study2[study2.treatment_arm != 'control']['platform_duration'], 
+                                                      study3[study3.treatment_arm != 'control']['platform_duration'],
+                                                      study4['platform_duration']], 
+                                                     ignore_index=True).mean())
+
+print('******')
+# Interface time spent each
+print('Study1 Interface:',study1[study1.treatment_arm != 'control']['platform_duration'].mean())
+print('Study2 Interface:',study2[study2.treatment_arm != 'control']['platform_duration'].mean())
+print('Study3 Interface:',study3[study3.treatment_arm != 'control']['platform_duration'].mean())
+print('Study4 Interface:',study4['platform_duration'].mean())
+
+
+### Plots
+
+# Enable the pandas-to-R conversion
+pandas2ri.activate()
+
+# Convert the DataFrame to R DataFrame
+w123_r = pandas2ri.py2rpy(study1)
+
+# Define the R code to create the plot and save it as an image
+r_code = """
+library(ggplot2)
+library(dplyr)
+
+# Filter the data
+w123_filtered <- w123 %>% filter(treatment_arm != "control")
+
+# Create the plot and save it as a PNG file
+surveytime_plot <- ggplot(w123_filtered) +
+    geom_histogram(aes(x = platform_duration, y = ..density.. / sum(..density..))) +
+    scale_x_continuous("Interface Time Taken (minutes),\nexcluding control respondents", breaks = seq(0, 100, 20), limits = c(-1, 101)) +
+    scale_y_continuous("Density") +
+    geom_vline(xintercept = mean(w123_filtered$platform_duration, na.rm = TRUE), linetype = "dashed", color = "red") +
+    annotate("text", x = mean(w123_filtered$platform_duration, na.rm = TRUE) + 1, y = 0.13, label = paste0("Average: ", round(mean(w123_filtered$platform_duration, na.rm = TRUE), 0), " minutes"), hjust = 0) +
+    geom_vline(xintercept = median(w123_filtered$platform_duration, na.rm = TRUE), linetype = "dotted", color = "red") +
+    annotate("text", x = median(w123_filtered$platform_duration , na.rm = TRUE) + 1, y = 0.16, label = paste0("Median: ", round(median(w123_filtered$platform_duration, na.rm = TRUE), 0), " minutes"), hjust = 0) +
+    theme_minimal()
+ggsave(surveytime_plot,filename = "../results/video_platform_duration_study1.pdf",height=3,width=5)
+"""
+
+# Load the DataFrame into the R environment & run
+robjects.globalenv['w123'] = w123_r
+robjects.r(r_code)
+#Image(filename="video_platform_duration_study1.png")
+
+# Enable the pandas-to-R conversion
+pandas2ri.activate()
+
+# Convert the dataframe to R DataFrame
+w123_r = pandas2ri.py2rpy(study2)
+
+# Define the R code to create the plot and save it as an image
+r_code = """
+library(ggplot2)
+library(dplyr)
+
+# Filter the data
+w123_filtered <- w123 %>% filter(treatment_arm != "control")
+
+# Create the plot and save it as a PNG file
+surveytime_plot <- ggplot(w123_filtered) +
+    geom_histogram(aes(x = platform_duration, y = ..density.. / sum(..density..))) +
+    scale_x_continuous("Interface Time Taken (minutes),\nexcluding control respondents", breaks = seq(0, 100, 20), limits = c(-1, 101)) +
+    scale_y_continuous("Density") +
+    geom_vline(xintercept = mean(w123_filtered$platform_duration, na.rm = TRUE), linetype = "dashed", color = "red") +
+    annotate("text", x = mean(w123_filtered$platform_duration, na.rm = TRUE) + 1, y = 0.13, label = paste0("Average: ", round(mean(w123_filtered$platform_duration, na.rm = TRUE), 0), " minutes"), hjust = 0) +
+    geom_vline(xintercept = median(w123_filtered$platform_duration, na.rm = TRUE), linetype = "dotted", color = "red") +
+    annotate("text", x = median(w123_filtered$platform_duration , na.rm = TRUE) + 1, y = 0.16, label = paste0("Median: ", round(median(w123_filtered$platform_duration, na.rm = TRUE), 0), " minutes"), hjust = 0) +
+    theme_minimal()
+
+ggsave(surveytime_plot,filename = "../results/video_platform_duration_study2.pdf",height=3,width=5)
+"""
+
+robjects.globalenv['w123'] = w123_r
+robjects.r(r_code)
+#Image(filename="video_platform_duration_study2.png")
+
+# Enable the pandas-to-R conversion
+pandas2ri.activate()
+w123_r = pandas2ri.py2rpy(study3)
+
+# Define the R code to create the plot and save it as an image
+r_code = """
+library(ggplot2)
+library(dplyr)
+
+# Filter the data
+w123_filtered <- w123 %>% filter(treatment_arm != "control")
+
+# Create the plot and save it as a PNG file
+surveytime_plot <- ggplot(w123_filtered) +
+    geom_histogram(aes(x = platform_duration, y = ..density.. / sum(..density..))) +
+    scale_x_continuous("Interface Time Taken (minutes),\nexcluding control respondents", breaks = seq(0, 100, 20), limits = c(-1, 101)) +
+    scale_y_continuous("Density") +
+    geom_vline(xintercept = mean(w123_filtered$platform_duration, na.rm = TRUE), linetype = "dashed", color = "red") +
+    annotate("text", x = mean(w123_filtered$platform_duration, na.rm = TRUE) + 1, y = 0.13, label = paste0("Average: ", round(mean(w123_filtered$platform_duration, na.rm = TRUE), 0), " minutes"), hjust = 0) +
+    geom_vline(xintercept = median(w123_filtered$platform_duration, na.rm = TRUE), linetype = "dotted", color = "red") +
+    annotate("text", x = median(w123_filtered$platform_duration , na.rm = TRUE) + 1, y = 0.16, label = paste0("Median: ", round(median(w123_filtered$platform_duration, na.rm = TRUE), 0), " minutes"), hjust = 0) +
+    theme_minimal()
+ggsave(surveytime_plot,filename = "../results/video_platform_duration_study3.pdf",height=3,width=5)
+"""
+
+# Load the DataFrame into the R environment
+robjects.globalenv['w123'] = w123_r
+robjects.r(r_code)
+#Image(filename="video_platform_duration_study3.png")
+
+
+pandas2ri.activate()
+w123_r = pandas2ri.py2rpy(study4)
+
+# Define the R code to create the plot and save it as an image
+r_code = """
+library(ggplot2)
+library(dplyr)
+
+# Filter the data
+w123_filtered <- w123 %>% filter(treatment_arm != "control")
+
+# Create the plot and save it as a PNG file
+surveytime_plot <- ggplot(w123_filtered) +
+    geom_histogram(aes(x = platform_duration, y = ..density.. / sum(..density..))) +
+    scale_x_continuous("Interface Time Taken (minutes)", breaks = seq(0, 100, 20), limits = c(-1, 101)) +
+    scale_y_continuous("Density") +
+    geom_vline(xintercept = mean(w123_filtered$platform_duration, na.rm = TRUE), linetype = "dashed", color = "red") +
+    annotate("text", x = mean(w123_filtered$platform_duration, na.rm = TRUE) + 1, y = 0.13, label = paste0("Average: ", round(mean(w123_filtered$platform_duration, na.rm = TRUE), 0), " minutes"), hjust = 0) +
+    geom_vline(xintercept = median(w123_filtered$platform_duration, na.rm = TRUE), linetype = "dotted", color = "red") +
+    annotate("text", x = median(w123_filtered$platform_duration , na.rm = TRUE) + 1, y = 0.16, label = paste0("Median: ", round(median(w123_filtered$platform_duration, na.rm = TRUE), 0), " minutes"), hjust = 0) +
+    theme_minimal()
+ggsave(surveytime_plot,filename = "../results/video_platform_duration_study4.pdf",height=3,width=5)
+"""
+
+robjects.globalenv['w123'] = w123_r
+robjects.r(r_code)
+#Image(filename="video_platform_duration_study4.png")
+
+
+
+

code/supplemental/increasingly extreme recommendations/10_partisanship_increase.py

@@ -0,0 +1,136 @@
+#!/usr/bin/env python
+# coding: utf-8
+
+# # Expected Value of Partisanship Increase
+
+# Libraries
+import pandas as pd
+import numpy as np
+import seaborn as sns
+import matplotlib.pyplot as plt
+import statsmodels.formula.api as smf
+import statsmodels.api as sm
+from scipy.stats import t
+
+print('=' * 80 + '\n\n' + 'OUTPUT FROM: supplemental/increasingly extreme recommendations/10_partisanship_increase.py' + '\n\n')
+
+
+# Data from Youtube Rec
+wage_data = pd.read_csv('../data/supplemental/metadata and ratings/metadata_with_lables_binary_only_checked_0410.csv')
+
+# GPT label (continuous) --> full tree data
+gpt_labels_cont = pd.read_csv("../data/supplemental/metadata and ratings/gpt_continuous_ratings_minwage_FULL_averaged (1).csv")[['naijia_originId','gpt_continuous']]
+gpt_labels_cont = gpt_labels_cont.rename(columns={'naijia_originId':'originID','gpt_continuous':'gpt_label'})
+
+# eliminate duplicates if any
+gpt_labels_cont = gpt_labels_cont.groupby('originID').agg({"gpt_label":"mean"}).reset_index()
+
+# ## Min Wage Increase - average recommendation on the average video
+
+pairs = wage_data[['originID','recID','step']]
+pairs = pairs.merge(gpt_labels_cont, how='left',on='originID').rename(columns={'gpt_label':'gpt_label_originID'})
+
+
+# data format: Each row is a (current, recommended) video pair
+# cols: cur video ID, cur video rating, cur video rec, cur video rec rating
+pairs = pairs.merge(gpt_labels_cont, 
+                    how='left',
+                    left_on='recID', 
+                    right_on='originID').rename(columns={'gpt_label':'gpt_label_recID',
+                                                         'originID_x':'originID'}).drop(columns=['originID_y'])
+
+pairs = pairs[(pairs.gpt_label_originID.isnull() == False) & 
+(pairs.gpt_label_recID.isnull() == False)]
+
+# weight videos by the number of recommendations they have
+weights = pairs.groupby('originID').agg({"recID":
+                                         "nunique"}).reset_index().rename(columns={"recID":
+                                                                                   "weight"})
+pairs = pairs.merge(weights, how='left',on='originID')
+
+# Difference = Recommended Score − Current Score
+pairs['difference'] = pairs['gpt_label_recID'] - pairs['gpt_label_originID']
+pairs['weighted_difference'] = pairs['difference'] / pairs['weight']
+
+## liberal/conservative categorization
+def label_category(row):
+    if row['gpt_label_originID'] > 0:
+        return 'conservative'
+    else:
+        return 'liberal'
+
+pairs['label_category'] = pairs.apply(label_category, axis=1)
+
+# Liberal Cur Videos
+liberal = pairs[pairs.label_category == 'liberal']
+
+
+# Constant for the intercept
+lib_X = sm.add_constant(liberal['gpt_label_originID'])
+
+# OLS model with two-way clustering
+model = sm.OLS(liberal['weighted_difference'], lib_X)
+lib_results = model.fit(cov_type='cluster', 
+                    cov_kwds={'groups': [liberal['originID'].tolist(), liberal['recID'].tolist()]})
+print(lib_results.summary())
+
+
+# Conservative Cur Videos
+conservative = pairs[pairs.label_category == 'conservative']
+
+# Constant for the intercept
+cons_X = sm.add_constant(conservative['gpt_label_originID'])
+
+# OLS model with two-way clustering
+model = sm.OLS(conservative['weighted_difference'], cons_X)
+cons_results = model.fit(cov_type='cluster', 
+                    cov_kwds={'groups': [conservative['originID'].tolist(), conservative['recID'].tolist()]})
+print(cons_results.summary())
+
+
+### SI Figure
+
+fig, ax = plt.subplots()
+
+lib_preds = lib_results.get_prediction(lib_X).summary_frame(alpha=0.05)
+cons_preds = cons_results.get_prediction(cons_X).summary_frame(alpha=0.05)
+
+ax.scatter(liberal['gpt_label_originID'], 
+            liberal['gpt_label_originID'] + liberal['weighted_difference'], 
+            color='lightblue', s=0.9, alpha=0.6)
+
+ax.scatter(conservative['gpt_label_originID'], 
+            conservative['gpt_label_originID'] + conservative['weighted_difference'], 
+            color='lightcoral', s=0.9, alpha=0.6)
+
+ax.fill_between(liberal['gpt_label_originID'], 
+                 liberal['gpt_label_originID'] + lib_preds['mean_ci_lower'], 
+                 liberal['gpt_label_originID'] + lib_preds['mean_ci_upper'], 
+                alpha=.4, color='blue')
+
+ax.fill_between(conservative['gpt_label_originID'], 
+                 conservative['gpt_label_originID'] + cons_preds['mean_ci_lower'], 
+                 conservative['gpt_label_originID'] + cons_preds['mean_ci_upper'], 
+                alpha=.4, color='red')
+
+ax.plot(liberal['gpt_label_originID'], 
+         lib_preds['mean'] + liberal['gpt_label_originID'], 
+         color='darkblue', linewidth=0.5)
+
+ax.plot(conservative['gpt_label_originID'],
+         cons_preds['mean'] + conservative['gpt_label_originID'], 
+         color='darkred', linewidth=0.5)
+
+ax.plot([-1, 1], [-1, 1], 'k--', linewidth=1.5)
+
+# Customize the plot
+ax.set_xlabel('Current Video Rating')
+ax.set_ylabel('Recommended Video Rating')
+ax.set_xlim(-1, 1)
+ax.set_ylim(-1, 1)
+ax.grid(False)
+ax.set_title('Current Video Rating vs \n Recommended Video Rating')
+plt.savefig('../results/video_rating_pairs.png', dpi=300, bbox_inches='tight')
+
+
+

code/supplemental/increasingly extreme recommendations/11_gpt_rating_plots.py

@@ -0,0 +1,76 @@
+#!/usr/bin/env python
+# coding: utf-8
+
+"""
+This file serves to plot Figures S12 and S13, which show the robustness of the GPT-generated ratings to different
+ways of quantifying a video's political extremeness (BERT ratings and Hosseinmardi's channel labels.)
+"""
+
+import pandas as pd
+import numpy as np
+import seaborn as sns
+import matplotlib.pyplot as plt
+
+print('=' * 80 + '\n\n' + 'OUTPUT FROM: supplemental/increasingly extreme recommendations/11_gpt_rating_plots.py' + '\n\n')
+
+# Read in the GPT continuous ratings, along with the other ways of quantifying political expremeness
+gpt_rating_wage = pd.read_csv("../data/supplemental/metadata and ratings/gpt_continuous_ratings_minwage.csv")
+wage_videos_full = pd.read_csv("../data/supplemental/metadata and ratings/bert_rated_wage_videos_all.csv")
+
+wage_all_merged = wage_videos_full.merge(gpt_rating_wage, on="naijia_originId")
+# deduplicate
+wage_all_merged = wage_all_merged[["naijia_originId", "homa_explanation_new", "gpt_label", "bert_score", "originCat"]].drop_duplicates()
+
+
+"""
+FIGURE S12: COMPARISON TO BERT MODEL FROM LAI ET AL. (2024)
+"""
+print("starting Figure S12...")
+
+plt.figure(figsize=(8, 6))
+colors = {'pro': 'blue', 'anti': 'red'}
+
+plt.scatter(wage_all_merged['gpt_label'], wage_all_merged['bert_score'], c=wage_all_merged['originCat'].map(colors), alpha=0.7)
+
+plt.axhline(0, color='gray', linestyle='--', linewidth=1.5)
+plt.axvline(0, color='gray', linestyle='--', linewidth=1.5)
+
+plt.xlabel('GPT Continuous')
+plt.ylabel('BERT Score')
+plt.title('Correlation between GPT Continuous and BERT Score')
+
+for cat, color in colors.items():
+    plt.scatter([], [], c=color, label=cat)
+plt.legend()
+
+plt.grid(True)
+plt.savefig('../results/figure_S12_comparison_to_Lai_et_al_bert_ratings.png')
+
+print("...done!")
+
+
+"""
+FIGURE S13: COMPAIRSON WITH HOSSEINMARDI ET AL. (2021)
+"""
+print("starting Figure S13...")
+
+fig, ax = plt.subplots(figsize=(12, 6))
+
+for category, data in wage_all_merged.groupby('homa_explanation_new'):
+    sns.histplot(data=data, x='gpt_label', label=category, kde=True, ax=ax, alpha=0.7)
+
+plt.title("Distribution of GPT Continuous Scores by Hosseinmardi et al. (2021)", fontsize=16)
+plt.xlabel('GPT Score', fontsize=14)
+plt.ylabel('Frequency', fontsize=14)
+plt.legend(title='Category', fontsize=12, title_fontsize=14)
+
+plt.xlim(-1, 1)
+plt.xticks(fontsize=12)
+plt.yticks(fontsize=12)
+
+ax.grid(True, linestyle='--', alpha=0.5)
+sns.despine()
+plt.tight_layout()
+plt.savefig('../results/figure_S13_comparison_to_Hosseinmardi_et_al_channel_ratings.png')
+
+print("...done!")

code/supplemental/thumbnails (first impressions)/12_thumbnail_analysis.py

@@ -0,0 +1,938 @@
+#!/usr/bin/env python
+# coding: utf-8
+
+# # THUMBNAILS EXPERIMENT
+
+# # Libraries
+
+import numpy as np
+import pandas as pd
+import matplotlib.pyplot as plt
+import seaborn as sns
+import scipy.stats as stats 
+import json
+import re
+import os, glob
+from collections import Counter
+from statistics import mode
+from sklearn.metrics import accuracy_score, precision_score, recall_score
+from datetime import datetime
+import gc
+import statsmodels.api as sm
+from stargazer.stargazer import Stargazer
+
+import warnings
+warnings.filterwarnings("ignore")
+
+print('=' * 80 + '\n\n' + 'OUTPUT FROM: supplemental/thumbnails (first impressions)/12_thumbnail_analysis.py' + '\n\n')
+
+# Reading Session Logs
+
+# Session logs --- pIDHash is our new respondent-id
+with open("../../data/platform session data/sessions.json") as json_file:
+    json_data = json.load(json_file)
+
+
+unique_topics = [] 
+real_data = []
+
+for item in json_data:
+
+    # Check if the session is completed
+    if item['sessionFinished']:
+    
+        # Check if the topic ID is 'min_wage' or 'gun_control'
+        if item['topicID'] in ['min_wage','gun_control']:
+
+            # Convert start time from milliseconds to seconds
+            unix_time_seconds_start = item['startTime'] / 1000
+            
+            # Convert the UNIX timestamp to a datetime object
+            normal_time_start = datetime.fromtimestamp(unix_time_seconds_start)
+
+            # Check if the session started on May 23, 2024, after 10 AM
+            if (normal_time_start.year == 2024 and
+                normal_time_start.month == 5 and
+                normal_time_start.day == 23 and
+                normal_time_start.hour >= 10):
+                
+                # Add the session to the real_data list
+                real_data.append(item)
+
+
+del json_data
+gc.collect()
+
+
+print('Total session count:', len(real_data))
+
+
+# # GPT Ratings
+
+
+# GPT Labels
+gpt_labels = pd.read_csv('../../data/supplemental/metadata and ratings/gpt_thumb_ratings_withHumanInfo.csv').drop_duplicates()
+
+# in case of duplicate labels by GPT, take the first one
+gpt_labels = gpt_labels.drop_duplicates(subset='originId', keep='first')
+
+gpt_labels['gpt_thumb_rating'] = gpt_labels['gpt_thumb_rating'].replace('pro.', 'pro') # 2 
+gpt_labels['gpt_thumb_rating'] = gpt_labels['gpt_thumb_rating'].replace('anti.', 'anti')# 3
+gpt_labels.gpt_thumb_rating.value_counts() # 1%
+
+
+# # Gold Standard Labels
+
+# these were the videos we actually provided to Jim
+labels_on_platform = pd.concat([pd.read_csv('../../data/supplemental/metadata and ratings/gun_thumbnails_updated_v4(gun_control).csv'), pd.read_csv('../../data/supplemental/metadata and ratings/wage_thumbnails_updated_v4(min_wage).csv')])
+
+# Actual ("Ground Truth") Labels
+gun_videos_all_metadata = pd.read_csv('../../data/supplemental/metadata and ratings/metadata_w_label_June_2021_NLversion.csv')
+wage_videos_all_metadata = pd.read_csv('../../data/supplemental/metadata and ratings/metadata_with_lables_binary_only_checked_0410.csv')
+gun_labels = gun_videos_all_metadata[['originID', 'originCat']].dropna().drop_duplicates().rename(columns={"originID": "originId"})
+wage_labels = wage_videos_all_metadata[['originID', 'originCat']].dropna().drop_duplicates().rename(columns={"originID": "originId"})
+gold_labels = pd.concat([gun_labels, wage_labels], axis = 0)
+
+
+# # Curate a Subset of "Easy to Rate" Videos
+
+bert_ratings = pd.read_csv('../../data/supplemental/metadata and ratings/bert_rated_wage_videos_all.csv')
+
+gpt_continuous_extremenss_ratings = pd.read_csv('../../data/supplemental/metadata and ratings/gpt_continuous_ratings_minwage.csv').drop_duplicates()
+gpt_continuous_extremenss_ratings = gpt_continuous_extremenss_ratings.rename(columns={"naijia_originId": "originID"})
+gpt_continuous_extremenss_ratings = gpt_continuous_extremenss_ratings[["originID", "gpt_label"]]
+
+bert_ratings = bert_ratings.rename(columns={"naijia_originId": "originID"})
+
+ratings_aggregated = pd.merge(gpt_continuous_extremenss_ratings, bert_ratings[["originID", "originCat", "bert_score"]], how= "inner", on = 'originID').drop_duplicates()
+
+# convert gpt label to 'pro' if it is < 0 and 'anti' otherwise
+ratings_aggregated['gpt_label'] = np.where(ratings_aggregated['gpt_label'] < 0, 'pro', 'anti')
+# do the same thing for bert_score
+ratings_aggregated['bert_score'] = np.where(ratings_aggregated['bert_score'] < 0, 'pro', 'anti')
+
+# get cases where all 3 agree
+ratings_aggregated['all_agree'] = np.where((ratings_aggregated['gpt_label'] == ratings_aggregated['bert_score']) & (ratings_aggregated['gpt_label'] == ratings_aggregated['originCat']), 1, 0)
+
+
+# This is a subset of minimum wage videos that is "easy to rate": GPT, BERT, and the human Turkers all got this correct
+
+MINWAGE_AGREED_VIDEOS = ratings_aggregated[ratings_aggregated['all_agree'] == 1]
+
+# get cases where GPT and BERT agree, but humans (originCat) disagrees
+ratings_aggregated['gpt_bert_agree'] = np.where((ratings_aggregated['gpt_label'] == ratings_aggregated['bert_score']) & (ratings_aggregated['gpt_label'] != ratings_aggregated['originCat']), 1, 0)
+
+# peek at specific videos
+gpt_continuous_extremenss_ratings[gpt_continuous_extremenss_ratings["originID"]=="Z_r5TlBdjEM"]
+
+# this happened in just 7 out of 154 cases...
+ratings_aggregated['gpt_bert_agree'].value_counts()
+
+# .... or 4.5%!
+ratings_aggregated['gpt_bert_agree'].value_counts()[1]/len(ratings_aggregated)
+
+ratings_aggregated[ratings_aggregated['gpt_bert_agree']==1]
+
+
+# # Session Level Performance I
+
+# Including videos without GPT labels (both no label and 'insufficient information' label cases)
+
+def session_rep_counts(data):
+    exp_sessions_wage, exp_sessions_gun = [], []
+    exp_indexes_wage, exp_indexes_gun = [], []
+    exp_indexes_wage_resp, exp_indexes_gun_resp = [], []
+    
+    for i in range(0, len(data)): # Iterate over all elements in the dictionary
+    
+        # Only completed sessions for our surveys
+        if data[i]['sessionFinished'] and data[i]['topicID'] in ['min_wage', 'gun_control'] and len(data[i]['ratingResults']) > 0:
+    
+            topic_id = data[i]['topicID']
+            resp_id = data[i]['pIDHash']
+
+            if data[i]['topicID'] == 'min_wage':
+                exp_sessions_wage.append(topic_id)
+                exp_indexes_wage_resp.append(resp_id)
+                exp_indexes_wage.append(i)
+        
+            elif data[i]['topicID'] == 'gun_control':
+                exp_sessions_gun.append(topic_id)
+                exp_indexes_gun_resp.append(resp_id)
+                exp_indexes_gun.append(i)
+    
+    print('Wage session count:',len(exp_indexes_wage))
+    print('Gun session count:',len(exp_indexes_gun))
+    
+    print('Unique respondents (authID) in Wage:',len(np.unique(exp_indexes_wage_resp)))
+    print('Unique respondents (authID) in Gun',len(np.unique(exp_indexes_gun_resp)))
+
+    return exp_indexes_wage, exp_indexes_gun
+
+def exp_analysis(data, index, total_count, pro_count, anti_count, insuf_count, nolabel_count, exc):
+
+    gold_matches = []
+    pro_matches = []
+    anti_matches = []
+    
+    for item in data[index]['ratingResults']:
+        
+        total_count += 1
+        exp_index = int(item['index'])
+
+        if data[index]['topicID'] == 'min_wage':
+            if exp_index == 1:
+                exp_label = 'pro'
+            elif exp_index == 2:
+                exp_label = 'anti'
+            elif exp_index == 3:
+                exp_label = 'insufficient data.'
+        elif data[index]['topicID'] == 'gun_control':
+            if exp_index == 1:
+                exp_label = 'anti'
+            elif exp_index == 2:
+                exp_label = 'pro'
+            elif exp_index == 3:
+                exp_label = 'insufficient data.'
+
+        try:
+            gpt_label = gpt_labels[gpt_labels.originId == item['vid']].gpt_thumb_rating.values[0]
+        except:
+            gpt_label = 'no_label'
+
+        try:
+            gold_label = gold_labels[gold_labels.originId == item['vid']].originCat.values[0]
+        except:
+            gold_label = 'no_label'
+    
+        if gold_label == 'pro':
+            pro_count += 1
+        elif gold_label == 'anti':
+            anti_count += 1
+        elif gold_label == 'insufficient data.':
+            insuf_count += 1
+        else:
+            nolabel_count += 1
+        
+        if exc == 0:
+            gold_matches.append(1 if exp_label == gold_label else 0)
+            # determine whether it's an anti or pro match
+            if exp_label == 'pro':
+                pro_matches.append(1 if exp_label == gold_label else 0)
+            elif exp_label == 'anti':
+                anti_matches.append(1 if exp_label == gold_label else 0)
+        elif exc == 1:
+            if gpt_label == 'insufficient data.' or pd.isnull(gold_label):
+                continue
+            else:
+                gold_matches.append(1 if exp_label == gold_label else 0)
+                
+    return gold_matches, pro_matches, anti_matches, pro_count, anti_count, insuf_count, nolabel_count, total_count
+
+def results(total_count,insuf_count,nolabel_count,pro_count,anti_count,gold_matches):
+    print('Number of videos:', total_count)
+    print('Number of labeled videos:', total_count - insuf_count - nolabel_count)
+    print('Number of pro videos:', pro_count)
+    print('Number of anti videos:', anti_count)
+    print('Number of vague videos:', insuf_count)
+    print('Number of non labeled videos:', nolabel_count)
+    print("***")
+    print('Total number of matches with GPT:', np.sum(gold_matches))
+    print("***")
+    print('Total % of matches with GPT %', np.round(np.sum(gold_matches) / len(gold_matches), 2) * 100)
+    print("***")
+    print('')
+
+# json_data is the json data
+# exc 0 if we want to include videos without GPT labels, 1 otw.
+def thumbnail_exp_check(data, exc=0):
+
+    result_df = pd.DataFrame(columns = ['session_id',
+                                        'topic_id',
+                                        'respondent_id',
+                                        'total_video_count',
+                                        'respondent_label_count',
+                                        'gold_insufficient_video_count',
+                                        'gold_nolabel_video_count',
+                                        'gold_pro_video_count',
+                                        'gold_anti_video_count',
+                                        'gold_match_count',
+                                        'gold_pro_match_count',
+                                        'gold_anti_match_count'
+                                       ])
+
+    print('Summary Statistics')
+    exp_indexes_wage, exp_indexes_gun = session_rep_counts(data) # session and unique respondent counts
+
+    # Check all matches for each experiment
+    for index in exp_indexes_wage + exp_indexes_gun:
+
+        if data[index]['sessionFinished']:
+
+            if len(data[index]['ratingResults']) > 0:
+
+                pro_count, anti_count, insuf_count, nolabel_count, total_count = 0, 0, 0, 0, 0
+        
+                gold_matches, pro_matches, anti_matches, pro_count, anti_count, insuf_count, nolabel_count, total_count = exp_analysis(data, index, total_count, pro_count, 
+                                                                           anti_count, insuf_count, nolabel_count, 
+                                                                           exc=exc)
+
+                resp_id = data[index]['pIDHash']
+
+                row = [index,
+                       data[index]['topicID'],
+                       resp_id, 
+                       len(data[index]['ratingResults']),
+                       total_count, 
+                       insuf_count, 
+                       nolabel_count, 
+                       pro_count, 
+                       anti_count, 
+                       np.sum(gold_matches),
+                       np.sum(pro_matches),
+                       np.sum(anti_matches)
+                      ]
+                row_df = pd.DataFrame(row).T
+                row_df.columns = ['session_id',
+                                        'topic_id',
+                                        'respondent_id',
+                                        'total_video_count',
+                                        'respondent_label_count',
+                                        'gold_insufficient_video_count',
+                                        'gold_nolabel_video_count',
+                                        'gold_pro_video_count',
+                                        'gold_anti_video_count',
+                                        'gold_match_count',
+                                        'gold_pro_match_count',
+                                        'gold_anti_match_count'
+                                       ]
+
+                result_df = pd.concat([result_df, row_df], axis=0)
+
+    result_df['gold_match_perc'] = result_df['gold_match_count'] / result_df['total_video_count']
+    ## look at the specific breakdown of pro versus anti
+    result_df['gold_pro_match_perc'] = result_df['gold_pro_match_count'] / result_df['gold_pro_video_count']
+    result_df['gold_anti_match_perc'] = result_df['gold_anti_match_count'] / result_df['gold_anti_video_count']
+    
+    return result_df
+
+# SESSION LEVEL RESULT
+result_df = thumbnail_exp_check(real_data, exc=0)
+
+print("number of unique participants")
+len(result_df)
+
+result_df["total_video_count"].value_counts()
+
+def calculate_quartiles(series):
+    rounded = []
+    for i in series.quantile([0.25, 0.5, 0.75]).to_list():
+        rounded.append(np.round(i,2))
+    return rounded
+
+result_df.groupby('topic_id').agg(
+    session_count=('session_id', 'nunique'),
+    respondent_count=('respondent_id', 'nunique'),
+    gold_match_mean=('gold_match_perc', 'mean'),
+    gold_match_std=('gold_match_perc', 'std'),
+    gold_match_quartiles=('gold_match_perc', calculate_quartiles)
+).reset_index()
+
+# examine the results by pro/anti
+results_by_pro_anti = result_df.groupby('topic_id').agg(
+    session_count=('session_id', 'nunique'),
+    respondent_count=('respondent_id', 'nunique'),
+    pro_gold_match_mean=('gold_pro_match_perc', 'mean'),
+    pro_gold_match_std=('gold_pro_match_perc', 'std'),
+    pro_gold_match_quartiles=('gold_pro_match_perc', calculate_quartiles),
+    anti_gold_match_mean=('gold_anti_match_perc', 'mean'),
+    anti_gold_match_std=('gold_anti_match_perc', 'std'),
+    anti_gold_match_quartiles=('gold_anti_match_perc', calculate_quartiles)
+).reset_index()
+
+melted_df = pd.melt(results_by_pro_anti, 
+                    id_vars=['topic_id', 'session_count', 'respondent_count'], 
+                    value_vars=[
+                        'pro_gold_match_mean', 
+                        'pro_gold_match_std', 
+                        'pro_gold_match_quartiles', 
+                        'anti_gold_match_mean', 
+                        'anti_gold_match_std', 
+                        'anti_gold_match_quartiles'],
+                    var_name='stat_type', value_name='value')
+melted_df['type'] = melted_df['stat_type'].apply(lambda x: x.split('_')[0])
+melted_df['statistic'] = melted_df['stat_type'].apply(lambda x: x.split('_')[-1])
+melted_df = melted_df.sort_values(by=['topic_id', 'statistic']).drop('stat_type', axis=1)
+melted_df
+
+print("t-test for gold match percentage (pooled)")
+all_gold_match_perc = np.asarray([float(num) for num in result_df["gold_match_perc"]])
+t_statistic, p_value = stats.ttest_1samp(a=all_gold_match_perc, popmean=0.5) 
+print("t-statistic:", t_statistic)
+print("p-value:", p_value)
+
+print("overall accuracy")
+print(np.mean(all_gold_match_perc))
+
+print("t-test for gold match percentage (liberal gun control; operationalizing random as 1/3)")
+gun_result_df = result_df[result_df["topic_id"] == "gun_control"]
+gun_gold_match_perc = np.asarray([float(num) for num in gun_result_df["gold_pro_match_perc"]])
+t_statistic, p_value = stats.ttest_1samp(a=gun_gold_match_perc, popmean=0.333) 
+print("t-statistic:", t_statistic)
+print("p-value:", p_value)
+
+result_df_broken_by_proanti = result_df[["topic_id", "gold_pro_match_perc", "gold_anti_match_perc"]]
+result_df_proanti_melted = pd.melt(result_df_broken_by_proanti, 
+                    id_vars=['topic_id'], 
+                    value_vars=[
+                        'gold_pro_match_perc',
+                        'gold_anti_match_perc'],
+                    var_name='stat_type', value_name='value')
+result_df_proanti_melted['Video Gold Label'] = result_df_proanti_melted['stat_type'].apply(lambda x: x.split('_')[1])
+result_df_proanti_melted = result_df_proanti_melted.drop('stat_type', axis=1)
+
+sns.boxplot(x='topic_id', y='gold_match_perc', data=result_df)
+plt.show()
+
+# make the labels more understandable
+result_df_proanti_melted["Video Gold Label"]= result_df_proanti_melted["Video Gold Label"].replace({"pro": "Liberal", "anti": "Conservative"})
+result_df_proanti_melted["topic_id"]= result_df_proanti_melted["topic_id"].replace({"min_wage": "Minimum Wage", "gun_control": "Gun Control"})
+
+sns.boxplot(x='topic_id', y='value', hue='Video Gold Label', data=result_df_proanti_melted)
+# add a horizontal line for the 50% mark
+plt.axhline(0.5, color='r', linestyle='--')
+plt.title("Individual Raters' Percentage Match by Topic and Gold Label")
+plt.xlabel("Topic")
+plt.ylabel("Percentage Match with Gold")
+plt.show()
+
+
+# # Video Level Performance
+exp_indexes_wage, exp_indexes_gun = session_rep_counts(real_data)
+
+indexes = exp_indexes_wage + exp_indexes_gun 
+videos = {}
+videos['gun_control'] = {}
+videos['min_wage'] = {}
+
+for index in indexes:
+    
+    ratings = real_data[index]['ratingResults']
+    for rating in ratings:
+        video = rating['vid']
+        exp_index = int(rating['index'])
+
+        if real_data[index]['topicID'] == 'min_wage':
+            if exp_index == 1:
+                exp_label = 'pro'
+            elif exp_index == 2:
+                exp_label = 'anti'
+            elif exp_index == 3:
+                exp_label = 'insufficient data.'
+        elif real_data[index]['topicID'] == 'gun_control':
+            if exp_index == 1:
+                exp_label = 'anti'
+            elif exp_index == 2:
+                exp_label = 'pro'
+            elif exp_index == 3:
+                exp_label = 'insufficient data.'
+
+        if real_data[index]['topicID'] == 'min_wage':
+            if video not in videos['min_wage'].keys():
+                videos['min_wage'][video] = [exp_label]
+            else:
+                videos['min_wage'][video].append(exp_label)
+                
+        elif real_data[index]['topicID'] == 'gun_control':
+            if video not in videos['gun_control'].keys():
+                videos['gun_control'][video] = [exp_label]
+            else:
+                videos['gun_control'][video].append(exp_label)
+
+for video, labels in videos.items():
+    print(video)
+
+majority_votes = {}
+vote_counts = []
+topics = []
+num_votes_for_majority = []
+majority_vote_drop_insuf = []
+
+for topicid, videolist in videos.items():
+
+    for video, labels in videolist.items():
+        majority_vote = mode(labels)
+        vote_count = len(labels)
+        votes_for_majority = [label for label in labels if label == majority_vote]
+        num_votes_for_majority.append(len(votes_for_majority))
+
+        # a version of majority_vote if we remove votes for "insufficient data."
+        majority_vote_drop_insuf.append(mode([label for label in labels if label != 'insufficient data.']))
+
+        # Debug what happens when we have a very small minority as the majority percentage
+        # how can the mode have only something like 35%?
+
+        # Answer -- it's because there's a 3-way split between 'pro,' 'anti,' and 'insufficient data.'
+
+        # if len(votes_for_majority) / vote_count < 0.4:
+        #     print('Majority vote:', majority_vote)
+        #     print('Votes for majority:', len(votes_for_majority))
+        #     print('Total votes:', vote_count)
+        #     print('Votes for majority %:', len(votes_for_majority) / vote_count)
+        #     print('Labels:')
+        #     print(pd.Series(labels).value_counts())
+        #     print('***')
+        
+        majority_votes[video] = majority_vote
+        vote_counts.append(vote_count)
+        topics.append(topicid)
+
+majority_voting = pd.concat([pd.Series(majority_votes.keys()), 
+                             pd.Series(topics), 
+                             pd.Series(vote_counts), 
+                             pd.Series(num_votes_for_majority),
+                             pd.Series(majority_votes.values()),
+                             pd.Series(majority_vote_drop_insuf)], 
+                            axis=1)
+majority_voting.columns = ['originId','topicID','vote_count', 'num_votes_for_majority', 'majority_label', 'majority_label_drop_insuf']
+print(majority_voting.shape)
+
+majority_voting = majority_voting.merge(gold_labels.drop_duplicates(), how='left',on='originId')
+majority_voting = majority_voting.merge(gpt_labels.drop_duplicates(), how='left',on='originId')
+
+majority_voting.head()
+majority_voting.originCat.value_counts()
+majority_voting.gpt_thumb_rating.value_counts()
+
+# Encode the labels
+majority_voting['gold_label_encoded'] = majority_voting['originCat'].map({'anti': 0, 
+                                                                  'pro': 1, 
+                                                                  'other': 2})
+
+majority_voting['majority_label_encoded'] = majority_voting['majority_label'].map({'anti': 0, 
+                                                                           'pro': 1, 
+                                                                           'insufficient data.': 2})
+
+majority_voting['majority_label_drop_insuf_encoded'] = majority_voting['majority_label_drop_insuf'].map({'anti': 0, 
+                                                                           'pro': 1})
+
+majority_voting['gpt_encoded'] = majority_voting['gpt_thumb_rating'].map({'anti': 0,
+                                                                  'pro': 1,
+                                                                  'insufficient data.': 2})
+
+def performance_metrics(filtered_df, topicID='OVERALL'):
+    print('**********')
+    print(f'{topicID}')
+
+    if topicID == 'OVERALL':
+        accuracy = accuracy_score(filtered_df['gold_label_encoded'], 
+                                  filtered_df['majority_label_encoded'])
+        precision = precision_score(filtered_df['gold_label_encoded'], 
+                                    filtered_df['majority_label_encoded'], 
+                                    labels = [0,1,2],
+                                    average='weighted',
+                                    zero_division=0)
+        recall = recall_score(filtered_df['gold_label_encoded'], 
+                              filtered_df['majority_label_encoded'],
+                              labels = [0,1,2], 
+                              average='weighted',
+                              zero_division=0)
+    else:
+        accuracy = accuracy_score(filtered_df[filtered_df.topicID == f'{topicID}']['gold_label_encoded'], 
+                                  filtered_df[filtered_df.topicID == f'{topicID}']['majority_label_encoded'])
+        precision = precision_score(filtered_df[filtered_df.topicID == f'{topicID}']['gold_label_encoded'], 
+                                    filtered_df[filtered_df.topicID == f'{topicID}']['majority_label_encoded'],
+                                    labels = [0,1,2], 
+                                    average='weighted',
+                                    zero_division=0)
+        recall = recall_score(y_true = filtered_df[filtered_df.topicID == f'{topicID}']['gold_label_encoded'], 
+                              y_pred = filtered_df[filtered_df.topicID == f'{topicID}']['majority_label_encoded'], 
+                              labels = [0,1,2],
+                              average= 'weighted',
+                              zero_division=0)
+                
+    print(f'Accuracy: {accuracy:.2f}')
+    print(f'Precision: {precision:.2f}')
+    print(f'Recall: {recall:.2f}')
+    print('**********')
+
+majority_voting_nonNullOriginCat = majority_voting[majority_voting.originCat.isnull() == False]
+
+majority_voting_nonNullOriginCat["majority_label"].value_counts()
+
+pro_only_maj_vote = majority_voting_nonNullOriginCat[majority_voting_nonNullOriginCat['originCat'] == 'pro']
+anti_only_maj_vote = majority_voting_nonNullOriginCat[majority_voting_nonNullOriginCat['originCat'] == 'anti']
+# flip all the labels for anti_only
+anti_only_maj_vote = anti_only_maj_vote.copy()
+anti_only_maj_vote.loc[:, 'majority_label_encoded'] = anti_only_maj_vote['majority_label_encoded'].replace({0: 1, 1: 0})
+anti_only_maj_vote.loc[:, 'gold_label_encoded'] = anti_only_maj_vote['gold_label_encoded'].replace({0: 1, 1: 0})
+anti_only_maj_vote.loc[:, 'majority_label_drop_insuf_encoded'] = anti_only_maj_vote['majority_label_drop_insuf_encoded'].replace({0: 1, 1: 0})
+
+print("number of videos that had a valid original rating")
+print(len(majority_voting_nonNullOriginCat))
+
+majority_voting_nonNullOriginCat.loc[:, "is_human_match"] = majority_voting_nonNullOriginCat["gold_label_encoded"] == majority_voting_nonNullOriginCat["majority_label_encoded"]
+
+print("t-test for gold match percentage (general, majority vote)")
+all_gold_match_perc = np.asarray([float(num) for num in majority_voting_nonNullOriginCat["is_human_match"]])
+t_statistic, p_value = stats.ttest_1samp(a=all_gold_match_perc, popmean=0.5) 
+print("t-statistic:", t_statistic)
+print("p-value:", p_value)
+
+print('Comparing MTURKERS AND OLD GOLD STANDARD LABELS')
+performance_metrics(majority_voting_nonNullOriginCat, topicID='OVERALL')
+performance_metrics(majority_voting_nonNullOriginCat, topicID='min_wage')
+performance_metrics(majority_voting_nonNullOriginCat, topicID='gun_control')
+
+print('PRO VIDEOS: Comparing MTURKERS AND OLD GOLD STANDARD LABELS')
+performance_metrics(pro_only_maj_vote, topicID='OVERALL')
+performance_metrics(pro_only_maj_vote, topicID='min_wage')
+performance_metrics(pro_only_maj_vote, topicID='gun_control')
+
+print('ANTI VIDEOS: Comparing MTURKERS AND OLD GOLD STANDARD LABELS')
+performance_metrics(anti_only_maj_vote, topicID='OVERALL')
+performance_metrics(anti_only_maj_vote, topicID='min_wage')
+performance_metrics(anti_only_maj_vote, topicID='gun_control')
+
+
+# # Gold Standard v. Majority Label (Dropping Insufficient Data)
+
+def performance_metrics_drop_insuf(filtered_df, topicID='OVERALL'):
+    print('**********')
+    print(f'{topicID}')
+
+    if topicID == 'OVERALL':
+        accuracy = accuracy_score(filtered_df['gold_label_encoded'], 
+                                  filtered_df['majority_label_drop_insuf_encoded'])
+        precision = precision_score(filtered_df['gold_label_encoded'], 
+                                    filtered_df['majority_label_drop_insuf_encoded'], 
+                                    labels = [0,1],
+                                    average='weighted',
+                                    zero_division=0)
+        recall = recall_score(filtered_df['gold_label_encoded'], 
+                              filtered_df['majority_label_drop_insuf_encoded'],
+                              labels = [0,1], 
+                              average='weighted',
+                              zero_division=0)
+    else:
+        accuracy = accuracy_score(filtered_df[filtered_df.topicID == f'{topicID}']['gold_label_encoded'], 
+                                  filtered_df[filtered_df.topicID == f'{topicID}']['majority_label_drop_insuf_encoded'])
+        precision = precision_score(filtered_df[filtered_df.topicID == f'{topicID}']['gold_label_encoded'], 
+                                    filtered_df[filtered_df.topicID == f'{topicID}']['majority_label_drop_insuf_encoded'],
+                                    labels = [0,1], 
+                                    average='weighted',
+                                    zero_division=0)
+        recall = recall_score(y_true = filtered_df[filtered_df.topicID == f'{topicID}']['gold_label_encoded'], 
+                              y_pred = filtered_df[filtered_df.topicID == f'{topicID}']['majority_label_drop_insuf_encoded'], 
+                              labels = [0,1],
+                              average= 'weighted',
+                              zero_division=0)
+                
+    print(f'Accuracy: {accuracy:.2f}')
+    print(f'Precision: {precision:.2f}')
+    print(f'Recall: {recall:.2f}')
+    print('**********')
+
+print('Comparing MTURKERS AND OLD GOLD STANDARD LABELS')
+performance_metrics_drop_insuf(majority_voting_nonNullOriginCat, topicID='OVERALL')
+performance_metrics_drop_insuf(majority_voting_nonNullOriginCat, topicID='min_wage')
+performance_metrics_drop_insuf(majority_voting_nonNullOriginCat, topicID='gun_control')
+
+print('PRO VIDEOS: Comparing MTURKERS AND OLD GOLD STANDARD LABELS')
+performance_metrics_drop_insuf(pro_only_maj_vote, topicID='OVERALL')
+performance_metrics_drop_insuf(pro_only_maj_vote, topicID='min_wage')
+performance_metrics_drop_insuf(pro_only_maj_vote, topicID='gun_control')
+
+print('ANTI VIDEOS: Comparing MTURKERS AND OLD GOLD STANDARD LABELS')
+performance_metrics_drop_insuf(anti_only_maj_vote, topicID='OVERALL')
+performance_metrics_drop_insuf(anti_only_maj_vote, topicID='min_wage')
+performance_metrics_drop_insuf(anti_only_maj_vote, topicID='gun_control')
+
+
+# # Gold Standard v. Majority Label ("Easy" Subset)
+majority_voting_easyOnly = majority_voting_nonNullOriginCat[majority_voting_nonNullOriginCat["originId"].isin(MINWAGE_AGREED_VIDEOS["originID"])]
+
+# create the pro- and anti-only sets
+pro_only_maj_vote_easy = majority_voting_easyOnly[majority_voting_easyOnly['originCat'] == 'pro']
+anti_only_maj_vote_easy = majority_voting_easyOnly[majority_voting_easyOnly['originCat'] == 'anti']
+# flip all the labels for anti_only
+anti_only_maj_vote_easy = anti_only_maj_vote_easy.copy()
+anti_only_maj_vote_easy.loc[:, 'majority_label_encoded'] = anti_only_maj_vote_easy['majority_label_encoded'].replace({0: 1, 1: 0})
+anti_only_maj_vote_easy.loc[:, 'gold_label_encoded'] = anti_only_maj_vote_easy['gold_label_encoded'].replace({0: 1, 1: 0})
+anti_only_maj_vote_easy.loc[:, 'majority_label_drop_insuf_encoded'] = anti_only_maj_vote_easy['majority_label_drop_insuf_encoded'].replace({0: 1, 1: 0})
+
+print('Comparing MTURKERS AND OLD GOLD STANDARD LABELS')
+performance_metrics(majority_voting_easyOnly, topicID='OVERALL')
+
+print('PRO VIDEOS: Comparing MTURKERS AND OLD GOLD STANDARD LABELS')
+performance_metrics_drop_insuf(pro_only_maj_vote_easy, topicID='OVERALL')
+
+print('ANTI VIDEOS: Comparing MTURKERS AND OLD GOLD STANDARD LABELS')
+performance_metrics_drop_insuf(anti_only_maj_vote_easy, topicID='OVERALL')
+
+
+# # Gold Standard versus GPT
+def performance_gpt(filtered_df, topicID='OVERALL'):
+    print('**********')
+    print(f'{topicID}')
+
+    if topicID == 'OVERALL':
+        accuracy = accuracy_score(filtered_df['gold_label_encoded'], 
+                                  filtered_df['gpt_encoded'])
+        precision = precision_score(filtered_df['gold_label_encoded'], 
+                                    filtered_df['gpt_encoded'], 
+                                    labels = [0,1,2], 
+                                    average='weighted',
+                                    zero_division=0)
+        recall = recall_score(filtered_df['gold_label_encoded'], 
+                              filtered_df['gpt_encoded'], 
+                              labels = [0,1,2], 
+                              average='weighted',
+                              zero_division=0)
+    else:
+        accuracy = accuracy_score(filtered_df[filtered_df.topicID == f'{topicID}']['gold_label_encoded'], 
+                                  filtered_df[filtered_df.topicID == f'{topicID}']['gpt_encoded'])
+        precision = precision_score(filtered_df[filtered_df.topicID == f'{topicID}']['gold_label_encoded'], 
+                                    filtered_df[filtered_df.topicID == f'{topicID}']['gpt_encoded'], 
+                                    labels = [0,1,2], 
+                                    average='weighted',
+                                    zero_division=0)
+        recall = recall_score(filtered_df[filtered_df.topicID == f'{topicID}']['gold_label_encoded'], 
+                              filtered_df[filtered_df.topicID == f'{topicID}']['gpt_encoded'], 
+                              labels = [0,1,2], 
+                              average='weighted',
+                              zero_division=0)
+        
+    print(f'Accuracy: {accuracy:.2f}')
+    print(f'Precision: {precision:.2f}')
+    print(f'Recall: {recall:.2f}')
+    print('**********')
+
+majority_voting_nonNullGPT = majority_voting_nonNullOriginCat[majority_voting_nonNullOriginCat.gpt_thumb_rating.isnull()==False]
+pro_only_maj_vote = majority_voting_nonNullGPT[majority_voting_nonNullGPT['originCat'] == 'pro']
+anti_only_maj_vote = majority_voting_nonNullGPT[majority_voting_nonNullGPT['originCat'] == 'anti']
+
+# flip all the labels for anti_only
+anti_only_maj_vote = anti_only_maj_vote.copy()
+anti_only_maj_vote.loc[:, 'gold_label_encoded'] = anti_only_maj_vote['gold_label_encoded'].replace({0: 1, 1: 0})
+anti_only_maj_vote.loc[:, 'gpt_encoded'] = anti_only_maj_vote['gpt_encoded'].replace({0: 1, 1: 0})
+
+print("number of videos that gpt rated")
+print(len(majority_voting_nonNullGPT))
+
+print('Comparing GPT AND OLD GOLD STANDARD LABELS')
+performance_gpt(majority_voting_nonNullGPT, topicID='OVERALL')
+performance_gpt(majority_voting_nonNullGPT, topicID='min_wage')
+performance_gpt(majority_voting_nonNullGPT, topicID='gun_control')
+
+print('PRO VIDEOS: Comparing GPT AND OLD GOLD STANDARD LABELS')
+performance_gpt(pro_only_maj_vote, topicID='OVERALL')
+performance_gpt(pro_only_maj_vote, topicID='min_wage')
+performance_gpt(pro_only_maj_vote, topicID='gun_control')
+
+print('ANTI VIDEOS: Comparing GPT AND OLD GOLD STANDARD LABELS')
+performance_gpt(anti_only_maj_vote, topicID='OVERALL')
+performance_gpt(anti_only_maj_vote, topicID='min_wage')
+performance_gpt(anti_only_maj_vote, topicID='gun_control')
+
+print("t-test for gold match percentage (general, majority vote)")
+majority_voting_nonNullGPT.loc[:, "is_human_match"] = majority_voting_nonNullGPT["gold_label_encoded"] == majority_voting_nonNullGPT["majority_label_encoded"]
+majority_voting_nonNullGPT.loc[:, "is_human_match_drop_insuf"] = majority_voting_nonNullGPT["gold_label_encoded"] == majority_voting_nonNullGPT["majority_label_drop_insuf_encoded"]
+majority_voting_nonNullGPT.loc[:, "is_gpt_match"] = majority_voting_nonNullGPT["gold_label_encoded"] == majority_voting_nonNullGPT["gpt_encoded"]
+
+print("Humans v. GPT")
+all_gold_match_perc = np.asarray([float(num) for num in majority_voting_nonNullGPT["is_human_match"]])
+t_statistic, p_value = stats.ttest_1samp(a=all_gold_match_perc, popmean=np.mean(majority_voting_nonNullGPT["is_gpt_match"])) 
+print("t-statistic:", t_statistic)
+print("p-value:", p_value)
+
+print("Humans (with 'Insufficient Data' Dropped) v. GPT")
+all_gold_match_perc = np.asarray([float(num) for num in majority_voting_nonNullGPT["is_human_match_drop_insuf"]])
+t_statistic, p_value = stats.ttest_1samp(a=all_gold_match_perc, popmean=np.mean(majority_voting_nonNullGPT["is_gpt_match"])) 
+print("t-statistic:", t_statistic)
+print("p-value:", p_value)
+
+
+# # What percentage of video thumbnails were 'clearly partisan?'
+majority_voting['Majority Vote Percentage'] = majority_voting['num_votes_for_majority'] / majority_voting['vote_count']
+
+# plot a histogram of the majority vote percentage
+sns.kdeplot(data=majority_voting, x='Majority Vote Percentage', hue='originCat')
+# add a vertical line at 0.5
+plt.axvline(0.5, color='r', linestyle='--')
+plt.axvline(0.8, color='lightpink', linestyle='--')
+plt.show()
+
+majority_voting_wage = majority_voting[majority_voting.topicID == 'min_wage']
+majority_voting_gun = majority_voting[majority_voting.topicID == 'gun_control']
+
+sns.kdeplot(data=majority_voting_gun, x='Majority Vote Percentage', hue='originCat')
+# add a vertical line at 0.5
+plt.axvline(0.5, color='r', linestyle='--')
+plt.axvline(0.8, color='lightpink', linestyle='--')
+plt.show()
+
+
+# ## Can we connect it with the original choices people made?
+
+# functions to get the original trees
+def explore_branches(tree_df, all_trees_by_ID, all_trees_by_channelID, row_index = 0, step = 0):
+    parent = tree_df.iloc[row_index]["originId"] # start with the first row
+    channel_id = tree_df.iloc[row_index]["originChannelId"]
+    
+    # create set of unique keys per step
+    if step not in all_trees_by_ID.keys():
+        all_trees_by_ID[step] = set() 
+    if step not in all_trees_by_channelID.keys():
+        all_trees_by_channelID[step] = set() 
+    
+    all_trees_by_ID[step].add(parent) # add the parent's video ID to the relevant step
+    all_trees_by_channelID[step].add(channel_id) # also store the channel ID
+    
+    for i in range(1, 4+1): # 4 + 1 because range() only prints up to n-1
+        child_node = tree_df.iloc[row_index]["rec"+str(i)]
+
+        # break if we hit a cycle
+        if(child_node in set().union(*all_trees_by_ID.values())):
+            break
+        else:
+            child_row_index = tree_df.index[tree_df['originId'] == child_node].tolist()
+            explore_branches(tree_df, all_trees_by_ID, all_trees_by_channelID, child_row_index[0], step+1) # call recursively to get all the tree levels
+
+def read_all_trees(tree_files):
+
+    all_trees_by_ID = {}
+    all_trees_by_channelID = {}
+    all_tree_files_df = pd.DataFrame()
+
+    for tree in tree_files:
+        print(tree)
+
+        # populate the tree
+        tree_df = pd.read_csv(tree)
+        
+        explore_branches(tree_df, all_trees_by_ID, all_trees_by_channelID) # recusrively parse out video ID's and channel ID's from the trees.
+
+        # save the dataframe to all_tree_files_df
+        if(all_tree_files_df.empty):
+            all_tree_files_df = tree_df
+        else:
+            all_tree_files_df = pd.concat([all_tree_files_df, tree_df], axis=0)
+
+    return all_trees_by_ID, all_trees_by_channelID
+
+tree_files_wage = glob.glob(os.path.join('../recommendation_trees/trees_wage/', '*.csv'))
+tree_files_gun = glob.glob(os.path.join('../recommendation_trees/trees_gun/', '*.csv'))
+
+
+all_trees_by_ID_wage, all_trees_by_channelID_wage = read_all_trees(tree_files_wage)
+
+all_trees_by_ID_gun, all_trees_by_channelID_gun = read_all_trees(tree_files_gun)
+
+# get the thing in parentheses as the topicid
+gun_topicids = [re.search(r'\((.*?)\)', filename).group(1) for filename in tree_files_gun]
+wage_topicids = [re.search(r'\((.*?)\)', filename).group(1) for filename in tree_files_wage]
+
+
+# # Thumbnail Distribution
+
+print('Total number of videos shown:', 
+      len(videos['gun_control']) + len(videos['min_wage']))
+
+print('Percentage of videos shown: %', 
+      np.round((len(videos['gun_control']) + len(videos['min_wage'])) / len(labels_on_platform), 3) * 100)
+
+unique_videos_shown = set(list(videos['gun_control'].keys()) + list(videos['min_wage'].keys()))
+unique_videos_in_platform_set = set(labels_on_platform['originId'])
+
+len(labels_on_platform)
+len(unique_videos_in_platform_set)
+unique_videos_in_platform_set.difference(unique_videos_shown)
+
+video_lengths = {}
+for vids, labellist in videos.items():
+    for vid, labels in labellist.items():
+        video_lengths[vid] = len(labels)
+
+
+average_number_of_ratings = np.mean([int(val) for val in video_lengths.values()])
+average_number_of_ratings
+
+# plot a histogram of video_lengths.values()
+plt.hist(video_lengths.values(), bins=40)
+# vertical line around the mean (average_number_of_ratings)
+plt.axvline(average_number_of_ratings, color='r', linestyle='dashed', linewidth=1)
+plt.xlabel('Number of Ratings')
+plt.ylabel('Number of Videos')
+
+
+# # Exploratory Analyses
+
+# ## What happens when we drop the cases where someone (either human or GPT) said there wasn't enough information?
+# When a rater abstains by saying there wasn't enough information, it deflates all the metrics, since the gold standard ratings are all binary.
+
+majority_voting_rated = majority_voting_nonNullGPT[(majority_voting_nonNullGPT.majority_label != "insufficient data.") & (majority_voting_nonNullGPT.gpt_thumb_rating != "insufficient data.")]
+len(majority_voting_rated) / len(majority_voting_nonNullGPT) # we now have 90% of the original videos
+
+len(majority_voting_rated)
+
+print('Comparing MTURKERS AND OLD GOLD STANDARD LABELS -- FILTERED')
+performance_metrics(majority_voting_rated, topicID='OVERALL')
+performance_metrics(majority_voting_rated, topicID='min_wage')
+performance_metrics(majority_voting_rated, topicID='gun_control')
+
+print('Comparing GPT AND OLD GOLD STANDARD LABELS -- FILTERED')
+performance_gpt(majority_voting_rated, topicID='OVERALL')
+performance_gpt(majority_voting_rated, topicID='min_wage')
+performance_gpt(majority_voting_rated, topicID='gun_control')
+
+
+# ## Are there any weird patterns that we should filter out?
+
+# For example, we might want to explore cases where people answered the same thing all 20 times, or always said 'insufficient information.'
+
+INDEXES_TO_EXCLUDE = set() # keep track of indices that we should exclude for various reasons
+
+exp_indexes_wage, exp_indexes_gun = session_rep_counts(real_data)
+indexes = exp_indexes_wage + exp_indexes_gun 
+
+for index in indexes:
+        
+    ratings = real_data[index]['ratingResults']
+    ratings_indices = [ratings[i]['index'] for i in range(0, len(ratings))]
+
+    # These people rated the same thing for all questions
+    if len(set(ratings_indices)) == 1:
+        INDEXES_TO_EXCLUDE.add(index)
+
+    # These people saw the same video multiple times but had inconsistent answers
+    rating_dict_for_individual = {}
+    for rating in ratings:
+        if(rating['vid'] not in rating_dict_for_individual.keys()):
+            rating_dict_for_individual[rating['vid']] = []
+        rating_dict_for_individual[rating['vid']].append(rating['index'])
+
+    # identify if any keys in rating_dict_for_individual have a length greater than 1
+    for key in rating_dict_for_individual.keys():
+        if len(rating_dict_for_individual[key]) > 1:
+            if(len(set(rating_dict_for_individual[key])) > 1): # these people had inconsistent responses when rating the same video
+                INDEXES_TO_EXCLUDE.add(index)
+
+
+# indices that we exclude for the above data quality reasons (button-smashing and inconsistent responses)
+INDEXES_TO_EXCLUDE
+
+indexes_cleaned = [index for index in indexes if index not in INDEXES_TO_EXCLUDE]
+
+# filter real_data to indexes_cleaned
+real_data_cleaned = [real_data[index] for index in indexes_cleaned]
+result_df_cleaned = thumbnail_exp_check(real_data_cleaned)
+
+# We don't actually do much better because there were only 7 participants dropped for data quality issues
+result_df_cleaned.groupby('topic_id').agg(
+    session_count=('session_id', 'nunique'),
+    respondent_count=('respondent_id', 'nunique'),
+    gold_match_mean=('gold_match_perc', 'mean'),
+    gold_match_std=('gold_match_perc', 'std'),
+    gold_match_quartiles=('gold_match_perc', calculate_quartiles)
+).reset_index()

code/supplemental/thumbnails (first impressions)/13_thumbnail_null_comparison.py

@@ -0,0 +1,893 @@
+#!/usr/bin/env python
+# coding: utf-8
+
+# # Re-analysis of Original Experiments to Evaluate Randomness of Video Recommendation Choices
+
+# # Libraries
+import numpy as np
+import pandas as pd
+import matplotlib.pyplot as plt
+import seaborn as sns
+import scipy.stats as stats 
+import json
+import re
+import os, glob
+from collections import Counter
+from statistics import mode
+from sklearn.metrics import accuracy_score, precision_score, recall_score
+from datetime import datetime
+import gc
+import statsmodels.api as sm
+from stargazer.stargazer import Stargazer
+import math
+from tqdm import tqdm
+from collections import defaultdict
+import random
+
+import warnings
+warnings.filterwarnings("ignore")
+
+print('=' * 80 + '\n\n' + 'OUTPUT FROM: supplemental/thumbnails (first impressions)/13_thumbnail_null_comparison.py' + '\n\n')
+
+with open("../../data/platform session data/sessions.json") as json_file:
+	json_data = json.load(json_file)
+
+# Actual ("Ground Truth") Labels
+gun_videos_all_metadata = pd.read_csv('../../data/supplemental/metadata and ratings/metadata_w_label_June_2021_NLversion.csv')
+wage_videos_all_metadata = pd.read_csv('../../data/supplemental/metadata and ratings/metadata_with_lables_binary_only_checked_0410.csv')
+gun_labels = gun_videos_all_metadata[['originID', 'originCat']].dropna().drop_duplicates().rename(columns={"originID": "originId"})
+wage_labels = wage_videos_all_metadata[['originID', 'originCat']].dropna().drop_duplicates().rename(columns={"originID": "originId"})
+gold_labels = pd.concat([gun_labels, wage_labels], axis = 0)
+
+# functions to get the original trees
+def explore_branches(tree_df, all_trees_by_ID, all_trees_by_channelID, row_index = 0, step = 0):
+	parent = tree_df.iloc[row_index]["originId"] # start with the first row
+	channel_id = tree_df.iloc[row_index]["originChannelId"]
+	
+	# create set of unique keys per step
+	if step not in all_trees_by_ID.keys():
+		all_trees_by_ID[step] = set() 
+	if step not in all_trees_by_channelID.keys():
+		all_trees_by_channelID[step] = set() 
+	
+	all_trees_by_ID[step].add(parent) # add the parent's video ID to the relevant step
+	all_trees_by_channelID[step].add(channel_id) # also store the channel ID
+	
+	for i in range(1, 4+1): # 4 + 1 because range() only prints up to n-1
+		child_node = tree_df.iloc[row_index]["rec"+str(i)]
+
+		# break if we hit a cycle
+		if(child_node in set().union(*all_trees_by_ID.values())):
+			break
+		else:
+			child_row_index = tree_df.index[tree_df['originId'] == child_node].tolist()
+			explore_branches(tree_df, all_trees_by_ID, all_trees_by_channelID, child_row_index[0], step+1) # call recursively to get all the tree levels
+
+def read_all_trees(tree_files):
+
+	all_trees_by_ID = {}
+	all_trees_by_channelID = {}
+	all_tree_files_df = pd.DataFrame()
+
+	for tree in tree_files:
+		print(tree)
+
+		# populate the tree
+		tree_df = pd.read_csv(tree)
+		
+		explore_branches(tree_df, all_trees_by_ID, all_trees_by_channelID) # recusrively parse out video ID's and channel ID's from the trees.
+
+		# save the dataframe to all_tree_files_df
+		if(all_tree_files_df.empty):
+			all_tree_files_df = tree_df
+		else:
+			all_tree_files_df = pd.concat([all_tree_files_df, tree_df], axis=0)
+
+	return all_trees_by_ID, all_trees_by_channelID
+
+
+tree_files_wage = glob.glob(os.path.join('../../data/recommendation trees/trees_wage/', '*.csv'))
+tree_files_gun = glob.glob(os.path.join('../../data/recommendation trees/trees_gun/', '*.csv'))
+
+
+all_trees_by_ID_wage, all_trees_by_channelID_wage = read_all_trees(tree_files_wage)
+all_trees_by_ID_gun, all_trees_by_channelID_gun = read_all_trees(tree_files_gun)
+
+# get the thing in parentheses as the topicid
+gun_topicids = [re.search(r'\((.*?)\)', filename).group(1) for filename in tree_files_gun]
+wage_topicids = [re.search(r'\((.*?)\)', filename).group(1) for filename in tree_files_wage]
+
+# Do a little filtering to get the set of "real" participants
+issue1 = pd.read_csv("../results/intermediate data/gun control (issue 1)/guncontrol_qualtrics_w123_clean.csv", dtype = {"urlid": str})
+
+print("(STUDY 1) Full length of data:")
+print(len(issue1))
+
+issue1 = issue1.dropna(subset=["treatment_arm"])
+
+print("number of treatment arm workers in complete data:")
+print(len(issue1))
+
+print("number once zero engagement is dropped:")
+issue1 = issue1.dropna(subset=["pro", "anti"])
+print(len(issue1))
+
+print("number of unique ID's:")
+print(len(issue1["worker_id"].drop_duplicates()))
+# identify the duplicates from the original issue1
+duplicate_workers_issue1 = issue1[issue1.duplicated(subset=["worker_id"], keep=False)]["worker_id"].drop_duplicates()
+# keep only the first response per unique worker_id
+issue1 = issue1.drop_duplicates(subset=["worker_id"], keep='first')
+
+
+print("number once NA topicID/urlID are dropped:")
+issue1 = issue1.dropna(subset=["topic_id", "urlid"])[["worker_id", "topic_id", "urlid"]]
+print(len(issue1))
+
+# merge in thirds
+thirds_workerid_i1 = pd.read_csv("../results/intermediate data/gun control (issue 1)/guncontrol_qualtrics_w12_clean.csv")[["thirds", "worker_id"]].drop_duplicates()
+issue1 = pd.merge(thirds_workerid_i1, issue1, on = "worker_id", how = "inner").drop_duplicates()
+
+print("number once we merge in thirds from wave 2 (they are not present in the dataframe):")
+print(len(issue1))
+
+# take a look at the responses of the repeated workers
+issue1_full = pd.read_csv("../results/intermediate data/gun control (issue 1)/guncontrol_qualtrics_w123_clean.csv", dtype = {"urlid": str})
+
+for worker in duplicate_workers_issue1:
+	worker_df = issue1_full[issue1_full["worker_id"]==worker][["gun_index_w3", "treatment_seed", "topic_id", "urlid"]].dropna()
+	if worker_df.empty:
+		continue
+	mismatched_columns = []
+	for col in worker_df.columns:
+		if not worker_df[col].eq(worker_df[col].iloc[0]).all():
+			mismatched_columns.append(col)
+			print(f"Worker: {worker};  Cols with mismatches: {mismatched_columns}")
+
+issue2 = pd.read_csv("../results/intermediate data/minimum wage (issue 2)/qualtrics_w12_clean.csv")
+
+print("(STUDY 2) Full length of data:")
+print(len(issue2))
+
+issue2 = issue2.dropna(subset=["treatment_arm"])
+
+print("number of treatment arm workers in complete data:")
+print(len(issue2))
+
+print("number once zero engagement is dropped:")
+issue2 = issue2.dropna(subset=["pro", "anti"])
+print(len(issue2))
+
+print("number of unique ID's:")
+print(len(issue2["worker_id"].drop_duplicates()))
+# identify the duplicates from the original issue2
+duplicate_workers_issue2 = issue2[issue2.duplicated(subset=["worker_id"], keep=False)]["worker_id"].drop_duplicates()
+# keep only the first response per unique worker_id
+issue2 = issue2.drop_duplicates(subset=["worker_id"], keep='first')
+
+print("number once NA topicID/urlID are dropped:")
+issue2 = issue2.dropna(subset=["topic_id", "urlid"])[["worker_id", "topic_id", "urlid", "thirds"]]
+print(len(issue2))
+
+# take a look at the responses of the repeated workers
+issue2_full = pd.read_csv("../results/intermediate data/minimum wage (issue 2)/qualtrics_w12_clean.csv")
+
+for worker in duplicate_workers_issue2:
+	worker_df = issue2_full[issue2_full["worker_id"]==worker][["mw_support_w2", "treatment_seed", "topic_id", "urlid"]].dropna()
+	if worker_df.empty:
+		continue
+	mismatched_columns = []
+	for col in worker_df.columns:
+		if not worker_df[col].eq(worker_df[col].iloc[0]).all():
+			mismatched_columns.append(col)
+			print(f"Worker: {worker};  Cols with mismatches: {mismatched_columns}")
+
+yougov_topicids = pd.read_csv("../results/intermediate data/minimum wage (issue 2)/yg_w12_clean.csv")
+
+print("(STUDY 3) Full length of data:")
+print(len(yougov_topicids))
+
+yougov_topicids = yougov_topicids.dropna(subset=["treatment_arm"])
+
+print("number of treatment arm workers in complete data:")
+print(len(yougov_topicids))
+
+print("number once zero engagement is dropped:")
+yougov_topicids = yougov_topicids.dropna(subset=["pro", "anti"])
+print(len(yougov_topicids))
+
+print("number of unique ID's:")
+print(len(yougov_topicids["caseid"].drop_duplicates()))
+# identify the duplicates from the original issue2
+duplicate_workers_yougov = yougov_topicids[yougov_topicids.duplicated(subset=["caseid"], keep=False)]["caseid"].drop_duplicates()
+# keep only the first response per unique worker_id
+yougov_topicids = yougov_topicids.drop_duplicates(subset=["caseid"], keep='first')
+
+print("number once NA topicID/urlID are dropped:")
+yougov_topicids = yougov_topicids.dropna(subset=["topic_id", "urlid"])[["caseid", "topic_id", "urlid", "thirds"]]
+print(len(yougov_topicids))
+
+# Get the topicIds assigned for people who were partisan (aka, filter out moderates)
+
+# these are the real people!
+
+# rename "worker_id" and "caseid" to "id"
+issue1 = issue1.rename(columns={"worker_id":"id"})
+issue2 = issue2.rename(columns={"worker_id":"id"})
+yougov_topicids = yougov_topicids.rename(columns={"caseid":"id"})
+
+# Decision (10/30): Drop Issue 1 entirely, because some people were recommended videos that were not pro/anti, and we don't have access to the rec set
+ALL_PARTICIPANTS = pd.concat([issue2[["id", "thirds", "urlid", "topic_id"]], yougov_topicids[["id", "thirds", "urlid", "topic_id"]]], axis = 0)
+
+len(ALL_PARTICIPANTS)
+
+
+# Parse out the condition (pro or anti) and the distribution (3-1 or 2-2) from the names and save them to the JSONs
+
+# Function to parse out the condition (pro or anti) and the distribution (3-1 or 2-2) from the names
+def parse_condition(topicId):
+	topic_components = topicId.split("_")[-4:]
+	distr = topic_components[1]
+	political_leaning = 'anti' if 'a' in topic_components[-1] else 'pro'
+
+	try:
+		return (int(distr), political_leaning)
+	except ValueError:
+		print("unable to extract distribution for: " + str(topicId))
+		# these are cases in which we can't extract a distribution; e.g., aTPMXi4EaKE_june2021_1_p
+		# exclude them from analysis for now by returning None
+		return (None, political_leaning)
+
+
+# Function to turn a video ID into a political leaning
+
+# this function converts from the longer video ID's (which have added numbers) to the "raw" format used in the gold labels
+def convert_vid_to_political_leaning(vidId):
+	# Keep up to the last non-numeric character and one trailing digit
+	vidId_mod = re.match(r'(.*\D\d)\d*$', vidId)
+	if vidId_mod:
+		vidId = vidId_mod.group(1)
+
+	matches = gold_labels[gold_labels["originId"] == vidId]["originCat"]
+
+	if not matches.empty:
+		return ', '.join(matches.astype(str))
+	else:
+		# If matches are empty, strip the last digit and try again
+		vidId_mod = re.match(r'(.*\D)\d$', vidId)
+		if vidId_mod:
+			vidId = vidId_mod.group(1)
+			matches = gold_labels[gold_labels["originId"] == vidId]["originCat"]
+			if not matches.empty:
+				return ', '.join(matches.astype(str))
+
+	return None
+
+
+# Filter the JSON blob to the LONGEST blob for each person
+
+topic_urlid_to_blob_map = defaultdict(list)
+for data_obj in json_data:
+	key = (str(data_obj["topicID"]), str(data_obj["urlid"]))
+	topic_urlid_to_blob_map[key].append(data_obj)
+	
+# get the longest data_obj per key
+topic_urlid_to_max_blob_map = defaultdict(dict)
+for key in topic_urlid_to_blob_map.keys():
+	blobs_list = []
+	vid_watch_times = []
+	for blob in topic_urlid_to_blob_map[key]:
+		try:
+			vid_watch_time = np.sum(list(blob['vidWatchTimes'].values()))
+		except KeyError:
+			vid_watch_time = 0
+		
+		blobs_list.append(blob) 
+		vid_watch_times.append(vid_watch_time)
+
+	max_blob = blobs_list[np.argmax(vid_watch_times)]
+	topic_urlid_to_max_blob_map[key] = max_blob
+
+
+# This is our main function for reading in the recommendations & what people chose
+
+def get_results_distribution(participant_url_identifiers, complete_only = False):
+
+	print("Number of participant identifiers:")
+	print(len(participant_url_identifiers))
+	
+	# create a set that appends the urlid to the topicId
+	urlid_topicid_set = {f"{row['urlid']}_{row['topic_id']}" for _, row in participant_url_identifiers.iterrows()}
+
+	# here, we're looking only at the MAX time blob for each person
+	orig_experiment_json = [obj for obj in topic_urlid_to_max_blob_map.values() if (obj['urlid'] + '_' + obj['topicID'] in urlid_topicid_set)]
+
+	# filter to only COMPLETE sessions
+	if(complete_only == True):
+		orig_experiment_json = [obj for obj in orig_experiment_json if obj["sessionFinished"] == True]
+
+	print("Number of participant JSON objects:")
+	print(len(orig_experiment_json))
+
+	# unpack information about the distribution and condition from the topicId
+	for obj in orig_experiment_json:
+		obj['distribution'], obj['political_leaning'] = parse_condition(obj["topicID"])
+
+	RESULTS_DICT = { # this is our main results dictionary
+		'pro': {22:[], 31: []},
+		'anti': {22:[], 31: []}
+	}
+
+	failures = { # log failures or reasons why we couldn't process all the data
+		"no_distribution": 0,
+		"videos_missing": 0,
+		"recs_incomplete": 0,
+		"activity_incomplete": 0,
+		"video_non_pro_anti": 0,
+		"reversed_politics_for_study1": 0}
+
+	NOBS_counted = 0
+
+	url_topic_id_processed = set()
+
+	for obj in orig_experiment_json:
+
+		participant_id = obj['urlid']
+
+		processed = False  # Keep track of whether we processed this participant
+		if(obj['distribution']) is None or obj['distribution'] not in {22, 31}:
+			failures["no_distribution"] += 1
+			continue  # We don't know the distribution; we can't analyze this
+
+		try:
+			recommendations = obj['displayOrders']
+		except KeyError:
+			failures["activity_incomplete"] += 1
+
+		if not processed:  # Only process recommendations if not already processed
+			try:
+				recs_keys = ['2-recs', '3-recs', '4-recs', '5-recs']
+				rec_info = {}
+				for i, key in enumerate(recs_keys):
+					rec_list = recommendations[key]
+					rec_list_leanings = [convert_vid_to_political_leaning(vid) for vid in rec_list]
+					rec_info[i+1] = {  # We start recs at level 2
+						"videos": rec_list,
+						"political_leaning": rec_list_leanings
+					}
+			except KeyError as e:
+				failures["recs_incomplete"] += 1 # Save what we have and move on to the next participant
+
+			watched_leanings = []
+			try:
+				p_watchlist = list(obj['vids'])
+				for rec_level, video in enumerate(p_watchlist):
+					if (rec_level == 0): continue  # Start recs at level 2
+					if(video in rec_info[rec_level]["videos"]):
+						political_leaning = convert_vid_to_political_leaning(video)
+						if(political_leaning not in ['pro', 'anti']):
+							continue
+						watched_leanings.append(political_leaning)
+			except KeyError as e:
+				failures["activity_incomplete"] += 1
+			
+			RESULTS_DICT[obj['political_leaning']][obj['distribution']].append({participant_id: watched_leanings})
+			processed = True  # Mark this participant as processed
+
+		# Counts the number of participants for which we were able to get valid data
+		if processed and (obj['topicID'], obj['urlid']) not in url_topic_id_processed:
+			NOBS_counted += 1
+			url_topic_id_processed.add((obj['topicID'], obj['urlid']))
+	
+	return RESULTS_DICT, NOBS_counted, failures, url_topic_id_processed
+
+
+# This is the version that includes PARTIAL data
+
+# run this for the liberals
+print("Liberals:")
+libs = ALL_PARTICIPANTS[ALL_PARTICIPANTS["thirds"] == 1]
+results_lib, nobs_lib, failures_lib, url_topic_id_processed_lib = get_results_distribution(libs)
+
+# run this for the conservatives
+print("Conservatives:")
+cons = ALL_PARTICIPANTS[ALL_PARTICIPANTS["thirds"] == 3]
+results_cons, nobs_cons, failures_cons, url_topic_id_processed_cons = get_results_distribution(cons)
+
+# (and just for kicks) run this for the moderates
+print("Moderates:")
+mods = ALL_PARTICIPANTS[ALL_PARTICIPANTS["thirds"] == 2]
+results_mods, nobs_mods, failures_mods, url_topic_id_processed_mods = get_results_distribution(mods)
+
+def turn_results_dict_to_dataframe(data):
+    rows = []
+
+    for top_level_key in data:
+        for second_level_key in data[top_level_key]:
+            for participant_dict in data[top_level_key][second_level_key]:
+                for participant_id, choices in participant_dict.items():
+                    row = {
+                        'pro': 1 if top_level_key == 'pro' else 0,
+                        'anti': 1 if top_level_key == 'anti' else 0,
+                        '22': 1 if second_level_key == 22 else 0,
+                        '31': 1 if second_level_key == 31 else 0,
+                        'participantID': participant_id,
+                        'choice_1': choices[0] if len(choices) > 0 else None,
+                        'choice_2': choices[1] if len(choices) > 1 else None,
+                        'choice_3': choices[2] if len(choices) > 2 else None,
+                        'choice_4': choices[3] if len(choices) > 3 else None,
+                    }
+                    rows.append(row)
+
+    return pd.DataFrame(rows)
+
+df_results_lib = turn_results_dict_to_dataframe(results_lib)
+df_results_cons = turn_results_dict_to_dataframe(results_cons)
+
+# EXPERIMENT: drop anyone who didn't finish
+df_results_lib_complete_only = df_results_lib.dropna(subset=["choice_1", "choice_2", "choice_3", "choice_4"])
+df_results_cons_complete_only = df_results_cons.dropna(subset=["choice_1", "choice_2", "choice_3", "choice_4"])
+
+# first, pivot the df to long
+df_libs_long = pd.melt(
+    df_results_lib,
+    id_vars=['pro', 'anti', '22', '31', 'participantID'], 
+    value_vars=['choice_1', 'choice_2', 'choice_3', 'choice_4'],  
+    var_name='choice_number',
+    value_name='choice'
+).dropna(subset=["choice"])
+
+df_cons_long = pd.melt(
+    df_results_cons,
+    id_vars=['pro', 'anti', '22', '31', 'participantID'], 
+    value_vars=['choice_1', 'choice_2', 'choice_3', 'choice_4'],  
+    var_name='choice_number',
+    value_name='choice'
+).dropna(subset=["choice"])
+
+# also pivot the complete only dfs
+df_libs_long_complete_only = pd.melt(
+    df_results_lib_complete_only,
+    id_vars=['pro', 'anti', '22', '31', 'participantID'], 
+    value_vars=['choice_1', 'choice_2', 'choice_3', 'choice_4'],  
+    var_name='choice_number',
+    value_name='choice'
+).dropna(subset=["choice"])
+
+df_cons_long_complete_only = pd.melt(
+    df_results_cons_complete_only,
+    id_vars=['pro', 'anti', '22', '31', 'participantID'], 
+    value_vars=['choice_1', 'choice_2', 'choice_3', 'choice_4'],  
+    var_name='choice_number',
+    value_name='choice'
+).dropna(subset=["choice"])
+
+# for lib, set 'pro' to 1 and 'anti' to 0 in choice_1, choice_2, choice_3, and choice_4
+df_libs_long[["choice"]] = df_libs_long[["choice"]].applymap(lambda x: 1 if x == 'pro' else 0)
+df_libs_long_complete_only[["choice"]] = df_libs_long_complete_only[["choice"]].applymap(lambda x: 1 if x == 'pro' else 0)
+
+# for cons, set 'pro' to 0 and 'anti' to 1 in choice_1, choice_2, choice_3, and choice_4
+df_cons_long[["choice"]] = df_cons_long[["choice"]].applymap(lambda x: 1 if x == 'anti' else 0)
+df_cons_long_complete_only[["choice"]] = df_cons_long_complete_only[["choice"]].applymap(lambda x: 1 if x == 'anti' else 0)
+
+df_cons_long_22 = df_cons_long[df_cons_long["22"] == 1]
+df_cons_long_31 = df_cons_long[df_cons_long["31"] == 1]
+
+# complete_only
+df_cons_long_complete_only_22 = df_cons_long_complete_only[df_cons_long_complete_only["22"] == 1]
+df_cons_long_complete_only_31 = df_cons_long_complete_only[df_cons_long_complete_only["31"] == 1]
+
+df_libs_long_22 = df_libs_long[df_libs_long["22"] == 1]
+df_libs_long_31 = df_libs_long[df_libs_long["31"] == 1]
+
+# complete_only
+df_libs_long_complete_only_22 = df_libs_long_complete_only[df_libs_long_complete_only["22"] == 1]
+df_libs_long_complete_only_31 = df_libs_long_complete_only[df_libs_long_complete_only["31"] == 1]
+
+
+# Linear Regression clustering by participant
+
+# run the regression for conservatives in 22
+y_cons22 = df_cons_long_22["choice"].astype(float)
+X_cons22 = sm.add_constant(pd.Series(1, index=y_cons22.index))
+model_cons22 = sm.OLS(y_cons22, X_cons22)
+results_cons22 = model_cons22.fit(cov_type='cluster', cov_kwds={'groups': df_cons_long_22['participantID']})
+print(results_cons22.summary())
+
+# run the regression for conservatives in 31
+y_cons31 = df_cons_long_31["choice"].astype(float)
+X_cons31 = sm.add_constant(pd.Series(1, index=y_cons31.index))
+model_cons31 = sm.OLS(y_cons31, X_cons31)
+results_cons31 = model_cons31.fit(cov_type='cluster', cov_kwds={'groups': df_cons_long_31['participantID']})
+print(results_cons31.summary())
+
+# run the regression for liberals in 22
+y_libs22 = df_libs_long_22["choice"].astype(float)
+X_libs22 = sm.add_constant(pd.Series(1, index=y_libs22.index))
+model_libs22 = sm.OLS(y_libs22, X_libs22)
+results_libs22 = model_libs22.fit(cov_type='cluster', cov_kwds={'groups': df_libs_long_22['participantID']})
+print(results_libs22.summary())
+
+# run the regression for liberals in 31
+y_libs31 = df_libs_long_31["choice"].astype(float)
+X_libs31 = sm.add_constant(pd.Series(1, index=y_libs31.index))
+model_libs31 = sm.OLS(y_libs31, X_libs31)
+results_libs31 = model_libs31.fit(cov_type='cluster', cov_kwds={'groups': df_libs_long_31['participantID']})
+print(results_libs31.summary())
+
+# now run the regressions for the complete only dfs
+# run the regression for conservatives in 22
+y_cons22_complete_only = df_cons_long_complete_only_22["choice"].astype(float)
+X_cons22_complete_only = sm.add_constant(pd.Series(1, index=y_cons22_complete_only.index))
+model_cons22_complete_only = sm.OLS(y_cons22_complete_only, X_cons22_complete_only)
+results_cons22_complete_only = model_cons22_complete_only.fit(cov_type='cluster', cov_kwds={'groups': df_cons_long_complete_only_22['participantID']})
+print(results_cons22_complete_only.summary())
+
+# run the regression for conservatives in 31
+y_cons31_complete_only = df_cons_long_complete_only_31["choice"].astype(float)
+X_cons31_complete_only = sm.add_constant(pd.Series(1, index=y_cons31_complete_only.index))
+model_cons31_complete_only = sm.OLS(y_cons31_complete_only, X_cons31_complete_only)
+results_cons31_complete_only = model_cons31_complete_only.fit(cov_type='cluster', cov_kwds={'groups': df_cons_long_complete_only_31['participantID']})
+print(results_cons31_complete_only.summary())
+
+df_libs_long_complete_only_22["choice"].dropna().mean().mean()
+
+df_libs_long_complete_only_22["choice"].describe()
+
+# complete only dfs
+# run the regression for liberals in 22
+y_libs22_complete_only = df_libs_long_complete_only_22["choice"].astype(float)
+X_libs22_complete_only = sm.add_constant(pd.Series(1, index=y_libs22_complete_only.index))
+model_libs22_complete_only = sm.OLS(y_libs22_complete_only, X_libs22_complete_only)
+results_libs22_complete_only = model_libs22_complete_only.fit(cov_type='cluster', cov_kwds={'groups': df_libs_long_complete_only_22['participantID']})
+print(results_libs22_complete_only.summary())
+
+# run the regression for liberals in 31
+y_libs31_complete_only = df_libs_long_complete_only_31["choice"].astype(float)
+X_libs31_complete_only = sm.add_constant(pd.Series(1, index=y_libs31_complete_only.index))
+model_libs31_complete_only = sm.OLS(y_libs31_complete_only, X_libs31_complete_only)
+results_libs31_complete_only = model_libs31_complete_only.fit(cov_type='cluster', cov_kwds={'groups': df_libs_long_complete_only_31['participantID']})
+print(results_libs31_complete_only.summary())
+
+
+# ### Get stats for how many observations of each seed we processed
+
+# Including INCOMPLETE data
+
+# how many observations did we get valid data from?
+print("Libs, Cons, Mods")
+print(nobs_lib, nobs_cons, nobs_mods)
+print("total:")
+print(nobs_lib + nobs_cons + nobs_mods)
+
+# ### Accounting by Topic ID + URL ID
+
+# of those observations, how many UNIQUE topicID + urlids did we get?
+### INCOMPLETE data included
+print("Libs, Cons, Mods")
+print(len(url_topic_id_processed_lib), len(url_topic_id_processed_cons), len(url_topic_id_processed_mods))
+print("total:")
+print(len(url_topic_id_processed_lib) + len(url_topic_id_processed_cons) +len(url_topic_id_processed_mods))
+
+
+# ### Look at Failures
+failures_lib, failures_cons, failures_mods
+
+# ### Summary Statistics
+def flatten_nested_dict(input_dict):
+    result = {
+        'pro': {22: [], 31: []},
+        'anti': {22: [], 31: []}
+    }
+    
+    for top_level_key, inner_dict in input_dict.items():
+        for second_level_key, participants in inner_dict.items():
+            for participant_dict in participants:
+                for choices in participant_dict.values():
+                    result[top_level_key][second_level_key].extend(choices)
+    
+    return result
+
+results_lib_flat = flatten_nested_dict(results_lib)
+results_cons_flat = flatten_nested_dict(results_cons)
+
+len(results_cons_flat['anti'][22])
+len(results_cons_flat['anti'][31])
+
+def print_summary_stats_for_results(RESULTS_DICT):
+	for seed, sub_dict in RESULTS_DICT.items():
+		print(f"Summary statistics for '{seed}' seed:")
+		
+		for key, labels in sub_dict.items():
+			total_labels = len(labels)
+			if total_labels == 0:
+				print(f"  List {key}: No labels to evaluate")
+				continue
+			
+			# Count how many labels agree with the parent category ('pro' or 'anti')
+			count_agree = sum(1 for label in labels if label == seed)
+			percent_agree = (count_agree / total_labels) * 100
+			
+			# Print the statistics
+			print(f"  [{key}]: {percent_agree:.2f}% selected videos with same partisanship as seed")
+			
+print("Liberals---------------------------")
+print_summary_stats_for_results(results_lib_flat)
+print("Conservatives----------------------")
+print_summary_stats_for_results(results_cons_flat)
+
+
+# ## Simulation of Random Guessing
+# 
+# This creates the baseline for which we compare everything
+
+def simulate_random_watching(N_ITER = 1000000, video_set = [1, 0, 0, 0]): # 1 is pro and 0 is anti
+
+    video_set_cur = video_set
+    all_watched_videos = []
+
+    for iter in range(N_ITER):
+    
+        watched_videos = []
+
+        for i in range(4): # we make 4 choices
+
+            # draws from the initial video distribution
+            random_index = np.random.choice(4, 1)[0]
+            random_video = video_set_cur[random_index]
+
+            if(random_video == 1):
+                video_set_cur = [1, 1, 1, 0]
+            else:
+                video_set_cur = [0, 0, 0, 1]
+
+            watched_videos.append(random_video)
+        
+        all_watched_videos.append(watched_videos)
+        # reset the video set
+        video_set_cur = video_set
+    
+    return all_watched_videos
+
+simulated_31_anti = simulate_random_watching(video_set=[1, 0, 0, 0])
+simulated_31_anti_flat =  [item for sublist in simulated_31_anti for item in sublist]
+print("Probability of selecting a pro video (given 3-1 distribution):")
+print(1-np.mean(simulated_31_anti_flat))
+p_31_anti = 1-np.mean(simulated_31_anti_flat)
+
+simulated_31_pro = simulate_random_watching(video_set=[0, 1, 1, 1])
+simulated_31_pro_flat = [item for sublist in simulated_31_pro for item in sublist]
+print("Probability of selecting a pro video (given 3-1 distribution):")
+print(np.mean(simulated_31_pro_flat))
+p_31_pro = np.mean(simulated_31_pro_flat)
+
+simulated_22 = simulate_random_watching(video_set=[1, 1, 0, 0])
+simulated_22_flat =  [item for sublist in simulated_22 for item in sublist]
+print("Probability of selecting a pro video (given 2-2 distribution):")
+print(np.mean(simulated_22_flat))
+p_22_pro = np.mean(simulated_22_flat)
+
+print("Probability of selecting an anti video (given 2-2 distribution):")
+print(1-np.mean(simulated_22_flat))
+p_22_anti = 1-np.mean(simulated_22_flat)
+
+
+# ### Statistical Tests
+
+# - whether conservative respondents, given a current conservative video, clicked a conservative recommendation at >.75 rate in the 3/1, or .5 in the 2/2
+# - whether conservative respondents, given a current liberal video, clicked a conservative recommendation at >.25 rate in the 3/1, or .5 in the 2/2
+# - whether liberal respondents, given a current liberal video, clicked a liberal recommendation at >.75 rate in the 3/1, or .5 in the 2/2
+# - whether liberal respondents, given a current conservative video, clicked a liberal recommendation at >.25 rate in the 3/1, or .5 in the 2/2
+
+simulated_baselines = {
+    "pro": {"22": p_22_pro, "31": p_31_pro},
+    "anti": {"22": p_22_anti, "31": p_31_anti}
+}
+
+# Function to calculate the proportion of 'pro' or 'anti' matches in the list
+def calculate_proportion_matches(results_list, target_key):
+	matches = [1 if value == target_key else 0 for value in results_list]
+	return matches
+
+# Collect the proportions and run t-tests for both keys (22 and 31)
+def run_stat_tests(RESULTS_DICT):
+	for parent_key, data in RESULTS_DICT.items():
+		print(f"Testing for parent key: {parent_key}")
+		
+		# For key 22, test against 0.5
+		if(len(data[22]) > 0):
+			proportion_22 = calculate_proportion_matches(data[22], parent_key)
+			t_statistic_22, p_value_22 = stats.ttest_1samp(a=proportion_22, popmean=simulated_baselines[parent_key]["22"])
+			print(f"t-test for 22 key (test {np.mean(proportion_22)} against {simulated_baselines[parent_key]['22']})")
+			print(f"t-statistic: {round(t_statistic_22, 5)}")
+			print(f"p-value: {round(p_value_22, 5)}")
+		
+		# For key 31, test against 0.75
+		if(len(data[31]) > 0):
+			proportion_31 = calculate_proportion_matches(data[31], parent_key)
+			t_statistic_31, p_value_31 = stats.ttest_1samp(a=proportion_31, popmean=simulated_baselines[parent_key]["31"])
+			print(f"t-test for 31 key (test {np.mean(proportion_31)} against {simulated_baselines[parent_key]['31']})")
+			print(f"t-statistic: {round(t_statistic_31, 10)}")
+			print(f"p-value: {round(p_value_31, 5)}")
+
+print("Liberals---------------------------")
+run_stat_tests(results_lib_flat)
+print("Conservatives----------------------")
+run_stat_tests(results_cons_flat)
+
+
+# # Figuring out data anomalies
+
+# First, there are some overlapping participants (326) between Study 1 and Study 2
+
+len(set(issue1["id"]).intersection(set(issue2["id"])))
+
+
+# These are the topicID's for everyone we expect to have data for
+
+ALL_PARTICIPANTS
+
+# Search for the participant in the JSON data and figure out whether we have at least one session for the participant in which they actually viewed data
+
+# these are the number of Topic/URL Id's among our participants
+unique_participant_topic_urlid = set([(str(row["topic_id"]), str(row["urlid"])) for _, row in ALL_PARTICIPANTS.iterrows()])
+len(unique_participant_topic_urlid)
+
+# these are the number of Topic/URL Id's within the JSON data
+unique_json_topic_urlid = set([(str(obj["topicID"]), str(obj["urlid"])) for obj in json_data])
+len(unique_json_topic_urlid)
+
+# Actually, every participant is in the data somewhere!
+len(unique_json_topic_urlid.intersection(unique_participant_topic_urlid))
+
+
+# Figure out what each of the participants did on the platform
+participant_session_counts = defaultdict(int)
+participant_vids_present_session_counts = defaultdict(int)
+participant_complete_session_counts = defaultdict(int)
+
+participants_set = set()
+
+topic_urlid_data_map = defaultdict(list)
+
+for data_obj in json_data:
+	key = (str(data_obj["topicID"]), str(data_obj["urlid"]))
+	topic_urlid_data_map[key].append({
+		"has_vids": 'vids' in data_obj.keys(),
+		"sessionFinished": data_obj.get("sessionFinished", False)
+	})
+
+for _, participant in tqdm(ALL_PARTICIPANTS.iterrows(), total=len(ALL_PARTICIPANTS), desc="Processing participants"):
+	
+	topicID = str(participant["topic_id"])
+	urlid = str(participant["urlid"])
+	participant_id = str(participant["id"])
+
+	# Use the (topicID, urlid) pair as the key
+	key = (topicID, urlid)
+
+	# Get the relevant data objects from the pre-built map
+	if key in topic_urlid_data_map:
+		
+		participants_set.add(participant_id)
+
+		for data_obj in topic_urlid_data_map[key]:
+			# Track if the participant has any appearances in the data at all
+			participant_session_counts[key] += 1
+
+			# Track how many times the participant saw valid videos
+			if data_obj["has_vids"]:
+				participant_vids_present_session_counts[key] += 1
+
+			# Track how many times they have complete data
+			if data_obj["sessionFinished"]:
+				participant_complete_session_counts[key] += 1
+
+
+# Who actually completed their sessions?
+len(participant_session_counts) # everyone is in the data somewhere
+
+len(participant_vids_present_session_counts) # everyone has some kind of video interaction
+
+len(participant_complete_session_counts) # only 5,573 people fully completed the study
+
+len(ALL_PARTICIPANTS)
+
+
+# But why is the number of unique participant ID's different?
+# 
+# **It turns out that it's because of the overlap in participants between Study 1 and Study 2!!!**
+
+# assert(len(ALL_PARTICIPANTS) - len(participants_set) == len(set(issue1["id"]).intersection(set(issue2["id"]))))
+
+
+# If we look only at "complete" sessions: the same participant has up to 3 complete sessions
+# 
+# If we include incomplete sessions: the same participant can have up to 14 partial sessions (!!!)
+
+pd.Series(participant_complete_session_counts.values()).value_counts()
+
+pd.Series(participant_vids_present_session_counts.values()).value_counts()
+
+pd.Series(participant_session_counts.values()).value_counts()
+
+# Who are the participants who didn't complete the study, and what were their other survey DV's like?
+
+no_session = unique_participant_topic_urlid.difference(set(participant_complete_session_counts.keys()))
+
+len(no_session)
+
+
+# 272 were in Study 1
+study1_topic_urlid = set([(str(row["topic_id"]), str(row["urlid"])) for _, row in issue1.iterrows()])
+len(study1_topic_urlid.intersection(no_session))
+
+# 196 were in Study 2
+study2_topic_urlid = set([(str(row["topic_id"]), str(row["urlid"])) for _, row in issue2.iterrows()])
+len(study2_topic_urlid.intersection(no_session))
+
+
+# 3 were in Study 3
+yougov_topic_urlid = set([(str(row["topic_id"]), str(row["urlid"])) for _, row in yougov_topicids.iterrows()])
+len(yougov_topic_urlid.intersection(no_session))
+
+issue1_full = pd.read_csv("../results/intermediate data/gun control (issue 1)/guncontrol_qualtrics_w123_clean.csv")
+issue2_full = pd.read_csv("../results/intermediate data/minimum wage (issue 2)/qualtrics_w12_clean.csv")
+yougov_full = pd.read_csv("../results/intermediate data/minimum wage (issue 2)/yg_w12_clean.csv")
+
+columns_to_collect = ["duration", "total_interactions", "gun_index_w2", "gun_index_2", "gun_index_w3", "stricter_laws_w3",
+					  "right_to_own_importance_w3", "assault_ban_w3", "handgun_ban_w3", "concealed_safe_w3",
+					  "gun_index_2_w3", "mw_index_w2", "trust_youtube_w2", "media_trust_w2", "media_trust_w3", "affpol_smart", 
+					  "smart_dems_w2", "smart_reps_w2"]
+
+collected_values = {col: [] for col in columns_to_collect}
+num_redundancies_by_study = {"Study_1": 0, "Study_2": 0, "Study_3": 0}
+
+def collect_values_from_study(df, id_column, participant_ids, collected_values):
+	filtered_df = df[df[id_column].isin(participant_ids)]
+	for col in columns_to_collect:
+		if col in filtered_df.columns:
+			collected_values[col].extend(filtered_df[col].dropna().tolist())
+
+for (topic_id, urlid) in no_session:
+	# Find matches in the datasets
+	matches_i1 = issue1[(issue1['topic_id'] == topic_id) & (issue1['urlid'] == urlid)]
+	matches_i2 = issue2[(issue2['topic_id'] == topic_id) & (issue2['urlid'] == urlid)]
+	matches_yg = yougov_topicids[(yougov_topicids['topic_id'] == topic_id) & (yougov_topicids['urlid'] == urlid)]
+
+	if not matches_i1.empty:
+		num_redundancies_by_study["Study_1"] += 1
+		collect_values_from_study(issue1_full, "worker_id", matches_i1["id"], collected_values)
+
+	if not matches_i2.empty:
+		num_redundancies_by_study["Study_2"] += 1
+		collect_values_from_study(issue2_full, "worker_id", matches_i2["id"], collected_values)
+
+	if not matches_yg.empty:
+		num_redundancies_by_study["Study_3"] += 1
+		collect_values_from_study(yougov_full, "caseid", matches_yg["id"], collected_values)
+
+# Filter columns with non-empty collected values
+non_empty_cols = {col: values for col, values in collected_values.items() if values}
+
+# Set up the plot grid based on the number of columns with data
+num_plots = len(non_empty_cols)
+n_cols = 3  # Maximum number of columns per row
+n_rows = math.ceil(num_plots / n_cols)  # Dynamically calculate number of rows
+
+# Set up the figure size dynamically based on number of plots
+plt.figure(figsize=(5 * n_cols, 4 * n_rows))
+
+# Plot histograms for non-empty columns
+for i, (col, values) in enumerate(non_empty_cols.items(), 1):
+	plt.subplot(n_rows, n_cols, i)
+	plt.hist(values, bins=20, color='blue', edgecolor='black')
+	plt.title(col)
+	plt.xlabel(col)
+	plt.ylabel('Frequency')
+
+plt.tight_layout()
+plt.show()
+
+num_redundancies_by_study
+

environment/Dockerfile

@@ -0,0 +1,45 @@
+# hash:sha256:44adf2d24cf24272656bd8a2cea0b2fc9a3de031215a3677db81b126432df80f
+FROM registry.codeocean.com/codeocean/py-r:python3.10.12-R4.3.2-JupyterLab4.0.10-RStudiorstudio-server-2023.12.0-369-ubuntu22.04
+
+ARG DEBIAN_FRONTEND=noninteractive
+
+RUN pip install -U --no-cache-dir \
+    matplotlib==3.10.0 \
+    numpy==1.26.4 \
+    pandas==2.2.3 \
+    rpy2==3.5.17 \
+    seaborn \
+    stargazer==0.0.7
+
+RUN Rscript -e 'remotes::install_version("car", "3.1-3")' \
+    && Rscript -e 'remotes::install_version("corrplot", "0.95")' \
+    && Rscript -e 'remotes::install_version("covr", "3.6.4")' \
+    && Rscript -e 'remotes::install_version("doParallel", "1.0.17")' \
+    && Rscript -e 'remotes::install_version("fastDummies", "1.7.5")' \
+    && Rscript -e 'remotes::install_version("feather", "0.3.5")' \
+    && Rscript -e 'remotes::install_version("ggtext", "0.1.2")' \
+    && Rscript -e 'remotes::install_version("janitor", "2.2.1")' \
+    && Rscript -e 'remotes::install_version("lubridate", "1.9.4")' \
+    && Rscript -e 'remotes::install_version("mockr", "0.2.1")' \
+    && Rscript -e 'remotes::install_version("psych", "2.4.12")' \
+    && Rscript -e 'remotes::install_version("randomizr", "1.0.0")' \
+    && Rscript -e 'remotes::install_version("sandwich", "3.1-1")' \
+    && Rscript -e 'remotes::install_version("stargazer", "5.2.3")' \
+    && Rscript -e 'remotes::install_version("systemfonts", "1.2.1")' \
+    && Rscript -e 'remotes::install_version("textshaping", "1.0.0")' \
+    && Rscript -e 'remotes::install_version("tidyverse", "2.0.0")'
+
+ADD "https://github.com/coder/code-server/releases/download/v4.95.3/code-server-4.95.3-linux-amd64.tar.gz" /.code-server/code-server.tar.gz
+	
+RUN cd /.code-server \
+    && tar -xvf code-server.tar.gz \
+    && rm code-server.tar.gz \
+    && ln -s /.code-server/code-server-4.95.3-linux-amd64/bin/code-server  /usr/bin/code-server
+
+RUN mkdir -p /.vscode/extensions \
+    && code-server --extensions-dir="/.vscode/extensions" --install-extension REditorSupport.R \
+    && code-server --extensions-dir="/.vscode/extensions" --install-extension continue.continue \
+    && code-server --extensions-dir="/.vscode/extensions" --install-extension ms-python.python \
+    && code-server --extensions-dir="/.vscode/extensions" --install-extension ms-toolsai.jupyter \
+    && code-server --extensions-dir="/.vscode/extensions" --install-extension reageyao.bioSyntax \
+    && code-server --extensions-dir="/.vscode/extensions" --install-extension saoudrizwan.claude-dev

metadata/metadata.yml

@@ -0,0 +1,34 @@
+metadata_version: 1
+name: Short-term exposure to filter-bubble algorithmic recommendations have limited
+  effects on polarization
+description: An enormous literature argues that recommendation algorithms drive political
+  polarization by creating "filter bubbles" and "rabbit holes." Using four experiments
+  with nearly 9,000 participants, we show that manipulating algorithmic recommendations
+  to create these conditions has limited effects on opinions. Our experiments employ
+  a custom-built video platform with a naturalistic, YouTube-like interface presenting
+  real YouTube videos and recommendations. We experimentally manipulate YouTube's
+  actual recommendation algorithm to simulate "filter bubbles" and "rabbit holes"
+  by presenting ideologically balanced and slanted choices. Our design allows us to
+  intervene in a feedback loop that has confounded the study of algorithmic polarization—the
+  complex interplay between *supply* of recommendations and user *demand* for content—to
+  examine downstream effects on policy attitudes. We use over 130,000 experimentally
+  manipulated recommendations and 31,000 platform interactions to estimate how recommendation
+  algorithms alter users' media consumption decisions and, indirectly, their political
+  attitudes. Our results cast doubt on widely circulating theories of algorithmic
+  polarization by showing that even heavy-handed (although short-term) perturbations
+  of real-world recommendations have limited causal effects on policy attitudes. Given
+  our inability to detect consistent evidence for algorithmic effects, we argue the
+  burden of proof for claims about algorithm-induced polarization has shifted. Our
+  methodology, which captures and modifies the output of real-world recommendation
+  algorithms, offers a path forward for future investigations of black-box artificial
+  intelligence systems. Our findings reveal practical limits to effect sizes that
+  are feasibly detectable in academic experiments.
+authors:
+- name: Naijia Liu, Xinlan Emily Hu, Yasemin Savas, Matthew Baum, Adam Berinsky, Allison
+    Chaney, Christopher Lucas, Rei Mariman, Justin de Benedictis-Kessner, Andrew Guess,
+    Dean Knox, and Brandon Stewart
+  affiliations:
+  - name: Multiple
+corresponding_contributor:
+  name: Dean Knox
+  email: dcknox@upenn.edu