|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| require(lme4)
|
| require(lfe)
|
| require(MatchIt)
|
| require(WeightIt)
|
| require(tjbal)
|
| require(optmatch)
|
| require(stargazer)
|
| require(cobalt)
|
| require(tidyverse)
|
| require(ggridges)
|
|
|
|
|
| rm(list = ls())
|
| gc()
|
|
|
| load('../Data/replication_data.RData')
|
|
|
|
|
|
|
| source('./helper_functions.R')
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| Y <- paste0(c("pct_","VAP_","pcttw_"),"sanders")
|
| D <- unlist(lapply(c("sc_","ln_"),function(x) paste0(x,paste0(c("county_","DMA_","state_"),"cases"))))
|
| FE <- c("0","DMA_CODE","date")
|
| covariates <- c("sc_CTY_LTHS","sc_CTY_CollUp",
|
| "sc_CTY_LT30yo","sc_CTY_60Up",
|
| "sc_CTY_Below_poverty_level_AGE_18_64","sc_CTY_Female_hher_no_husbandhh",
|
| "sc_CTY_Unem_rate_pop_16_over","sc_CTY_Labor_Force_Part_Rate_pop_16_over",
|
| "sc_CTY_Manufactur","sc_CTY_Md_inc_hhs",
|
| "sc_CTY_POPPCT_RURAL","sc_CTY_Speak_only_English","sc_CTY_White","sc_CTY_Black_or_African_American",
|
| "ln_CTY_tot_pop","sc_turnout_pct_20")
|
|
|
|
|
|
|
|
|
|
|
| set.seed(123)
|
| treatInd <- "March17Cases"
|
| vd <- '2020-03-17'
|
| d <- 'ln_DMA_cases'
|
| est <- 'meanfirst'
|
| treatInd <- paste0("DMA_",treatInd)
|
| forTjbalAnal$treatGroup <- ifelse(forTjbalAnal$voteDate == vd & forTjbalAnal[[treatInd]] > 0,1,0)
|
| forTjbalAnal$treat <- ifelse(forTjbalAnal$treatGroup == 1 & forTjbalAnal$date > as.Date("2020-04-17"),1,0)
|
|
|
| inds <- which(forTjbalAnal$treatGroup == 1)
|
| inds <- unique(forTjbalAnal$stcou[inds])
|
| ctrlInds <- unique(forTjbalAnal$stcou[-c(which(forTjbalAnal$stcou %in% inds),which(forTjbalAnal$voteDate >= as.Date(vd)))])
|
|
|
| tmpTjbalAnal <- forTjbalAnal %>% filter(stcou %in% c(inds,ctrlInds))
|
| tjout <- try(tjbal(data = as.data.frame(tmpTjbalAnal),Y = d,D = "treat",X = unique(c(covariates,"pct_sanders16")),
|
| index = c("stcou","dateNum"),estimator = est,demean = T,vce = "fixed"))
|
|
|
|
|
| toplot <- as.data.frame(tjout$Y.bar)
|
| toplot$date <- seq.Date(from = as.Date("2020-01-23"),to = as.Date("2020-05-01"),length.out = nrow(tjout$Y.bar))
|
| getwd()
|
| pdf("../Figures/SI_figure1.pdf")
|
| toplot %>%
|
| gather(type,value,-date) %>%
|
| ggplot(aes(x = date,y = value,color = type,size = type,linetype = type,alpha = type)) +
|
| geom_line() +
|
| geom_vline(xintercept = as.Date(unique(finalDat$Date)[which(as.Date(unique(finalDat$Date)) < as.Date("2020-03-17"))]),linetype = "dashed",alpha = .6) +
|
| geom_vline(xintercept = as.Date("2020-03-17"),size = 1.2) +
|
| xlab("Date") + ylab("Number of Cases (logged)") +
|
| annotate(geom = "text",label = c("Iowa Caucuses","New Hampshire","Nevada","South Carolina","Super Tuesday","March 10","March 17"),
|
| x = as.Date(c("2020-02-03","2020-02-11","2020-02-22","2020-02-29",
|
| "2020-03-03","2020-03-10","2020-03-17")),y = Inf,angle = 90,vjust = 1,hjust = 1,size = 3.5,color = "grey30") +
|
| annotate(geom = "text",label = c("Treated","Weighted","Control"),
|
| y = c(7.8,7.5,6.6),x = as.Date("2020-05-02"),angle = 0,
|
| parse = F,hjust = 0,vjust = .3,size = 3.5,color = "grey30") +
|
| scale_linetype_manual(values = c("solid","dotted","solid")) +
|
| scale_size_manual(values = c(1.2,1.2,2.5)) +
|
| scale_color_manual(values = c("red","black","black")) +
|
| scale_alpha_manual(values = c(1,1,.3)) +
|
| xlim(as.Date("2020-01-23"),as.Date("2020-05-10")) + theme_ridges() +
|
| theme(legend.position = "none") +
|
| theme(panel.grid.major = element_blank(), panel.grid.minor = element_blank(),
|
| panel.background = element_blank(), axis.line = element_line(colour = "black"))
|
| dev.off()
|
|
|
|
|
|
|
|
|
| set.seed(123)
|
| tjBalWgts <- list()
|
| for(vd in c("2020-03-03","2020-03-10","2020-03-17")) {
|
| for(d in c("county_cases")) {
|
| for(est in c("kernel")) {
|
|
|
| if(vd == "2020-03-03") {
|
| treatInd <- "March3Cases"
|
| } else if(vd == "2020-03-10") {
|
| treatInd <- "March10Cases"
|
| } else {
|
| treatInd <- "March17Cases"
|
| }
|
| if(grepl("county",d)) {
|
| treatInd <- paste0("county_",treatInd)
|
| } else {
|
| treatInd <- paste0("DMA_",treatInd)
|
| }
|
|
|
| forTjbalAnal$treatGroup <- ifelse(forTjbalAnal$voteDate == vd & forTjbalAnal[[treatInd]] > 0,1,0)
|
| forTjbalAnal$treat <- ifelse(forTjbalAnal$treatGroup == 1 & forTjbalAnal$date > as.Date("2020-04-17"),1,0)
|
|
|
| inds <- which(forTjbalAnal$treatGroup == 1)
|
| inds <- unique(forTjbalAnal$stcou[inds])
|
| ctrlInds <- unique(forTjbalAnal$stcou[-c(which(forTjbalAnal$stcou %in% inds),which(forTjbalAnal$voteDate >= as.Date(vd)))])
|
|
|
| ests <- finalDat %>% select(stcou)
|
| for(i in 1:length(inds)) {
|
| tmpTjbalAnal <- forTjbalAnal %>% filter(stcou %in% c(inds[-i],ctrlInds))
|
| sink <- capture.output(tjout <- try(tjbal(data = as.data.frame(tmpTjbalAnal),Y = d,D = "treat",
|
| X = unique(c(covariates,"pct_sanders16")),seed = 123,
|
| index = c("stcou","dateNum"),estimator = est,demean = T,vce = "fixed")))
|
|
|
| if(class(tjout) == "try-error") { next }
|
| tjout$data.wide$w <- tjout$w
|
| ests <- ests %>% left_join(tjout$data.wide %>% select(stcou = unit,w),by = 'stcou')
|
| colnames(ests)[which(colnames(ests) == "w")] <- paste0("w",i)
|
| cat('.')
|
| }
|
| cat('\n')
|
|
|
| tjBalWgts[[vd]][[d]][[est]]$jacknife <- ests
|
| tmpTjbalAnal <- forTjbalAnal %>% filter(stcou %in% c(inds,ctrlInds))
|
| sink <- capture.output(tjout <- try(tjbal(data = as.data.frame(tmpTjbalAnal),Y = d,D = "treat",X = unique(c(covariates,"pct_sanders16")),
|
| index = c("stcou","dateNum"),estimator = est,demean = T,seed = 123,
|
| vce = "fixed")))
|
|
|
| if(class(tjout) == "try-error") { next }
|
| tjout$data.wide$w <- tjout$w
|
| tjBalWgts[[vd]][[d]][[est]]$full <- tjout$data.wide %>% select(stcou = unit,w)
|
| }
|
| }
|
| }
|
|
|
| save(tjBalWgts,file = "../Data/Results/tjbalWgtsNEW.RData")
|
| load("../Data/Results/tjbalWgtsNEW.RData")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| resTjbal <- list()
|
| for(y in Y) {
|
| for(geo in c("county_","DMA_")) {
|
| for(d in c("March3Cases","March10Cases","March17Cases")) {
|
| if(d == "March3Cases") {
|
| int <- "2020-03-03"
|
| } else if(d == "March10Cases") {
|
| int <- "2020-03-10"
|
| } else {
|
| int <- "2020-03-17"
|
| }
|
| for(balCases in names(tjBalWgts[[int]])) {
|
| for(est in names(tjBalWgts[[int]][[balCases]])) {
|
|
|
| if(is.null(tjBalWgts[[int]][[balCases]][[est]]$full)) { next }
|
|
|
| finalDat %>% left_join(tjBalWgts[[int]][[balCases]][[est]]$full) %>%
|
| left_join(tjBalWgts[[int]][[balCases]][[est]]$jacknife) %>% filter(!is.na(w)) -> tmpAnal
|
| tmpAnal$post <- ifelse(tmpAnal$date == as.Date(int),1,0)
|
| tmpAnal$treatGroup <- ifelse(tmpAnal[[paste0(geo,d)]] > 0,1,0)
|
| tmpAnal$treat <- ifelse(tmpAnal$post == 1 & tmpAnal$treatGroup == 1,1,0)
|
| tmpAnal %>% select(post,treatGroup,treat,date)
|
|
|
| ests <- NULL
|
| for(i in colnames(tmpAnal %>% select(matches("^w\\d")))) {
|
| ests <- c(ests,sum(tmpAnal[[y]]*tmpAnal[[i]],na.rm=T))
|
| }
|
| resTjbal[[y]][[geo]][[d]][[balCases]][[est]]$bootstrapped <- ests
|
| resTjbal[[y]][[geo]][[d]][[balCases]][[est]]$lmCont <- summary(lm(as.formula(paste0(y," ~ ",balCases)),tmpAnal,weights = abs(tmpAnal$w)))$coefficients
|
| resTjbal[[y]][[geo]][[d]][[balCases]][[est]]$lmBin <- summary(lm(as.formula(paste0(y," ~ treat")),tmpAnal,weights = abs(tmpAnal$w)))$coefficients
|
| }
|
| }
|
| }
|
| }
|
| }
|
|
|
|
|
| toplot <- NULL
|
| for(period in names(resTjbal$pcttw_sanders$DMA_)) {
|
| toplot <- bind_rows(toplot,
|
| data.frame(bs = resTjbal$pcttw_sanders$county_[[period]]$county_cases$kernel$bootstrapped,
|
| period = period))
|
| }
|
|
|
|
|
|
|
|
|
|
|
| pdf('../Figures/figure6.pdf',width = 7,height = 5)
|
| toplot %>%
|
| mutate(period = factor(gsub('Cases','',
|
| gsub('March','March ',period)),levels = rev(c('March 3','March 10','March 17')))) %>%
|
| ggplot(aes(x = bs,y = period)) +
|
| geom_density_ridges(alpha = .7,color = 'black') +
|
| theme_ridges() +
|
| geom_vline(xintercept = 0,linetype = 'dashed') +
|
| xlab('ATT of Exposure on Sanders Support') + ylab('Outbreak Date')
|
| dev.off() |
