analysis/scripts/rq2.R

library(tidyverse) 
library(brms)      
library(lme4)      

# 1. Data Preparation
emoji_df <- read.csv("RQ2_data.csv")

# Ensure proper factoring
emoji_df$two_emojis <- fct_relevel(as.factor(emoji_df$two_emojis), "False")

# Scale and Center variables (as.numeric prevents matrix issues in CSV export)
emoji_df <- emoji_df %>% mutate(
  z_speaker_variability = as.numeric(scale(percentage_different)),
  z_speaker_emoji_use   = as.numeric(scale(speaker_emoji_usage)),
  z_listener_emoji_use  = as.numeric(scale(listener_age)), # Fixed based on your previous logic
  z_listener_age        = as.numeric(scale(listener_age))
)

# Subset to only pairs judged as different
distinct_pairs_df <- emoji_df %>% filter(similar_sounding == "no")

# 2. Frequentist Models (lme4)
freq_rq2_accuracy <- glmer(
  is_correct ~ z_speaker_variability + two_emojis + 
    (z_speaker_emoji_use * z_listener_emoji_use) + 
    (1 | utterance_id) + (1 | listener_id),
  data = distinct_pairs_df, family = binomial,
  control = glmerControl(optimizer = "bobyqa")
)

freq_rq2_pass <- glmer(
  is_pass ~ z_speaker_variability + two_emojis + 
    (z_speaker_emoji_use * z_listener_emoji_use) + 
    (1 | utterance_id) + (1 | listener_id),
  data = distinct_pairs_df, family = binomial,
  control = glmerControl(optimizer = "bobyqa")
)

summary(freq_rq2_accuracy)
summary(freq_rq2_pass)


# 3. Bayesian Models (brms)
bayesian_rq2_accuracy <- brm(
  is_correct ~ z_speaker_variability + two_emojis + 
    (z_speaker_emoji_use * z_listener_emoji_use) + 
    (1 | utterance_id) + (1 | listener_id),
  data = distinct_pairs_df, family = bernoulli(),
  warmup = 1000, iter = 2000, seed = 11, chains = 4, cores = 4,
  prior = c(set_prior('normal(0,0.5)', class = "b"))
)

bayesian_rq2_pass <- brm(
  is_pass ~ z_speaker_variability + two_emojis + 
    (z_speaker_emoji_use * z_listener_emoji_use) + 
    (1 | utterance_id) + (1 | listener_id),
  data = distinct_pairs_df, family = bernoulli(),
  warmup = 1000, iter = 2000, seed = 11, chains = 4, cores = 4,
  prior = c(set_prior('normal(0,0.5)', class = "b"))
)

summary(bayesian_rq2_accuracy)
summary(bayesian_rq2_pass)


# --- Frequentist Results ---
write.csv(as.data.frame(summary(freq_rq2_accuracy)$coefficients), 
          "RQ2_Freq_Accuracy_singleemoji.csv")

write.csv(as.data.frame(summary(freq_rq2_pass)$coefficients), 
          "RQ2_Freq_Pass_singleemoji.csv")

# --- Bayesian Results ---
write.csv(as.data.frame(fixef(bayesian_rq2_accuracy)), 
          "RQ2_Bayes_Accuracy_singleemoji.csv")

write.csv(as.data.frame(fixef(bayesian_rq2_pass)), 
          "RQ2_Bayes_Pass_singleemoji.csv")