estimate_contrasts.R

library(tidyverse)
library(modelbased)
library(magrittr)

overview <- readRDS("data/overview.rds")

## helper function
get_plabs <- function (pval) {
  dplyr::case_when(is.na(pval) ~ "", pval < 0.001 ~ "***", 
                   pval < 0.01 ~ "**", pval < 0.05 ~ "*", 
                   pval < 0.1 ~ "", TRUE ~ "")
}

## estimate model
reach_mod <- lm(reach ~ targeting * party  + engagement, data = overview) 


## estimate contrasts
contrasts_nobreak <- estimate_contrasts(reach_mod, 
                                        contrast = c("targeting", "party"),
                                        at = c("targeting",  "party")) %>% 
  as.data.frame()  %>%  
  mutate(Contrast = paste(Level1, "-", Level2)) %>%  
  #### in the following I filter down the data to only the comparisons I am interested in ####
  ## only keep within partyy comparisons (i.e. if party appears twice)
  filter(
    str_count(condition_comparison, "VVD") == 2 |
    str_count(condition_comparison, "GroenLinks") == 2 |
    str_count(condition_comparison, "PvdA") == 2 
  ) %>% 
  ## only keep comparison between targeting pairs (i.e. targeting appears twice)
  filter(
    str_count(condition_comparison, "Environment") == 2 |
    str_count(condition_comparison, "Economy") == 2 |
    str_count(condition_comparison, "Politics") == 2 |
    str_count(condition_comparison, "Education") == 2 
  ) %>% 
  # in case the comparison is in wrong direction, change around
  mutate_at(vars(Difference, CI_low, CI_high), ~ifelse(str_detect(Level1, "excluded"), .x*-1, .x)) %>%
  mutate_at(vars(Difference, CI_low, CI_high), ~ifelse(str_detect(Level1, "Low"), .x*-1, .x)) %>%
  ## get labels for p-values
  mutate(plabel = get_plabs(p)) %>% 
  mutate(diff_label = paste0(round(Difference), plabel)) %>%
  ## some more labelling for plot
  mutate(party = str_extract(condition_comparison, "VVD|GroenLinks|PvdA")) %>% 
  mutate(condition_comparison = str_remove_all(condition_comparison, "VVD|GroenLinks|PvdA")) %>% 
  mutate(condition_comparison = case_when(
    str_detect(condition_comparison, "Economy") ~ "Economy (vs. not)",
    str_detect(condition_comparison, "Education") ~ "Higher Educated (vs. lower)",
    str_detect(condition_comparison, "Politics") ~ "Politics (vs. not)",
    str_detect(condition_comparison, "Environment") ~ "Environment (vs. not)"
    )) %>% 
  mutate(condition_comparison = fct_reorder(condition_comparison, Difference)) 
  
## create visualization
contrasts_nobreak %>%  
  ggplot(aes(condition_comparison, Difference, color = party)) + 
  geom_point(position = position_dodge(width = 0.9)) +
  geom_errorbar(aes(ymin = CI_low, ymax = CI_high), width = 0, position = position_dodge(width = 0.9)) +
  coord_flip() +
  geom_text(aes(label = diff_label, x = as.factor(condition_comparison) %>% as.numeric() %>% magrittr::add(0.15)), show.legend = F, position = position_dodge(width = 0.9)) +
  geom_hline(yintercept = 0, linetype = "dashed") +
  labs(y = "Estimated reach differences",
       x = "Targeting Comparisons") +
  theme_minimal() +
  theme(legend.position = "bottom",
        plot.title = element_text(size = 19, face = "bold"),
        strip.text.y = element_blank(), 
        axis.text.y = element_text(face = "bold"),
        strip.background = element_rect(fill = "lightgrey"))   +
  ggtitle("Reach differences per party")