In xmarquez/AuthoritarianismBook: Replication Package for the Book Non-Democratic Politics by Xavier Marquez

knitr::opts_chunk$set(echo = TRUE,
                      cache = TRUE,
                      cache.rebuild = FALSE,
                      message = FALSE,
                      warning = FALSE)

Preparing the data

We first gather the data on personal power and put it in the right form. The first step is to add the VDem variables:

library(tidyverse)
library(PoliticalDatasets)
library(reshape2)

personalism_data <- vdem %>% 
  select(country_name,GWn,year,in_system,v2xlg_legcon,v2x_jucon,v2x_execorr) %>%
  melt(id.vars = c("country_name","GWn","year","in_system"),na.rm=TRUE)

Then we add the magaloni personalism measure lindex:

personalism_data <- bind_rows(personalism_data, magaloni %>%
                                select(country_name,GWn,year,in_system,lindex) %>%
                                melt(id.vars = c("country_name","GWn","year","in_system"),na.rm=TRUE)) 

And Wahman, Teorell, and Hadenius's persnagg1ny measure:

personalism_data <- bind_rows(personalism_data, wahman_teorell %>%
                                select(country_name,GWn,year,in_system,persagg1ny) %>%
                                melt(id.vars = c("country_name","GWn","year","in_system"),na.rm=TRUE)) 

And Kailitz's measure of whether the regime has a "personal" component:

personalism_data <- bind_rows(personalism_data, kailitz.yearly %>%
                                select(country_name,GWn,year,in_system,personal) %>%
                                melt(id.vars = c("country_name","GWn","year","in_system"),na.rm=TRUE)) 

Then Polity IV's measure of executive constraints, which we need to clean up to eliminate the special codes:

polity_annual <- polity_annual %>%  mutate(exconst = ifelse(exconst < -10, NA, exconst))

personalism_data <- bind_rows(personalism_data, polity_annual %>%
                                select(country_name,GWn,year,in_system,exconst) %>%
                                melt(id.vars = c("country_name","GWn","year","in_system"),na.rm=TRUE)) 

rm(polity_annual)

Then Geddes, Wright, and Frantz's measure of whether a regime has a personal component:

all_gwf_extended_yearly <- all_gwf_extended_yearly %>%
  select(country_name,GWn,year,in_system,gwf_full_regimetype) %>%
  mutate(gwf_personal_single = grepl("^personal",gwf_full_regimetype),
         gwf_personal_hybrid  = grepl("personal",gwf_full_regimetype),
         gwf_personal = gwf_personal_single + gwf_personal_hybrid) %>%
  select(-gwf_personal_single,-gwf_personal_hybrid)

personalism_data <- bind_rows(personalism_data, all_gwf_extended_yearly %>%
                                select(-gwf_full_regimetype) %>%
                                melt(id.vars = c("country_name","GWn","year","in_system"),na.rm=TRUE)) 

rm(all_gwf_extended_yearly)

And we put it in the right shape:

personalism_data <- dcast(personalism_data, country_name + GWn + year + in_system ~ variable)

This is what the data should look like:

personalism_data %>%
  filter(!is.na(personal), !is.na(lindex), !is.na(persagg1ny)) %>%
  head() %>%
  knitr::kable()

summary(personalism_data)

We can check the correlations:

library(GGally)

ggcorr(data= personalism_data %>% select(exconst:v2xlg_legcon), label = TRUE, label_round = 2)

We then cut the continuous variables into intervals:

personalism_data <- personalism_data %>% 
  mutate_each(funs(unclass(cut(., 
                               breaks = 20, 
                               include.lowest = TRUE, 
                               right = FALSE, 
                               ordered_result = TRUE))),
              v2x_execorr:v2xlg_legcon)

personalism_data %>%
  filter(!is.na(personal), !is.na(lindex), !is.na(persagg1ny)) %>%
  head() %>%
  knitr::kable()

personalism_data <- personalism_data %>% 
  mutate(lindex = unclass(cut(percent_rank(lindex), 
                              20, 
                              include.lowest=TRUE, 
                              ordered_result=TRUE)), 
         persagg1ny = unclass(cut(percent_rank(persagg1ny),
                                  20,
                                  include.lowest=TRUE, 
                                  ordered_result=TRUE)))

personalism_data %>%
  filter(!is.na(personal), !is.na(lindex), !is.na(persagg1ny)) %>%
  head() %>%
  knitr::kable()

personalism_data <- personalism_data %>% 
  mutate_each(funs(unclass(factor(., 
                                  ordered = TRUE))), 
              v2xlg_legcon:gwf_personal)

personalism_data %>%
  filter(!is.na(personal), !is.na(lindex), !is.na(persagg1ny)) %>%
  head() %>%
  knitr::kable()

personalism_data <- personalism_data %>% 
  mutate_each(funs(as.numeric(.)), 
              exconst:v2xlg_legcon)

personalism_data %>%
  filter(!is.na(personal), !is.na(lindex), !is.na(persagg1ny)) %>%
  head() %>%
  knitr::kable()

We can check the correlations again:

library(GGally)

ggcorr(data= personalism_data %>% select(exconst:v2xlg_legcon), label = TRUE, label_round = 2)

Finally we create the model:

library(mirt)
personal_model <- mirt(personalism_data %>% 
                         select(exconst:v2xlg_legcon), 
                       model = 1, 
                       itemtype = "graded", 
                       SE = TRUE, 
                       verbose = FALSE)

personal_model

summary(personal_model)

personal_scores <- fscores(personal_model, full.scores = TRUE, full.scores.SE = TRUE) %>% 
  data.frame() %>% 
  dplyr::rename(z1 = F1, se.z1 = SE_F1) %>% 
  mutate(pct975 = z1 + 1.96 * se.z1, pct025 = z1 - 1.96 * se.z1)

personal_scores <- bind_cols(personalism_data %>% select(country_name,GWn,year,in_system),personal_scores)

And do some visual testing

# Visual test
leader_data <- archigos2014 %>% 
  group_by(GWn,leader,startdate,enddate) %>%
  summarise(year = year(startdate))


personal_data <- left_join(personal_scores,leader_data)

for(country in unique(personal_data$country_name)) {
  message(country)

  data <- personal_data %>% filter(country_name %in% country, year > 1900)

  if(length(data$year[!is.na(data$leader)]) > 0) {
  p <- ggplot(data = data, aes(x=year,y = -z1, ymin = -pct025, ymax = -pct975, fill = in_system)) +
    geom_path() +
    geom_ribbon(alpha=0.2) +
    theme_bw() +
    scale_x_continuous(breaks = data$year[!is.na(data$leader)],
                       labels = paste(data$year[!is.na(data$leader)], data$leader[!is.na(data$leader)])) +
    theme(axis.text.x = element_text(angle = 90, hjust = 1, vjust = 0.5)) +
    facet_wrap(~country_name, scales="free")

  print(p)

  }

}

Finally, we save the data for package use:

devtools::use_data(personal_scores, overwrite = TRUE)

xmarquez/AuthoritarianismBook documentation built on May 4, 2019, 1:24 p.m.