Ministerial exits from government

exits.R

### Load libraries
library(tidyverse)
library(countrycode)
library(hrbrthemes)

### Download the files if they do not exist already
govf <- "Seki-Williams Governments--Version 2.0.csv"
minf <- "Seki-Williams Ministers--Version 2.0.csv"

if (!file.exists(govf)) {
    download.file("http://faculty.missouri.edu/williamslaro/Seki-Williams%20Governments--Version%202.0.csv",
                  destfile = govf)
}

if (!file.exists(minf)) {
    download.file("http://faculty.missouri.edu/williamslaro/Seki-Williams%20Ministers--Version%202.0.csv",
                  destfile = minf)
}

### Read in the files
gov <- read.csv(govf)
min <- read.csv(minf)

### Handle dates
gov <- gov %>%
    mutate(gstart = paste(startday, startmonth, startyear,
                           sep = "/"),
           gstart = as.Date(gstart,
                            format = "%d/%m/%Y"),
           gend = gstart + duration)

min <- min %>%
    mutate(mstart = paste(mstart_day, mstart_month, mstart_year,
                          sep = "/"),
           mstart = as.Date(mstart,
                          format = "%d/%m/%Y"),
           mend = paste(mend_day, mend_month, mend_year,
                         sep = "/"),
           mend = as.Date(mend,
                          format = "%d/%m/%Y")) 

### Handle countries
gov$iso3c <- countrycode(gov$country,
                         "country.name",
                         "iso3c")
min$iso3c <- countrycode(min$ccode,
                         "country.name",
                         "iso3c")

### Select only some variables
gov <- gov %>%
    select(iso3c, govtseq, prime, gstart, gend)
min <- min %>%
    select(iso3c, govtseq, minister_id, mstart, mend)

### Merge
min <- merge(min, gov,
             all.x = TRUE,
             all.y = FALSE)

### Remove those spells which terminated at the same time as the government
min <- min %>%
    filter(mend < gend)

### Convert multiple ministerial spells or multiple portfolios to a
### single entry

### Case (1): two spells w/ duplicated start, end-dates
### (b/c multiple portfolios)
### select one unique row
nrow(min)
min <- min %>%
    group_by(iso3c, govtseq, prime, minister_id, gstart, gend) %>%
    distinct(mstart, mend)
nrow(min)

### Case (2): two spells w/ duplicated end-dates
### (b/c multiple nonconsecutive portfolios)
### select row with longer span
nrow(min)
min <- min %>%
    group_by(iso3c, govtseq, prime, minister_id, mend) %>%
    filter(mstart == min(mstart))
nrow(min)

### Case (3): two spells w/ duplicated start-dates
### (b/c multiple nonconsecutive portfolios)
### select row with longer span
nrow(min)
min <- min %>%
    group_by(iso3c, govtseq, prime, minister_id, mstart) %>%
    filter(mend == max(mend))
nrow(min)

### Case (3): two spells which overlap
### What to do? First, check for continuity
### I'll take any spell longer than a week as evidence of a break
checkContinuous <- function(start, end) {
    if (any(is.na(end))) {
        return(NA)
    }
    if (any(start > end)) {
        return(NA)
    }
    
    outseq <- lapply(1:length(start), function(i) {
        seq.Date(from = start[i], to = end[i], by = 1)
    })
    outseq <- sort(unique(do.call("c", outseq)))
    diffs <- diff(outseq)
    isBroken <- any(diff(outseq) > 7)
    return(!isBroken)
}

  
min <- min %>%
    group_by(iso3c, govtseq, minister_id) %>%
    arrange(mstart) %>% 
    mutate(contig = checkContinuous(mstart, mend))

### Deal with contiguous and non-contiguous separately
### Treat non-contiguous as though they were all non-overlapping
min_a <- min %>%
    filter(contig) %>%
    group_by(iso3c, govtseq, prime, minister_id, gstart, gend) %>%
    summarize(mstart = min(mstart),
              mend = max(mend))

min_b <- min %>%
    filter(!contig)

min <- merge(min_a, min_b, all = TRUE)
rm(min_a, min_b)

plot.df <- min %>%
    select(iso3c, govtseq, prime, mend, gstart, gend) %>% 
    group_by(iso3c, govtseq) %>%
    arrange(mend) %>%
    mutate(tally = 1:n(),
           daysAfter = as.numeric(mend - gstart)) %>%
    group_by(iso3c, govtseq) %>%
    filter(daysAfter != max(daysAfter))


### Add on a zero row
addon <- subset(plot.df, tally == 1)
addon$mend <- addon$gstart
addon$tally <- 0
addon$daysAfter <- 0

plot.df <- rbind(plot.df, addon)

### Add on the May cabinet
maydf <- data.frame(iso3c = "GBR",
                    prime = "May",
                    govtseq = 999,
                    tally = 0:3,
                    mend = as.Date(c("2017-06-11",
                                     "2017-11-01",
                                     "2017-11-08",
                                     "2017-12-20")),
                    gstart = as.Date("2017-06-11"))

maydf$daysAfter <- as.numeric(maydf$mend - maydf$gstart)

plot.df <- merge(plot.df, maydf, all = TRUE)

### Create a single value for each day
plot.df <- plot.df %>%
    group_by(iso3c, govtseq, prime, gstart, gend, daysAfter) %>%
    summarize(tally = max(tally))

ann.df <- plot.df %>%
    group_by(iso3c, govtseq, prime, gstart, gend) %>%
    summarize(tally = max(tally),
              daysAfter = max(daysAfter)) %>%
    filter(tally > 17 | tally == 15 & daysAfter < 30) %>%
    mutate(label = paste0(countrycode(iso3c, "iso3c", "country.name"),
                         "\n",
                         prime,
                         "\n(",
                         format(gstart, "%Y"),
                         "-",
                         format(gend, "%y"),
                         ")")) %>%
    arrange(daysAfter)

ann.df$adj <- c(1, rep(0, nrow(ann.df) - 1))

p <- ggplot(plot.df, aes(x = daysAfter, y = tally,
                         group = interaction(iso3c, govtseq))) +
    geom_line(colour = "#99999977") +
geom_point(data = subset(plot.df, govtseq == 999 & tally == 3),
          color = 'red') +
geom_line(data = subset(plot.df, govtseq == 999),
          color = 'red') +
    geom_point(data = ann.df, color = "#99999977" ) + 
    scale_x_continuous("Days after government formed") +
    scale_y_continuous("Numbers of government exits") +
    labs(title = "Ministerial exits, all governments 1990 to 2014",
         subtitle = "Second May cabinet highlighted in red",
         caption = "Data: Seki and Williams (2014), 'Updating the Party Government Data Set'. Electoral Studies.") +
    geom_text(data = ann.df,
              aes(label = label, adj = adj),
              nudge_x = c(0, 0, 20, 0)) + 
    theme_ipsum_rc()

ggsave(p, file = "p.png", width = 12, height = 10)
                                        #

summary.df <- plot.df %>%
    group_by(iso3c, govtseq) %>%
    filter(tally == max(tally)) %>%
    mutate(rate = daysAfter / tally) %>%
    arrange(rate)

summary.df$durat <- as.numeric(summary.df$gend - summary.df$gstart)

summary(lm(log1p(durat) ~ rate, data = summary.df))

with(summary.df, cor(rate,
                     as.numeric(gend - gstart),
                     use = "pairwise"))

which(summary.df$govtseq == 999)


ann.df <- plot.df %>%
    filter(iso3c == "GBR") %>% 
    group_by(iso3c, govtseq, prime, gstart, gend) %>%
    summarize(tally = max(tally),
              daysAfter = max(daysAfter)) %>% 
    mutate(label = paste0(prime,
                         "\n(",
                         format(gstart, "%Y"),
                         "-",
                         format(gend, "%y"),
                         ")")) %>%
    arrange(daysAfter)
    
p3 <- ggplot(subset(plot.df, iso3c == "GBR"),
             aes(x = daysAfter, y = tally,
                 group = interaction(iso3c, govtseq))) +
    geom_point(data = ann.df, color = "#99999999") +
    geom_point(data = subset(ann.df, govtseq == 999), color = "red") + 
    geom_text(data = ann.df, aes(label = label), adj = 0, nudge_x = 2) + 
    geom_line(colour = "#99999999") +
    geom_line(data = subset(plot.df, govtseq == 999),
              color = 'red') +
        scale_x_continuous("Days after government formed") +
        scale_y_continuous("Numbers of ministers moved, fired, resigned, or dead") +
        labs(title = "Ministerial replacements, UK governments Major to Brown",
               subtitle = "Second May cabinet highlighted in red",
               caption = "Data: Seki and Williams (2014), 'Updating the Party Government Data Set'. Electoral Studies.") + 
        theme_ipsum_rc()

ggsave(p3, file = "p3.png", width = 12, height = 10)