Load libraries and data

library(magrittr)
library(dplyr)
library(feather)
library(tidyr)
library(purrr)
library(broom)
library(lubridate)
library(ggplot2)
library(ggstance)
library(gridExtra)
library(scales)
library(Cairo)

source(file.path(PROJHOME, "Analysis", "lib", "graphic_functions.R"))

full.data <- read_feather(file.path(PROJHOME, "Data", "data_processed",
                                    "full_data.feather"))

my.seed <- 1234
set.seed(my.seed)

General data summary

Number of countries:

length(unique(full.data$Country))
## [1] 130

Range of years:

min(full.data$year.num)
## [1] 1991
max(full.data$year.num)
## [1] 2015

Understanding and visualizing civil society environment

The V-Dem Bayesian measurement model is cool and all, but it’s really hard to interpret by itself. Showing a few example countries can help.

cs.plot.all <- full.data %>%
  filter(year.num > 2000) %>%
  group_by(Country) %>%
  summarise(env.mean = mean(cs_env_sum, na.rm=TRUE)) %>%
  filter(!is.na(env.mean)) %>%
  arrange(env.mean) %>%
  ungroup()

cs.plot <- bind_rows(cs.plot.all %>% slice(1:5),
                     data_frame(Country = "—", env.mean = 0),
                     cs.plot.all %>% slice((nrow(.) - 4):nrow(.))) %>%
  mutate(Country = factor(Country, levels=unique(Country), ordered=TRUE))

The usual suspects get their scores

cs.plot
Country env.mean
North Korea -6.096056
Cuba -3.142303
Syria -3.015231
Libya -3.011029
Myanmar -2.682016
— 0.000000
Brazil 5.799338
Estonia 6.086975
Costa Rica 6.215544
Norway 6.232822
France 6.807109

Average volatility/range by regime type:

cs.plot.all %>% 
  left_join(select(full.data, Country, polity_ord), by="Country") %>%
  group_by(polity_ord) %>% summarise(index.change.avg = mean(env.mean))
polity_ord index.change.avg
Autocracy -0.316877
Anocracy 1.626376
Democracy 3.732983
NA 1.859263 
cs.plot.all %>% 
  left_join(select(full.data, Country, polity_ord2), by="Country") %>%
  group_by(polity_ord2) %>% summarise(index.change.avg = mean(env.mean))
polity_ord2 index.change.avg
Autocracy 0.5179618
Democracy 3.4138055
NA 1.8592626 
cs.plot.all %>% 
  left_join(select(full.data, Country, gwf.binary), by="Country") %>%
  group_by(gwf.binary) %>% summarise(index.change.avg = mean(env.mean))
gwf.binary index.change.avg
Autocracy 0.8720814
Democracy 3.7391560
NA 2.0991161 
plot.csre.top.bottom <- ggplot(cs.plot, aes(x=Country, y=env.mean)) +
  geom_point(size=2) + 
  geom_segment(aes(yend=0, xend=Country), size=1) + 
  labs(x=NULL, y="Mean civil society regulatory environment index (CSRE)") +
  coord_flip() +
  theme_ath()
plot.csre.top.bottom

fig.save.cairo(plot.csre.top.bottom, filename="1-csre-top-bottom", 
               width=5, height=2.5)

Visualizing basic correlation between regime type and CSRE

Regime type and CSRE are quite correlated

plot.data <- full.data %>%
  select(cs_env_sum, uds_mean, e_polity2) %>% 
  na.omit()

plot.data %>%
  summarise_each(funs(cor(., plot.data$cs_env_sum)), -cs_env_sum)
uds_mean e_polity2
0.8479438 0.83553

You can see that visually too

plot.polity <- ggplot(plot.data, aes(x=e_polity2, y=cs_env_sum)) + 
  geom_vline(xintercept=0, colour="#EA2E49", size=0.5) +
  geom_point(size=0.25, alpha=0.25) + 
  geom_smooth(method="lm", se=TRUE, colour="#014358") + 
  labs(x="Polity IV score", 
       y="Civil society regulatory environment\n(CSRE)") +
  scale_y_continuous(breaks=seq(-6, 6, 2)) +
  coord_cartesian(ylim=c(-6, 6)) +
  theme_ath()

plot.uds <- ggplot(plot.data, aes(x=uds_mean, y=cs_env_sum)) + 
  geom_vline(xintercept=0, colour="#EA2E49", size=0.5) +
  geom_point(size=0.25, alpha=0.25) + 
  geom_smooth(method="lm", se=TRUE, colour="#014358") +
  labs(x="Mean UDS score", y=NULL) +
  scale_y_continuous(breaks=seq(-6, 6, 2)) +
  coord_cartesian(ylim=c(-6, 6)) +
  theme_ath()

# GWF regime type and Polity/UDS and CSRE
gwf.summary <- full.data %>%
  group_by(gwf.binary) %>%
  summarise(env.avg = mean(cs_env_sum, na.rm=TRUE),
            env.sd = sd(cs_env_sum, na.rm=TRUE),
            env.se = env.sd / sqrt(n()),
            env.upper = env.avg + (qnorm(0.975) * env.se),
            env.lower = env.avg + (qnorm(0.025) * env.se)) %>%
  na.omit

plot.gwf.csre <- ggplot(gwf.summary, aes(x=gwf.binary, y=env.avg)) + 
  geom_pointrange(aes(ymin=env.lower, ymax=env.upper), size=0.5) +
  geom_point(data=full.data, aes(y=cs_env_sum), size=0.25, alpha=0.05) +
  labs(y=NULL, x="Geddes et al. categorization") +
  scale_y_continuous(breaks=seq(-6, 6, 2)) +
  coord_cartesian(ylim=c(-6, 6)) +
  theme_ath()

plot.regime.csre <- arrangeGrob(plot.polity, plot.uds, plot.gwf.csre, nrow=1)
## Warning: Removed 1483 rows containing missing values (geom_point).
grid::grid.draw(plot.regime.csre)

fig.save.cairo(plot.regime.csre, filename="1-regime-csre", 
               width=5, height=2)

Understanding and visualizing ICRG

Conceptualizing the political risk measure is a little tricky. Showing a few example countries can help. Figure out which countries have change the least/most since 2000:

risk.stats <- full.data %>%
  filter(year.num > 2000) %>%
  group_by(Country) %>%
  summarise(index.change = max(icrg.pol.risk.internal.scaled) - 
              min(icrg.pol.risk.internal.scaled)) %>%
  filter(!is.na(index.change)) %>%
  arrange(index.change) %T>%
  {print(head(., 5))} %>% {print(tail(., 5))}
## # A tibble: 5 x 2
##     Country index.change
##       <chr>        <dbl>
## 1     Congo     2.794715
## 2    Norway     3.302846
## 3     Gabon     3.607724
## 4   Myanmar     4.065041
## 5 Congo, DR     4.573171
## # A tibble: 5 x 2
##   Country index.change
##     <chr>        <dbl>
## 1   Libya     19.76626
## 2 Ireland     19.81707
## 3   Egypt     20.07114
## 4 Algeria     20.37602
## 5   Syria     27.23577

Norway and Congo are the most consistent; Syria saw the biggest change.

Average volatility/range by regime type:

risk.stats %>% 
  left_join(select(full.data, Country, polity_ord), by="Country") %>%
  group_by(polity_ord) %>% summarise(index.change.avg = mean(index.change))
polity_ord index.change.avg
Autocracy 22.73120
Anocracy 20.23374
Democracy 19.81707
NA 20.94131 
risk.stats %>% 
  left_join(select(full.data, Country, polity_ord2), by="Country") %>%
  group_by(polity_ord2) %>% summarise(index.change.avg = mean(index.change))
polity_ord2 index.change.avg
Autocracy 22.06790
Democracy 19.97935
NA 20.94131 
risk.stats %>% 
  left_join(select(full.data, Country, gwf.binary), by="Country") %>%
  group_by(gwf.binary) %>% summarise(index.change.avg = mean(index.change))
gwf.binary index.change.avg
Autocracy 21.86230
Democracy 19.81707
NA 21.45325

Visualize changes:

example.countries <- c(652, 385, 484)
example.stability <- full.data %>%
  filter(year.num > 2000, cowcode %in% example.countries) %>%
  mutate(country.plot = factor(Country,
                               levels=c("Norway", "Syria", "Congo"), 
                               ordered=TRUE))

# example.stability$icrg.pol.grade.internal
# c(0, 49.99, 59.99, 69.99, 79.99, Inf)
plot.icrg.examples <- ggplot(example.stability, 
                             aes(x=year.actual, 
                                 y=icrg.pol.risk.internal.scaled,
                                 colour=country.plot)) + 
  geom_line(size=1.5) + 
  labs(x=NULL, y="Internal political risk (ICRG)") + 
  coord_cartesian(xlim=ymd(c("2000-01-01", "2015-01-01"))) +
  scale_colour_manual(values=c("#014358", "#FD7401", "#BEDB3A"), name=NULL) +
  theme_ath()
plot.icrg.examples

fig.save.cairo(plot.icrg.examples, filename="1-icrg-examples", 
               width=5, height=2)

ICRG across regime type

Are autocracies necessarily more unstable than democracies? They are more volatile, as shown above with risk.stats summaries…

plot.data <- full.data %>%
  group_by(gwf.binary, year.actual) %>%
  summarise(icrg = mean(icrg.pol.risk.internal.scaled, na.rm=TRUE),
            icrg.sd = sd(icrg.pol.risk.internal.scaled, na.rm=TRUE),
            icrg.se = icrg.sd / sqrt(n()),
            icrg.upper = icrg + (qnorm(0.975) * icrg.se),
            icrg.lower = icrg + (qnorm(0.025) * icrg.se)) %>%
  na.omit() %>%
  ungroup() %>%
  mutate(gwf.binary = factor(gwf.binary, levels=c("Democracy", "Autocracy"),
                             labels=c("Democracies    ", "Autocracies")))

plot.icrg.regime <- ggplot(plot.data, aes(x=year.actual, y=icrg, 
                                          colour=gwf.binary)) + 
  geom_ribbon(aes(ymin=icrg.lower, ymax=icrg.upper, fill=gwf.binary), 
              alpha=0.3, colour=NA) +
  geom_line(size=1.5) + 
  labs(x=NULL, y="Mean internal political risk (ICRG)") + 
  scale_colour_manual(values=c(col.dem, col.auth), name=NULL) +
  scale_fill_manual(values=c(col.dem, col.auth), name=NULL, guide=FALSE) +
  theme_ath()
plot.icrg.regime

fig.save.cairo(plot.icrg.regime, filename="1-icrg-regime", 
               width=5, height=3)

Check if the difference in means is significant in each year

year.diffs <- full.data %>%
  select(gwf.binary, year.num, icrg.pol.risk.internal.scaled) %>%
  na.omit() %>%
  group_by(year.num) %>%
  do(tidy(t.test(icrg.pol.risk.internal.scaled ~ gwf.binary, data=.)))

year.diffs %>% select(1:6) %>% print(n=nrow(.))
## Source: local data frame [20 x 6]
## Groups: year.num [20]
## 
##    year.num   estimate estimate1 estimate2 statistic      p.value
##       <dbl>      <dbl>     <dbl>     <dbl>     <dbl>        <dbl>
## 1      1991 -14.628779  47.01362  61.64240 -4.916415 4.057853e-06
## 2      1992 -14.015009  49.81629  63.83130 -4.887999 3.898728e-06
## 3      1993 -12.457222  52.27290  64.73013 -4.929161 3.470461e-06
## 4      1994 -12.184781  53.85163  66.03641 -5.027880 2.534530e-06
## 5      1995 -10.794577  55.74752  66.54209 -4.551420 1.799641e-05
## 6      1996 -11.094540  57.20621  68.30075 -4.569834 1.586503e-05
## 7      1997 -11.596676  58.82114  70.41781 -4.697399 8.642876e-06
## 8      1998 -14.224174  57.79202  72.01619 -5.952591 3.753401e-08
## 9      1999 -12.103957  57.41870  69.52266 -5.101275 1.384338e-06
## 10     2000 -11.343099  57.23826  68.58136 -4.675969 8.132532e-06
## 11     2001 -12.464450  57.85869  70.32314 -5.471518 2.771680e-07
## 12     2002 -11.717572  56.74584  68.46341 -4.921817 2.969395e-06
## 13     2003 -12.254173  56.45247  68.70664 -5.091675 1.567803e-06
## 14     2004 -10.175838  58.49286  68.66870 -4.459799 2.194878e-05
## 15     2005  -9.653980  59.12312  68.77710 -4.196184 6.165725e-05
## 16     2006  -8.501555  59.49549  67.99704 -3.784545 2.673390e-04
## 17     2007  -8.683644  59.06131  67.74496 -3.796688 2.528076e-04
## 18     2008  -8.981813  58.60279  67.58460 -3.920897 1.646835e-04
## 19     2009  -7.634443  58.92995  66.56440 -3.399936 9.935484e-04
## 20     2010  -8.360392  57.95905  66.31944 -3.723539 3.333923e-04

Yup. They are.

Other authoritarian stability variables and CSRE

All three variables seem moderately correlated with the CSRE

plot.data <- full.data %>%
  select(cs_env_sum, yrsoffc, years.since.comp, opp1vote) %>% 
  na.omit()

plot.data %>%
  summarise_each(funs(cor(., plot.data$cs_env_sum)), -cs_env_sum)
yrsoffc years.since.comp opp1vote
-0.4754157 -0.5129839 0.5151119

Plot the correlations

plot.yrs.offc <- ggplot(plot.data, aes(x=yrsoffc, y=cs_env_sum)) + 
  geom_point(size=0.25, alpha=0.25) + 
  geom_smooth(method="lm", se=TRUE, colour="#014358") + 
  labs(x="Years executive has been in office", 
       y="Civil society regulatory environment\n(CSRE)") +
  scale_y_continuous(breaks=seq(-6, 6, 2)) +
  theme_ath()

plot.yrs.since.comp <- ggplot(plot.data, 
                              aes(x=years.since.comp, y=cs_env_sum)) + 
  geom_point(size=0.25, alpha=0.25) + 
  geom_smooth(method="lm", se=TRUE, colour="#014358") +
  labs(x="Years since a competitive election", y=NULL) +
  scale_y_continuous(breaks=seq(-6, 6, 2)) +
  theme_ath()

plot.opp.vote <- ggplot(plot.data, aes(x=opp1vote, y=cs_env_sum)) + 
  geom_point(size=0.25, alpha=0.25) + 
  geom_smooth(method="lm", se=TRUE, colour="#014358") +
  labs(x="Vote share for largest opposition party", y=NULL) +
  scale_y_continuous(breaks=seq(-6, 6, 2)) +
  theme_ath()

plot.auth.vars <- arrangeGrob(plot.yrs.offc, plot.yrs.since.comp, 
                              plot.opp.vote, nrow=1)
grid::grid.draw(plot.auth.vars)

fig.save.cairo(plot.auth.vars, filename="1-auth-vars", 
               width=6, height=2)

ICRG stability and competition

plot.data <- full.data %>%
  select(icrg.pol.risk.internal.scaled, yrsoffc, years.since.comp, opp1vote) %>% 
  na.omit()

plot.data %>%
  summarise_each(funs(cor(., plot.data$icrg.pol.risk.internal.scaled)),
                 -icrg.pol.risk.internal.scaled)
yrsoffc years.since.comp opp1vote
-0.1941299 -0.2782814 0.3723299 
plot.yrs.offc <- ggplot(plot.data, aes(x=yrsoffc, y=icrg.pol.risk.internal.scaled)) + 
  geom_point(size=0.25, alpha=0.25) + 
  geom_smooth(method="lm", se=TRUE, colour="#014358") + 
  labs(x="Years executive has been in office", 
       y="ICRG") +
  scale_y_continuous(breaks=seq(-0, 100, 20)) +
  theme_ath()

plot.yrs.since.comp <- ggplot(plot.data, 
                              aes(x=years.since.comp, y=icrg.pol.risk.internal.scaled)) + 
  geom_point(size=0.25, alpha=0.25) + 
  geom_smooth(method="lm", se=TRUE, colour="#014358") +
  labs(x="Years since a competitive election", y=NULL) +
  scale_y_continuous(breaks=seq(-0, 100, 20)) +
  theme_ath()

plot.opp.vote <- ggplot(plot.data, aes(x=opp1vote, y=icrg.pol.risk.internal.scaled)) + 
  geom_point(size=0.25, alpha=0.25) + 
  geom_smooth(method="lm", se=TRUE, colour="#014358") +
  labs(x="Vote share for largest opposition party", y=NULL) +
  scale_y_continuous(breaks=seq(-0, 100, 20)) +
  theme_ath()

plot.auth.vars <- arrangeGrob(plot.yrs.offc, plot.yrs.since.comp, 
                              plot.opp.vote, nrow=1)
grid::grid.draw(plot.auth.vars)

Neighboring and regional ICRG risk

Example of Kenya’s neighborhood in 2012

kenya.neighbors <- full.data %>%
  filter(year.num == 2012, cowcode == 501) %>%
  mutate(neighbors.cow = strsplit(neighbors.cow, ",")) %>%
  select(neighbors.cow) %>% unlist %>% map_dbl(as.numeric)

full.data %>%
  filter(year.num == 2012, cowcode %in% kenya.neighbors) %>%
  select(Country, icrg.pol.risk.internal.scaled)
Country icrg.pol.risk.internal.scaled
Uganda 47.61179
Tanzania 58.63821
Somalia 22.86585
Ethiopia 45.93496
NA NA

Visualize neighbor and subregional instability

plot.data <- full.data %>%
  select(cs_env_sum, neighbor.pol.risk.min, icrg.pol.risk.subregional)

Plot the correlations

plot.neighbor.risk <- ggplot(plot.data, aes(x=neighbor.pol.risk.min, 
                                            y=cs_env_sum)) + 
  geom_point(size=0.25, alpha=0.25) + 
  geom_smooth(method="lm", se=TRUE, colour="#014358") + 
  labs(x="Minimum political risk in neighboring country", 
       y="Civil society regulatory environment\n(CSRE)") +
  scale_y_continuous(breaks=seq(-6, 6, 2)) +
  theme_ath()

plot.subregion.risk <- ggplot(plot.data, aes(x=icrg.pol.risk.subregional, 
                                             y=cs_env_sum)) + 
  geom_point(size=0.25, alpha=0.25) + 
  geom_smooth(method="lm", se=TRUE, colour="#014358") + 
  labs(x="Mean political risk in subregion", 
       y=NULL) +
  scale_y_continuous(breaks=seq(-6, 6, 2)) +
  theme_ath()

plot.ext.vars <- arrangeGrob(plot.neighbor.risk, plot.subregion.risk, nrow=1)
## Warning: Removed 935 rows containing non-finite values (stat_smooth).
## Warning: Removed 935 rows containing missing values (geom_point).
## Warning: Removed 1227 rows containing non-finite values (stat_smooth).
## Warning: Removed 1227 rows containing missing values (geom_point).
grid::grid.draw(plot.ext.vars)

fig.save.cairo(plot.ext.vars, filename="1-ext-risk-vars", 
               width=6, height=2)

The two external risk variables are fairly correlated though

cor(plot.data$icrg.pol.risk.subregional, 
    plot.data$neighbor.pol.risk.min, use="complete.obs")
## [1] 0.7044036
plot.risk.cor <- ggplot(plot.data, aes(x=icrg.pol.risk.subregional, 
                                       y=neighbor.pol.risk.min)) + 
  geom_point(size=0.25, alpha=0.25) + 
  geom_smooth(method="lm", se=TRUE, colour="#014358") + 
  labs(x="Mean political risk in subregion", 
       y="Minimum political risk in neighboring country") +
  theme_ath()
plot.risk.cor
## Warning: Removed 1401 rows containing non-finite values (stat_smooth).
## Warning: Removed 1401 rows containing missing values (geom_point).

Visualize coups

plot.data <- full.data %>%
  select(cs_env_sum, neighbor.coups.activity.bin, coups.activity.subregional) %>%
  na.omit %>%
  mutate(neighbor.coups.activity.bin = factor(neighbor.coups.activity.bin, 
                                              labels=c("No coup activity",
                                                       "Coup activity")),
         coups.activity.subregional = factor(coups.activity.subregional))

plot.coups.neighbor <- ggplot(plot.data, aes(x=neighbor.coups.activity.bin, 
                                             y=cs_env_sum)) + 
  geom_violin() +
  geom_point(size=0.15, alpha=0.15, position="jitter") +
  geom_point(stat="summary", fun.y="mean", size=2) +
  labs(x="Coup activity in neighboring countries", 
       y="Civil society regulatory environment\n(CSRE)") +
  scale_y_continuous(breaks=seq(-6, 6, 2)) +
  theme_ath()

plot.coups.subregion <- ggplot(plot.data, aes(x=coups.activity.subregional, 
                                              y=cs_env_sum)) + 
  geom_violin() +
  geom_point(size=0.15, alpha=0.15, position="jitter") +
  geom_point(stat="summary", fun.y="mean", size=2) +
  labs(x="Coup attemps in subregion", 
       y=NULL) +
  scale_y_continuous(breaks=seq(-6, 6, 2)) +
  theme_ath()

plot.coup.vars <- arrangeGrob(plot.coups.neighbor, plot.coups.subregion, nrow=1)
grid::grid.draw(plot.coup.vars)

fig.save.cairo(plot.coup.vars, filename="1-ext-coup-vars", 
               width=6, height=2)

Event data visualization

This is all old stuff that I’ve replaced with more accurate data.

# plot.data <- full.data %>%
#   select(Country, year.num, year.actual, gwf.binary,
#          icews.conflict.severity.abs, icews.pct.shame,
#          icews.conflict.severity.abs.ingos, icews.pct.shame.ingos) %>%
#   na.omit %>%
#   group_by(year.actual, gwf.binary) %>%
#   summarise(shame = mean(icews.pct.shame),
#             shame.sd = sd(icews.pct.shame, na.rm=TRUE),
#             shame.se = shame.sd / sqrt(n()),
#             shame.upper = shame + (qnorm(0.975) * shame.se),
#             shame.lower = shame + (qnorm(0.025) * shame.se),
#             shame.ingo = mean(icews.pct.shame.ingos),
#             shame.ingo.sd = sd(icews.pct.shame.ingos, na.rm=TRUE),
#             shame.ingo.se = shame.ingo.sd / sqrt(n()),
#             shame.ingo.upper = shame.ingo + (qnorm(0.975) * shame.ingo.se),
#             shame.ingo.lower = shame.ingo + (qnorm(0.025) * shame.ingo.se),
#             severity = mean(icews.conflict.severity.abs),
#             severity.sd = sd(icews.conflict.severity.abs, na.rm=TRUE),
#             severity.se = severity.sd / sqrt(n()),
#             severity.upper = severity + (qnorm(0.975) * severity.se),
#             severity.lower = severity + (qnorm(0.025) * severity.se),
#             severity.ingo = mean(icews.conflict.severity.abs.ingos),
#             severity.ingo.sd = sd(icews.conflict.severity.abs.ingos, na.rm=TRUE),
#             severity.ingo.se = severity.ingo.sd / sqrt(n()),
#             severity.ingo.upper = severity.ingo + (qnorm(0.975) * severity.ingo.se),
#             severity.ingo.lower = severity.ingo + (qnorm(0.025) * severity.ingo.se)) %>%
#   mutate(gwf.binary = factor(gwf.binary, levels=c("Democracy", "Autocracy"),
#                              labels=c("Democracies    ", "Autocracies")))
# 
# #' ## Interstate shame percent and severity
# plot.states.shame <- ggplot(plot.data, aes(x=year.actual, y=shame, 
#                                            colour=gwf.binary)) + 
#   geom_ribbon(aes(ymin=shame.lower, ymax=shame.upper, fill=gwf.binary), 
#               alpha=0.3, colour=NA) +
#   geom_line(size=1.5) +
#   labs(x=NULL, y="Mean percent of all interstate\nevents that are conflictual") + 
#   scale_colour_manual(values=c(col.dem, col.auth), name=NULL) +
#   scale_fill_manual(values=c(col.dem, col.auth), name=NULL, guide=FALSE) +
#   scale_y_continuous(labels=percent, limits=c(0, 1)) +
#   theme_ath()
# 
# plot.states.severity <- ggplot(plot.data, aes(x=year.actual, y=severity, 
#                                               colour=gwf.binary)) + 
#   geom_ribbon(aes(ymin=severity.lower, ymax=severity.upper, fill=gwf.binary), 
#               alpha=0.3, colour=NA) +
#   geom_line(size=1.5) + 
#   labs(x=NULL, y="Mean intensity of\ninterstate conflictual events") + 
#   scale_colour_manual(values=c(col.dem, col.auth), name=NULL) +
#   scale_fill_manual(values=c(col.dem, col.auth), name=NULL, guide=FALSE) +
#   scale_y_continuous(limits=c(0, 10)) +
#   theme_ath()
# 
# plot.events.states <- arrangeGrob(plot.states.shame, plot.states.severity, nrow=1)
# grid::grid.draw(plot.events.states)
# 
# fig.save.cairo(plot.events.states, filename="1-events-states", 
#                width=6, height=2)
# 
# #' Check if the difference in average shame percent is significant in each year
# year.diffs <- full.data %>%
#   select(gwf.binary, year.num, icews.conflict.severity.abs) %>%
#   na.omit() %>%
#   group_by(year.num) %>%
#   do(tidy(t.test(icews.conflict.severity.abs ~ gwf.binary, data=.)))
# 
# year.diffs %>% select(1:6) %>% print(n=nrow(.))
# #' Nope, not really.
# #' 
# 
# #' Check if the difference in average severity is significant in each year
# year.diffs <- full.data %>%
#   select(gwf.binary, year.num, icews.pct.shame) %>%
#   na.omit() %>%
#   group_by(year.num) %>%
#   do(tidy(t.test(icews.pct.shame ~ gwf.binary, data=.)))
# 
# year.diffs %>% select(1:6) %>% print(n=nrow(.))
# #' Again, nope.
# #' 
# 
# #' ## INGO shaming percent and severity
# plot.ingos.shame <- ggplot(plot.data, aes(x=year.actual, y=shame.ingo, 
#                                           colour=gwf.binary)) + 
#   geom_ribbon(aes(ymin=shame.ingo.lower, ymax=shame.ingo.upper, fill=gwf.binary), 
#               alpha=0.3, colour=NA) +
#   geom_line(size=1.5) +
#   labs(x=NULL, y="Mean percent of all INGO-state\nevents that are conflictual") + 
#   scale_colour_manual(values=c(col.dem, col.auth), name=NULL) +
#   scale_fill_manual(values=c(col.dem, col.auth), name=NULL, guide=FALSE) +
#   scale_y_continuous(labels=percent, limits=c(0, 1)) +
#   theme_ath()
# 
# plot.ingos.severity <- ggplot(plot.data, aes(x=year.actual, y=severity.ingo, 
#                                              colour=gwf.binary)) + 
#   geom_ribbon(aes(ymin=severity.ingo.lower, ymax=severity.ingo.upper, fill=gwf.binary), 
#               alpha=0.3, colour=NA) +
#   geom_line(size=1.5) + 
#   labs(x=NULL, y="Mean intensity of\nINGO-state conflictual events") + 
#   scale_colour_manual(values=c(col.dem, col.auth), name=NULL) +
#   scale_fill_manual(values=c(col.dem, col.auth), name=NULL, guide=FALSE) +
#   scale_y_continuous(limits=c(0, 10)) +
#   theme_ath()
# 
# plot.events.ingos <- arrangeGrob(plot.ingos.shame, plot.ingos.severity, nrow=1)
# grid::grid.draw(plot.events.ingos)
# 
# fig.save.cairo(plot.events.ingos, filename="1-events-ingos", 
#                width=6, height=2)
# 
# #' Check if the difference in average shame percent is significant in each year
# year.diffs <- full.data %>%
#   select(gwf.binary, year.num, icews.conflict.severity.abs.ingos) %>%
#   na.omit() %>%
#   group_by(year.num) %>%
#   do(tidy(t.test(icews.conflict.severity.abs.ingos ~ gwf.binary, data=.)))
# 
# year.diffs %>% select(1:6) %>% print(n=nrow(.))
# #' Nope.
# #' 
# 
# #' Check if the difference in average severity is significant in each year
# year.diffs <- full.data %>%
#   select(gwf.binary, year.num, icews.pct.shame.ingos) %>%
#   na.omit() %>%
#   group_by(year.num) %>%
#   do(tidy(t.test(icews.pct.shame.ingos ~ gwf.binary, data=.)))
# 
# year.diffs %>% select(1:6) %>% print(n=nrow(.))
# #' Nope.
# #' 
# 
# #' All four plots at the same time
# plot.events <- arrangeGrob(plot.states.shame + theme(legend.position="none"), 
#                            plot.states.severity + theme(legend.position="none"), 
#                            plot.ingos.shame, plot.ingos.severity)
# grid::grid.draw(plot.events)
# 
# fig.save.cairo(plot.events, filename="1-events-both", 
#                width=6, height=4)