perf <- foreign::read.dta("jahn/PoEP_Replication_Data/Environmental_Performance_Chapter5.dta") %>%
  as_tibble() %>%
  select(Country = country, Year = year, 
         General = PolGen100,
         Water = PolWat100,
         Mundane = Mundane100,
         Successfully = Success100,
         Specific1980 = LUPI82_1200,
         Specific2010 = LUPI07_1200) %>%
  mutate(Country = as.character(Country)) %>%
  mutate(Country = ifelse(Country == "UK", "United Kingdom", Country)) %>%
  mutate(Country = ifelse(Country == "US", "United States", Country)) %>%
  filter(Year %in% 1980:2010)

load("wdi/wdi-tax.RData")

tax.rev.l <- tax.rev %>%
  select(Country = country, tax.revenue = GC.TAX.TOTL.GD.ZS, Year = year) %>%
  group_by(Country) %>%
  summarize(tax.revenue = median(tax.revenue, na.rm = TRUE))

countries <- as.character(levels(ca$Country))
load("wdi/wdi.RData")
wdi <- wdi[,c("country", "year", "gdp", "population", "gdp.capita", "trade")]
wdi <- subset(wdi, year >= 1976 & year <=2005)
wdi$country[wdi$country=="Korea, Rep."] <- "Korea, Republic of"
wdi <- subset(wdi, country %in% countries)

# GDP pc in Ireland is bad
ireland.wdi <- subset(wdi, country=="Ireland")

# So we use the combination of GDP and population 
# to make a regression against the observed GDP per capita 
# and impute accordingly.
ireland.wdi <- cbind(ireland.wdi, gdp.capita.div = ireland.wdi$gdp/ireland.wdi$population)
m.ireland <- lm(gdp.capita ~ gdp.capita.div, data=ireland.wdi)
wdi$gdp.capita[wdi$country=="Ireland" & is.na(wdi$gdp.capita)] <-
  predict(m.ireland, ireland.wdi)[1:(length(predict(m.ireland, ireland.wdi)) - (length(predict(m.ireland))))]

ireland.wdi <- subset(wdi, country=="Ireland")
ireland.wdi <- cbind(ireland.wdi, gdp.capita.div = ireland.wdi$gdp/ireland.wdi$population)
m.ireland <- lm(gdp.capita ~ year, data=ireland.wdi)
wdi$gdp.capita[wdi$country=="Ireland" & is.na(wdi$gdp.capita)] <-
  predict(m.ireland, ireland.wdi)[1:(length(predict(m.ireland, ireland.wdi)) - (length(predict(m.ireland))))]

# Switzerland is not so bad, but still problematic until 1979. 
# But the procedure does not work, because GDP is also missing. 
# So a simple imputation based on evolution over time is performed.
switzerland.wdi <- subset(wdi, country=="Switzerland")
switzerland.wdi <- cbind(switzerland.wdi, gdp.capita.div = switzerland.wdi$gdp/switzerland.wdi$population)
m.switzerland <- lm(gdp.capita ~ year, data=switzerland.wdi)
wdi$gdp.capita[wdi$country=="Switzerland" & is.na(wdi$gdp.capita)] <-
  predict(m.switzerland, switzerland.wdi)[1:(length(predict(m.switzerland, switzerland.wdi)) - (length(predict(m.switzerland))))]

#New Zealand only misses 1976' GDP per capita, 
# so the same procedure than with Switzerland is used.
newzealand.wdi <- subset(wdi, country=="New Zealand")
newzealand.wdi <- cbind(newzealand.wdi, gdp.capita.div = newzealand.wdi$gdp/newzealand.wdi$population)
m.newzealand <- lm(gdp.capita ~ year, data=newzealand.wdi)
wdi$gdp.capita[wdi$country=="New Zealand" & is.na(wdi$gdp.capita)] <-
  predict(m.newzealand, newzealand.wdi)[1:(length(predict(m.newzealand, newzealand.wdi)) - (length(predict(m.newzealand))))]


switzerland.wdi <- subset(wdi, country=="Switzerland")
m.switzerland <- lm(trade ~ year, data=switzerland.wdi)
wdi$trade[wdi$country=="Switzerland" & is.na(wdi$trade)] <-
  predict(m.switzerland, switzerland.wdi)[1:(length(predict(m.switzerland, switzerland.wdi)) - (length(predict(m.switzerland))))]

# GDP per capita growth
# Another way at looking at resources, is to calculate 
# how many times is the overall wealth per capita at the 
# end of the period compared to the beginning.
wdi.gdp.capita.ratio <- subset(wdi[,c("country", "year", "gdp.capita")], year==min(year) | year==max(year))
wdi.gcr.w <- wdi.gdp.capita.ratio %>%
  spread(year, gdp.capita)
wdi.gcr.w <- cbind(wdi.gcr.w, gdpc.ratio=wdi.gcr.w$`2005`/wdi.gcr.w$`1976`)


# Data is averaged by country through all years.
wdi.c <- wdi %>%
  gather(variable, value, -country, -year) %>%
  group_by(country, variable) %>%
  summarize(m = median(value, na.rm = TRUE)) %>%
  ungroup()


# Include GDP per capita growth, ratio
wdi.l <- wdi.gcr.w %>%
  select(country, gdpc.ratio) %>%
  mutate(variable = "gdpc.ratio") %>%
  rename(m = gdpc.ratio) %>%
  select(country, variable, m) %>%
  bind_rows(wdi.c)

load("vdem/v_dem-ra-1950_2018.RData") # loads vdem

load("wgi/government_effectiveness.RData")

gov.eff.original <- gov.eff
gov.eff.original$country <- as.character(gov.eff.original$country)
gov.eff.original$country[gov.eff.original$country=="Korea, Rep."] <- "Korea, Republic of"
gov.eff.original$country <- as.character(gov.eff.original$country)

gov.eff.l <- gov.eff.original %>%
  rename(gov.eff = value) %>%
  gather(variable, value, -country, -year) %>%
  group_by(country, variable) %>%
  summarize(m = median(value, na.rm = TRUE))

load("./polcon/polcon2017.RData") # loads polcon

load("./manifesto/cpm-consensus.RData")
cpm$country <- as.character(cpm$country)
cpm$country[cpm$country=="Korea"] <- "Korea, Republic of"
cpm$country[cpm$country=="Great Britain"] <- "United Kingdom"
cpm <- subset(cpm, country %in% countries)
cpm$country <- factor(cpm$country)
cpm <- subset(cpm, date>="1970-01-01" & date<="2005-12-31")

# Take only Green parties and Socialist=social democrats + communists
cpm$family <- as.character(cpm$family)
cpm$family[cpm$family=="Social democratic"] <- "Socialist"
cpm$family[cpm$family=="Communist"] <- "Socialist"
cpm <- subset(cpm, family=="Green" | family=="Socialist")
cpm$family <- factor(cpm$family)

# Aggregate duplications in Socialist
cpm <- cpm %>%
  group_by(country, date, family) %>%
  summarize(p.seats=sum(p.seats))

# Calculate the weighted means.
# Unfortunately, a ddply approach would be too complicated and a loop solves it quite quickly.
families <- c("Green", "Socialist")
wmsf <- data.frame(country=countries, Green=NA, Socialist=NA)
for (C in 1:nC) {
  for (F in 1:length(families)) {
    series <- subset(cpm, country==countries[C] & family==families[F])[,c(2, 4)]
    series <- series[order(series$date),]
    #v <- weighted.mean(series$p.seats, diff(c(as.Date("1976-01-01"), series$date)))
    v <- weighted.mean(series$p.seats, as.numeric(diff(c(as.Date("1976-01-01"), series$date))))
    v[is.nan(v)] <- 0
    wmsf[C, 1+F] <- v
  }
}

d.perf <- perf %>%
  # Delete Turkey and Korea
  filter(!Country %in% c("Korea, Republic of", "Turkey")) %>%
  # Delete Years for which we don't have data
  filter(Year >= 1980 & Year <= 2005) %>%
  gather(Indicator, Performance, -Country, -Year) %>%
  # Select specific performance indicators
  filter(Indicator %in% c("General", "Specific1980")) %>%
  droplevels()

Y <- reshape2::acast(d.perf, Indicator ~ Country ~ Year, value.var = "Performance")
nS <- dim(Y)[1]
nC <- dim(Y)[2]
nY <- dim(Y)[3]


country.label <- dimnames(Y)[[2]]
nC <- length(country.label)

indicator.label <- dimnames(Y)[[1]]
sector.label <- dimnames(Y)[[2]]
nI <- length(indicator.label)

year.label <- dimnames(Y)[[3]]
year.label.numeric <- as.integer(as.numeric(year.label))
nY <- length(year.label)



# Function to assign zeros to the fake countries (mean value)
zero.fk <- function(x, id = id.fake.countries) { # zero to fake countries
  x[id] <- 0
  return(x)
}

source("get-eu_time.R") # generates eu.ms

diversity <- ca %>%
  select(Sector, Country, Year, Diversity) %>%
  # Delete Turkey and Korea
  filter(!Country %in% c("South Korea", "Turkey")) %>%
  filter(Sector == "Environmental") %>%
  filter(Year >= 1980 & Year <= 2005) %>%
  select(-Sector) %>%
  droplevels() %>%
  reshape2::acast(Country ~ Year, value.var = "Diversity")
if ( length(which(!dimnames(diversity)[[1]] == country.label)) > 0) stop("Ep! There is a mistake here")


GDPpc <- wdi %>%
  select(country, year, gdp.capita) %>%
  filter(country %in% country.label) %>%
  filter(year >= 1980 & year <= 2005) %>%
  mutate(gdp.capita = std(gdp.capita)) %>%
  reshape2::acast(country ~ year, value.var = "gdp.capita")
if ( length(which(!dimnames(GDPpc)[[1]] == country.label)) > 0) stop("Ep! There is a mistake here")

load("wdi/wdi-gdpgrowth.RData") # gdp.growth
gdp.growth <- gdp.growth %>%
  select(country, year, gdp.growth = NY.GDP.MKTP.KD.ZG) %>%
  filter(country %in% country.label) %>%
  filter(year >= 1980 & year <= 2005) %>%
  mutate(gdp.growth = std(gdp.growth)) %>%
  reshape2::acast(country ~ year, value.var = "gdp.growth")
if ( length(which(!dimnames(gdp.growth)[[1]] == country.label)) > 0) stop("Ep! There is a mistake here")

load("wdi/wdi-urban.RData") # urban
urban <- urban %>%
  select(country, year, urban = SP.URB.TOTL.IN.ZS) %>%
  filter(country %in% country.label) %>%
  filter(year >= 1980 & year <= 2005) %>%
  mutate(urban = std(urban)) %>%
  reshape2::acast(country ~ year, value.var = "urban")
if ( length(which(!dimnames(urban)[[1]] == country.label)) > 0) stop("Ep! There is a mistake here")

load("wdi/wdi-industry.RData") # industry
industry <- industry %>%
  select(country, year, industry = NV.IND.TOTL.ZS) %>%
  filter(country %in% country.label) %>%
  filter(year >= 1980 & year <= 2005) %>%
  mutate(industry = std(industry)) %>%
  reshape2::acast(country ~ year, value.var = "industry")
if ( length(which(!dimnames(industry)[[1]] == country.label)) > 0) stop("Ep! There is a mistake here")
industry.means <- apply(industry, 1, mean, na.rm = TRUE)

trade <- wdi %>%
  select(country, year, trade) %>%
  filter(country %in% country.label) %>%
  filter(year >= 1980 & year <= 2005) %>%
  mutate(trade = std(trade)) %>%
  reshape2::acast(country ~ year, value.var = "trade")
if ( length(which(!dimnames(trade)[[1]] == country.label)) > 0) stop("Ep! There is a mistake here")

deliberation <- vdem %>%
  mutate(Country = as.character(Country)) %>%
  mutate(Country = ifelse(Country == "United States of America", "United States", Country)) %>%
  mutate(Country = ifelse(Country == "Germany (RDA)", "Germany", Country)) %>%
  filter(Democracy == "Deliberative") %>%
  filter(Country %in% country.label) %>%
  filter(Year >= 1980 & Year <= 2005) %>%
  mutate(value = std(value)) %>%
  select(Country, Year, value) %>%
  reshape2::acast(Country ~ Year, value.var = "value")
if ( length(which(!dimnames(deliberation)[[1]] == country.label)) > 0) stop("Ep! There is a mistake here")

constraints <- polcon %>%
  mutate(country.polity = as.character(country.polity)) %>%
  mutate(country.polity = ifelse(country.polity == "Germany West", "Germany", country.polity)) %>%
  filter(country.polity %in% country.label) %>%
  filter(year >= 1980 & year <= 2005) %>%
  mutate(polcon = std(polcon)) %>%
  select(country.polity, year, polcon) %>%
  reshape2::acast(country.polity ~ year, value.var = "polcon")
if ( length(which(!dimnames(constraints)[[1]] == country.label)) > 0) stop("Ep! There is a mistake here")


gov.eff <- expand.grid(country = country.label, year = year.label.numeric) %>%
  left_join(gov.eff.original) %>%
  filter(country %in% country.label) %>%
  filter(year >= 1980 & year <= 2005) %>%
  mutate(value = std(value)) %>%
  select(country, year, value) %>%
  reshape2::acast(country ~ year, value.var = "value")
if ( length(which(!dimnames(gov.eff)[[1]] == country.label)) > 0) stop("Ep! There is a mistake here")


portfolio.size <- ca %>%
  filter(Sector == "Environmental") %>%
  filter(Country %in% country.label) %>%
  filter(Year >= 1980 & Year <= 2005) %>%
  mutate(Size = std(logit(Size))) %>%
  ungroup() %>%
  reshape2::acast(Country ~ Year, variable.var = "Size")
if ( length(which(!dimnames(portfolio.size)[[1]] == country.label)) > 0) stop("Ep! There is a mistake here")

eu <- expand.grid(country = country.label, year = 1958:2020) %>%
  as_tibble() %>%
  mutate(eu = 0) %>%
  left_join(eu.ms, by = c("country" = "ms")) %>%
  mutate(eu = ifelse(year == ms.y, 1, eu)) %>%
  mutate(eu = ifelse(is.na(eu), 0, eu)) %>%
  group_by(country) %>%
  arrange(country, year) %>%
  mutate(eu = cumsum(eu)) %>%
  ungroup() %>%
  select(country, year, eu) %>%
  filter(country %in% country.label) %>%
  filter(year %in% year.label.numeric) %>%
  reshape2::acast(country ~ year, value.var = "eu")
if ( length(which(!dimnames(eu)[[1]] == country.label)) > 0) stop("Ep! There is a mistake here")

green.socialist <- cpm %>%
  ungroup() %>%
  mutate(calendar.year = as.numeric(format(date, "%Y"))) %>%
  mutate(calendar.month = as.numeric(format(date, "%m"))) %>%
  mutate(year = ifelse(calendar.month > 6, calendar.year + 1, calendar.year)) %>%
  # Manually delete some elections that are held too close
  # and assigned both to the same year
  # Delete the first one, as in case of instability it is unlikely 
  # that they have passed legislation
  filter(!(country == "Japan" & date == as.Date("1979-10-07"))) %>%
  filter(!(country == "Greece" & date == as.Date("1989-11-05"))) %>%
  filter(!(country == "Denmark" & date == as.Date("1987-09-08"))) %>%
  #
  select(country, year, family, p.seats) %>%
  full_join(expand.grid(country = country.label, 
                        year = (min(year.label.numeric)-5):max(year.label.numeric),
                        family = c("Socialist", "Green"))) %>%
  group_by(country, family) %>%
  arrange(country, family, year) %>%
  # A bit clumsy, but works
  mutate(p.seats = ifelse(is.na(p.seats), lag(p.seats), p.seats)) %>%
  mutate(p.seats = ifelse(is.na(p.seats), lag(p.seats), p.seats)) %>%
  mutate(p.seats = ifelse(is.na(p.seats), lag(p.seats), p.seats)) %>%
  mutate(p.seats = ifelse(is.na(p.seats), lag(p.seats), p.seats)) %>%
  mutate(p.seats = ifelse(is.na(p.seats), lag(p.seats), p.seats)) %>%
  ungroup() %>%
  select(country, year, family, p.seats) %>%
  filter(country %in% country.label) %>%
  filter(year %in% year.label.numeric) %>%
  mutate(p.seats = ifelse(is.na(p.seats), 0, p.seats)) %>%
  group_by(family) %>%
  mutate(p.seats = std(p.seats)) %>%
  ungroup() %>%
  unique() %>%
  reshape2::acast(family ~ country ~ year, value.var = "p.seats")
if ( length(which(!dimnames(green.socialist)[[2]] == country.label)) > 0) stop("Ep! There is a mistake here")
if (!dimnames(green.socialist)[[1]][1] == "Green" & sector.label[1] == "Environmental") stop("Ep! There is a mistake here")

# Use only the green dimension
green <- green.socialist[1,,]


DJ <- list(
  Y = unname(Y), 
  diversity = unname(diversity),
  GDPpc = unname(GDPpc),
  trade = unname(trade),
  gdp.growth = unname(gdp.growth),
  urban = unname(urban),
  industry = unname(industry),
  industry.means = industry.means,
  deliberation = unname(deliberation),
  constraints = unname(constraints),
  gov.eff = unname(gov.eff),
  gov.eff.mean.observed = unname(apply(gov.eff, 1, mean, na.rm = TRUE)),
  portfolio.size = unname(portfolio.size),
  eu = unname(eu),
  green = green,
  nC = nC,
  nI = nI,
  nY = nY)

nC   # Number of countries

nS   # Number of sectors

years  # Range of years

M <- "performance-tscs"
M.lab <- "Baseline"
m <- "model {
  #
  # Data part at the observational level
  #
  for (i in 1:nI) {
    for (c in 1:nC) {
      for (t in 2:nY) {
        Y[i,c,t] ~ dnorm(mu[i,c,t], tau[i,c,t])
        mu[i,c,t] <- #alpha[i]
                   delta[i,c] * Y[i,c,t-1]
                 + beta[1,i] * GDPpc[c,t-1]
                 + beta[2,i] * trade[c,t-1]
                 + beta[3,i] * eu[c,t-1]
                 + beta[5,i] * gdp.growth[c,t-1]
                 + beta[6,i] * urban[c,t-1]
                 + beta[7,i] * industry[c,t-1]
                 + theta[1,i] * portfolio.size[c,t]
                 + theta[2,i] * diversity[c,t]
                 + rho[i,c] * (Y[i,c,t-1] - mu[i,c,t-1] )
        tau[i,c,t] <- 1 / sigma.sq[i,c,t]
        sigma.sq[i,c,t] <- exp(
                                 lambda_c[i,c])
        resid[i,c,t] <- Y[i,c,t] - mu[i,c,t]
      }
      mu[i,c,1] <- alpha[i]
                 + beta[1,i] * GDPpc[c,1]
                 + beta[2,i] * trade[c,1]
                 + beta[3,i] * eu[c,1]
                 + beta[5,i] * gdp.growth[c,1]
                 + beta[6,i] * urban[c,1]
                 + beta[7,i] * industry[c,1]
                 + theta[1,i] * portfolio.size[c,1]
                 + theta[2,i] * diversity[c,1]
      rho[i,c] ~ dunif(-1, 1)
      delta[i,c] ~ dunif(0, 1)
      resid[i,c,1] <- Y[i,c,1] - mu[i,c,1]
    }

    #
    # Priors for variance component
    #
    lambda[1,i] ~ dnorm(0, 2^-2)
    lambda[2,i] ~ dnorm(0, 2^-2)
    
    #
    # Priors for the intercept
    #
    alpha[i] ~ dunif(0, 100)

    #
    # Priors for the control variables
    #
    for (b in 1:8) {
      beta[b,i] ~ dnorm(0, 1^-2)
    }

    #
    # Priors for main effects
    #
    for (t in 1:2) {
      theta[t,i] ~ dnorm(0, 1^-2)
    }
  }

  # Variance component, intercepts by country
  for (i in 1:nI) {
    for (c in 1:nC) {
      lambda_c[i,c] ~ dt(0, 0.1^-2, 3)
    }
  }

  #
  # Missing data
  #
  for (c in 1:nC) {
    for (t in 1:nY) {
      industry[c,t] ~ dnorm(industry.means[c], 0.05^-2)
    }
  }
}"
write(m, file= paste("models/model-", M, ".bug", sep = ""))
par <- NULL
par <- c(par, "alpha", "beta", "theta", "sigma")
par <- c(par, "lambda", "lambda_c")
par <- c(par, "delta")
par <- c(par, "nu")
par <- c(par, "rho")
par <- c(par, "resid")
inits <- list(
  list(.RNG.name="base::Super-Duper", .RNG.seed=1),
  list(.RNG.name="base::Super-Duper", .RNG.seed=2),
  list(.RNG.name="base::Wichmann-Hill", .RNG.seed=3),
  list(.RNG.name="base::Wichmann-Hill", .RNG.seed=2))

t0 <- proc.time()
rj <- run.jags(model = paste("models/model-", M, ".bug", sep = ""),
  data = dump.format(DJ, checkvalid=FALSE),
  inits = inits,
  modules = "glm",
  n.chains = 4, adapt = 1e2, burnin = 1e3, sample = 5e2, thin = 10,
  monitor = par, method = "parallel", summarise = FALSE)
s <- as.mcmc.list(rj)
save(s, file = paste("sample-", M, ".RData", sep = ""))
proc.time() - t0

load(file = paste("sample-", M, ".RData", sep = ""))

ggmcmc(ggs(s, family = "alpha|beta|theta|delta|lambda|rho"), 
       param_page = 10, file = paste("ggmcmc-full-", M, ".pdf", sep = ""))

ggmcmc(ggs(s, family = "sigma|alpha|beta|lambda|nu|rho"), 
       plot = c("traceplot", "crosscorrelation", "caterpillar"),
       param_page = 8, file = paste("ggmcmc-partial-", M, ".pdf", sep = ""))

L.lambda <- plab("lambda", list(Variable = c("(Intercept)", "Portfolio size"),
                                Sector = sector.label))
S.lambda <- ggs(s, family = "lambda\\[", par_labels = L.lambda)
ggs_caterpillar(S.lambda) +
  ggtitle("Variance component")

L.lambda.c <- plab("lambda_c", list(Indicator = indicator.label, Country = country.label))
S.lambda.c <- ggs(s, family = "lambda_c\\[", par_labels = L.lambda.c)
ggs_caterpillar(S.lambda.c, label = "Country") +
  facet_grid(~ Indicator) +
  ggtitle("Variance component (countries)")

L.rho <- plab("rho", list(Indicator = indicator.label, Country = country.label))
S.rho <- ggs(s, family = "rho", par_labels = L.rho)
ggs_caterpillar(S.rho, label = "Country") +
  facet_wrap(~ Indicator, scales="free") +
  aes(color=Indicator) +
  expand_limits(x = c(-1, 1)) +
  ggtitle("Auto-regressive component (countries)") +
  scale_colour_xfim()

L.delta <- plab("delta", list(Indicator = indicator.label, Country = country.label))
S.delta <- ggs(s, family = "delta", par_labels = L.delta)
ggs_caterpillar(S.delta, label = "Country") +
  facet_wrap(~ Indicator, scales="free") +
  aes(color=Indicator) +
  ggtitle("Lagged dependent variable (countries)") +
  scale_colour_xfim()

L.thetas <- plab("theta", list(Variable = c("Portfolio size", "Diversity"),
                        Indicator = indicator.label))

L.betas <- plab("beta", list(Variable = c("GDP pc", "Trade", "EU", "Green",
                                          "GDP growth", "Urban", "Industry",
                                          "Consensus"),
                        Indicator = indicator.label))
L.betas <- bind_rows(L.thetas, L.betas)

S.betas <- ggs(s, family = "^beta\\[|^theta\\[", par_labels = L.betas) %>%
  filter(value != 0)
S.betas <- filter(S.betas, !Variable %in% c("Green", "Consensus"))

S.betas %>%
  group_by(Variable, Indicator) %>%
  summarize(median = median(value), sd = sd(value),
           `Prob > 0` = length(which(value > 0)) / n(),
    `Prob < 0` = length(which(value < 0)) / n(),
    `Mean expected effect` = mean(value)) %>%
  kable()

ggs_caterpillar(S.betas, label = "Variable") +
  facet_wrap(~ Indicator, scales="free") +
  aes(color = Indicator) +
  geom_vline(xintercept = 0, lty = 3) +
  scale_colour_xfim()

S.betas %>%
  filter(Indicator %in% c("General", "Specific1980")) %>%
  mutate(Variable = factor(as.character(Variable), 
                              levels = rev(c("Diversity", "Portfolio size",
                                         "EU", 
                                         "GDP pc", "GDP growth",
                                         "Industry", "Urban",
                                         "Trade")))) %>%
ggs_caterpillar(label = "Variable", sort = FALSE) +
  facet_wrap(~ Indicator, scales="free") +
  geom_vline(xintercept = 0, lty = 3) #+

S.betas %>%
  filter(Indicator %in% c("General")) %>%
  mutate(Variable = factor(as.character(Variable), 
                              levels = rev(c("Diversity", 
                                         "Portfolio size",
                                         "EU", 
                                         "GDP pc", "GDP growth",
                                         "Industry", "Urban",
                                         "Trade")))) %>%
ggs_caterpillar(label = "Variable", sort = FALSE) +
  geom_vline(xintercept = 0, lty = 3) +
  ggtitle("General performance")

S.betas %>%
  filter(Indicator %in% c("Specific1980")) %>%
  mutate(Variable = factor(as.character(Variable), 
                              levels = rev(c("Diversity", 
                                         "Portfolio size",
                                         "EU", 
                                         "GDP pc", "GDP growth",
                                         "Industry", "Urban",
                                         "Trade")))) %>%
ggs_caterpillar(label = "Variable", sort = FALSE) +
  geom_vline(xintercept = 0, lty = 3) +
  ggtitle("Specific 1980 performance")

S.betas %>%
  filter(Indicator %in% c("General", "Specific1980")) %>%
  mutate(Variable = factor(as.character(Variable), 
                              levels = rev(c("Diversity", 
                                         "Portfolio size",
                                         "EU", 
                                         "GDP pc", "GDP growth",
                                         "Industry", "Urban",
                                         "Trade")))) %>%
ggs_caterpillar(label = "Variable", sort = FALSE) +
  facet_wrap(~ Indicator, scales="free") +
  geom_vline(xintercept = 0, lty = 3) #+

ci(S.betas) %>%
ggplot(aes(x = Variable, 
              y = median,
              group = Indicator, color = Indicator)) +
  coord_flip() +
  geom_point(size = 3, position = position_dodge(width = 0.5)) +
  geom_linerange(aes(ymin = Low, ymax = High), size = 1.5, position = position_dodge(width = 0.5)) +
  geom_linerange(aes(ymin = low, ymax = high), size = 0.5, position = position_dodge(width = 0.5)) +
  ylab("HPD") + xlab("Parameter") +
  geom_vline(xintercept = 0, lty = 3)

ggplot(S.betas, aes(x = value, color = Indicator, fill = Indicator)) +
  geom_density(alpha = 0.5) +
  facet_wrap(~ Variable, ncol = 5, scales = "free") +
  xlab("HPD") +
  geom_vline(xintercept = 0, lty = 3)

S.betas %>%
  filter(value != 0) %>%
  group_by(Variable, Indicator) %>%
  summarize(`Prob > 0` = length(which(value > 0)) / n(),
    `Prob < 0` = length(which(value < 0)) / n(),
    `Mean expected effect` = mean(value)) %>%
  group_by(Variable, Indicator) %>%
  mutate(max = max(abs(`Prob > 0`), abs(`Prob < 0`))) %>%
  arrange(desc(max)) %>%
  select(-max) %>%
  kable()

S.betas.gral <- S.betas %>%
  filter(Indicator == "General") %>%
  mutate(Parameter = Variable)

ggs_caterpillar(S.betas.gral) +
  geom_vline(xintercept = 0, lty = 3)

L.data <- plab("resid", list(Indicator = indicator.label,
                             Country = country.label,
                             Year = year.label))

Obs.sd <- Y %>%
  as.data.frame.table() %>%
  as_tibble() %>%
  rename(Indicator = Var1, Country = Var2, Year = Var3, value = Freq) %>%
  mutate(Year = as.integer(as.numeric(as.character(Year)))) %>%
  group_by(Indicator) %>%
  summarize(obs.sd = sd(value, na.rm = TRUE))

S.rsd <- ggs(s, family = "resid", par_labels = L.data) %>%
  filter(!str_detect(Country, "Z-")) %>%
  group_by(Iteration, Chain, Indicator) %>%
  summarize(rsd = sd(value))

ggplot(S.rsd, aes(x = rsd)) +
  geom_histogram(binwidth = 0.001) +
  geom_vline(data = Obs.sd, aes(xintercept = obs.sd)) +
  facet_grid(Indicator ~ .) +
  expand_limits(x = 0)