poll.conf

poll.conf <- function(url, inc.und=T){
  library(rvest)
  library(magrittr)
  library(stringr)
  
  poll.page <- html(url)
  
  poll.r <- poll.page %>% html_node("table[class='poll-results-table']") %>% html_table()
  poll.n <- poll.page %>% html_node("div[class='subpop-description']") %>% html_text()
  tmp <- gregexpr("\\d", poll.n)[[1]]
  st <- tmp[[1]]
  en <- tmp[[1]] + sum(attr(tmp, "match.length")) - 1
  
  N <- as.integer(substr(poll.n, start=st, stop=en))
  
  c.names <- gsub(" \\(R\\)","", poll.r$X1)
  per <- as.integer(gsub("%","", poll.r$X2))
  
  tot <- round(per * N/100)
  
  poll.clean <- data.frame(N=c.names, P=per, Tot=tot)
  if(!inc.und){
    poll.clean <- subset(poll.clean, N!="Undecided")
    c.names <- as.character(poll.clean$N)
    per <- poll.clean$P
    tot <- poll.clean$Tot
  }
  full.data <- character()
  for(ii in 1:nrow(poll.clean)){
    full.data <- append(full.data, rep(as.character(poll.clean[ii, "N"]), poll.clean[ii, "Tot"]))
  }  
  
  sim.n <- 1000
  sim <- matrix(NA, sim.n, length(c.names))
  colnames(sim) <- c.names
  sim.winner <- character(sim.n)
  
  
  for(ii in 1:sim.n){
    tmp.votes <- sample(full.data, N, replace=T)
    tmp.table <- table(tmp.votes)/N
    for(jj in 1:length(c.names)){
      tmp.name <- c.names[jj]
      if(tmp.name %in% names(tmp.table)){
        sim[ii, tmp.name] <- tmp.table[tmp.name]
      } else {
        sim[ii, tmp.name] <- 0
      }
    }
    sim.winner[ii] <- names(which.max(tmp.table))
  }
  
  sim.table <- table(sim.winner)/sim.n
  
  par(mar=c(2,8,1,1), mfrow=c(2,1))
  max.x <- ((max(sim)*100) %/% 5 + 1) *5
  plot.new()
  plot.window(ylim=c(1,length(c.names)+.2), xlim=c(0,max.x/100))
  od <- rev(order(apply(sim, 2, median)))
  for(ii in 1:length(c.names)){
    lines(y=c(ii,ii), x=quantile(sim[,od[ii]], c(0.025,0.975)), lwd=3)
    points(y=ii, x=per[od[ii]]/100, cex=1.5)
    dens <- density(sim[,od[ii]])
    lines(y=dens$y/50+ii, x=dens$x, lty=2)
  }
  legend("right", legend=c("95% Confidence Int", "Density", "Reported Percent"), lty=c(1,2, NA), 
         lwd=c(3,1,NA), pch=(NA,NA,1))
  
  xax <- seq(0, max.x, by=5)
  xlab <- paste(xax, "%", sep="")
  axis(1, at=xax/100, labels=xlab)
  axis(2, at=1:length(c.names), c.names[od], las=1)
  title(main="Bootstrapped Confidence Intervals")
  
  barplot(sim.table, horiz=T, las=1, main="Freq of Simulated Wins")
  
}