coefplot.R

gistfile1.r

require("RColorBrewer")
# not run:
# model = lm(y~x1+x2+x3+x4...)

# Construct quantile table for varialbes of interest
# t value: n-p=df.residual=634, alpha=5%,50%. t(1-alpha/2, n-p)
t.05 <- qt(1-0.025, 634, lower.tail = TRUE, log.p=FALSE)
t.5 <- qt(1-0.25, 634, lower.tail = TRUE, log.p=FALSE)
coef.quantil <- data.frame(matrix(0, nrow=length(model$coefficients)-1, ncol=8))
names(coef.quantil) <- c("coef", "mean", "sd", "X2.5", "X25", "X50", "X75", "X97.5")
for(i in 1:nrow(coef.quantil)) {
	varname <- row.names(summary(model)$coefficients)[i+1]
	varname <- gsub("genre", "", varname)
	substring(varname , 1, 1) <- toupper(substring(varname, 1, 1))
  	substring(varname, 2) <- tolower(substring(varname , 2))
	coef.quantil[i,1] <- varname
	coef.quantil[i,c(2,3,6)] <- summary(gross_model)$coefficients[i+1,c(1,2,1)]
	coef.quantil[i, c(4,8)] <- coef.quantil[i,2]+c(-1, 1)*t.05*coef.quantil[i,3]
	coef.quantil[i, c(5,7)] <- coef.quantil[i,2]+c(-1, 1)*t.5*coef.quantil[i,3]
}

# order data to make pretty pattern
coef.quantil <- coef.quantil[order(coef.quantil$mean, decreasing=FALSE),]


################ plot coefficients ###############################
# set colors
dark_blue = brewer.pal(4,"Blues")[4]
dark_blue2 = brewer.pal(4,"Blues")[3]
light_blue = brewer.pal(4,"Blues")[2]
blues = c(dark_blue, dark_blue2, light_blue)

dark_red = brewer.pal(4,"Reds")[4]
dark_red2 = brewer.pal(4,"Reds")[3]
light_red = brewer.pal(4,"Reds")[2]
reds = c(dark_red, dark_red2, light_red)


png(filename = "~/Dropbox/PredictiveModel/coefplot.png", 
	width = 640, 
	height = 480, 
	units = "px", 
	pointsize = 12)

par(family="OpenSans-Semibold",  #specify the font to use
	bg="white", #specify the background color
 	fg="grey50", #specify the color of the "foreground" - axes, labels, etc.
 	col.axis=dark_blue, #specify the color of the "axis ticks"
 	col.lab=dark_blue, #specify the color of axis labels
 	ljoin=0, #make the axes have rounded edges
 	col.main="grey50", # specify the color of the title
 	mai=c(0,0.05,.3,0.05),
	cex=1,
	cex.main=1.5,
	bg="white")

plot(0,0, 
	xlim=c(min(coef.quantil[,4])-1.3, 
	max(coef.quantil[,8])), 
	ylim=c(-1, 22.5), 
	type="n", 
	axes=F, 
	ylab="", 
	xlab="Gross (in millions of US dollars")
title("Coefficient ranges of a simple box office prediction model") 

# vertical line
lines(c(0,0), c(22, 0), lwd=2, lty=1, col="grey75")

# ticks and text 
range(coef.quantil[,4])
range <- c(-4, -3, -2, -1, 0, 1, 2, 3, 4,5)
n.range <- length(range)
for (i in 1:n.range){
  # bottom ticks and text:
 	text(range[i], 0, range[i], pos=1, cex=0.8, col="grey50")
  #top ticks and text:
  	text(range[i], 22.0, range[i], pos=3, cex=0.8, col="grey50") # this will have to be set custom based on the number of variables in your model
  # gridlines:
	segments(range[i], 0, range[i], 22.0, lwd=0.3, lty=1, col="grey50")
}
# Coef names, 50% & 95% intervals
# conditioned on significant variables 
# I hand coded these.
for (i in 1:nrow(coef.quantil)) {
  if (coef.quantil$coef[i]=="Sequel" |
  	  coef.quantil$coef[i]=="Rotten" |
  	  coef.quantil$coef[i]=="Budget"){
  	cols = reds
}else{
	cols = blues
}
#label
text(min(coef.quantil[,4])-1.5, i, 
		coef.quantil[i,1], 
		adj=0, 
		cex=0.9,
		col=cols[1])
#95% CI
lines(c(coef.quantil[i,4], 
  		coef.quantil[i,8]), 
  		rep(i, 2), 
  		lwd=5, 
  		col=cols[3])
#50% CI
lines(c(coef.quantil[i,5], 
  		coef.quantil[i,7]), 
  		rep(i,2), 
  		col=cols[2], 
  		lwd=8)
#coef
points(coef.quantil[i,2], i, 
  		pch=20, 
  		col=cols[1], 
  		cex=2.75)	
}
dev.off()