##
## Drew Linzer
## dlinzer@ucla.edu
## January 6, 2005
## 
## Linzer-CArecall.R
## 
## Statistical analyses for chapter by Drew Linzer (2005):
## "Statistical Regularities in the Recall Results." 
## In Shawn Bowler and Bruce Cain, eds., "Clicker Politics: 
## Essays on the California Recall." New York: Prentice Hall.
## 

library(foreign)
library(systemfit)
recall <- dget("C:/.../CArecalldata.put")
par(ask=T)

## Figure 3.1. Trend of Job Approval Rating of Gray Davis as Governor
## Source: California Field Poll 2003
cat("Figure 3.1 \n")
davis.app <- data.frame(polldate=as.Date(c("1999-03-01","1999-08-28","1999-10-27","2000-02-24","2000-06-23","2000-08-22","2001-01-19","2001-05-19","2001-09-16","2001-12-15","2002-01-14","2002-04-14","2002-07-13","2002-09-11","2003-04-09","2003-07-08","2003-08-07")),
                        Approve=c(54,59,54,62,61,56,57,36,38,38,39,39,41,39,24,23,22),
                        Disapprove=c(15,26,27,20,24,28,34,55,52,51,53,54,49,49,65,66,70))
matplot(davis.app[,1],davis.app[,2:3],ylim=c(0,80),type="l",col=c("gray","black"),lty=1,lwd=2,
        xaxt="n",yaxt="n",xlab="",ylab="Percent approving/disapproving")
matplot(davis.app[,1],davis.app[,2:3],ylim=c(0,80),col=c("gray","black"),pch=c(18,15),cex=c(2,1.5),add=T)
axis(2,seq(0,80,10),las=1)
axis.Date(1,at=davis.app$polldate,format="%m-%Y",las=3)
text(as.Date("1999-08-28"),63,"Approve")
text(as.Date("1999-08-28"),31,"Disapprove")

## Figure 3.3. Scatterplot of percent voting for Schwarzenegger for Governor
## versus percent voting in favor of the recall for fifty-eight counties.
cat("Figure 3.3 \n")
plot(recall$YESRP,recall$ARNOLDP,xlim=c(10,90),ylim=c(10,70),cex=1.3,las=1,
    xlab="Percent voting in favor of recall",ylab="Percent voting Schwarzenegger for Governor")
axis(1,seq(10,90,10))
text(20,19,"San Francisco",pos=4)

## Figure 3.4. Scatterplot of percent voting in favor of the recall
## versus percent registered Republican for fifty-eight counties.
cat("Figure 3.4 \n")
plot(recall$PCTREP,recall$YESRP,xlim=c(10,55),ylim=c(10,90),cex=1.3,las=1,
    xlab="Percent registered Republican",ylab="Percent voting in favor of recall")
axis(1,seq(10,55,5))
axis(2,seq(10,90,10),las=2)
abline(0,1)
abline(coefficients(lm(recall$YESRP~recall$PCTREP)),lwd=2)
text(15,32,"Linear fit")
text(45,30,"Percent voting in favor of recall if all \n Republicans, and only Republicans, \n voted for recall")

## Table 3.1. Counties in which the Davis/Bustamante administration
## suffered its worst losses between 2002 and 2003.
cruz03adj <- recall$CRUZV * (recall$GV02/recall$GV03)
falloff.cradj <- cruz03adj - recall$DAVIS02V
falloff.cradjv <- falloff.cradj/recall$REGVOTER
falloff.cradjp <- falloff.cradj/recall$DAVIS02V
o <- order(falloff.cradj)
tab31a <- data.frame(County=recall$COUNTY[o],BustaFalloff=round(falloff.cradj[o],0),
                     FalloffPct=round(100*falloff.cradjv[o],1),PctChange=round(100*falloff.cradjp[o],1))[1:10,]
o <- order(falloff.cradjp)
tab31b <- data.frame(County=recall$COUNTY[o],BustaFalloff=round(falloff.cradj[o],0),
                     FalloffPct=round(100*falloff.cradjv[o],1),PctChange=round(100*falloff.cradjp[o],1))[1:10,]
cat("\n Table 3.1. Counties in which the Davis/Bustamante administration \n suffered its worst losses between 2002 and 2003.\n")
cat("\n Sorted by Net Votes Lost \n")
print(tab31a)
cat("\n Sorted by Percent Change \n")
print(tab31b)
cat("\n")

## Figure 3.5. Scatterplot of percentage difference between recall support in 2003
## and Davis opposition in 2002 versus percent voting for Davis in 2002 for fifty-
## eight counties.  California's ten southern counties are denoted with solid triangles.
cat("Figure 3.5 \n")
d.davisopp <- recall$DAVIS02P-recall$NORP
plot(recall$DAVIS02P[recall$SOUTH==0],d.davisopp[recall$SOUTH==0],cex=1.3,xlim=c(20,70),ylim=c(-20,20),
     xlab="2002 Davis vote share",ylab="Change in opposition to Davis")
points(recall$DAVIS02P[recall$SOUTH==1],d.davisopp[recall$SOUTH==1],cex=1.3,pch=17)

## Table 3.2. OLS estimation. DV is logit-transformed estimated percent of
## 2002 voters who voted in favor of the recall in 2003, by county.
davrec.logit <- log(recall$EIDAVREC/(1-recall$EIDAVREC))
reg32 <- lm(davrec.logit~PCTREP+log(INCOME01)+LATINO+SOUTH,recall)
cat("\n Table 3.2. OLS estimation. DV is logit-transformed estimated \n percent of 2002 voters who voted in favor of the recall in \n 2003, by county.")
print(summary(reg32))

## Table 3.3. SUR estimation, fifty-eight observations.  DVs are log-ratios
## of countywide vote shares for all candidates, calculated with respect to
## Bustamante's vote shares.
## NOTE: standard errors differ slightly from published results because of 
## computing differences in the estimation algorithm.
lr.arnold <- log(recall$ARNOLDP/recall$CRUZP)
lr.mcclint <- log(recall$MCCLP/recall$CRUZP)
lr.camejo <- log(recall$CAMEJOP/recall$CRUZP)
lr.others <- log((100-(recall$ARNOLDP+recall$CRUZP+recall$MCCLP+recall$CAMEJOP))/recall$CRUZP)

f.arnold <- lr.arnold~PCTREP+log(INCOME01)+LATINO+SOUTH
f.mcclint <- lr.mcclint~PCTREP+log(INCOME01)+LATINO+SOUTH
f.camejo <- lr.camejo~PCTREP+log(INCOME01)+LATINO+SOUTH
f.others <- lr.others~PCTREP+log(INCOME01)+LATINO+SOUTH
eqsystem <- list(f.arnold,f.mcclint,f.camejo,f.others)

reg33 <- systemfit("SUR",eqsystem,data=recall)
reg33.coeffmat <- round(cbind(reg33$b,reg33$se,reg33$t,reg33$p),3)
colnames(reg33.coeffmat) <- c("Coeff.","SE","t","p>t")
cat("\n Table 3.3. SUR estimation. DVs are log-ratios of countywide \n vote shares for all candidates, calculated with respect \n to Bustamante's vote shares. \n")
print(reg33.coeffmat)
cat("\n \n")

# end of file.