-
Notifications
You must be signed in to change notification settings - Fork 0
/
electiogeddon.r
311 lines (269 loc) · 15.2 KB
/
electiogeddon.r
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
#####################################################
## ##
## electiogeddon ##
## Benjamin J. Radford ##
## November 24, 2016 ##
## http://www.benradford.com ##
## https://www.github.com/benradford/electiogeddon ##
## ##
## Data and code to evaluate vote discrepencies in ##
## the 2016 U.S. presidential election. ##
## ##
#####################################################
## Load necessary libraries
library(rgdal)
library(lme4)
library(sjPlot)
library(MASS)
library(RColorBrewer)
predictions <- function(model,n=500000)
{
## Function to draw random coefficients for lm predictions.
## args
## model: a model object from lm.
## returns
## A matrix of size 500,000 x D where D is the number of coefficients.
se <- summary(model)$coefficients[,2]
beta <- summary(model)$coefficients[,1]
draws <- mvrnorm(n,beta,diag(se^2))
draws
}
## Vectors for converting state name to state abbreviation
state_list <- c('Alabama','Alaska','Arizona','Arkansas','California','Colorado','Connecticut','Delaware','Florida','Georgia','Hawaii','Idaho','Illinois','Indiana','Iowa','Kansas','Kentucky','Louisiana','Maine','Maryland','Massachusetts','Michigan','Minnesota','Mississippi','Missouri','Montana','Nebraska','Nevada','New Hampshire','New Jersey','New Mexico','New York','North Carolina','North Dakota','Ohio','Oklahoma','Oregon','Pennsylvania','Rhode Island','South Carolina','South Dakota','Tennessee','Texas','Utah','Vermont','Virginia','Washington','West Virginia','Wisconsin','Wyoming','District of Columbia')
abbr_list <- c('AL','AK','AZ','AR','CA','CO','CT','DE','FL','GA','HI','ID','IL','IN','IA','KS','KY','LA','ME','MD','MA','MI','MN','MS','MO','MT','NE','NV','NH','NJ','NM','NY','NC','ND','OH','OK','OR','PA','RI','SC','SD','TN','TX','UT','VT','VA','WA','WV','WI','WY','DC')
## Read the datasets
pop <- read.csv("data/PopulationEstimates.csv",stringsAsFactors=F)
results <- read.csv("data/election_results.csv",stringsAsFactors=F)
machine <- read.csv("data/verifier-search.csv",stringsAsFactors=F)
unemployment <- read.csv("data/unemployment.csv",stringsAsFactors=F)
education <- read.csv("data/Education.csv",stringsAsFactors=F)
# census <- read.csv("data/census.csv",stringsAsFactors=F)
census <- read.csv("data/census_cleaned.csv",stringsAsFactors=F)
## Clean up the census data
# census <- census[census$AGEGRP==0 & census$YEAR==8,]
# census$percent_white <- (census$WA_MALE + census$WA_FEMALE) / census$TOT_POP
# census$abbr <- abbr_list[match(census$STNAME,state_list)]
# census$state_county <- paste(census$abbr, census$CTYNAME)
## Clean up the election results data
results$dem <- results$votes_dem > results$votes_gop
results$gop_ratio <- results$votes_gop / (results$votes_gop + results$votes_dem)
results$state_county <- paste(results$state_abbr, results$county_name)
results$percent_white <- census$percent_white[match(results$state_county, census$state_county)]
## Clean up the Verified Voter data
machine$FIPS <- as.numeric(unlist(lapply(machine$FIPS.code, FUN=function(x)ifelse(nchar(x)==9,substr(x,1,4),substr(x,1,5)))))
machine$Make <- toupper(machine$Make)
electric <- machine[grepl("DRE",machine$Equipment.Type),]
## Read in the US map and merge it with the other data by state and county
us <- readOGR(dsn="cb_2015_us_county_500k", layer="cb_2015_us_county_500k")
us$FIPS <- as.numeric(as.character(paste0(us$STATEFP,us$COUNTYFP)))
us$state_abbr <- results$state_abbr[match(us$FIPS,results$combined_fips)]
us$dem <- results$dem[match(us$FIPS,results$combined_fips)]
us$gop_ratio <- results$gop_ratio[match(us$FIPS,results$combined_fips)]
us$pop <- log(as.numeric(gsub(",","",pop$POP_ESTIMATE_2015[match(us$FIPS,pop$FIPS)])))
us$std_pop <- (us$pop - min(us$pop,na.rm=T))/(max(us$pop,na.rm=T)-min(us$pop,na.rm=T))
us$dre <- us$FIPS %in% electric$FIPS
us$unemployment <- unemployment$Unemployment_rate_2015[match(us$FIPS, unemployment$FIPS_Code)]
us$college <- education$Percent.of.adults.with.a.bachelor.s.degree.or.higher..2010.2014[match(us$FIPS,education$FIPS.Code)]
us$percent_white <- results$percent_white[match(us$FIPS,results$combined_fips)]
test <- us$state_abbr
us <- us[!us$state_abbr %in% c("AK","HI"),]
us <- us[!is.na(us$state_abbr),]
## Output election map
png("election_map.png",width=1024,height=612)
myPal <- colorRampPalette(rev(brewer.pal(7,"RdBu")[c(1,2,4,6,7)]))
par(bg="#444444")
plot(us, col=myPal(100)[us$gop_ratio * 99 + 1], border=NA)
rect(grconvertX(seq(0,0.25,length.out=10), from="npc", to="user"),
grconvertY(0,from = "npc", to = "user"),
grconvertX(seq(0,0.25,length.out=10)+(0.25/10),from = "npc", to = "user"),
grconvertY(0.05,from = "npc", to = "user"),col=myPal(10)[1:10],border=NA)
text(grconvertX(seq(0,0.25,length.out=3) + .25/20, from = "npc", to = "user"), grconvertY(0.075,from = "npc", to = "user"), c("Dem","Split","Rep"), col="white")
mtext("2016 Election Results by County",side=3, line=1, cex=2, col="white")
mtext("Benjamin J. Radford", side=3, line=-1, cex=1, col="white")
dev.off()
## Output header image
png("header_map.png",width=1200,height=600)
myPal <- colorRampPalette(rev(brewer.pal(7,"RdBu")[c(1,1,2,4,6,7,7)]))
par(bg="#444444")
plot(us, col=myPal(100)[us$gop_ratio * 99 + 1], border=NA) #ifelse(us$dre,"white",NA))
plot(us[us$dre==1,], border="white",add=T)
dev.off()
## Output population map
png("population_map.png",width=1024,height=612)
myPal <- colorRampPalette(brewer.pal(7,"YlGnBu"))
par(bg="#444444")
plot(us, col=myPal(100)[us$std_pop * 99 + 1], border=NA)
mtext("Population per County (logged)",side=3, line=1, cex=2, col="white")
mtext("Benjamin J. Radford", side=3, line=-1, cex=1, col="white")
dev.off()
## Output unemployment map
png("unemployment_map.png",width=1024,height=612)
myPal <- colorRampPalette(brewer.pal(7,"YlGnBu"))
par(bg="#444444")
plot(us, col=myPal(100)[(us$unemployment-min(us$unemployment,na.rm=T))/(max(us$unemployment,na.rm=T)-min(us$unemployment,na.rm=T)) * 99 + 1], border=NA)
mtext("Unemployment by County",side=3, line=1, cex=2, col="white")
mtext("Benjamin J. Radford", side=3, line=-1, cex=1, col="white")
dev.off()
## Output electronic voting map
png("electronic_voting.png",width=1024,height=612)
par(bg="#444444")
plot(us, col=ifelse(us$dre, "#b2182b", "#e0e0e0"), border=NA)
mtext("Electronic Voting Machines",side=3, line=1, cex=2, col="white")
mtext("Benjamin J. Radford", side=3, line=-1, cex=1, col="white")
dev.off()
## Run models
us_data <- data.frame("gop_vote"=us$gop_ratio, "electronic"=as.numeric(us$dre), "state"=as.factor(us$state_abbr), "population"=us$pop, "unemployment"=us$unemployment, "college"=us$college, "percent_white"=us$percent_white)
wi_data <- us_data[us_data$state=="WI",]
pa_data <- us_data[us_data$state=="PA",]
model_a <- lmer(gop_vote ~ electronic + (1|state), data=us_data)
model_b <- lmer(gop_vote ~ electronic + population + unemployment + college + percent_white + (1|state), data=us_data)
model_wi_a <- lm(gop_vote ~ electronic, data=wi_data)
model_wi_b <- lm(gop_vote ~ electronic + population + unemployment + college + percent_white, data=wi_data)
model_pa_a <- lm(gop_vote ~ electronic, data=pa_data)
model_pa_b <- lm(gop_vote ~ electronic + population + unemployment + college + percent_white, data=pa_data)
## Output model tables
sjt.lmer(model_a,model_b,file="national_table.html")
sjt.lm(model_wi_a,model_wi_b,model_pa_a,model_pa_b,file="wi_pa_table.html")
## Out-of-sample prediction
sample <- sample(1:nrow(us_data),1000,replace=F)
in_sample <- us_data[sample,]
out_sample <- us_data[!1:nrow(us_data) %in% sample,]
model_c <- lmer(gop_vote ~ electronic + population + unemployment + college + percent_white + (1|state), data=in_sample)
preds <- predict(model_c,us_data)
us$pred <- preds
us$pred_density <- -1
us$pred_density[sample] <- 20
## Out-of-sample prediction map
png("predicted_map.png",width=1024,height=612)
myPal <- colorRampPalette(rev(brewer.pal(7,"RdBu")[c(1,2,4,6,7)]))
par(bg="#444444")
plot(us, col=myPal(100)[us$pred * 99 + 1], border=NA, density=us$pred_density)
rect(grconvertX(seq(0,0.25,length.out=10), from="npc", to="user"),
grconvertY(0,from = "npc", to = "user"),
grconvertX(seq(0,0.25,length.out=10)+(0.25/10),from = "npc", to = "user"),
grconvertY(0.05,from = "npc", to = "user"),col=myPal(10)[1:10],border=NA)
text(grconvertX(seq(0,0.25,length.out=3) + .25/20, from = "npc", to = "user"), grconvertY(0.075,from = "npc", to = "user"), c("Dem","Split","Rep"), col="white")
mtext("2016 Out-of-Sample Predicted Results",side=3, line=1, cex=2, col="white")
mtext("Benjamin J. Radford", side=3, line=-1, cex=1, col="white")
dev.off()
## Subset to states that have both electronic (DRE) and non-electronic voting
states_with_dre <- sort(unique(us$state_abbr[us$dre!=0]))
states_without_dre <- sort(unique(us$state_abbr[us$dre==0]))
states_with_dre <- intersect(states_with_dre, states_without_dre)
## Run models for states with both electronic and non-electronic voting
model_list <- list()
for(ss in states_with_dre)
{
state_data <- us_data[us_data$state==ss,]
state_data[complete.cases(state_data),]
model_list <- c(model_list,list(lm(gop_vote ~ electronic + population + unemployment + college + percent_white, data=state_data)))
}
names(model_list) <- states_with_dre
demo_dre <- demo_nodre <- as.data.frame(t(colMeans(us_data[,!names(us_data) %in% c("state","gop_vote")],na.rm=T)))
demo_dre$electronic <- 1
demo_nodre$electronic <- 0
## Output state-wise maps
png("dre_states.png",width=1024,height=1024)
myPal <- colorRampPalette(rev(brewer.pal(7,"RdBu")[c(1,2,4,6,7)]))
par(mfrow=c(ceiling(sqrt(length(states_with_dre))), ceiling(sqrt(length(states_with_dre)))), mar=c(3.1,1.1,3.1,1.1), bg="#444444")
for(ss in states_with_dre)
{
plot(us[us$state_abbr==ss,],col=ifelse(us$dre[us$state_abbr==ss],"#b2182b", "#999999"), border=NA)
mtext(ss,3,line=0,cex=2,col="white")
if(sum(is.na(coef(model_list[[ss]]))) == 0)
{
draws <- predictions(model_list[[ss]])
sims_dre <- draws %*% as.matrix(unlist(c(1,demo_dre)))
sims_nodre <- draws %*% as.matrix(unlist(c(1,demo_nodre)))
# sims_dre <- sort(sims_dre[round(.025*500000):round(0.975*500000)])
# sims_nodre <- sort(sims_nodre[round(.025*500000):round(0.975*500000)])
dens_dre <- density(sims_dre)
dens_dre <- data.frame(x=dens_dre$x, y=dens_dre$y)
dens_nodre <- density(sims_nodre)
dens_nodre <- data.frame(x=dens_nodre$x, y=dens_nodre$y)
standardY <- max(c(dens_dre$y,dens_nodre$y))
dens_dre <- dens_dre[dens_dre$x >= 0 & dens_dre$x <= 1,]
dens_dre <- rbind(c(0,0),dens_dre,c(1,0))
dens_nodre <- dens_nodre[dens_nodre$x >= 0 & dens_nodre$x <= 1,]
dens_nodre <- rbind(c(0,0),dens_nodre,c(1,0))
dens_dre$y <- dens_dre$y/standardY/4
dens_nodre$y <- dens_nodre$y/standardY/4
polygon(grconvertX(dens_nodre$x, from="npc", to="user"),grconvertY(dens_nodre$y, from="npc", to="user"), col="#e0e0e099", border="white")
polygon(grconvertX(dens_dre$x, from="npc", to="user"),grconvertY(dens_dre$y, from="npc", to="user"), col="#b2182b99", border="white")
mtext(sprintf("beta: %.2f | s.e.: %.2f | t: %.2f", coef(model_list[[ss]])["electronic"], summary(model_list[[ss]])$coefficients[,2]['electronic'], summary(model_list[[ss]])$coefficients[,3]['electronic']),1,line=0.25,col="white")
}
}
plot(0,0,type="n",frame=F,xaxt="n",yaxt="n",xlim=c(0,1),ylim=c(0,1))
rect(0,0,0.25,0.25,col="#e0e0e0",border="white")
rect(0,0.5,0.25,0.75,col="#b2182b",border="white")
text(0.3,0.25/2,"No E-Voting",cex=2,pos=4,col="white")
text(0.3,0.25/2+0.5,"E-Voting",cex=2,pos=4,col="white")
dev.off()
##
## Quasibinomial regression models
##
predictionsPart2 <- function(preds, model, newdata)
{
predicted_value <- list()
for(ii in 1:nrow(preds))
{
model$coefficients <- preds[ii,]
predicted_value <- c(predicted_value,list(predict(model, newdata, type="response")))
}
predicted_value <- do.call(c,predicted_value)
predicted_value
}
## Run models
## Using glm instead of glmer because of availability of quasipoisson and convergence issues with glmer binomial modesl.
us_data <- data.frame("gop_vote"=us$gop_ratio, "electronic"=as.numeric(us$dre), "state"=as.factor(us$state_abbr), "population"=us$pop, "unemployment"=us$unemployment, "college"=us$college, "percent_white"=us$percent_white*100)
model_a <- glm(gop_vote ~ electronic + as.factor(state), data=us_data, family=quasibinomial)
model_b <- glm(gop_vote ~ electronic + population + unemployment + college + percent_white + as.factor(state), data=us_data, family=quasibinomial)
summary(model_a)
summary(model_b)
## Subset to states that have both electronic (DRE) and non-electronic voting
states_with_dre <- sort(unique(us$state_abbr[us$dre!=0]))
states_without_dre <- sort(unique(us$state_abbr[us$dre==0]))
states_with_dre <- intersect(states_with_dre, states_without_dre)
## Run models for states with both electronic and non-electronic voting
model_list <- list()
for(ss in states_with_dre)
{
state_data <- us_data[us_data$state==ss,]
state_data[complete.cases(state_data),]
model_list <- c(model_list,list(glm(gop_vote ~ electronic + population + unemployment + college + percent_white, family=quasibinomial, data=state_data)))
}
names(model_list) <- states_with_dre
demo_dre <- demo_nodre <- as.data.frame(t(colMeans(us_data[,!names(us_data) %in% c("state","gop_vote")],na.rm=T)))
demo_dre$electronic <- 1
demo_nodre$electronic <- 0
## Output state-wise maps
png("dre_states_quasibinomial.png",width=1024,height=1024)
myPal <- colorRampPalette(rev(brewer.pal(7,"RdBu")[c(1,2,4,6,7)]))
par(mfrow=c(ceiling(sqrt(length(states_with_dre))), ceiling(sqrt(length(states_with_dre)))), mar=c(3.1,1.1,3.1,1.1), bg="#444444")
for(ss in states_with_dre)
{
plot(us[us$state_abbr==ss,],col=ifelse(us$dre[us$state_abbr==ss],"#b2182b", "#999999"), border=NA)
mtext(ss,3,line=0,cex=2,col="white")
if(sum(is.na(coef(model_list[[ss]]))) == 0)
{
draws <- predictions(model_list[[ss]],n=5000)
sims_dre <- predictionsPart2(draws, model_list[[ss]], demo_dre)
sims_nodre <- predictionsPart2(draws, model_list[[ss]], demo_nodre)
sims_dre <- c(round(sims_dre * 20),0:20)/20
sims_nodre <- c(round(sims_nodre * 20),0:20)/20
dens_dre <- table(sims_dre)-1
dens_nodre <- table(sims_nodre)-1
scaling <- max(c(dens_dre,dens_nodre))
dens_dre <- dens_dre/scaling/4
dens_nodre <- dens_nodre/scaling/4
rect(grconvertX((0:20)/21,"npc","user"),grconvertY(dens_nodre,"npc","user"),grconvertX((1:21)/21,"npc","user"),grconvertY(0,"npc","user"),col="#e0e0e099",border=NA)
rect(grconvertX((0:20)/21,"npc","user"),grconvertY(dens_dre,"npc","user"),grconvertX((1:21)/21,"npc","user"),grconvertY(0,"npc","user"),col="#b2182b99",border=NA)
mtext(sprintf("beta: %.2f | s.e.: %.2f | t: %.2f", coef(model_list[[ss]])["electronic"], summary(model_list[[ss]])$coefficients[,2]['electronic'], summary(model_list[[ss]])$coefficients[,3]['electronic']),1,line=0.25,col="white")
}
}
plot(0,0,type="n",frame=F,xaxt="n",yaxt="n",xlim=c(0,1),ylim=c(0,1))
rect(0,0,0.25,0.25,col="#e0e0e0",border="white")
rect(0,0.5,0.25,0.75,col="#b2182b",border="white")
text(0.3,0.25/2,"No E-Voting",cex=2,pos=4,col="white")
text(0.3,0.25/2+0.5,"E-Voting",cex=2,pos=4,col="white")
dev.off()