Home Assignment 3

Student ID: 1A182901-2

---

1. Set up environment

rm(list=ls(all=TRUE))
setwd("~/Desktop/R/polimetrics")
library(manifestoR)
library(quanteda)
library(ggplot2)
library(DT)
library(cowplot)

2. Download Data

#Download Data
mp_setapikey(key.file = NULL, key = "e51dc314a21ce75bd8221fcb338e2ee5")
mp_use_corpus_version("2017-2")
cmp <- mp_maindataset()
cmp$vanilla_i <- vanilla(cmp,invert = TRUE)
sa <- mp_corpus(countryname=="South Africa" & date == 201405)
quanteda_sa <- corpus(sa)

3. Construct dfm

dfm_sa <- dfm(quanteda_sa, stem = TRUE, tolower=TRUE, remove_punct = TRUE, remove = stopwords("english"),  remove_numbers=TRUE)
dfm_sa2 <- dfm_group(dfm_sa, "manifesto_id") 

4. 5 most frequent words

ffc <- textstat_frequency(dfm_sa2, n = 5, group = "manifesto_id") 
sa_cmp <- cmp[ which(cmp $countryname=="South Africa"  & cmp $date == 201405),]
ffc <- mutate(ffc,manifesto_id = factor(group, labels = sa_cmp$partyname)) 
ffc$name <- paste(LETTERS[1:25],ffc$feature,sep="")
ggplot(ffc,aes(x = reorder(name,frequency), y = frequency, fill = manifesto_id)) +
  geom_col(width = 0.8, show.legend = FALSE) +
  theme_bw()+
  labs(x = NULL, y = "5 most frequent words for each party") +
  facet_wrap(vars(manifesto_id), ncol = 2, scales = "free") +
  scale_fill_brewer(palette = "Set1")+
  scale_x_discrete(labels = setNames(ffc$feature, ffc$name))+
  coord_flip()

5. Wordfish

wfmsa <- textmodel_wordfish(dfm_sa2, dir = c(2, 3))
v_wf <- wfmsa$theta
df_wf <- as.data.frame(v_wf)
df_wf$min <- wfmsa$theta-wfmsa$se.theta
df_wf$max <- wfmsa$theta+wfmsa$se.theta
df_wf$title <- sa_cmp$partyname
##Graph
gr_wf <- ggplot(df_wf,aes(y=reorder(title,v_wf),x=v_wf,color=title))
gr_wf_res <- gr_wf + geom_point(size=3) +  ylab("") + xlab("") + theme_light() + theme(legend.position="none")
gr_wf_res

6. Wordscore

ws <- textmodel_wordscores(dfm_sa2, c(NA,5.2,10.5,NA,NA))
pr_raw <- predict(ws, newdata = dfm_sa2)
df_raw <- as.data.frame(pr_raw)
df_raw$title <- sa_cmp$partyname
##Graph
gr_ws <- ggplot(df_raw,aes(y=reorder(title,pr_raw),x=pr_raw,color=title))
gr_ws_res <- gr_ws + geom_point(size=3) + ylab("") + xlab("")+theme_light() + theme(legend.position="none")
gr_ws_res

7. CMP data

sa_cmp$right <- sa_cmp$per104+sa_cmp$per201+sa_cmp$per203+sa_cmp$per305+sa_cmp$per401+sa_cmp$per402+sa_cmp$per407+sa_cmp$per414+sa_cmp$per505+sa_cmp$per601+sa_cmp$per603+sa_cmp$per605+sa_cmp$per606
sa_cmp$left <- sa_cmp$per103+sa_cmp$per105+sa_cmp$per106+sa_cmp$per107+sa_cmp$per202+sa_cmp$per403+sa_cmp$per404+sa_cmp$per406+sa_cmp$per412+sa_cmp$per413+sa_cmp$per504+sa_cmp$per506+sa_cmp$per701
sa_cmp$RILE <- sa_cmp$right-sa_cmp$left
sa_cmp$RATIO <- (sa_cmp$right-sa_cmp$left)/(sa_cmp$right+sa_cmp$left)
sa_cmp$LOGIT <- mp_scale(sa_cmp, scalingfun = logit_rile)

8. Graph CMP result

df_sa <- data.frame(party=sa_cmp$partyname,
                     rile=sa_cmp$RILE,
                     ratio=sa_cmp$RATIO,
                     logit=sa_cmp$LOGIT,
                     vanilla_i=sa_cmp$vanilla_i)
gr_rl <- ggplot(df_sa,aes(x=rile,y=reorder(party,rile),color=party))+
  geom_point(size=3)+
  theme_gray()+
  theme(legend.position="none")+
  theme(axis.ticks = element_blank(), axis.text.y = element_blank())+
  ylab("")
gr_rl_l <- ggplot(df_sa,aes(x=rile,y=reorder(party,rile),color=party))+
  geom_point()+
  theme(axis.ticks = element_blank(), axis.text.y = element_blank())+
  theme(legend.text = element_text(size = 14))+
  ylab("")
gr_ra <- ggplot(df_sa,aes(x=ratio,y=reorder(party,ratio),color=party))+
  geom_point(size=3)+
  theme_gray()+
  theme(legend.position="none")+
  theme(axis.ticks = element_blank(), axis.text.y = element_blank())+
  ylab("")
gr_lo <- ggplot(df_sa,aes(x=logit,y=reorder(party,logit),color=party))+
  geom_point(size=3)+
  theme_gray()+
  theme(legend.position="none")+
  theme(axis.ticks = element_blank(), axis.text.y = element_blank())+
  ylab("")
gr_va <- ggplot(df_sa,aes(x=vanilla_i,y=reorder(party,vanilla_i),color=party))+
  geom_point(size=3)+
  theme_gray()+
  theme(legend.position="none")+
  theme(axis.ticks = element_blank(), axis.text.y = element_blank())+
  xlab("vanilla(invert)")+
  ylab("")
gr_res <- plot_grid(gr_rl,gr_ra,gr_lo,gr_va,labels=colnames(df_sa)[2:5],label_size = 12, nrow = 4,ncol = 1)
legend <- get_legend(gr_rl_l)
gr_res2 <- plot_grid(gr_res,legend,rel_widths = c(1,.8))
gr_res2

9. Combine Wordfish, Wordscore, and CMP

gr_wf_res2 <- gr_wf_res + theme(axis.ticks = element_blank(), axis.text.y = element_blank())
gr_ws_res2 <- gr_ws_res + theme(axis.ticks = element_blank(), axis.text.y = element_blank())
gr_right <- plot_grid(gr_wf_res2,gr_ws_res2,legend,nrow = 3,labels = c("Wordfish","Wordscore",""),label_size = 12)
gr_com <- plot_grid(gr_res,gr_right)
gr_com

10. Discussion

• According to the graph of Wordfish, Wordscore, CMP data of South Africa in 2014, we could have several conclusions.
  • Despite the order changed sometimes, African National Congress(red), Congress of people(brown), Economic Freedom Figters(blue) were on the left in every result, comparing to Democratic Alliance(green) and Inkatha Freedom Party(pink).
  • In the left part of the graph,
    1. The divergence between RILE and RATIO results, represented the effect of changing analytical tools.
    2. The result of vanilla(inverted), revealed the distinct pattern, which meant that those manifestoes contained more than left-right scale.
  • Comparing CMP and Wordfish, wordscore,
    1. The results of RATIO and LOGIT were similar to the result of wordscore. Those three left-right scale analyses represented some robustness, regardless using sentences of whole manifestoes.
    2. The two broad methods, vanilla and wordfish, not only presented different patterns, but also suggested the distinct outcomes from other left-right analyses. The further implications or discussions require more information, such as the political coalitions or seats, or the analyses of other scales.

---