Home Assignment 2

Student ID: 1A182901-2

---

1. Set up environment

rm(list=ls(all=TRUE))
setwd("~/Desktop/R/polimetrics")
library(RColorBrewer)
library(readtext)
library(quanteda)
library(ggplot2)
library(cowplot)
setwd("~/Desktop/r/polimetrics/IE")

---

2. Read file

myText <- readtext("~/Desktop/r/polimetrics/IE/*.txt",
                   docvarsfrom = "filenames", dvsep = " ", docvarnames = c("Year", "Party"))
testCorpus <- corpus(myText)
testCorpus <- corpus(myText, docid_field = "doc_id")
docnames(testCorpus) <- gsub(".txt", "", docnames(testCorpus))
myDfm <- dfm(testCorpus , remove = stopwords("english"), tolower = TRUE, stem = TRUE,
             remove_punct = TRUE, remove_numbers=TRUE)

---

3. Wordfish

wfm <- textmodel_wordfish(myDfm, dir = c(1, 3))
v_wf <- wfm$theta
df_wf <- as.data.frame(v_wf)
df_wf$min <- wfm$theta-wfm$se.theta
df_wf$max <- wfm$theta+wfm$se.theta
df_wf$title <- gsub("19","",wfm$docs)
df_wf$title <- with(df_wf,reorder(title,v_wf))

---

4. Wordscore

ws <- textmodel_wordscores(myDfm, c(4.5,13.13,15,6.88,17.63,rep(NA,5)))
#Predict wordscore
pr_raw <- predict(ws, newdata = myDfm[c(6,7,8,9,10), ])
pr_raw_a <- predict(ws, newdata = myDfm[c(1:10), ])
pr_lbg <- predict(ws, rescaling = "lbg", newdata = myDfm[c(6,7,8,9,10), ], interval = "confidence")
#Raw dataframe
v_raw <- c(c(4.5,13.13,15,6.88,17.63),pr_raw)
df_raw <- as.data.frame(v_raw,row.names = docnames(myDfm))
df_raw$title <- gsub("19","",rownames(df_raw))
df_raw$title <- with(df_raw,reorder(title,v_raw))
#All raw dataframe
df_raw_a <- as.data.frame(pr_raw_a)
df_raw_a$title <- gsub("19","",rownames(df_raw_a))
df_raw_a$title <- with(df_raw_a,reorder(title,pr_raw_a))
#LBG dataframe
df_lbg <- as.data.frame(pr_lbg)
df_lbg$title <- gsub("19","",rownames(df_lbg))
colnames(df_lbg) <- c("score","lb","ub","title")
df_lbg$g <- "pr"
df_ref <- data.frame(
  value1 = c(4.5,13.13,15,6.88,17.63),
  value2 = c(4.5,13.13,15,6.88,17.63),
  value3 = c(4.5,13.13,15,6.88,17.63)
)
rownames(df_ref) <- gsub(".txt","",myText$doc_id[1:5])
df_ref$title <- gsub("19","",rownames(df_ref))
colnames(df_ref) <- c("score","lb","ub","title")
df_ref$g <- "ref"
df_lbg_res <- as.data.frame(rbind(df_ref,df_lbg))
df_lbg_res$title <- with(df_lbg_res, reorder(title,score))

---

5. Graph dataframe

#Raw
gr <- ggplot(df_raw,aes(y=title,x=v_raw))
gr_res<- gr + geom_point() + ylab("") + xlab("")+theme_light()
#All raw
gr_ra <- ggplot(df_raw_a,aes(y=title,x=pr_raw_a))
gr_ra_res <- gr_ra + geom_point() + ylab("") + xlab("")+theme_light()
#LBG
gr_lbg <- ggplot(df_lbg_res,aes(score,title,color = g)) +
  geom_point()+
  geom_errorbarh(aes(xmin=lb,xmax=ub,height = .1)) +
  xlab("")+
  ylab("")+
  theme_light()+
  theme(legend.position="none")+
  scale_color_manual(values=c("navy", "black"))
#wordfish
gr_wf <- ggplot(df_wf,aes(y=title,x=v_wf))
gr_wf_res <- gr_wf + geom_point() +  ylab("") + xlab("")+theme_light()

---

6. Raw wordscores(with reference) versus Wordfish

plot_grid(gr_res, gr_wf_res,labels = c('Raw Wordscores', 'Wordfish'))

• The result of raw scores without any transformation could not display the trends of parties between 1992 and 1997. The following two graphs are two kinds of transformation versus wordfish result.

---

7. ALL wordscores versus Wordfish

plot_grid(gr_ra_res, gr_wf_res,labels = c('All Raw Wordscores', 'Wordfish'))

• The prediction of all documents with reference text still remained the same trends to the previous graph. The LBG transformation may have further information.

---

8. LBG transformed wordscores(with reference) versus Wordfish

plot_grid(gr_lbg,gr_wf_res,labels = c('LBG Wordscores', 'Wordfish'))

• According to the Wikipedia, the coalition changed since the breakdown of FF-Lab government in 1994. In 1997’s election, two coalitions, FF-PDs and DL-Lab-FG, were competing for the majority. This competition of government was not revealed by the social left-right scale.

• However, in the right part, the result of wordfish had some patterns. FG and Lab were the two that changed a lot in the relative positions, they were also the members of the new coalition. Moreover, the coalition of DL-Lab-FG (Rainbow coalition) in 1997 shared close relative positions.

• In conclusion, the wordfish result revealed more dynamics of political competition and coalition. One possible argument could be that the wordfish revealed the general, multiple scales of the context, which corresponded better to the relationship of parties in elections.

---