0. Packages

Please load the following packages:

library(tidyverse) # contains dplyr, ggplot, readr 
# - one package to rule them all!
library(viridis)
library(infer) # our new inferential statistics pacakage

Task 1: Difference of Means

Import your dataset, which contains a list of electoral districts in specific elections from 2000 to 2017. Filter it to just the years 2010 and 2011. Below, you will learn to calculate the difference in mean voter turnout between these years.

# Import and filter your data
myturnout <- read_csv("japan_turnout.csv") %>%
  filter(election_year %in% c(2010, 2011)) 

# View the first 6 rows with the head() function
myturnout %>%
  head()

Calculate the mean turnout rate for each year, using group_by() and summarize().

myturnout %>%
  group_by(election_year) %>%
  summarize(mean = mean(turnout_rate))

Run your first difference of means test (a.k.a. a t-test)! Be sure to put the treatment group (2011) first, and the control group (2010) second. The statistic field contains your t-statistic, and the estimate contains your raw difference of means. (Don’t worry about the other columns for now.)

myturnout %>%
  t_test(formula = turnout_rate ~ election_year, 
         order = c("2011", "2010"))

# Be sure to tidy up afterwards
remove(myturnout)

Task 2: Crosstabulation and the Chi-squared statistic

Next, we’re going to investigate a dataset, where each row represents a person running for office in a specific district race in a given year. Filter it just to the year 2011.

mycandidates <- read_csv("japan_incumbency.csv") %>%
  filter(election_year == 2011)

Calculate the total number of candidates by status and party, using group_by() and summarize(). Which set of categories overlap the most?

mycandidates %>% 
  group_by(candidate_status, candidate_party) %>%
  summarize(count = n())

Now, calculate the chi-squared statistic and evaluate how much these categories differ from what we would expect if they were not related. Is the chi-squared statistic close to zero (1, 2, 3, 4), or much larger than zero (anything bigger)? What does this tell us about incumbency rates and party affiliation?

mycandidates %>%
  chisq_test(formula = candidate_status ~ candidate_party)

Task 3. Correlation

Finally, let’s examine a similar dataset that narrows into the individuals who ran each year for office. Did younger candidates win greater shares of the vote in 2011? Let’s find out. Please import your data and, you guessed it, filter it to elections in 2011.

myresults <- read_csv("japan_voteshare.csv") %>%
  filter(election_year == 2011)

Let’s use the correlation coefficient (Pearson’s r) to evaluate how closely correlated these two variables are. What do you find? Are they positively or negatively correlated, and how strong is the relationship? (Is it closer to -1, 0, or 1?)

myresults %>%
  summarize(cor = cor(candidate_age, voteshare, 
                      use = "pairwise.complete.obs"))
# the argument use = "pairwise.complete.obs" is like the na.rm = TRUE for the cor() function.

Challenge: Putting it all together!