Question 1

Q1. Report the relationship between three issues scales (Guaranteed Jobs and Incomes-VCF0809, Government Health Insurance Scale-VCF0806, Government Services-Spending Scale-VCF0839). Which is the most strongly re- lated to presidential vote choice?

library(plyr)
library(tidyverse)
library(magrittr)
library(purrrlyr)
library(broom)

d1 <- "https://github.com/thomasjwood/ps7160/raw/master/anes_cdf_20211118.rds" %>% 
  url %>% 
  gzcon %>% 
  readRDS %>% 
  tbl_df

t1  <- d1 %>% 
  select(
    VCF0704a, VCF0809, VCF0806, VCF0839
  ) %>% 
  filter(
    VCF0704a %>% 
      str_detect("1. |2. ")
  ) %>% 
  modify_at(
    1,
    function(i)
      i %>% 
      str_sub(, 1) %>% 
      as.numeric %>% 
      subtract(1)
    ) %>% 
  pivot_longer(
    -VCF0704a, 
    names_to = "scale", 
    values_to = "ans", 
    values_drop_na = T
  ) %>% 
  mutate(
    ans = ans %>% 
      str_sub(, 1) %>% 
      as.numeric,
    scale = scale %>% 
      mapvalues(
        c("VCF0809", "VCF0806", "VCF0839"),
        c("Guaranteed jobs and income",
          "Govt health insurance scale",
          "Govt spending services scale")
        )
    ) %>% 
  filter(
    ans %>%
      is_in(1:7)
    )

library(DAMisc)

t1 %>% 
  group_by(
    scale
  ) %>% 
  nest %>% 
  mutate(
    pre = data %>%
      map(
        \(i)
        glm(
          VCF0704a ~ ans, data = i, family = binomial()
        ) %>% 
          pre %>% 
          extract2(., "pre")
      ) %>% 
      unlist
  )
## # A tibble: 3 × 3
## # Groups:   scale [3]
##   scale                        data                    pre
##   <chr>                        <list>                <dbl>
## 1 Guaranteed jobs and income   <tibble [21,655 × 2]> 0.391
## 2 Govt health insurance scale  <tibble [19,498 × 2]> 0.415
## 3 Govt spending services scale <tibble [17,757 × 2]> 0.438

The pre function reports the proportional reduction in error–how much more predictive a GLM model is, following the inclusion of a predictor variable, compared to a GLM with only an intercept. Bigger values indicate larger improvements in predictiveness. Accordingly, the VCF0839–government spending and services scale–is the most strongly related.

Question 2

Q.2 Working still with these issue scales, does the relationship with presidential vote change over time?

t2 <- d1 %>% 
  select(
    VCF0704a, VCF0004, VCF0809, VCF0806, VCF0839
  ) %>% 
  filter(
    VCF0704a %>% 
      str_detect("1. |2. ")
  ) %>% 
  modify_at(
    1,
    function(i)
      i %>% 
      str_sub(, 1) %>% 
      as.numeric %>% 
      subtract(1)
  ) %>% 
  pivot_longer(
    -c(VCF0704a, VCF0004), 
    names_to = "scale", 
    values_to = "ans", 
    values_drop_na = T
  ) %>% 
  mutate(
    ans = ans %>% 
      str_sub(, 1) %>% 
      as.numeric,
    scale = scale %>% 
      mapvalues(
        c("VCF0809", "VCF0806", "VCF0839"),
        c("Guaranteed jobs and income",
          "Govt health insurance scale",
          "Govt spending services scale")
      )
    ) %>% 
  filter(
    ans %>%
      is_in(1:7)
  )

Now we’ll just do the version of the PRE test we did above, now just doing the test separately by year

t2 %>% 
  group_by(
    scale, VCF0004
  ) %>% 
  nest %>% 
  pmap(
    function(VCF0004, scale, data)
      tibble(
        VCF0004 = VCF0004,
        scale = scale,
        pre = glm(
          VCF0704a ~ ans, data = data, family = binomial()
        ) %>%
          pre %>%
          extract2(., "pre")
      )
  ) %>% 
  list_rbind %>% 
  ggplot(
    aes(VCF0004, pre)
  ) +
  geom_line() +
  facet_wrap(~scale) +
  theme_minimal() +
  labs(
    x = "",
    y = "Poportional reduction in error"
  )

Gawsh! Look at how Govt health insurance jumps in 2012? With whose theory of mass attitudes does this comport?

Question 3

Q.3 Working still with these issue scales, does the relationship with presidential vote interact with respondent education?

Now we only need to trade out years for education!

t3 <- d1 %>% 
  select(
    VCF0704a, VCF0110, VCF0809, VCF0806, VCF0839
  ) %>% 
  filter(
    VCF0704a %>% 
      str_detect("1. |2. ")
  ) %>% 
  modify_at(
    1,
    function(i)
      i %>% 
      str_sub(, 1) %>% 
      as.numeric %>% 
      subtract(1)
  ) %>% 
  pivot_longer(
    -c(VCF0704a, VCF0110),
    names_to = "scale", 
    values_to = "ans",
    values_drop_na = T
  ) %>% 
  mutate(
    ans = ans %>% 
      str_sub(, 1) %>% 
      as.numeric,
    scale = scale %>% 
      mapvalues(
        c("VCF0809", "VCF0806", "VCF0839"),
        c("Guaranteed jobs and income",
          "Govt health insurance scale",
          "Govt spending services scale")
      )
  ) %>% 
  filter(
    ans %>%
      is_in(1:7) &
    VCF0110 %>% 
      str_detect("0. DK") %>% 
      not
  )

t3 %>%
  group_by(
    scale, VCF0110
  ) %>% 
  nest %>% 
  pmap(
    function(VCF0110, scale, data)
      tibble(
        VCF0110 = VCF0110,
        scale = scale,
        pre = glm(
          VCF0704a ~ ans, data = data, family = binomial()
        ) %>%
          pre %>%
          extract2(., "pre")
      )
  ) %>%
  list_rbind %>% 
  ggplot(
    aes(scale, pre, fill = VCF0110)
  ) +
  geom_col(
    color = "black",
    size = .6,
    width = .75,
    position = "dodge"
    ) +
  scale_x_discrete(
    breaks = t3$scale %>% 
      factor %>% 
      levels,
    labels = t3$scale %>% 
      factor %>% 
      levels %>% 
      str_wrap(15)
  ) +
  scale_fill_grey(
    guide = guide_legend(nrow = 2),
    labels = t3$VCF0110 %>%
      levels %>%
      extract(-1) %>% 
      str_sub(4)
    ) +
  theme_minimal() +
  theme(
    panel.grid = element_blank(),
    legend.position = "bottom"
    ) +
  labs(
    fill = "",
    x = "",
    y = "Poportional reduction in error"
  )

Question 4

Q.4 Consider the Lewis-Beck et al. criteria for issue voting:

  • The public reports they have thought about some issue
  • The public adopts a non-neutral position
  • THe public can discriminate between the parties’ respective positions on an issue

Among the ANES CDF’s 7 point issue scales, report an issue where a higher proportion of the public met the issue voting criteria in 2016, compared to prior years.

Ok, so I need to cheat a lot on this, and do some stuff compactly which we’ll only demo later this semester.

t4 <- list(
  c("govhealth", "VCF0806", "VCF0508", "VCF0509"),
  c("guarntjob", "VCF0809", "VCF0513", "VCF0514"),
  c("govserv","VCF0839", "VCF0541"  , "VCF0542")
  ) %>% 
  map(
    function(i)
      
      d1 %>%
      select(
        VCF0006a, VCF0004, i[-1]
      ) %>% 
      gather(actor, ans, -c(1:2), na.rm = T) %>% 
      mutate(
        ans = ans %>% 
          str_sub(, 1) %>% 
          as.numeric,
        actor = actor %>% 
          plyr::mapvalues(
            i[-1],
            c("respondent", "dem_party", "rep_party")
          ),
        scale = i[1]
      ) %>% 
      spread(actor, ans)
  ) %>% 
  list_rbind

t5 <- t4 %>% 
  group_by(
    VCF0004, scale
  ) %>% 
  nest %>% 
  pmap_dfr(
    function(VCF0004, scale, data)
      
      tibble(
        VCF0004 = VCF0004,
        scale = scale,
        perc = data %>% 
      filter(
        dem_party %>% 
          is_in(1:7) &
          rep_party %>% 
          is_in(1:7) &
          respondent %>% 
          is_in(1:7)
      ) %>% 
      mutate(
        issue_voter = ifelse(
          respondent != 4 &
            dem_party != rep_party,
          "prospective issue voter",
          "non-prospective issue voter"
        )
      ) %>% 
      group_by(
        issue_voter
      ) %>% 
      tally %>% 
      mutate(
        prop = n %>% 
          divide_by(
            n %>% 
              sum
          )
      ) %>% 
      select(-n) %>% 
      filter(
        issue_voter == "prospective issue voter"
      ) %>% 
      use_series(prop)
      )
    ) %>% 
  ungroup

t5 %>% 
  ggplot(
    aes(VCF0004, perc)
  ) +
  geom_point() +
  geom_smooth(
    method = "lm"
  ) +
  facet_wrap(~scale) +
  theme_minimal() +
  labs(
    x = "",
    y = "Voters qualifying under American Voter's\nissue voting criteria"
  )

That answers was way way more technical than I wanted to be–but we’ll get to this stuff later in the semester!