Is ideological polarization in the US happening equally on both sides of the party system, or do the parties behave differently in terms of their distance from the ideological middle? Let’s try to get an answer to this that cleans up some of the background noise by using a regression-discontinuity design. We’ll compare close elections that Republicans just barely win to close elections that Democrats just barely win and look at the absolute value of the resulting ideological distance from the middle of the elected representatives.

Start by loading relevant data:

library(tidyverse)
## ── Attaching core tidyverse packages ──────────────────────── tidyverse 2.0.0 ──
## ✔ dplyr     1.1.4     ✔ readr     2.1.6
## ✔ forcats   1.0.1     ✔ stringr   1.6.0
## ✔ ggplot2   4.0.1     ✔ tibble    3.3.0
## ✔ lubridate 1.9.4     ✔ tidyr     1.3.1
## ✔ purrr     1.2.0     
## ── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──
## ✖ dplyr::filter() masks stats::filter()
## ✖ dplyr::lag()    masks stats::lag()
## ℹ Use the conflicted package (<http://conflicted.r-lib.org/>) to force all conflicts to become errors
#Download data on US House elections
housevotes <- read.csv("https://raw.githubusercontent.com/jnseawright/PS312/refs/heads/main/Data/housevotes.csv")

#Discard data about third-party and independent candidacies
housevotes <- housevotes %>% filter(party %in% c("DEMOCRAT","REPUBLICAN"))

#Calculate the two-party vote margin for each district in each cycle
election_totals <- housevotes %>%
  group_by(year, state, district) %>%
  summarise(
    dem_votes = sum(candidatevotes[party == "DEMOCRAT"], na.rm = TRUE),
    rep_votes = sum(candidatevotes[party == "REPUBLICAN"], na.rm = TRUE),
    .groups = "drop"
  ) %>%
  mutate(
    two_party_total = dem_votes + rep_votes,
    # Compute Democratic share of two‑party vote (set to NA if no votes at all)
    dem_two_party_share = if_else(two_party_total > 0,
                                  dem_votes / two_party_total,
                                  NA_real_),
    # Margin: Democratic share minus Republican share = 2*dem_share - 1
    two_party_margin = 2 * dem_two_party_share - 1
  )

#Read in data about US House members' ideology. These data come from https://voteview.com/data Information about each variable is here: https://voteview.com/articles/data_help_members
Hall_members <- read.csv("https://raw.githubusercontent.com/jnseawright/PS312/refs/heads/main/Data/Hall_members.csv")

#Filter the time period for the ideology variables to match our election data.
members_1970sforward <- Hall_members %>% filter(congress >= 95)

# Create a named vector mapping ICPSR codes (as integers) to uppercase state names
icpsr_state_names <- c(
  "1"  = "CONNECTICUT",        "2"  = "MAINE",
  "3"  = "MASSACHUSETTS",      "4"  = "NEW HAMPSHIRE",
  "5"  = "RHODE ISLAND",       "6"  = "VERMONT",
  "11" = "DELAWARE",           "12" = "NEW JERSEY",
  "13" = "NEW YORK",           "14" = "PENNSYLVANIA",
  "21" = "ILLINOIS",           "22" = "INDIANA",
  "23" = "MICHIGAN",           "24" = "OHIO",
  "25" = "WISCONSIN",          "31" = "IOWA",
  "32" = "KANSAS",             "33" = "MINNESOTA",
  "34" = "MISSOURI",           "35" = "NEBRASKA",
  "36" = "NORTH DAKOTA",       "37" = "SOUTH DAKOTA",
  "40" = "VIRGINIA",           "41" = "ALABAMA",
  "42" = "ARKANSAS",           "43" = "FLORIDA",
  "44" = "GEORGIA",            "45" = "LOUISIANA",
  "46" = "MISSISSIPPI",        "47" = "NORTH CAROLINA",
  "48" = "SOUTH CAROLINA",     "49" = "TEXAS",
  "51" = "KENTUCKY",           "52" = "MARYLAND",
  "53" = "OKLAHOMA",           "54" = "TENNESSEE",
  "55" = "DISTRICT OF COLUMBIA", "56" = "WEST VIRGINIA",
  "61" = "ARIZONA",            "62" = "COLORADO",
  "63" = "IDAHO",              "64" = "MONTANA",
  "65" = "NEVADA",             "66" = "NEW MEXICO",
  "67" = "UTAH",               "68" = "WYOMING",
  "71" = "CALIFORNIA",         "72" = "OREGON",
  "73" = "WASHINGTON",         "81" = "ALASKA",
  "82" = "HAWAII"
)

# Function to convert numeric codes to state names
icpsr_to_state <- function(codes) {
  # Convert codes to character for safe indexing (handles both integer and factor input)
  char_codes <- as.character(codes)
  # Return mapped names, with NA for any unmapped codes
  unname(icpsr_state_names[char_codes])
}

members_1970sforward$state_name <- icpsr_to_state(members_1970sforward$state_icpsr)

#Calculate which Congress number each election year chose
election_totals$congress <- floor((election_totals$year - 1789) / 2) + 2

#Combine the datasets together
elections_and_ideology <- election_totals %>% left_join(members_1970sforward, 
                                                        by = c("state" = "state_name", 
                                                               "district" = "district_code",
                                                               "congress" = "congress"))

At this point, you’re ready to start carrying out your regression-discontinuity design. Fill in the sample code below with the outcome variable and the assignment variable, as makes sense for your analysis.

library(rdrobust)
ideology_party <- rdrobust(abs(PUT YOUR DEPENDENT VARIABLE HERE), PUT THE VARIABLE THAT SPLITS CASES BETWEEN DEMOCRATIC AND REPUBLICAN WINNERS HERE)
summary(ideology_party)

rdplot(abs(elections_and_ideology$PUT YOUR DEPENDENT VARIABLE HERE), elections_and_ideology$PUT THE VARIABLE THAT SPLITS CASES BETWEEN DEMOCRATIC AND REPUBLICAN WINNERS HERE)

As a group, evaluate the credibility of your results and interpret their meaning. Send your figure and a paragraph explaining your estimator and explaining your results to your TA.