What is the correlation between swing and change in votes cast from the 2016 to 2020 US presidential elections among major parties?

Question

In the past, it was said that turnout benefits the Democratic Party because populations that tend to vote Democratic are underrepresented among actual voters relative to the voting eligible population of the USA.

The 2020 election gives us an opportunity to test this assumption. Every state recorded a larger vote total for the two major parties in 2020 than in 2016. This is because it had the highest turnout out of any election in the last century: 2 out of 3 eligible voters voted in the 2020 presidential election compared to the turnout in 2016 which was roughly 11 points lower.

What is the correlation between a state's two party vote swing towards Biden (in percentage points) and the percent increase in votes cast for either major party in 2020 vs 2016?

Answer 1

The data for this question, and many you ask, is readily available from the MIT Election Lab, specifically here I used the "U.S. President 1976-2020" dataset.

The correlation between the figures you ask is r=-.08, p=.6. If you remove Utah, which had a successful third party candidate in Evan McMullin in 2016 (receiving 21.5% of the vote), the correlation becomes r=-.11, p=.4. This means there is no statistically significant correlation between increase in turnout and swing toward the Democrats.

R Code:

library(tidyverse)
library(ggrepel)
library(Hmisc)

setwd("~/")

load("1976-2016-president.RData")

potus <- read_csv("1976-2020-president.csv") %>%
  filter(
    year %in% c(2016, 2020),
    party_simplified %in% c("DEMOCRAT", "REPUBLICAN")
  ) %>%
  select(year, state_po, party_simplified, candidatevotes) %>%
  group_by(year, state_po, party_simplified) %>%
  summarize(
    # A few places where some states have two DEM/REP lines
    candidate_votes = sum(candidatevotes),
  ) %>%
  ungroup() %>%
  group_by(year, state_po) %>%
  mutate(
    # Create new two-party vote total
    two_party_vote = sum(candidate_votes)
  )

potus_wide <- potus %>%
  pivot_wider(id_cols = c(year, state_po, two_party_vote), 
              names_from = party_simplified,
              values_from = candidate_votes) %>%
  mutate(
    pct_dem = DEMOCRAT / two_party_vote,
    pct_gop = REPUBLICAN / two_party_vote
  )

years_wide <- potus_wide %>%
  pivot_wider(id_cols = state_po, 
              values_from = c(two_party_vote, DEMOCRAT, REPUBLICAN, 
                              starts_with("pct_")),
              names_from = year) %>%
  mutate(
    pct_inc_totalvotes = (two_party_vote_2020 / two_party_vote_2016) * 100 - 100,
    pct_dem_swing = (pct_dem_2020 - pct_dem_2016) * 100,
    dem_win = DEMOCRAT_2020 > REPUBLICAN_2020
  ) %>%
  ungroup()

years_wide %>%
  select(pct_inc_totalvotes, pct_dem_swing) %>%
  as.matrix() %>%
  rcorr()

ggplot(years_wide, aes(x = pct_inc_totalvotes, y = pct_dem_swing)) +
  geom_point(aes(color = dem_win)) +
  geom_text_repel(aes(label = state_po, color = dem_win)) +
  geom_smooth(method = "lm") +
  scale_x_continuous(limits = c(0, NA)) +
  theme_minimal() +
  theme(legend.position = "none") +
  labs(x = "% increase total two-party vote", y = "Dem swing (pp)")

years_wide %>%
  filter(state_po != "UT") %>%
  select(pct_inc_totalvotes, pct_dem_swing) %>%
  as.matrix() %>%
  rcorr()