Virginia 2023 State House Election Rural Postcard Campaign Evaluation

Swing Blue Alliance Research Group

 

Summary
In 2023, Swing Blue Alliance (SBA) joined with Rural Ground Game (RGG)and the National Women’s Political Caucus to test the impact of candidate postcards in rural areas on both turnout and vote share. We know of no previous studies of candidate cards on turnout or vote share. The target districts were two House districts and a non-overlapping Senate district in the 2023 Virginia State House Election. Turnout impact was evaluated at the individual level. Since individual vote choice is private, we evaluated vote share in the smallest areas for which it is available, voting precincts. Each precinct was randomly assigned a percentage of registered voters to whom a card would be sent, and within precincts, individuals were randomly assigned to receive a card. Differences in the characteristics of precincts and individuals that appeared despite random assignment were controlled statistically using regression adjustment.

The intervention targeted two groups of registered voters who had not voted in all four of the most recent statewide elections: registered Democrats and registered non-partisans (unaffiliated voters) under the age of 51.

The study yielded three main findings:

• Sending candidate cards did not increase turnout. In one district, postcards were associated with roughly a half percentage point decrease in the probability of voting (0.46%; (p= 0.024).
• In contrast, candidate cards did increase the Democratic share of the vote. For every 10 percent increase in the penetration rate, the Democratic candidate vote share increased by roughly two-thirds of a percentage point (0.65%; p = 0.03).
• Statistical adjustment for differences between treatment and control that remained despite randomization was critically important. Failing to make these adjustments actually reversed the sign of one treatment effect.

Although the impact of uncontrolled differences between treatment and control groups is expected to be smaller in large samples, this study illustrates that one cannot safely ignore differences simply because of sample size. Despite large samples, this study found treatment/control differences large enough to badly bias findings if not controlled statistically. This has important implications for future studies, as many experimental studies of GOTV campaigns and vote share have not included the necessary statistical controls.