Experiment 1: Postcards to low-turnout, non-primary participating Republican voters in Florida and Pennsylvania swing districts
In the 2024 election cycle I designed and implemented a couple of experiments to measure the impact of postcards on voters’ behavior. In this post, I summarize the results of the postcard experiment to low-turnout Republican voters in heavily-Republican precincts in Florida and Pennsylvania.
Postcards to lower-turnout Republican voters did not significantly reduce Republican turnout nor induce Republicans to change their party affiliation, with no significant variation either comparing all postcard recipients (n=1,613) to a holdout sample or the sample universe overall, nor comparing the holdout sample and samle universe overall to each of four messaging conditions (each about n=400 apiece, totaling n=1,613). Results are marginally in the direction one would expect from prior experimental and theoretical work - with the “contrast” condition producing a statistically marginal result in the expected direction and the “pro-Democrats” condition producing a statistically marginal backlash, but no results from this experiment produced statistically significant differences from the holdout sample or control condition.
Experiment 2: Postcards to low-turnout Democratic voters in Florida and Pennsylvania swing districts
I observed no statistically significant differences between sending handwritten, typed, or “facsimile handwritten” (i.e., with handwritten messages printed onto the postcards) to encourage turnout among low-propensity Democratic voters (n=1,500). Those who received any postcard, or any particular type of postcard, were no more or less likely to turn out to vote than were holdout observations within the sample geography or to the sample universe overall.
Together, these results demonstrate limited payoff to using postcards to attempt to change voter behavior.
In the November 2024 election cycle, I designed and implemented two experiments that intended to use postcards to induce changes in voter behavior. These experiments measured very different things.
What I will refer to as Experiment 1 targeted postcards to low-propensity Republican voters who both haven’t recently participated in primaries and who have skipped at least a couple recent general elections (while remaining active on their state’s voterfile). The full design is summarized here. Of the 46,439 Pennsylvanians who qualified for this study, n=1,613 were randomized into one of four messaging conditions, which determined what type of postcard they received. I also kept track of a holdout sample (n=400) drawn from the list of treatment precincts within the overall sample geography (full description in linked post but, basically, very very Republican precincts with a Republican representative in the State House, State Senate, Congress, and that gave a majority of their vote to Trump in the 2020 election). Regressing what is available on the voterfile against assignment into the holdout versus the sample universe overall didn’t really produce anything (i.e., the groups are demographically pretty analogous) but I knew people would ask so here we are.
The