Explaining the 2024 Election

I ran a Bayesian logistic regression using data from the 2024 American National Election Study (ANES) to understand who shifted to Donald Trump late in the campaign. I focused on three groups: voters who had previously supported Kamala Harris, voters who had been leaning toward another candidate, and those who were undecided earlier in the year. The model helps identify which attitudes and perceptions best predicted movement into Trump’s camp.

The chart below shows coefficient estimates (points) with 95% credible intervals (bars).

Right of zero (no overlap): higher values associated with greater movement toward Trump
Left of zero (no overlap): higher values associated with less movement toward Trump
Crosses zero: no clear effect of the variable

Key Findings

Lower confidence that one's vote would be counted accurately was a powerful predictor of late movement toward Trump. As the election progressed, Trump consolidated support among those with doubts about the integrity of the electoral system. This effect persists even after controlling for institutional (media and government) and interpersonal trust, suggesting that mistrust in the mechanics of voting itself operate as a distinct and potent driver of political behavior.
People who perceived Trump as holding liberal views on abortion were also more likely to move toward him. Despite appointing the justices who were pivotal to ending Roe, Trump's mixed messaging on the issue throughout his lifetime may have created space for different voters to see what they wanted to see.
Perceptions that the national economy was doing poorly were strongly associated with movement toward Trump. By contrast, personal financial circumstances were not, despite popular narratives about inflation’s impact on household budgets.
Foreign policy issues, such as the wars in Gaza and Ukraine, had surprisingly little influence. One possible explanation is that late-deciders, who are often less politically engaged, may not have been closely following these conflicts.

The interactive map below comes from a multilevel Bayesian logistic regression I ran using the 2024 Cooperative Election Study (CES), a large, nationally representative survey. The model predicts the probability of voting for Donald Trump based on a variety of factors, including attitudes toward immigration.

Because a state's political context can shape how issues matter, I estimated a hierarchical Bayesian logistic regression with random slopes for immigration attitudes by state. That means the model estimates not just one national effect, but allows the effect of immigration attitudes on Trump support to vary across states. States with more data “pull” the estimates toward their observed patterns, while states with smaller samples are “partially pooled” toward the national average.

The results are shown on the map in terms of odds ratios. Odds ratios capture how much the odds of supporting Trump change with a one-unit increase on the immigration attitudes index. The index ranges from 0 to 4, with higher scores denoting a more conservative position. Darker shades represent states where immigration attitudes had a stronger effect on vote choice.

An odds ratio above 1 means that more conservative views on immigration are associated with greater odds of supporting Trump in that state.
An odds ratio below 1 would indicate the opposite (though in practice, all states showed positive associations).
The farther the odds ratio is from 1, the stronger the effect of immigration attitudes on vote choice.

An odds ratio of 1.5 means the odds of supporting Trump are 50% higher for each step toward a more conservative immigration view. For example, if someone started with odds of 2-to-1 (about a 67% chance), one step more conservative would raise their odds to 3-to-1 (about a 75% chance).

Key Findings

Immigration mattered everywhere: In every state, more conservative views on immigration increased the likelihood of supporting Trump
There was substantial interstate variation in the strength of this relationship. In some states, (e.g., Missouri, Colorado, Pennsylvania), immigration attitudes were especially powerful predictors of vote choice, while in others the effect was more modest.
Immigration views were not as determinative as expected in states like California and Texas. It is possible that in states where the issue has long been highly salient, immigration is so deeply ingrained in partisan politics that there is little additional variation for immigration attitudes to explain. By contrast, in states where immigration has not historically dominated political debate, differences in immigration views can exert a stronger independent effect on voter choice.