A model of vote intent in 2020 — half of women voters will not vote for Trump

After estimating model of vote intent in 2020 — based on responses from over 230,000 registered U.S. voters — I noticed that its confidence hinges on respondents’ gender.

• Among the group of female respondents who are estimated to be the least likely to support Trump (the bottom quartile), the probability of any person voting for Trump is 2.2% or less.
• Among the most unlikely male Trump voters (again, the bottom quartile), the probability of supporting Trump is 6.1% or less.

For predicting a non-vote for Trump, the model has nearly three times more confidence in correctly classifying women’s choices.

When it comes to the most pro-Trump group of women (the top quartile based on their probability of supporting Trump), the probability interval is 76.7% and above. The model is less sure that women who appear to be Trump supporters will actually acts Trump voters, compared to men. Specifically, 25% of men have a 89.3% probability — or higher — of supporting the president.

Consider also these numbers for men:

• For 20% of male respondents, the model suggests that the probability of a Trump vote is at least 92.7%.
• Among the men in the top (most pro-Trump) decile the predicted probability of supporting Trump 97.1% or higher.

This is the full distribution of probabilities for women and men:

The median probability of voting for Trump among women is 13.3%. That is, half of women have this (or lower) probability of supporting the president. The median probability among male respondents is about 40%.

Some of the low-probability women’s votes will be cast for the current president, but the histogram shows that there is a significant mass of people with Prob(2020 Trump vote) somewhere between 13% and 25% too. We can expect that most of these voters, if they cast a ballot, will support Biden.

Given the current data and the specification I used (details below), the current prediction for November is:

• Biden wins the popular vote, but loses among men.
• Biden wins among women.
• It’s conceivable that 58% of Biden voters could be women. (Someone should check whether there is a precedent for that!)

Note on methods / Model estimation:

• Survey data: Nationscape waves 1 through 50, restricted to registered voters.
• Link function: logit.
• Predictors of vote intent: partisanship, education, age, income, gender, race, views on economic and social policy, views on immigration, and racial attitudes.
• Sample: 113,782 men and 123,877 women.
• 96,660 respondents stated they would vote for Trump against Joe Biden.