In my last post I reported on how median income and the % having at least some college in a county predicted support for Donald Trump in that county in Pennsylvania. These were univariate correlations with Trump’s vote. One sociologist I know commented that median income would not be significant if I controlled for more variables. So I ran a multiple linear regression model to see which of the County Health Ranking (CHR) variables were the most robust predictors of his vote. The model settled on 7 predictors out of more than 60 that explain 92.9% of the variability in Trump’s vote:

• The average number of physically unhealthy days reported in the last 30 days
• The percentage receiving a flu vaccine
• The percentage with access to exercise opportunities
• Severe housing cost burden
• Some college
• Census participation
• The percentage of non Hispanic white.

My sociologist friend was right that median income would be no longer significant in a multiple regression model while a college education would be. This post will focus on the average number of unhealthy days as a predictor of Trump’s vote.

This graph shows the correlation between the average number of physically unhealthy days and Trump’s vote in that county. The model states that for every one day increase in this measure for a county there is a predicted 10% increase in Trump’s vote in that county. The lone outlier in this measure is Philadelphia county with an average of 4.6 unhealthy days and 20% of the vote for Trump.

This measure states that unhealthy counties (except for Philadelphia) were more likely to support Donald Trump than healthy ones. This effect was still present after the other 60 measures were considered. My next posts will look at the other predictive measures of Trump’s vote. When considering all of the variables at once, we can be more confident of the number of unhealthy days as a predictor of Trump’s vote in PA.