Thursday, June 22, 2017

Education and Race Account for Outliers in Trump's Vote at the State and PA County Level

I've written about how things like education, the uninsured, and the concentration hate groups predict Trump's % of the vote at the US state level and the county level in Pennsylvania.  At the state level Washington, DC was an outlier relative to the rest of the states and at the county level Philadelphia was an equal outlier relative to the rest of the counties in PA.  One variable I haven't considered directly is race.  The exit poll in 2016 showed that 60% white women without a college education voted for Trump while 71% of white men with out a college education did.  74% of nonwhites voted for Clinton.

For PA Counties
For this post I will look at how the introduction of the % white as a predictor variable in the state and county models.  The table below shows the model for PA counties with % of the county with a bachelors degree and % white as predictor variables.  Both predictors were significant and accounted for 86.4% of the variability in Trump's of the vote.  For every 1% increase in the % bachelors in a county there was a 0.84% decrease in the predicted % of the vote for Trump.  For every 1% increase in the % white there was a 0.78%increase in Trump's % of the vote.  The plot above shows that only Forest county is an outlier with a higher than expected % of the vote for Trump.  The other counties are well fit by the model.

PA Counties
Coefficients
Standard Error
t Stat
P-value
Lower 95%
Upper 95%
Intercept
10.76
7.08
1.52
0.13
-3.38
24.91
% Bachelors
-0.84
0.08
-10.87
3.57E-16
-1.00
-0.69
% white
0.78
0.07
11.34
5.93E-17
0.64
0.91
  
For US States


For the model for the states there were three three significant variables accounting for 78.5% of the variability: % bachelors, % white, and % uninsured.   For every 1% increase in the % bachelors there was a 1.11% decrease in the predicted Trump %.  Likewise there was a 0.31% increase for Trump for a 1% increase in % white and a 0.74% increase for a 1% increase in the % uninsured.  The chart above shows few points were poorly fit by the model.  The concentration of hate groups was no longer significant when % white was added to the model.


US States 
Coefficients
Standard Error
t Stat
P-value
Lower 95%
Upper 95%
Intercept
51.55
8.92
5.78
5.75E-07
33.61
69.48
% bachelors
-1.11
0.15
-7.55
1.2E-09
-1.41
-0.82
% White
0.31
0.06
4.95
1.01E-05
0.18
0.43
% uninsured
0.74
0.26
2.86
0.006319
0.22
1.26

The addition of % white population in the states of the US and the counties of PA has greatly improved the predictive power of the model and reduced the effect of outliers in the model.  At the state level the effect of the uninsured shows that an increase in the uninsured predicts an increase in Trump's vote.  This is troubling given that the health care reform law considered by congress will likely increase the uninsured.


**Related Posts**

How is Washington DC an outlier? Let's count the ways. (Repost from Data Driven Journalism)

 

Educational Attainment and the % Uninsured Explain Trump's % of the Vote with Philly Considered

 

Hate Groups and Trump's Vote%: Predictive effect present when education and poverty are considered