Nothing But Numbers

There has been a lot of debate on how exactly Donald Trump was able to defy the pundits and win the electoral college in 2016.  Pre-election polls had shown Clinton with a moderate yet consistent lead in the national popular vote over Trump, based on strong support among whites with a college degree and non-white voters.  On election night, however, the national exit poll showed Trump narrowly winning among white voters with a college degree and outperforming Mitt Romney among Hispanic voters, contrary to the pre-election polling.  This led to the so-called “Shy” Trump voter hypothesis: the idea that the polls were wrong because some well-educated white and non-white Trump voters were too embarrassed to admit to pollsters that they supported Trump.   Indeed, many have insisted that it was actually wealthy, well-educated voters – not working class whites – who were the core of Trump’s voter base.  A closer examination of the 2016 results, however, calls into question the validity of the exit poll and the “Shy” Trump voter hypothesis, and points to a surge in white working class support as the most important factor in Trump’s victory.

In order to get a better understanding of how race and education impacted the 2016 election, I categorized each US county into six demographic groups based on US Census and American Community Survey data: non-Hispanic whites with a bachelor’s degree or higher, non-Hispanic whites without a bachelor’s degree, Black, Hispanic or Latino, Native American, and Asian/Pacific Islander.  I used Pew Research data to estimate the share of Hispanic voters who were ineligible to vote, and I used the Center for American Progress’s (CAP) research to adjust for differing voter turnout rates among groups.  Based on this data, I was able to estimate each group’s share of the electorate in every county in the country outside of Alaska (Alaska reports election results by state house districts rather than by county).  After estimating the demographic make-up of the 2016 electorate in each county, I then conducted a regression analysis to determine how each group’s voting patterns shifted from 2012 to 2016, weighted by the total number of votes cast in each county in 2016.  Specifically, I examined how race and education impacted both voter turnout as well as the vote margin between the two parties.

It should be noted that there are limitations with this type of analysis in terms of precision.  While demographics are highly predictive of voting behavior, no demographic group is monolithic: college educated white voters in West Virginia almost certainly voted differently than college educated white voters in Manhattan, and the regression analysis misses this distinction.  By treating these groups as monolithic, it’s likely that the regression analysis is underestimating Clinton’s support among Republican-leaning groups in Democratic-leaning areas, and underestimating Trump’s support among Democratic-leaning groups in Republican-leaning areas.  Despite this, the general trends captured in the regression analysis should still be accurate.  An analysis of county voting patterns can also provide some information that polls cannot: election results in Starr County, Texas, for example – where 93% of the electorate is Hispanic – can provide a better understanding of the Hispanic vote than public polls, which may not be able to reach a representative sample of Hispanic voters.

2016 Voters Were Whiter and Less Educated than the 2016 Exit Poll Showed

Despite the country becoming more diverse and educated in recent years, white voters still make up a sizable majority of the electorate – 74% of all 2016 voters – and whites without a college degree are by far the largest individual group, according to a Center for American Progress analysis.  This differs dramatically from the national exit poll, which showed college educated whites making up a larger share of the electorate than non-college whites.  Frankly, college educated whites being the largest voting bloc – as the exit poll found – simply doesn’t pass the smell test: according to the Census’s Current Population Survey, nearly two thirds of non-Hispanic whites age 25 and older do not have a bachelor’s degree.  Even with higher voting rates among college educated voters, there are just not enough whites with a college degree for them to represent a larger share of the electorate than those without a degree.

Demographic Composition of the 2016 Electorate, CAP Estimates versus Exit Poll

Demographic Group	Estimated Share of 2016 Voters (CAP)	Share of 2016 Voters, National Exit Poll
Whites without College Degree	45%	34%
Whites with College Degree	29%	37%
Black	12%	12%
Latino	9%	11%
Other*	5.5%	7%

Data rounded to the nearest half percent.  Due to rounding, the sum of the groups may not add up to 100%. *To compare with exit poll, Native and Asian/Pacific Islander voters are grouped together as “Other”.

Differing pollster response rates between college graduates and non-college graduates is likely the explanation for this error in the exit poll.  Research has shown that voters with a college degree are more likely to participate in polls than voters without a college degree, and not properly adjusting for this bias would lead to college educated voters being over-represented in the exit poll and non-college educated voters being under-represented.

No Shy Trump Vote: Clinton Surged with College Educated Whites and Hispanics

A detailed analysis of county level data reveals that it’s unlikely that there was a “shy” Trump vote among non-whites and college educated whites as the exit poll showed.  College educated whites, Hispanics, and Asians/Pacific Islanders all supported Clinton by significantly larger margins than Obama, and there was no statistically significant drop-off in the Democratic margin of victory among black and Native American voters.

Change in Voter Turnout and Democratic Margin of Victory, Based on Regression of County Election Results

Demographic Group	Estimated Change in Total Votes, 2012 to 2016	Estimated Change in Democratic Margin of Victory, 2012 to 2016
White without College Degree	+2%	-31%
White with College Degree	+13%	+37%
Black	-9%	-1%*
Hispanic	+28%	+14%
Asian/Pacific Islander	+6%	+9%
Native American	-1%*	-5%*

*Not statistically significant

In particular, Clinton’s surge with white college educated voters is striking, with the regression estimating a 37 point gain in Clinton’s margin among this group compared to Obama.  This shift is also evident when looking exclusively at the most heavily educated counties: in counties where whites with a college degree made up at least 55% of the electorate, Clinton won by an average of 24 points, compared to a 13 point margin of victory for Obama, an 11 point improvement.  No matter how you measure it, Clinton performed just as well – or perhaps even better – among college educated whites compared to pre-election polls, casting serious doubt on the “Shy” Trump voter hypothesis.

Unlike other non-white voters, however, black voters did not support Clinton in larger numbers than Obama, and turnout in heavily black counties declined, which some analysts have said is the true culprit for Clinton’s loss.  Still, it’s unclear whether this decline was decisive or the fault of Clinton; according to Pew Research, while black turnout declined compared to 2012, it roughly matched that of 2004, and exceeded turnout in 1996 and 2000.  So while it’s possible – although not certain – that Clinton would have won the electoral college with black turnout at 2012 levels, that was probably not a realistic expectation with Obama no longer the Democratic candidate.

Non-College Whites Who Supported Obama in 2012 Swung The Election to Trump

More so than anything else, it was a large improvement among white voters without a college degree – an even larger shift than predicted by pre-election polls – that propelled Trump to his electoral college victory.  The regression analysis found that white voters without a college degree voted for Trump in 2016 by a margin 31 points greater than they did for Romney in 2012.  This can be seen in the counties where non-college whites are most prominent: in counties where the share of whites without a college degree was at least 2 standard deviations above average, Trump improved on Romney’s margin by a staggering 21 points – winning those counties by an average of 53 points, compared to 32 points for Romney:

Average Change in Voter Turnout and Democratic Margin of Victory by County, By Non-College White Share of Electorate

Non-College White Share of County Electorate, Standard Deviations Above/Below Average	Average Change in Voter Turnout, 2012 to 2016	Average Change in Democratic Margin of Victory, 2012 to 2016
Greater than 2	4%	-21%
1 to 2	3%	-13%
0 to 1	3%	-6%
-1 to 0	2%	-1%
Fewer than -1	2%	-1%

Interestingly, it appears that Trump did not inspire a particularly large voter turnout among his most favorable demographic; the regression analysis found just a 2% increase in total votes compared to 2012 among non-college whites, and turnout only increased by 4% in the most heavily non-college white counties.   This indicates that Trump did not bring about an unprecedented surge of new voters motivated by his racially charged rhetoric, as some have suggested.  On the contrary, Trump won largely by winning over voters who had previously supported Obama in 2012, perhaps suggesting that it would be unwise for Democrats to write off these voters as un-winnable.

Indeed, there are signs that Trump’s support among non-college whites is already wavering; in the 2018 midterm elections, GOP support declined most in areas with large numbers of whites without a college degree, and recent 2020 polls show Biden outperforming Clinton among this demographic by nearly 10 points, a decline large enough to all but extinguish Trump’s chances of winning re-election.