Educated Voters are Propelling Clinton to Swing State Leads  


It is no secret that Donald Trump has few viable paths to 270 electoral votes.  Those that do exist hinge on a few swing states: Iowa, Ohio, Florida, Nevada, North Carolina, Pennsylvania, and New Hampshire.  A map with those states classified as toss-ups and all other states classified a la 2012 (except for Maine’s second congressional district, where recent polls have shown a double-digit Trump lead) leaves Trump 78 electoral votes shy of winning the presidency.  From there, it’s a game of math – the most credible and likely paths necessitate that he win Florida, Pennsylvania, and a couple of other states.  Losing Pennsylvania greatly diminishes Trump’s prospects: Without the Keystone State, he needs to sweep the remainder to earn exactly the magic number.

And so it becomes natural to ask: How is Trump doing in said swing states?  What are his prospects for winning?  Is he notably doing better in some swing states than in others?  It is the last question I attempt to answer in order to gain insight into electoral coalitions and divisions.

Ohio —0.990.850.700.990.980.980.91
Iowa0.99 —0.920.790.980.940.990.96
Florida0.700.790.87    —0.660.550.710.93
Nevada0.990.980.880.66      —0.971.000.89
National0.910.960.950.930.890.810.92        —

Table 1: Correlation index between the swing states.

To flesh out the leading question – how is Trump doing in state X compared to  state Y – I regressed each swing state results on the others as well as the national outcome, one at a time, to generate a formula used to predict Trump’s vote share in each swing state.[1]  I use Trump’s RealClearPolitics polling average as the x-variable for each equation.[2]  The results, pictured in table 2, are highlighted red or blue to denote whether the predicted Trump value is greater than Hillary Clinton’s polling average for that state.

 Ohio (18)Iowa (6)PA (20)Florida (29)Nevada (6)NC (15)NH (4)NationalResult
Clinton Avg0.4080.3870.4470.4330.4150.4220.4270.409 

Table 2: This shows Trump’s projected vote shares based on his polling average in each state shown and his national polling average.  Each row is the projected vote share for the swing states based on the row’s state header.   Simulations are then compared to Clinton’s polling average (bottom); if Trump is projected to have a higher vote share, then the cell is highlighted red.  A blue cell means Clinton is ahead.  Electoral votes are then added – Trump needs 78 to win.  Highlighted color shows the electoral winner.

Trump needs 78 electoral votes to clinch the presidency; he reaches that number in only one simulation – that with state predictions based on his current Nevada polling average.  Besides again showing the arduous task Trump faces in accumulating the needed electoral votes, this model shows whether Trump is systematically over/underperforming expectations in some states based on his polling in others.  A few examples quickly jump out.  Trump is notably outperforming expectations in Iowa and Nevada while coming up short in New Hampshire and the all-important Pennsylvania.  In both Iowa and Nevada, his current polling average (bolded and as of this writing) greatly exceeds predictions for those states based on his performance elsewhere.  The opposite holds true in Pennsylvania and New Hampshire.  There, Trump’s polling average rests well below where we would expect him to be given his standing in other states.  Why might this be and what does it tell us?

Race, a default heuristic, tells part, but not all of, the story.  Trump might be outperforming expectations in Iowa because of his strength among whites (and Iowa is around 90 percent white), but what about Nevada, with its growing population of Hispanics (who have no love for the candidate)?  Pennsylvania is around 80 percent white, but Trump shows few signs of strength there.  Furthermore, New Hampshire is another overwhelmingly white state and yet Trump is very much underperforming expectations there.  Though race is an important factor in Trump’s performances in these swing states, the discrepancy in his polling numbers can be further explained by another variable: Education.

Nevada, as noted columnist and state expert Jon Ralston noted, is not a particularly well-educated state.  In fact, according to the 2010 Census, only 21.7 percent of adults in the state had at least a bachelor’s degree.  Iowa fares somewhat better — 24.9 percent of Iowan adults fit the criteria.  By comparison, in Pennsylvania and New Hampshire, 27.1 percent and 32.8 percent, respectively, have a least a bachelor’s degree.

The numbers are even more disparate when looking at exit polls.  As education increases, so does voter turnout, so the differences noted by the census are exasperated at the polls.  In 2012, 43 percent of Iowa voters and 42 percent of those in Nevada were college graduates as opposed to 48 percent in Pennsylvania and 51 percent in New Hampshire.  While these numbers might not seem dramatic, a few percentage points means tens to hundreds of thousands of voters.

In 2012, Barack Obama won voters who graduated college by just two percentage points.  Today, according to Reuters polling, Clinton leads Trump by 20 points among college graduates, an advantage which extends to white voters.  She could be the first Democrat to win white voters in 60 years.  Strength among college educated voters is empowering Clinton even in states where she might otherwise be at a disadvantage given the state’s racial composition (combined with Trump’s strong showing among whites, especially white men).  Holding all else equal, educational attainment differences between swing states likely explains why Trump is beating expectations in some while falling short in others.

And yet it’s a deficit that could be overcome if Trump had a sophisticated ground game capable of registering and mobilizing non-college graduates inclined to support him.  Luckily for Democrats, anemic investment on the ground means that Trump will be playing catchup if he hopes to mitigate the effects of his non-appeal to non-college graduates.


[1] This exercise, of course, has a very small n.  However, because many swing states have highly correlated results, the regressions were generally statistically significant and, as will be shown, the predictions from the resultant equations almost always pass the eye test for reasonability.

[2] Throughout this example, I use four-way polling averages from RCP.

Leave a Reply

Your email address will not be published. Required fields are marked *