The conventional wisdom going into next Tuesday’s election is that Donald Trump has a built-in advantage in the Electoral College, and that if he can narrow the gap in the popular vote to 2 or 3% or maybe even 4%, he can retain the Presidency.
This expectation results partly from his 2016 victory, as Hillary Clinton had a 2.1% popular vote lead but still lost handily in the Electoral College. It also results from an analysis of state by state polls, which appear to show a closer election in the battleground states than the national polls reflect.
Seven days out from the election, according the 538 polling average, Biden leads Trump by 9.1% nationally. In the key states of Wisconsin and Pennsylvania, however, the lead is only 7.1% and 5.3%, respectively. Pennsylvania, by 538’s calculations, is the “tipping point” state. Thus, if each state’s preferences shifted uniformly toward Trump by 5.4%, Pennsylvania would move into the Republican column and give Trump the win, even though Biden would still lead the popular vote by 3.7%.
A deeper analysis of the state polls of both Pennsylvania and Wisconsin suggests that this narrative is false.
The truth is that Biden’s lead is larger in those states than the polling averages indicate.
To understand why, it’s helpful to think about the various calculations a pollster must make in computing the results of its surveys. The demographic breakdown of the 500 or 1,000 people who respond to a poll never precisely matches the demographic breakdown of the people who turn out to vote. And the demographics of the people who vote have a huge impact on the outcome of the election.
The pollster, therefore, has to weight each respondent’s answer differently, so that the total weighting matches the expected demographics of the electorate. That job is tricky because the demographics of the electorate change with each election.
Since there are no “correct weightings,” each pollster must make their own estimate. What you see, when you look carefully, is substantial variance from poll to poll in the demographic weightings they’ve applied. For example, the Trafalgar Pennsylvania poll assumes 82% of voters are white, while the Suffolk University poll has only 77% white voters. As we’ll see below, this sort of difference can significantly affect the outcome of the poll.
What’s absolutely fascinating is that, when you account for these different weightings, most of the differences between the polls disappears. While Trafalgar shows Biden ahead by 1%, and Suffolk has him ahead by 7%, within each demographic, the responses to each poll are actually quite similar. The main source of difference is the assumed makeup of the electorate.
What I mean by this is that, for example, the proportion of whites with a college degree who tell Trafalgar they prefer Trump is essentially the same as the proportion who tell Suffolk they prefer Trump. The difference between the polls is the number of such people expected to vote.
For any given poll, it can be dangerous to read too much into these subgroup preferences, because the margin of error for each subgroup is quite high. (A poll of 500 people may include only 100 age 65 and up, so drawing conclusions of how voters in that demographic are planning to vote is fraught with error.)
This is where having a lot of polls is very useful. By taking the average across a number of polls, you can obtain a reliable estimate for each major demographic subgroup.
I looked at the ten most recent polls of Pennsylvania and found that, on average, white voters prefer Trump by a margin of 51–46. The range is very narrow: the most Trump-favorable poll had him at 53–44 among white voters, while the least Trump-favorable poll had him at 49–47.
Nonwhite voters are less easy to predict, for two reasons: Black voters prefer Biden by a wider margin than other nonwhite voters, and the relatively small number of nonwhite voters makes the sampling error wider. Still, the average across these polls, a Biden advantage of 76–19, is reasonably consistent.
Using these two ratios, it’s possible to layout a variety of outcomes depending on the ultimate makeup of the electorate. Here is a chart that shows the outcome in Pennsylvania based on various possible proportions of white to nonwhite voters. The more nonwhites who vote, the better Biden does.
As the size of the white vote increases, the election tightens. If the nonwhite vote is only 14%, Trump comes within 4% of winning the state. But if the nonwhite vote is 25%, Biden cruises to a ten point victory.
For historical context, it’s helpful to look at the exit polls from Pennsylvania in 2016 and 2018. In 2016, when Trump won the state by a margin of less than 1%, the electorate was 81% white. In 2018, a more favorable year for Democrats, the electorate was 77% white. The range from 81% white to 77% white suggests a Biden lead of between 6.8% and 9.2%.
Why do the polls show a Biden lead of only 5.1%? Because they are assuming an extremely white electorate for 2020. They have baked in an electorate that is 84% white.(By the way, the overall percentage of nonwhites in Pennsylvania has grown by about 6% since 2016, so all things being equal, the electorate ought to be slightly more nonwhite than in the past).
We can see this dynamic even more powerfully when we look at party registration.
It’s no surprise that most Democrats are planning to vote for Biden and most Republicans are planning to vote for Trump. Democrats in Pennsylvania are for Biden by 91–7, and Republicans are for Trump 90–7. Independents, though, are substantially more likely to vote for Biden, by a margin of 53–37. It’s obvious that, if these numbers hold true, Trump can only win if there are a lot more Republicans who vote than Democrats.
Here is a chart that shows the outcome in Pennsylvania based on various possible proportions of Democrat to Republican to Independent voters. (Note that I’ve kept Independents constant at 20%. Because more independents prefer Biden, a higher proportion of them slightly favors him.)
Once again, let’s look to the exit polls from 2016 and 2018 to see what is a reasonable expectation for 2020. In 2016, when Trump won, Democrats outnumbered Republicans by a ratio of 42–39. In 2018, that difference grew to 42–35. (It’s also useful to know that overall, in the state of Pennsylvania, voter registration is 47% Democrat, 39% Republican, and 14% Independent.)
How can the polls show a Biden lead of only 5%? Given the party breakdowns, it can only be that they are anticipating an electorate similar to 2016, where Democrats only outnumber Republicans by 2–3%.
That may prove correct, but even if it does, all it accomplishes for Trump is to buy him a narrower loss. To win in Pennsylvania, he needs Republican voters to outnumber Democratic voters by roughly 42–37. That’s a very heavy lift, considering that, even with all the things that went right for him in 2016, there were still more Democrats than Republicans who voted in Pennsylvania.
Given the various signals suggesting strong Democratic turnout this year, a party breakdown closer to 2018’s seems more likely. That would result in a Biden victory of 9–10%.
We can see very similar dynamics at play in Wisconsin. Here Biden’s lead is 7.1%, two points closer to his national lead of 9.1%.
Wisconsin is a little bit whiter than Pennsylvania. But unlike in Pennsylvania, the white voters in Wisconsin prefer Biden, by a margin of 49–46. Nonwhite voters support Biden by a wider margin, 68–25. Here is a chart that shows the outcome in Wisconsin based on various possible proportions of white to nonwhite voters.
It’s clear from this chart that, to compute a Biden lead of just over 7%, the pollsters must be assuming an electorate that is 88–89% white. But when we look at the exit polls, we find that the 2016 electorate was 86% white, and the 2018 electorate was only 83% white. If the pollsters had used the average of those two numbers, Biden’s lead would be nearly 9%.
It’s a similar story with party registration, though with one little twist. Republicans support Trump by a margin of 90–8, roughly comparable with the breakdown in Pennsylvania. Democrats, though, are even more lopsided in their candidate’s favor: they support Biden by 96–3. That means Trump is at a disadvantage even before considering Independents, who are for Biden by a margin of 50–35.
Here is a chart that shows the outcome in Wisconsin based on various possible proportions of Democrat to Republican to Independent voters.
For Trump to make this election close, he needs A LOT more Republicans to show up than Democrats. Historically, those numbers have been quite even. In 2016, Democrats had a slight 35–34% advantage, and in 2018, both parties had 34%. Depending on the number of Independent voters, Biden looks to be ahead by 8–9% if Democratic and Republican voters show up in equal numbers.
All four charts tell a very similar story: Biden has a lead in both Wisconsin and Pennsylvania that is roughly 8–9%.
If that’s the case, why have the pollsters set their models so that he appears to be leading by less?
It does not appear to be a result of respondents’ answers regarding their likelihood of voting. Voters from both parties report high enthusiasm and unprecedented expectations for voting. Turnout seems likely to set records. In most years, polls that distinguish between registered voters and likely voters lean more Republican in the “likely voter” version.
This year, the opposite is true. The CNN/SSRS Pennsylvania poll gives Biden a 5% lead with registered voters and a 10% lead with likely voters.
I’m going to speculate here, but I think what’s happening is that the pollsters got burned in these states in 2016, underestimating Republican turnout and overestimating Democratic turnout. They have set their assumptions this year to avoid making the same mistake. In doing so, they have built their polls so that they reflect a “best case” scenario for Trump.
Because the polling misses in most other states were generally not as large, the polling models in those states may better reflect the reality of the race. If that’s the case, Trump’s electoral college advantage is indeed a mirage. A national swing toward him of 5 or 6%, driven by either a change in the sentiment of the electorate or some widespread polling error, would not be enough for him to win the election. Pennsylvania and Wisconsin would stay in the Biden column, if only by a margin of 2–3%, within a point or two of his popular vote victory.