Readers of my Post piece predicting Donald Trump will win and Republicans will have a good night in congressional contests may wonder how I derived the numbers underlying those calls.
Here I examine that in detail and show why getting the balance between Democrats and Republicans among voters — partisan preference — is the key unlocking the polls.
Polling’s theoretical accuracy relies on the statistics underlying the relation between a random sample and the broader population it’s drawn from. But surveyors can no longer get truly random samples because cellphones and the Internet have changed how people live.
Pollsters have reacted to this in a variety of ways, but they all rely on something called weighting the sample. That means they use different ways to get their sample — calling a mixture of cellphones and landlines, for example, or using online samples. It also means taking those raw data and assigning different values — “weights” — to each respondent based on what share of the likely electorate that person possesses.
Follow the NY Post’s live updates on the 2024 presidential race as Election Day looms
You might ask, “How do pollsters know that value?” They don’t, at least not with perfect precision. They can only estimate that from the expected share certain demographic groups have historically had within an electorate, adjusted for things like population change.
A recent article walks readers through this conundrum. Using the raw data from a large-sample national poll he conducted in early October, the author shows how selecting between competing ways to weight the data can shift the margin by as many as 8 points.
That’s huge.
There are no obvious answers to any of the questions the author raised, which is why different pollsters use different methods to weight polls. It’s also why the polls show such a wide variation in outcomes.
Two recent national polls show this dilemma.
An Atlas Intel poll shows Trump up by 2 points. It also estimates Republicans will outnumber Democrats by 3 points (R+3 in polling lingo).
A Morning Consult survey shows Harris up by 3 points. It reserved details about its weighting methodology for subscribers, but I infer it shows an electorate either even or D+1.
I can do that because both pollsters released their estimates for how partisans and independents replied. The similarities are striking.
Atlas found Democrats preferred Harris by 87 points; Morning Consult had Harris by 90 with Democrats. Morning Consult has Trump up by 86 with Republicans, while Atlas had him up by 84. Atlas had the two candidates tied with independents, while Morning Consult had Harris up by 6.
You cannot get a 5-point difference between these polls simply from these raw data. They are too close to one another for that to happen. You could get that wide variance, however, if they weighted their likely voter profiles differently, yielding a significantly different partisan breakdown.
If we look at nearly 100 years of history, it would be obvious Morning Consult is right, and Atlas is wrong. There have been more Democrats than Republicans in America in every election since 1936, though the margin shrunk dramatically after Ronald Reagan’s re-election.
But that has arguably changed during Biden’s presidency. Gallup found Republicans outnumbered Democrats by 3 points in September, the first time ever. The shift started in 2021, and the drift away from the Democrats has continued ever since.
Other pollsters have found this, too. Quinnipiac University’s final 2020 national poll had a D+6 sample, but a recent 2024 national poll had only a D+2 sample. The Wall Street Journal’s polls also show a 4-point partisan shift since 2021.
The question I had to answer, then, is what I think the partisan breakdown will be Election Day.
To do that, I took each pollster’s findings for each group and averaged them. The result is Harris led with Democrats by 89 points (94% to 5%) while Trump led with Republicans by 87 (93-6). Harris led among independents by a mere 2 points (48-46).
With this in hand, it’s a simple matter to compute the national popular vote totals under different partisan scenarios. Trump-era exit polls show independents were either 30% or 31% of the total electorate in three of the last four elections, averaging 29.5%. I rounded this total up to 30% for my calculations.
The result shows Harris needs a D+2 electorate to win the popular vote by 3 points (+3.06 to be exact). A D+1 electorate gives her only a 2.18 margin, while an even electorate gives her a 1.3-point lead.
A Republican-leaning electorate spells doom for her. An R+1 scenario gives her a scant 0.42-point lead, basically a rounding error. And an R+2 electorate gives Trump the popular-vote victory by a 0.46-point margin.
To be conservative, I went with the even-partisan-split scenario. My gut tells me the GOP-leaning electorate is likelier to be the case, but I didn’t want to rely on instinct for something so important.
This allowed me to predict the popular vote percentages. Third-party candidates and write-ins received a shade under 2% in 2020 and 2012, and more than 6% in 2016. It’s reasonable to presume they’ll receive around 2% combined this year.
If Harris wins the popular vote by about 1.3 points, that means she’ll get about 49.6% to Trump’s 48.3%, with 2.1% going to the others.
This was crucial for calling the Electoral College because each of the seven swing states has voted to the right of the nation in the Trump era. This difference ranged in 2020 between 5.79 points (North Carolina) and 1.67 (Michigan).
If you compare these margins with my predicted 1.3-point Harris win, Trump carries every one. That would give him 312 electoral votes to Harris’ 229. To be conservative, I am calling Michigan’s 15 electoral votes for Harris based on its historic pro-Democratic lean, but I could easily be wrong.
I’ve mentioned three reasons that could happen already: a D-friendly electorate, Harris does better with independents than the averages show or a dramatic reduction in the Electoral College gap. Let me explain why I reject each of these possibilities.
The partisan shift towards Republicans is evident in voter-registration data across the country over the past two years. John Couvillon, a Louisiana political strategist and pollster, tracks such numbers in the 30 states that require people to choose a party. His data show that the share of registered Democrats has dropped by 1.6 points since November 2022 while the share of Republicans has increased by 0.5 points. That’s a net shift towards Republicans of 2.1 points.
This has happened in all four of the swing states with partisan registration too.
Pennsylvania has shifted the most, moving 2.6 points in the GOP’s direction, followed by Arizona with a 2.4-point change and North Carolina’s 2.3-point movement, with Nevada showing only a 1.8-point move towards the GOP.
It’s possible these demonstrable shifts in partisan attitudes won’t manifest at the voting booth. That’s what a number of polls suggest, showing Trump leading among all registered voters but trailing among so-called likely voters.
The trouble with this analysis is it presumes pollsters can predict turnout likelihood with a high degree of certainty. The difficulty in doing this is very similar to the challenges inherent in weighting a poll. The model the pollster creates might be right — but if it’s wrong in a close election like this, it sends a very misleading signal.
The possibility Harris might run stronger among independents is very real, but one must again cherry pick polls to make the case. She wins independents by 5 or 6 points in three of the polls I consulted, but she also loses independents to Trump in three other polls. I am not confident either extreme is correct, although either could be. The average is the most reasonable place to be absent strong proof to the contrary.
Finally, Harris could run so much better in the blue-wall states than she does nationally that the sizable Electoral College gap could vanish or at least markedly decline. This might be the most probable pro-Harris lens to adopt, so let me explain how that might arise before I discuss why I don’t think it’ll happen.
The national numbers mask significant underlying changes in key demographic groups. Cook Political Report keeps a running average of poll crosstabs for whites with and without a college degree, blacks and Hispanics, groups that combine to more than 90% of the electorate.
The data are clear: Trump is gaining votes relative to 2020 with blacks and Hispanics while he’s losing them among whites with a college degree. Nationally this trade is good for Trump, as he’s gaining more among the nonwhites who will be 22% to 25% of the electorate than he’s losing among the college-educated whites who’ll be a tad more than 30%.
But these groups are not evenly distributed across the country. Blacks and Hispanics are much larger shares of the electorate in Georgia, North Carolina, Arizona and Nevada than they are in Wisconsin, Michigan and Pennsylvania. They are also much larger shares of the electorates in safely blue states like New York and California or securely red ones like Texas and Florida.
This means Trump will likely have more “wasted votes” — votes that do not affect whether he wins a state — than he previously did. That alone would shrink the gap between the national popular vote and the swing states, especially in the low-minority blue wall.
Harris’ apparent gains with college-educated whites should further shrink this gap. College-educated whites cast between 29% (Michigan) and 36% (Pennsylvania) of the vote in 2020 here, and natural population changes should slightly increase those shares this year as older, less-educated whites pass away and younger, more-educated whites replace them as voters.
The question is not whether this is happening; the question is whether it’s large enough to allow Harris to win if she wins the popular vote by fewer than 2 points. Noted political analyst Nate Silver’s model says no; it shows Trump with a 53.8% chance to win the Electoral College with a predicted 2.1-point Harris national-popular-vote margin.
Silver’s model also shows Harris’ chance of winning drops significantly with every tenth of a point her margin drops. If she wins nationally by between 1 and 2 points, he says she has about a 26% chance of winning. Win by between 2 and 3 points, and her chances rise to a slim majority.
Again, this could happen, and she could win. The safe thing to do is to throw my hands up and say the race is a jump ball. Having committed to a hard prediction, however, I can’t do that. And it’s clear the cumulative weight of the evidence on hand today suggests Trump has the better chance of prevailing than does Harris.
Henry Olsen, a political analyst and commentator, is a senior fellow at the Ethics and Public Policy Center.
