2 weeks Ago
With just one week to go until Election Day, we asked back BRIAN GAINES , the Honorable W. Russell Arrington Professor in State Politics for the University of Illinois system, for a breakdown of the race for the White House between VP Kamala Harris and former President Donald Trump . In the final days of the 2024 presidential election campaign, poll results are ubiquitous.
As of today, many show a tie in key swing states, perhaps with Trump momentum. Below are a few tips on how seriously to take those polls. Pennsylvania is a priority for both campaigns.
Squint at a figure showing poll averages and you will note multiple lead changes. Over the last two weeks, Trump has opened up a small edge. Before you panic or celebrate, notice that the margin has been within 1 percentage point since early August.
TIP 1: Beware of over-interpretation. The safest thing to say about Pennsylvania is that it looks tied. We are highly confident, but not certain, that actual vote intentions are in the vicinity of the poll estimates, and reported margins of error are usually too small, not too big, because they are based on optimistic assumptions about how weights fix biases and errors.
Do not take too seriously leads under 3 percent. Pennsylvania predictions were not great in 2016, but in 2020 they were pretty good — the averages matched the actual outcome almost perfectly. Is it clear which firms do well, in which states? Can a shrewd consumer pick out the believable wheat from the misleading chaff? Firms vary in past accuracy, but year to year, there is not a consistent frontrunner.
And some pollsters that can boast of having nailed, say, Georgia, have to admit that they botched Arizona. The logic of accuracy is far from obvious. Regard anyone’s model that converts raw polling data into accurate forecasts with much suspicion.
The national polls show ...
TIP 3: If you are preoccupied with predicting the presidential result, ignore the national polls. Some argue that Trump support has been under-estimated in all of his runs, so a seemingly tied Trump-Harris poll is probably a small Trump lead. TIP 4: The premise is about right, but the conclusion is too crude.
Trump’s support is probably being underestimated in some states, but past errors have not been uniform. And the polling industry has not stood still — they have tinkered with weighting schemes and how to identify likely voters, and technology and real-world events keep altering what are the biggest challenges of polling in any case. Yesterday’s errors need not show up today, or tomorrow in identical form.
In states with lots of polling on another statewide race — say, for a U.S. Senate seat — comparison can be useful to try to separate sampling effects (who is willing to be polled just now) from shifts and swings.
TIP 5: Don’t forget Tip 1. Even with many polls and multiple races, and even with substantial early voting, meaning that some of the responses are not merely intentions, but reports of actions already taken, polls are noisy measures. Two Republican candidates seeming to gain at the same time can reflect independents moving in both contests, a sample that was a bit too red, or accidentally correlated random error.
There is no free lunch, and polls and poll averages, however constructed, are structured guesses, not guarantees..
