The polls were way off again? Not so fast

As I write this on November 6, 2020, many people are already arguing that the presidential election polls for the election three days ago — which is still undecided at the time of writing — were systematically wrong yet again. The crux of this argument is that polls made it look like Clinton would beat Trump in the Electoral College in 2016, which did not happen, and that polls also made it look like Biden would win comfortably, which has not happened (though it remains entirely possible for Biden to win the Electoral College and is, in fact, more likely that he will win it than Trump; as I write this, Biden has 253 electoral votes and Trump has 214).

Here I argue that the conclusion that polls are worthless, or even that presidential election polls this year systematically overestimated how well Biden would do, is premature at best. I am making this argument for the following reasons:

1. Many votes have not yet been counted both in called and uncalled states, rendering it premature to compare the results in those states to what the polls had predicted. Yet some “presidential polls can’t be trusted” pundits (e.g. the hosts of Pod Save America’s November 4 episode) have irresponsibly compared the results of states like Florida, Michigan, and Wisconsin based only on the votes already counted at that time, ignoring the fact that many uncounted votes remained and still remain at the time of writing. This is likely to be particularly true in Ohio, where, though Trump has indeed won, per the New York Times, “Postmarked ballots have until Nov. 13 to arrive, but no additional results will be reported reported for at least a week and a half after the election.” In addition, the top 4 counties with the largest percent of their estimated total votes still unreported are all Biden counties (Lucas, Cuyahoga, Montgomery, and Hamilton), making it even likelier that Trump’s lead in Ohio, while it will not go away completely, will shrink significantly when all votes are counted (which, again, will not happen for roughly a week from when I write this). A similar point applies to the national popular vote, which Biden was forecast to win by about 8 points. While he is currently ahead of Trump in the national popular vote by only about 3 points, given the numerous votes left to count in Democratic strongholds. Most notably, this is the case in California, where, by my back of the envelope calculations, over 3 million votes remain to be counted, and roughly twice as many of those should be for Biden as for Trump, judging by the results of the counted ballots in that state. But Biden can also expect to grow his popular vote lead significantly from many of the uncounted ballots in such blue states as New York, Maryland, Illinois, and New Jersey — each of which still has over 10% of the vote uncounted as I write this (in fact, Maryland still has 30% of its vote uncounted).
2. There is a fundamental difference between “Biden has a 90% chance of winning” and “There is a 90% chance that Biden will win within a few hours of the last polls closing” or “There is a 90% chance that calling enough states for Biden to win the Electoral College will happen in a rapid and stress-free manner”. Personally, I don’t like probabilistic forecasts of elections in general, given recent research indicating that they mislead and confuse potential voters, but it is important not to misinterpret their meaning.
3. We still don’t know who won the election. If we were at a point where Trump had won the Electoral College despite having only a 10% chance of doing so in the FiveThirtyEight forecast, then it would make relatively more sense to criticize the forecast and the polls on which it is based. However, it clearly makes much more sense to be patient and see if the thing there was (supposedly) a 90% chance of happening has actually happened before passing judgment on polls, forecasts and the like.
4. Related to point 2, there is a difference between an election being close — i.e. depending on, say, a few hundred votes in Florida — and it taking a long time to decide. This election was obviously always going to take much longer to decide than normal, so the fact that we have had to wait for a day or two longer than normal to be sure should not be surprising, nor should it be misinterpreted as an unexpected “polling error”. In particular, political observers knew before Election Day that some states would take longer to decide than others: as FiveThirtyEight’s Nathaniel Rakich and Elena Mejia wrote on October 31, “We should know the winner in Wisconsin by Wednesday morning; Michigan and Pennsylvania, by contrast, will probably take until the end of the week.”
5. Not only were mail ballots expected to overwhelmingly favor Biden, they were expected to be counted relatively late in key swing states like Pennsylvania. Likewise, mail-in ballots can’t be counted until election day in Pennsylvania, Michigan, or Wisconsin, but they could be counted weeks before then in Florida and North Carolina. Not to mention that 20 states will still accept mail ballots some amount of time after election day, so there’s every reason to expect the vote counts to change after the polls close in these states. This is why NBC News warned before the election that “it will be extremely hazardous to read into the results before individual counties report that they have completed counting.” Therefore, given that many counties are still significantly short of completing counting, it is clearly premature to jump to such conclusions as “the polls were wrong” or “they can’t be trusted ever again” or “the polling industry is a scam!”