Why it's still too soon to write off the U.S. election polls

For the sake of the pollsters — and American democracy — let the counting finish.

Roasting the pollsters is a popular pastime after each election. Sometimes it's richly deserved. Sometimes the ire is misplaced. The current situation in the U.S. might fall somewhere in between.

But, much like the election itself, we don't quite know where things will land at this point — because millions of votes still have to be counted.

It was obvious as soon as election night began that perceptions were going to be a problem. Americans (as well as voters in most democracies) are used to having a result on election night. The media is hardwired to try to figure out what that result will be as quickly as possible.

When it was clear that Joe Biden was not going to win Florida, it followed that U.S. networks would be unlikely to project a winner before the sun rose the next day. Then the early counts put Donald Trump ahead in Wisconsin, Michigan and Pennsylvania. Texas was out of reach for Biden, as was Ohio. Taking the lead in North Carolina meant Trump's electoral map looked a lot like the one that gave him the White House in 2016.

You might remember that election. It was the one that everyone said the pollsters got horribly wrong.

Except that isn't exactly what happened — the polls called a few states badly and Trump beat his polling at the national level by a modest degree. At the state level, the polls performed slightly worse than they normally do in an average U.S. election. At the national level, they were slightly better.

In the end, that's probably what happened this time as well.

It wasn't supposed to happen again, of course. The pollsters had changed their methodologies to account for the missing Trump voters from 2016. It seems that those tweaks didn't do the job.

There were also many fewer undecided voters this time than there were in 2016. These voters swung to Trump four years ago, accounting for some of the polling error. Exit polls (which are not entirely reliable until they are weighted against the actual results) so far suggest that the undecideds might have broken for Trump again in 2020.

Had there been more of them, perhaps that would have been enough to make the difference for Trump. But Biden's lead in the polls was predicted to be big enough to withstand a 2016-sized error. If he secures the presidency, that prediction will turn out to be true.

Count the votes, elect a president, then judge the polls

And while there's a lot of talk now about a polling industry in "crisis", we actually don't know for certain how the polls performed in the U.S. election. That's because a lot of votes still haven't been counted — and as we have seen in Wisconsin, Michigan, Pennsylvania and Georgia (to name just a few), counting all the ballots makes a big difference.

At the national level, the margin between Biden and Trump was 2.4 percentage points at about 1:30 p.m. on Thursday — wider than the final margin between Hillary Clinton and Trump in 2016 (which wasn't actually finalized until a month after election day).

An average of over 96 per cent of the estimated vote has been counted in states that were deemed "Safe Republican" by the CBC's Presidential Poll Tracker. In "Safe Democratic" states, only about 89 per cent of ballots have been counted.

Meanwhile, big Democratic states like California and New York have seen lower percentages of their vote totals counted so far. Because of this, Biden's lead will only grow as more votes are counted, with estimates putting the final margin of victory at between four and six points.

That margin is quite a bit smaller than what the polls were suggesting. The Presidential Poll Tracker put the margin at 8.6 points. That means the national error could end up being somewhere between three and five points — roughly in line with the average error in the polls since 1972.

That's not a disastrous performance — but had Biden's margin over Trump been as narrow as Clinton's was in 2016, that could have cost him an anticipated victory.

What's up with Wisconsin, again?

The problem in 2016, though, was in a handful of states that made the difference in the electoral college. It appears the polls had some problems in a few states this time — some of them the same states that bedevilled pollsters in 2016.

The Midwest seems to be the biggest challenge for pollsters. While Biden has won Michigan by just under three percentage points, the polls were suggesting he had a lead of about 7.5 points. Wisconsin looked very safe for Biden, with an average lead of 8.4 points. Biden eked out a win there by only 0.7 points.

The polls appear to have been off by similarly large margins in Iowa and Ohio, which were supposed to be tighter Trump wins than they have turned out to be. But, as is the case elsewhere, a lot of votes still have to be counted in those two states.

It could be that many Trump supporters just aren't being polled. This isn't a case of "shy Trump" voters — there's no evidence of significant numbers of Trump supporters who don't want to admit that support to pollsters.

But it may be that some of the Americans most likely to vote for Trump — non-college graduates and those who distrust the media and other institutions — are unlikely to participate in a poll to begin with. How to get around this problem remains a puzzle.

Florida was another miss — perhaps the most consequential one. It's the only state projected by the Presidential Poll Tracker as a Biden win that has gone instead to Trump.

The polls gave Biden a lead of just over three points in Florida. Instead, Trump won it by just over three points. That's a big miss pollsters will have to figure out. Early evidence suggests that Trump's performance among Cuban Americans might have been underestimated.

But the polls did quite a bit better in other swing states. North Carolina, Arizona and Georgia were toss-ups that likely will be decided by a point or two. By the time counting is done in Pennsylvania, it's possible that Biden's projected five-point win could turn out to have been overestimated by only two or three points. The same goes for Trump in Texas, where he had a narrow lead in the polls.

Narrative of failure set early

There is no doubt that pollsters will be trying to figure out how to improve their methodologies over the next four years — and that they will continue to fall short of perfection. But some things specific to the 2020 election were outside of their control.

The number of mail ballots is one of them. Someone answering a poll and saying they voted by mail expects that their vote will be delivered (not all were) and that their vote will not be rejected. But mail ballots (cast disproportionately by Democrats) are generally rejected at a higher rate than ballots cast in person (cast disproportionately by Republicans). There are indications that mail ballots cast by Black voters — who overwhelmingly vote for the Democrats — are rejected at an even higher rate.

Many critiques of pollsters' performances argue that the polls were pointing toward a Biden landslide. While some polls did indicate that was a possibility, others suggested Biden's national lead concealed a closer race that would be decided in a few swing states. In other words, there's a bit of selective memory at work among those focusing on one scenario the polls suggested could happen (but didn't) while ignoring the one the polls also suggested could happen — and actually came to pass.

Because of the electoral college, it was always going to be a "fine line between a landslide and a nail-biter." If Biden is able to win Nevada and Arizona, he'll squeak by with 270 electoral college votes. If he also wins Georgia and Pennsylvania by close margins, which seems possible, he'll have 306 — similar to where the polls suggested he had a good chance of ending up.

Had things gone a little better for Biden, Florida, North Carolina, Ohio and maybe Texas would have inflated his count to over 350. That's landslide territory — and the polling consistently indicated it was a possibility.

Our understanding of this election was shaped in part by the polls. That understanding wasn't incorrect. Biden was favoured by a margin wide enough to become a big win — but not wide enough that a very close result couldn't be ruled out.

If the first scenario had happened, there would be no discussion of a "crisis" in polling. It might not be fair, but that's the way it is. Polling, like politics, is a tough business.