Too close to call? Why US pollsters might have some explaining to do

Presidential candidates Kamala Harris and Donald Trump.

Polls are an important part of our democracy. When some accuse politicians of being “poll-driven” they usually mean it as a criticism. But try to imagine another circumstance in which politicians didn’t care what ordinary voters think. Done properly, polls should and mostly do express the democratic will of ordinary people.

Admittedly, compared to a surgeon or an airline pilot, the consequences of a pollster getting it right or wrong are hardly life and death.

Pollsters might grumble that when we get it right nine times out of 10 that should just be expected. But if we get it wrong once, the whole industry is called into question. The fact people do care shows why polls matter.

The Australian’s Newspoll has an outstanding record by global standards. Two misses in three decades is pretty good going.

Some of our competitors haven’t been so fortunate, like the one that tipped “prime ministers” Kim Beazley and Mark Latham two elections in a row. Or the other one that confidently said Tasmania would vote Yes at last year’s Indigenous voice referendum.

There’s a reason politicians, corporates and ordinary punters pay more attention to Newspoll than any other poll.

Kim Beazley

Mark Latham

Spare a thought for our American equivalents, though. After errors outside the margins in the swing states in 2016 and then again in 2020, it’s understandable that many would be a bit gun shy. The result is an amazing amount of congruence in swing state polls that show the states that delivered Joe Biden the White House in 2020 are improbably all “too close to call” this time around.

As a pollster I’d love to be able to peek under the hood of the US pollsters’ swing-state polls, to see what’s really going on. I’ve learned not to speak authoritatively on polls I’m not responsible for. It’s certainly possible the race is in fact really close in all these states, but it seems “off” that none of them would be clearly leaning to either Donald Trump or Kamala Harris this time around.

To be honest, the close resemblance between polls this time around and the results from the last election have caused me to raise my eyebrows more than a bit.

One explanation for this behaviour is called hedging. Basically, if you put out a poll that leans heavily one way or the other, it’s like putting your head up in the trenches – and you’ve got a much greater chance of having it blown off. If everything is either 49, 50 or 51 it gives you much greater cover regardless of the result.

This is a different phenomenon from “herding” which is where some pollsters allegedly make their poll results resemble those from other, better-credentialed polling companies.

If every single swing state is “too close to call” one might legitimately ask, what’s the point? ­People commission polls to get an answer not to have pollsters hedge towards the middle.

I don’t know how this is happening, and I’m not suggesting anyone is acting improperly.

There’s a simple explanation that is legitimate and could produce this effect. Many US pollsters weight their polls by past vote or voter ID. This is not wrong, but you want to be cautious about doing it dogmatically. The reason is people don’t always recall accurately what they did four hours ago, let alone four years ago.

Mark McGowan speaks to the media in Perth.

One US pollster who certainly isn’t hedging is Ann Selzer, who put out a poll that showed Harris winning Iowa. That’s a real challenge to the rest of the industry. If Harris is in fact winning Iowa, this race isn’t close.

A similar situation arose for Newspoll at the 2021 West Australian election. Our poll showed a record landslide of unheard of proportions for Mark McGowan, while other polls and the collective wisdom of the “experts” and bookies suggested the race was much closer. Since we have confidence in our data we released it and it turned out to be right.

I don’t know whether the Iowa poll is right or wrong. However, it is unwise to conclude that because the polls understated Trump’s vote in 2016 and 2020 it necessarily means they’ll do the same again. Perhaps the efforts they’ve gone to fix that problem have overcorrected, causing errors in the other direction. One thing that is clear, however, is that if Selzer’s poll is right, all this hedging is going to be really transparent once all the votes are counted.

That’s a critical point. We’ll only be able to tell whether the polls were right or wrong after all the votes are counted. That’s likely to take days or weeks if it is in fact really close.

The pattern of results is likely to be that the early figures from small rural counties will look very good for Trump. Harris is likely to perform better when big urban vote centres are counted.

But if the result is known by the time we go to bed in Australia, then whichever way it went, the US polling industry will have some explaining to do.

Campbell White is the founder of Pyxis Polling & Insights and has administered Newspoll since 2019. He has more than 25 years experience in political polling at all levels of government in Australia, New Zealand and Southeast Asia, including both public and internal polling. He holds a PhD in social psychology and is a former chair of the Australian Polling Council.