7

Leading up to the 2019 election, the polls suggested a Labor win, but they lost. What was the cause of the difference?

I realize it may take some time before a good answer is available.

  • 4
    538 notes that, on average, polls just before a US presidential election are out by 2 percentage points. Since the polls in australia narrowed to 52-48 before the election, this could be seen as a "normal polling error" – James K May 19 at 7:37
  • 1
    Usually such gaps are caused by a different voter turnout than predicted. Polls are quite often wrong and have error margins, sometimes quite big ones. – Trilarion May 21 at 8:00
4

FiveThirtyEight.com took a swing at an explanation (at the end of the article):

Polls showed the conservative-led coalition trailing the Australian Labor Party approximately 51-49 in the two-party preferred vote. Instead, the conservatives won 51-49. That’s a relatively small miss: The conservatives trailed by 2 points in the polls, and instead they won by 2, making for a 4-point error. The miss was right in line with the average error from past Australian elections, which has averaged about 5 points. Given that track record, the conservatives had somewhere around a 1 in 3 chance of winning.

So the polls were actually off by slightly less than average.

Dig in deeper, and you can find things to criticize in the polls. In particular, they showed signs of herding: all the polls showed almost exactly the same result in a way that’s statistically implausible. If Labor was ahead by only 2 points, a few polls should have shown conservatives winning just by chance alone because of sampling error.

Herding suggests that pollsters were sitting on some of the results. So if everyone else was showing Labor ahead but this particular poll showed the Liberals ahead, the pollster assumed that there was polling error and did not release the poll result. Which then causes it to look like everyone is saying that Labor is ahead. So the next pollster who sees the Liberals ahead...

If pollsters only show polls that confirm their preexisting beliefs, that makes the situation ripe for a polling error. If the missing polls had been included, the Labor lead would have looked even smaller and people would have been more prepared for the result. Heck, for all we know, the real polling result with all polls included may have had the Liberals ahead.

In general, it is easy to think of a poll that has Labor winning by one vote when the Liberals actually won by one vote as wrong. But that's not how polls work. They don't know anything about who is winning. They are guessing a number. Their guess will tend to be within three or four percentage points of the actual result, but they are not more accurate than that. So when you see a poll that says Labor is up by two percentage points, you should realize that about a third of the time, the Liberals will actually be winning. That's what a two point lead means.

There was a close election. Polls are simply bad at predicting those. Because the margin of error in the poll is simply larger than the result it is finding. A 51-49 polling result should not suggest that the side with 51 is a sure thing. It should tell you that anything could happen.

While there does seem to have been some bad behavior by pollsters, there was truly wretched behavior by journalists. They should have noted the herding problem before the election. And they should not have been so confident that a 51-49 lead in polls would also appear in the results. Even without the herding bias, a Liberal win should have been seen as possible given the polling results.

2

Sampling error. The error margin for published opinion polling in Australia is about 2.5%, and was out by a lot more - in Queensland, as a notable example, it was out by 8%.

It's known that the sampling error was continued right up until the close of polls, because the exits polls matched the opinion polls in predicting a Labor win:

Voter turnout normally isn't a factor; Australia has compulsory voting, with formal voting rates in the low 90s such as at the previous Australian Federal election in 2016. The data currently published, which suggests the 2019 turnout is down over 9% points; this is because postal and declaration (or pre-poll) vote counting hasn't finished yet, but if it stays that low, that's your answer - 10% of the population who have voted election on election failed to turn up to this one particular election (an unlikely story; the AEC Twitter feed kept talking about the 400,000+ votes being recorded every day in pre-polling).

A representative of one of the polling companies said it was sampling error, noting they get an over-representative level of Green support (which skews their two-party preferred calculations) and engaged voters. She points out, however, that they got the primary vote of Labor dead right. Primary votes rarely win elections, so preference flows are vital, and the opinion poll results have shown that predicting preference flows is problematic.

The Australian experience of opinion poll failure is a reflection of what's happening elsewhere. To improve sampling, the tele-research companies want access to the electoral roll. I don't think that's the answer; taking the pulse of the nation needs to move into the 21st century and use election prediction techniques that are shown to work:

In the five days before election night, Professor Bela Stantic analysed 2 million social media comments, from more than half a million unique accounts, relating to 50 key terms, and predicted that Scott Morrison would win.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.