Is exit-polling more accurate than pre-election polling?

Question

It has been claimed so in a comment which cites an FT article that I can't read (right now). Presumably the topic is interesting enough to have been discussed elsewhere.

I'm mainly interested in the US and the UK.

Some of the theoretical advantages and disadvantages in the US case are outlined in an ABC paper:

Each of these sources has different advantages and disadvantages. Exit polls are conducted with actual voters (including a small national sample of absentee voters contacted by RDD telephone survey), and the national exit poll is based on a very large sample size (12,219 in 2004). At the same time, exit polls are based on cluster samples and therefore have larger design effects than telephone surveys. Good-quality pre-election polls conducted via RDD are based on modeled rather than actual voters, have smaller sample sizes, and almost no design effect.

That's for the theory; the real/tough question is: does one advantage/disadvantage outweigh the other in practice?

Answer 1

There's at least one counterexample to the claim that exit-polls are more accurate. From the exact same ABC paper:

The final vote estimate released by ABC News, based on data collected October 29-31, was within two points on the Bush percentage and exactly matched the Kerry percentage of actual vote. [...]
In overall vote, the ABC [pre-election] data were more accurate. The NEP national exit poll overestimated Kerry’s vote by three percentage points and underestimated the Bush vote by the same amount, before the data were weighted to the final election result (Edison Media Research and Mitofsky International, 2004). Among subgroups, there’s close correspondence between the final vote-weighted exit poll and ABC’s final pre-election poll (which is not weighted to actual vote).

But this of course could just be a statistical aberration (an outlier), the proverbial exception that confirms the rule; it could well be that on average exit polls are more accurate. If anyone has data to prove that, please post your better answer.

I also found a stats page on UK which says

Prior to 2001 the methods used had been different, and the accuracy of predictions based on exit polls was rather variable — sometimes good, sometimes poor. Perhaps the most famous failure was in 1992, when the forecasts made by both BBC and ITV based on separate exit polls were that the Conservatives would only just get more seats than Labour, resulting in a predicted hung parliament. It turned out that the Conservatives had actually done much better than that in the 1992 election, winning 65 more seats than Labour and achieving a parliamentary majority of 21 seats.

And it doesn't look like simply a matter of gerrymandered seats; the page only give data for 2005 percentage-wise, but it does look impressive

So exit polling can be very good as well; it looks like it all hinges on methodology. Judging by last US presidential election and its exit polls, it doesn't look like this UK methodological innovation has made over the Atlantic just yet. Also note that this latter (UK) page does not compare exit-polling with any pre-election polling.

A case where the pre-election polls got it all wrong but the exit poll was right was in the 2015 UK election:

The unveiling of the results of the exit poll at 10pm on 7th May 2015 has already become part of election television folklore in the UK. Throughout the election campaign the opinion polls had suggested that the Conservatives and Labour were neck and neck with each other. However, the exit poll forecast that the Conservatives would win 316 seats, while Labour would win just 239. If the exit poll was right, the opinion polls would be seen to have called the election ‘wrong’.

By 6am the following morning, it was clear that the polls had indeed overestimated Labour and underestimated Conservative support. On average the final estimates of the polling companies put the Conservatives on 34% and Labour on 34%. No individual poll put David Cameron’s party more than a point ahead. Yet in the event the Conservatives won 38% of the vote in Great Britain, Labour 31%.

This probably explains the FT article. And since people may be curious why the 2015 pre-election polls in the UK were (so) wrong, the same source conclued:

Our conclusion is that the primary cause of the polling miss in 2015 was unrepresentative samples. The methods the pollsters used to collect samples of voters systematically over-represented Labour supporters and under-represented Conservative supporters. The statistical adjustment procedures applied to the raw data did not mitigate this basic problem to any notable degree. The other putative causes can have made, at most, only a small contribution to the total error.

Answer 2

Yes. Exit polling is considered to be about as accurate as possible for election polls, under traditional circumstances. Whether that applies to the 2020 election cycle is certainly up for debate, give the large first-time shift towards greater mail-in voting. But under the more traditional circumstances:

#1 - pre-election polls are about how people are feeling, at that time. Minds can change, moods can swing. Exit polling is a tally of what people actually did. Unless they are intentionally lying, it is going to align perfectly with the actual results because it is an accounting of what actually happened.

#2 - Polling experts say they can pretty reliability account for slight skews for sampling issues, or even the often-cited difficulties of getting a hold of younger people who don't have land lines. What they state is their biggest challenge is determining the likelihood that respondents are going to vote. Who is an "eligible" voter vs who is a "likely" voter vs who is an almost certain voter. Exit polling removes that uncertainty. They know they can count the person as a valid part of the sample because they know they already voted. They also don't have to worry about whether anyone (for the most part) shouldn't be counted, because who goes to the polls to not vote, and then answers questions about the vote they didn't cast?

By it's very nature, it makes perfect sense that exit polling would be an extremely accurate sampling of voters, much moreso than pre-election polling, and the results bear that out.

That's why when one is looking for signs of potential errors or election fraud, a large deviation from election polling is a telltale sign.