# What is the difference between a late and exit poll?

When glancing at the Polish presidential elections last Sunday, I saw that the results after the closure of the voting offices (21:00) were tagged as being "late poll" results, with an uncertainty of ~2%.

Then at midnight, a new set of results, called "exit polls", were provided, with an uncertainty of ~1%.

Both seem to come from polls done during the day (and not after 21:00) so I do not understand why they would yield different results.

This is certainly not a matter of logistics: results with information on the person asked (for the representativeness of the sample) can be sent real time to a centralized system, and even if they needed say an hour to gather them, the voters between 20:00 and 21:00 are not likely to introduce such a biais.

How are they different?

• I think you mean 'poll', not 'pool'. If so, maybe edit your question. // A poll with about $2500$ subjects has a margin of error of about $\pm 2\%$ if the proportion being estimated is not too far from $50\%.$ Exit polls are regarded as more accurate because you know the subject being interviewed (a) really did just finish voting (b) for the person stated with no time lag for change of mind. So it's like getting info directly from the population. Also, using historical records from that particular polling place helps to give early warning of unexpected shifts of opinion from prior elections. – BruceET Jul 15 '20 at 1:59
• @BruceET: thanks for the typo, I was influenced by the tag name. Now, I infer from your comment that the "late polls" are not done on the day of the elections? This was not the explanation I saw (where both seemed to happen on the same day). If this is just a poll of the populations shortly before (and possibly during) the elections then I understand where the difference can come from. The fact that the exit polls are not used right after the voting ends (the dramatized apparition of the winner) is something I do not understand, though. – WoJ Jul 15 '20 at 7:18
• In the US, TV networks often have a good idea from exit polls early on election day who will win, But by agreement none of this info is released until all voting is done in a state. This is to avoid premature broadcasting of info that might effect voter turnout or choices. – BruceET Jul 15 '20 at 8:12

## 1 Answer

Exit poll is a survey research, where you sample voting stations, and then pollsters ask random voters who did they vote. Late poll collects actual voting results from random voting stations and uses them to estimate the election result. As you can learn from this interview (in Polish) with Paweł Predko, from Ipsos research company that did the polls, in practice they seem to be using some kind of statistical model that combines the samples from exit polls and actual results from sample of voting stations and this is what they call the late poll.

It is hard to comment about the uncertainties without having access to the actual methodologies they used (I failed to find it online), but there are some obvious differences between the methodologies, so they have different pros and cons. In exit polls, you know the demographics of the people who took part in the survey, but on another hand, you only know what they said about their vote. With collecting results of random voting stations, you know the actual results from those stations, but it is just a sample of all the stations. Moreover, there's likely (and actual, in those elections) clustering in those results, so for example, people voting in a voting station in major city would be similar to each other, but the station as a whole would be different from a voting station located on a countryside. So if you "just" summed up the votes from the stations, this might not give adequate estimate of the election result.

• Late poll collects actual voting results ..." → ahhh, these are ACTUAL results, not any kind of questions asked to voters. This is clear now, thank you. – WoJ Jul 15 '20 at 8:41