Hi,
With all the polls in, as everybody else, I come to the conclusion that the polls cannot predict a winner in this election This means that that the only way for polls to be wrong would be that the difference between Kamala Harris and Donald Trump is substantial to the point that many polls would be outside their margin of error. This is unlikely because there is not much difference between pollsters.
And even then, such a difference could be due to people changing their minds -- changing preferences, deciding to go vote although they were not supposed to, etc. In order to conclude that this happened, it would be necessary to examine whether we can show that last-minute shifts explain the discrepancy between the polls and the vote. This means pollsters would have to recontact the respondents of their last poll to ask them whether they voted, for whom and when they took their decision.
The following graph shows the forecast of the vote. It shows a perfect equality. Although I have seen such equality in other elections in Canada, I checked preceding US presidential elections and I do not see such equality. All of the polls of the last week vary between 48% and 52% for one or the other candidate. However, voting intention for Harris is often higher than for Trump.
In addition, there is still a tendency for web op-in polls to estimate support for Harris higher than the other polls but the difference is quite small -- about half a percentage point --, as we can see in the following graph. So there does not seem to be much difference left between pollsters that could be explained by modes of administration and sampling sources. It may be -- and it is expected -- that pollsters improved their methods since 2020 and 2022.
However, it is interesting to notice that, in the last week, the proportion of polls conducted using web opt-in is lower than in previous weeks. There are eight web op-in polls during the last week compared to 9 mixed mode and quasi-random polls like live interviewer polls and Web probabilistic polls. Eleven of these polls put Trump ahead of Harris or at par with her. Five of them were conducted using web opt-in and five using quasi-random single mode polls.
In conclusion
Right now, there are no reasons to believe that polls will be wrong or that some polls will be wrong because all the polls say about the same thing. Some attribute this situation to herding (pollsters aligning their estimates to other pollsters) but it is way more likely that it is explained by the weighting procedures used, particularly those that use voter recall or party id.
Personnally, I am quite surprised that, in such a situation, the sample sizes remained rather small. I guess polls that are conducted in the swing states are more important for the pollsters and for the media if they want to predict the final winner.
