There is no such thing as a poll of likely voters

[Starting a sometimes meta thread on the current state of polling.]

There are polls of the general population and polls of registered voters, but there has never been a poll of likely voters. What we do have do have is polls of registered voters that have been weighted to favor the respondents whom the pollsters think are likely to turn out on election day.

The polls and the models used to determine how likely different people are to vote are entirely different creatures, supported by different assumption, prone to fail in entirely different ways for entirely different reasons. If we are going to pay attention to electoral polling (and I have very mixed feeling about whether we should), we need to be aware of these things.

If you have a candidate like Trump or an issue like Dobbs bringing in large numbers of people who otherwise probably wouldn't have voted, inevitably you will screw over likely voter models. This is doubly true if the model puts heavy weight on past voting history. Dobbs is particularly interesting in this respect since it seems to be causing a surge not just in registrations, but in registrations of young people. If a likely voter model looked at voting history and age these are the last people it would flag as likely.

Of course, outside of a handful of special elections which don't really tell us much, we don't know how many of those young people who registered will actually vote. We won't know until November and from the standpoint of prediction, that will be too late.

Here's my take. At this point, I would not put any weight whatsoever on likely voter models. Not this year and not in 2024. Every warning light relevant to predictive models is flashing proceed with caution and we are so far out of the range of observed data that it is no longer visible on the horizon.