Sunday, November 03, 2024

I Do Not Think That Word Means What You Think It Means

I am not an expert on this stuff (though it isn't complicated), but it's pretty clear that in 2024 polling is much more modeling than polling. Is the point of a poll to make inferences based on the data collected, or is the point of a poll to make election predictions? "A bit of both" has long been the answer, but now the data is so bad that it's much more the latter than the former in the way these things are presented.

Response rates are so shitty (and I suspect most of the right wing polls are just made up for a variety of grifter reasons), that the polls are generally "unskewed," so to speak, based on some belief about the underlying population and electorate.

The results are presented as weighted, and none of the "margin of errors" reflect the margin of error of the weights, which are just guesstimates like the rest of it.

This is not a new practice, but the noise is overwhelming the data.

I'm not unskewing the polls one way or another.  I'm not going to waste my beautiful mind having that argument.  I'm just saying what is happening isn't really polling. I am not faulting the main pollsters or arguing they are being dishonest or anything like that, I'm just saying that I think it's a mistake to take too much from them.

Scroll down to the bottom here and you can see how it's done.