2024 US Presidential Election

[sc94597]

Cerebralbore101 said:

sc94597 said:

https://www.vox.com/policy-and-politics/2022/11/27/23475262/midterm-elections-2022-results-red-wave-democrats

If you were looking at polling averages that included Republican polls, “you were looking at a completely different election than we were looking at,” he added.

When Rosenberg stripped out the partisan polling, he foresaw an election in which New Hampshire, Arizona, Georgia, and Pennsylvania were leaning Democrat, Nevada was too close to call, and Ohio, North Carolina, and Wisconsin were leaning a little Republican. That’s consistent with what actually transpired.

 

As to the second sentence, there is no binary of "being wrong" and "being right" here. The point of polling isn't to get it exactly right at the individual poll level, but to create an aggregate picture that better reflects the population group that we want to estimate. If we cut-off the tails due to the individual poll being potentially biased (but without showing it), which is what pollsters seem to be doing currently (because of the rating system), we can possibly lose part of the picture and can have a worse aggregate because of it. This is called "herding" where the pollsters are refusing to publish results that show a large difference from their priors. It helps their individual rating and error rate, but hurts our over-all predictability. 

https://aapor.org/wp-content/uploads/2022/12/Herding-508.pdf

What do you mean cut off the tails? This sounds to me like pollsters are more worried about being right than just reporting the poll numbers as is. But being more worried about being right and then adjusting for it would just make you wrong anyway. Either report the poll numbers as is, or find out how you are missing so many different types of voters that your poll is useless.

An assumption of inferential parametric statistics is that the distribution of sample means is roughly normally distributed, with their mean approximating the population mean, becoming more normal and a closer approximation of the population mean as the number of samples increase. See: Central limit Theorem When I say "cutoff the tails" I mean that they're not including polls that would be at the tail-ends of said normal distribution of sample means. 

Again, there is no "right" or "wrong" as a binary here. "Rightness" and "wrongness" are continuous values, defined by how closely the mean of the aggregation of samples matches the population mean (and therefore the result of an election.) 

It's impossible to get a poll that perfectly represents the population distribution, for many reasons, but mainly because that population is a moving target and you don't know what the population mean is. That is why statisticians do the next best thing, form a distribution of samples that approximate the population distribution's value at their mean. This is the basis of the field of statistical inference.

Weighing the sample is a useful technique when you know where the bias is, so I don't think they should abandon it. It does make polling more accurate so long as there is a good enough threshold of what "knowing where the bias is" means. Of course, that isn't always sound, which is why we see things like herding. 
.