Like most everyone else, I started yesterday confident of significant Democratic victories. The polls showed a consistent national margin for the Democratic ticket of 7%. I went to bed last night sick at heart as everything seemed to be falling apart, just as it did in 2016.
Things look better this morning, and I now believe we will be saying “ President-elect Biden” by the end of the week.
But...the national numbers aren’t even close to the final polling aggregate. In state after state, the polls appear to have under-estimated tRump votes. Instead of a 7% national margin, it’s looking more like 2%. The final outcome looks better this time around than in 2016, but it’s the same phenomenon we saw then: the polls were consistently wrong. Yes, I know, they were in the margin of error, but for essentially all of the polls to under-state the GOPer vote in the same direction suggests a more fundamental problem at work.
First, I don’t believe tRump voters “lie” to the polls. They are bullies and filled with boastful bravado. There’s no evidence to support they might do this, and it doesn’t fit their mindset.
As I reflected on this, I recalled that I hung up on at least a dozen phone “polls” in the last month where the framing of the questions told me it was a push poll—a poll where a built-in bias in the questions meant it was designed to produce a particular result. I did respond to polls where the questions had a neutral framing.
Now think about the tRump cult. In the first place, they hate and reject science. This makes them less likely to respond to a scientific poll. However, a *scientific* poll will have questions with neutral framing. My experience with tRump supporters is that they find that very neutrality offensive. What if they respond to scientific polls—ones with neutrally framed questions—the way I respond to push polls? If that describes a significant part of the cult, that could account for what appear to be the systematic errors we’ve seen in the polling.
This creates a problem for legitimate polling organizations. If this is right, what we’ve got is a fairly large group of voters—maybe as much as 5% of the electorate—which refuses to participate in scientific polls precisely because they reject the very neutrality of the questions. This would be a group which self-selects out of scientific polls. Because they self-select out, it’s impossible to predict, based on a sample, the size of this group because they never get into any sample.
However, when we have an actual election, these people vote. Biden’s 7% national lead has morphed to a 2% victory margin. So, it looks to me like as much 5% of the electorate self-selects out of scientific polls. If this part of the electorate is distributed across demographic groups, tweaking the sample to adjust for education or other factors won’t help.
My conclusion, unless someone can suggest something else, is to add some number, probably 5%, to GOPer numbers in all polls in the future before drawing any conclusions.
I believe in evidence. I’ve taught statistics for years at the University level. I know that when polls are consistently off in the same direction, something is amiss. The most likely problems are non-representative samples and inadvertent question bias. My hypothesis is that cult members self-select out of unbiased, scientific polls resulting in non-representative samples..
Meantime, I’m holding my breath and hoping that Democrats succeeds at the real polls, the ones where people vote.