Big swings in polls reflect response bias, not true change

Since nobody has remembered anything about polling from the last election, it's worth emphasizing that there is no such thing as a wild swing in polling support.

A recent Monmouth poll showed Biden dropping from his usual support of just over 30% to 19%. That had the effect of putting the frontrunner in 3rd place, behind Bernie and Warren, each of whom had 20% support.

Harris' average on Real Clear Politics was steady at 7%, then more than doubled to over 15% after the first debate where she slammed Biden over his busing record, only to tumble back to 7% after the second debate where she was the one getting slammed by Tulsi over her severely punitive record as a prosecutor.

Neither of those two changes is real, although they may reflect a subtle change in the same direction. Biden's continuing cognitive meltdown may be making some supporters less supportive -- but only by a small amount, a few points perhaps.

I reviewed this stuff during the final stage of the 2016 election, which really turned out to be important after the pussy tape made everyone think Trump was done for. He did take a little hit, but not the plunge that most polls suggested.

The only reason that wild swings show up in polls is that some people become more likely to participate, or less likely, so that you don't have a snapshot of the same overall group of people before and after some major event or series of events. This is response bias -- how willing various demographic or partisan groups are to participate in the poll in the first place.

This can only be corrected by tracking the same panel of people over time. By recruiting them for a long-haul series of polls, you're sure not to miss some group of them who might go into hiding when their preferred candidate gets womped in a debate, or to overcount some group who is eager to participate because their candidate did the womping and it got them all hyped up.

Usually these major events are debates, as in Harris' case. In reality, support for her has been constant at 7% -- the apparent pump and then dump are back-to-back illusions, due only to her two debate performances hyping up her supporters and then demoralizing them into hiding.

But in Biden's case, it could be a week or two of heavy media coverage of his brain going haywire. That will demoralize his supporters, who will be less willing to take part in the poll. They don't want to have to say, "Yeah, I support that guy whose brain is melting right before our eyes." But that doesn't mean they don't still support him, or won't wind up voting for him in the primary. Could these malfunctions cost him a couple points? Sure, but not double digits.

There's more detail in my old post, though if you want the technical analysis, read the source article by Gelman et al (2016), "The Myth of the Swing Voter".