I'm going out on a limb here, since we'll know pretty soon whether the presidential election polling has been reasonably accurate. But I will say that if it is correct, and we have a very close election, that will be more luck than science.
When I was getting my M.A. in environmental policy at Tufts back in the 1980s, I took a course in survey research. We even did our own poll of one of the state elections, I forget which one now. Anyway, that wasn't really so long ago, even though there are people who have been old enough to vote for 20 years now who weren't even born then. So in case you didn't know, or haven't really thought about it lately, in 1984 we weren't ruled by Big Brother, but we also didn't have cell phones. Telephones were plugged into the wall. Also, the area code and first three digits of your phone number specified the location of the wall into which your telephone was plugged -- either a town, or a specific neighborhood within a larger city.
Voicemail did not exist. Answering machines -- tape recorders -- existed but were quite uncommon. There was no called ID either. This means that when the phone rang, people answered it. There was very little telemarketing. There wasn't the technology to do it robotically, so you had to pay people and it wasn't much worth it. People weren't perpetually annoyed by spam calls and they weren't predisposed to blow off pollsters. So we could draw a good quality probability sample of households, stratified by location, and expect people to answer the phone and talk to us. That meant we could confidently assume that the data we started with was representative of the population we wanted to describe.
Those days are gone. Phone numbers don't correspond to locations, most people don't answer the phone when they don't recognize the caller's number. They just assume it's spam and they let it go to voicemail. Obviously, the proclivity to do this is not randomly distributed in the population. The result is that pollsters have no confidence that the people they manage to speak to constitute a representative sample. So they are forced to weight their sample by whatever variables they think will make it more like the actual population of likely voters. Those can only be variables they ask people about, since they have no other way of knowing anything about them, which means that questions like income and education may not get honest answers. They also don't know if their sample is representative in terms of the likelihood of voting, and answers to that question may also be wrong, either because people don't want to answer it honestly or genuinely don't know themselves that well. So they have to try to use respondent characteristics to predict that as well.
So nowadays, pollsters are factoring all sorts of assumptions into their predictions, and by tweaking any or many of them just a bit they could get a very different result. If their result looks a lot different from other reputable pollsters, that will make them nervous and worry that their assumptions are off, so they're likely to recast their predictions to look more like the consensus. That we started out with a near tie and the polls barely moved from that since Harris became the Democratic nominee suggests that what we are seeing is a herd effect. I don't have any way of directly unskewing the polls, but I do believe based on other information that the outcome of this election will not actually be very close. Then maybe the corporate media will decide they need to come up with a different way of covering elections in the future.
No comments:
Post a Comment