I’ve been looking at the polls and predictions of a “blue wave” in the mid-term elections. I’ve done surveys, and looked for biases that might affect the accuracy of my results – even found one that might eventually change the approach to surveys in Indian Country. And what I keep coming back to is the idea of non-response bias.
A lot of years have passed since I noted that non-response bias as an explanation of the difference between the poll results on restricting abortion in South Dakota and the ballot results that contradicted the pollsters. A few less since my hypothesis on non-response bias on Reservations was confirmed by Mary Kills a Hundred.
It wasn’t that the pollsters (then or now) weren’t competent professionals. The problem with polling is that the non-responders aren’t responding. That seems like a statement that deserves “Well, Duh.” as an answer. The thing is, we’re accustomed to non-responders cancelling out – and, somewhere on one side of the bell curve, the non-responders tend to vote in opposition to the people who answer polls.
In the most recent election, the bias within the model seemed to run about 3 percent more in favor of the Dems than the elections did – but I don’t know if the algorithm has changed in the last couple of years. If I were running the polls, at the least I would be adjusting my results – but I’m an old retiree, and the new kids are playing the game now.
This evening, I got a pollster call – offering that if I would hit one on my phone, they wouldn’t call again. On one hand, that sounds like cooperating with non-response bias. On the other hand, if it isn’t really non-response, if there is a consistent group that gives consistent misinformation, getting rid of some respondents is probably a good idea.
My inclination is that there are folks out there that vote for Trump, folks that vote MAGA, who aren’t strong supporters – they’re just people who regard Trump and MAGA as the lesser of two evils . . . and they don’t want to talk about it. I doubt if the polling models really have it all that well figured. We do know that in the last election there were people who voted for Trump, who voted MAGA, but didn’t want to tell pollsters about it. And three, three and a half percent of that sort of non-responders cuts down on the reliability of survey da
Leave a Reply