In a world of perfect polling – and there are spots where polling sample correlation with reality approaches perfection – there would be no suspense or surprises on election day. As I look at today’s polls, I recall an election where the polls showed South Dakota voters supported a very restrictive law on abortion – but that proposed law failed miserably at the polls. Since I worked with populations and polls, I heard the question “Why?”
Simply enough, a poll samples a population. If we ask about eye color, our poll responses will reflect reality within the population we have sampled – plus or minus our statistical margin of error. So what happens to make political polling less reliable?
Political polls don’t sample a single cohesive population. If I were to oversimplify the challenge, I would say they sample two somewhat different populations that are intermingled. Both the response rates and the answers differ by ideology. The South Dakota polls that showed great support for limiting abortion access were telephone polls – and a significant number of respondents were untruthful. I joked that it was because they couldn’t be sure that the pollster wouldn’t tell Aunt Mary their answers – but it was joking on the square. We do have people that will answer a question one way at Thanksgiving dinner and the opposite in the privacy of the voting booth.
Usually, I hang up on pollsters. Caller ID has made that even easier. Over a lifetime, I’ve learned that telephone pollsters take my time and provide me no benefit. When you read the poll numbers, they don’t include the folks who didn’t answer the phone – whatever the reason. My general assumption is that the folks who respond to pollsters are somewhat more likely to support the democrat ideology. Folks who reject both ideologies, or support the republican ideology are a bit more likely to let the phone ring until the answering machine comes on, and then ignore the message.
Political polls sample at least 3 intermingled populations – potential democrat voter, potential republican voters, and folks who despise both groups of candidates. Think back 6 years – it was Donald Trump v. Hillary Clinton. It took ideological purity to wholeheartedly support either . . . kind of like the Disraeli quote: “Damn your principles! Stick to your party.” The not-so-ideologically pure said, “I guess I’ll hold my nose and vote for * .”
We have plenty of differences between the two dominant parties – age, education, income, employment, race all come to mind – yet when the polling results come out, they are presented as if a single population answered the question. If I were tasked with developing an accurate polling method, I’d be using an intensely selected focus group, and getting the answers would cost more than anyone is willing to pay. So I have to accept that inaccurate polls are the best I can get.
Still, there is precision and there is accuracy. If a rifle puts all the bullets into the same spot, even if I can’t adjust the sights to the bull, I can use Kentucky windage – I sight in on the spot that will put my bullets into the bull. Same thing with political polls. Most are precise, yielding similar results with each new poll. The models are fairly good. The spot we have troubles is in how we adjust that model to infer reality.
In most cases, this year it doesn’t make a whole lot of difference. The entire House of Representatives, all 435 of them, are up for election. (Here in Montana, the chance of getting a democrat representative is somewhere between slim and none) The US Senate is a bit different: 34 seats are up in 2022 – and by my count, only 14 seats are held by democrats. This is the spot where we can look at polling results. RealClearPolitics shows:
As you look at these numbers, the highest democrat senator (Murray) is +8.7% (and projected to win) while the highest republican is Rubio at +4.7% (and projected to win). Vance at +1.2 is the lowest republican projected to win, while Warnock, at +3.8 is the lowest dem projected to win.
As a guesstimate, RealClearPolitics regards a +3.5 polling lead for a democrat candidate as somewhere approaching a tie. Every pollster creates a model, but the model samples several distinctly different populations with the same questions.
Polling results, of course, can’t lead us astray in Lincoln County, where the bastards have taken choice away from the voters who participate in the general election this November.