The U.S. Senate election in Alabama on Tuesday settled not only the identity of that state’s new senator but also some open-ended questions about polls.
There’s Idaho resonance in that, since polling results lately have become a topic of discussion in the Gem state as they are elsewhere.
Several writers, including me, have been contacted about the results of a recent poll on the Republican nomination contest for Idaho governor. That poll showed the three-way race fairly close, with a high level of undecideds. (The latter point was the main focus of my column.)
The poll has been criticized since by one of the campaigns. I’ll not get into the weeds on that here (and circle around as to why), but it is good reason to talk a little about political polls . . . since we’ll surely be hearing more rather than less about them, locally and beyond, in the coming year.
Polls can be highly useful political tools, a good way to tell any candidate where they are and, especially, what they need to do, mainly in the case of well-run and highly detailed polls. Those pollsters who keep you on the phone for a longer time with more questions are likely to produce polls of more value to somebody.
But there are all kinds of somebodies, and of pollsters, and differentiating between them isn’t always easy. My website has a simple one-question opt-in poll which is fun to watch, but I make no pretensions about its scientific accuracy. When I worked for the newspaper in Pocatello, it used an informal supermarket poll – people who came by the store could “vote a ballot” – and we’d report the results. For local elections, it often proved remarkably accurate, though it would have passed no tests for professional standards.
Mainly, what you see are independent and candidate polls. Independent polls often are run by news media (though much less often than they once were) and sometimes other organizations, including interest groups. Candidate polls are, as you might expect, run for the use of candidates, who sometimes figure they have reason to release them, or release part of them, to the public.
Both kinds of polls can have issues. Usually, I tend to pay more attention to independent polls for two reasons. The more obvious is that campaigns tend to release results that are beneficial to themselves, and only those. The less obvious is that some pollsters will provide feel-good results to candidates whose business they would like to have; it’s not a common-place marketing tactic and many professional pollsters are careful not to do it, but it does happen: I’ve seen it. Independent polls can be subject to their own problems. Since money often is an object, some independent polls (not all) can wind up with cost-cutting that reduces their accuracy.
And any of these polls can vary by the way they’re conducted. Do they rely on telephone contacts or opt-in online surveys? Do they account for the change to cells phones, and if so, how? These elements and many more matter a lot.
Those are some of the reasons one independent poll in the Alabama race showed one candidate ahead by about nine points in one poll, and another poll showed the other ahead by 10. Many polls also weigh their results to match demographic (gender, race and other elements) of a population. One pollster showed how, in that Alabama race, you could shift those assumptions, each time in a plausible way, and drastically shift the poll’s bottom line outcome. It took the same polling data and applied a different filter to give of the two Senate candidates a big lead, depending on which assumptions were adopted.
So what to do? Best thing to do is to balance or average out the results from a bunch of polls, and recent history shows this tends to yield closely accurate results.
The catch in Idaho, of course, is that polling is sparse. There are neither a lot of polls nor a lot of providers. That creates a problem in depending on them. It’s why I was willing to use just one poll in writing that earlier column: There’s not a lot else available. But I’d be a lot more comfortable talking about Idaho poll results if there were.