Trego's Mountain Ear

The first challenge to political polls is sampling bias. Most polls sample the whole population. In Montana, most are US citizens – but in states with larger immigrant populations, there are spots where up to 18% (writing from memory) are immigrants of one sort or another. Elections are decided by voters – so a poll will come closer for predictive purposes if it samples only registered voters. On the other hand, if you want a poll result to show that people are dissatisfied with ICE and Kristi Noem, conduct your surveys in areas that have more than the normal share of immigrants. I’m reasonably sure that Donald Trump will poll better with registered voters than with the general population.

It’s not so different from the 1948 telephone polls that showed Dewey leading Truman – in 1948, Republicans were more likely to own telephones. Today, cell phones are changing the business – but I suspect Republican voters will continue to be overrepresented on land lines. I recall polls in South Dakota that showed greater support for laws limiting abortion than showed up on election day. My conclusion on that was that the people who were telephoned that mildly opposed the legislation were more likely to lie to the pollster. They can’t be sure that the voice on the phone won’t take the story back to Grandma.

Then there’s another selection bias – I got my first polling call of 2026 last night. Very noticeable south of the border accent, telling me she needed of her polling connections and wanting to talk to Michael McCurry. Some folks are challenged by the economy of vowels in Scots names – it makes it easy to spot people I don’t want to keep me on the call list. So I told her he was dead, and went on with a description of how the poor man died. I know, it makes polling unreliable – but the phone I pay for is for my convenience, not that of telemarketers or pollsters.