Election polls questionable; could mislead voters

 

Gordon L. Weil

Note:  My column this week will focus on the reasons to mistrust the polls, because this is a different election, one that may escape normal analysis. 

 

The New York Times/Sienna poll, considered one of the best, was just published, and it shows a 48%-48% tie between Kamela Harris and Donald Trump.

It touts the closeness of the race, but also reveals at least a couple of factors that can raise serious doubts about its reliability. And, like other polls, it fails to mention one.

To arrive at a sample of 2,516 responses, the number needed to bring the margin of error down to 2.2% when the sample is used to forecast the entire universe of voters, the pollsters made 260,000 calls to 80,000 voters.  It’s likely that many chose not to participate.

Unlike the earlier days of surveying, when it was possible to contact a random sample of voters, the net has to be cast much wider and the results have to be subjected to an analysis according to the pollster’s concepts.  Each pollster may use a different questionnaire and a different set of screening standards, but that is lost when polling averages are created.

These are the adjusting factors used by the Times/Sienna poll:

Sex, age, educational level, home ownership, marital status, party, race, region, voting history and intent to vote.

The net result is supposed to present an accurate profile of the American voting population, while recognizing that a sample of this size will have a margin of error around each of the major results.  That means that candidate A could be off by 2.2% and, at the same time, candidate B could be off by 2.2%.  At the extreme, the survey could really be showing a 4.4% gap between the candidates, i.e., it could vary significantly from the message publicized.   A difference of this magnitude between the forecast and the result has happened.

The reported results for each candidate leave 4% unaccounted for.  Some of that would go to other candidates, but maybe this includes some truly undecided voters.  Even one or two percent going one way or the other could make a big difference.

The unstated factor?  The forecast margin of error is only assured 19 out of 20 times.  The operation of the statistical method cannot do better.  This may not matter, but the average person is not told about this factor.

None of these observations suggest the Times/Siena poll produced a biased result between Harris and Trump.  It does suggest that this is a close race, because of the manner in which the data is managed and reflecting the decisions of the pollsters.  My column later this week  will question that assumption. And it also suggests that voters are invited to pay too much attention to polls.

 

That said, here is a personal observation on what people may tell pollsters.

In the Times/Sienna survey, a Wisconsin woman is quoted as saying that her concern about illegal immigration is what has determined her to vote for Trump.  This is a statement about how a single issue decided her.

Suppose you asked her how she could overlook some of Trump’s promises to take extreme action on matters ranging from high tariffs to arresting his political opponents.  Perhaps she would say that he should not be taken seriously, but that such statements are merely the way he talks.  Nobody really expects he would or could do such things.

Such thinking ignores two points.

First, the Supreme Court has given the president a blank check for the exercise of their powers.  The only real checks would be a Congress that could override their veto or impeachment and conviction.

Second, the person most likely to believe Trump is Trump himself.  While voters might not take his extreme promises seriously, if he is elected, he could well say, “I told them what I would do, and they gave me a mandate.”  It takes only one person to take him seriously – himself – and any thinking that he would not go to the promised extremes would not matter.