Are Polls Showing a Win for Accused Molester Roy Moore Accurate?

A steady drumbeat of poll numbers over the last ten days had shown a comfortable four to five point lead, on average, for accused child molester Roy Moore over Democrat Doug Jones in Tuesday’s special election to replace Jeff Sessions as U.S. Senator from Alabama.

Until yesterday.

Fox News’ polling is far less partisan than their news reporting, and it dropped a bombshell into the mix showing a double digit 50%-40% lead for Doug Jones. A Monmouth University poll followed which, while varying according to turnout modelling, is generally being pegged at a 46%-46% tie. Emerson University polling, often unreliable, has Moore leading by 9%.

Our method here at CounterPunch of averaging all polls in the last ten days without adjustments and including undecided voters has more accurately called two major elections (2016’s U.S. Presidential and 2017’s UK General) than other well-known poll aggregators. That method currently shows a 2.6% lead for Moore over Jones.

Those numbers may well be on target, even perhaps underestimating how committed GOP Alabamans are to the conviction that diddling kids is less bad than snuffing out their lives in the womb.

While the numbers in my polling methodology provide a strong guide, I have never felt slavishly tied to them, as when it was clear to me that Clinton was in more trouble in Michigan than polls allowed there ahead of November 2016.

In this case, I also think the polling average has it wrong. Some pollsters have done better than others at calling special elections since November 2016. Likely voter models, in my view, that assume Democrats are more motivated to vote than Republicans in Alabama today display the same kind of insight as pollsters who were more accurate in certain contests involving, for instance, Bernie Sanders, Donald Trump, Jeremy Corbyn (UK), and the recent Virginia election.

Taking, then, the average of better recent pollsters (without getting bogged down in the details of each decision to include or exclude by pollster and while rejecting Gravis’ late change in methodology) produces a 2% lead for Jones.

It’s always dangerous to pick and choose numbers, which is why I’ve included the unmolested numbers above. But the heart of the argument comes down to voter turnout modelling.

For a more in-depth argument on one of the particular issues, I’m weighing, see Nate Silver’s article of yesterday here on especially the differences between automated calling, traditional calling, and online polling. In essence, I’m going with a particular way of reading the Survey Monkey data because it fits at larger pattern of what I take to be the better way of handling likely voter modelling in situations where one side is more enthusiastic than the other (trust what voters are saying they will do, but do make reasonable adjustments for a variety of demographics in the population and with respect to recent previous elections).

In short, I’m going with a choice among numbers that matches what makes best sense of the narrative and a hard look at differences under the hood in dealing with data.

Doug Jones to win by 2%.

[CDATA[ $('input[type="radio"]
[CDATA[ $('input[type="radio"]
[CDATA[ $('input[type="radio"]
[CDATA[ $('input[type="radio"]