Paradox of the Polls

AMAC Exclusive – By Barry Casselman

It is now the home stretch of the 2022 national midterm election cycle, and speculation is all over the political map about who will win and what each major party might gain or lose at the federal level and in the various states with elections this year.

There are, of course, no precise tools to aid in this speculation, but there are inexact devices on which most speculation is based. These include historical patterns and precedents, media accounts, the dollar numbers of campaign money raised, the volume of ad buys paid for by that fundraising, subjective analyses of candidate debate performance, turnout at candidate events, quality and effectiveness of arguments — and, most of all, campaign polls.

The contemporary problem with polling, however, is its recent track record of wrongly predicting outcomes. Some of this is the result of pollster bias, bad judgment, poor procedures or even deliberate falsification. But one other major reason for this defect is that, as public opinion polling has evolved over the past nine decades, the statistical accuracy of polling voters has been compromised by a growing reluctance of voters to participate in polls — causing pollsters to make more random contacts, thus increasing the margin of error in the poll. Traditional dependence on phone contact for the polls has skewed accuracy since many voters no longer have land lines. Many pollsters also ask long lists of questions that require voters to spend too much of their time taking the poll, causing incomplete polling sessions or attracting only the most politically engaged to complete to questionnaire.

Yet polls remain the principal tool for most media pundits and publications to speculate about competitive races and how well or poorly candidates might do when all the votes are counted. Individual campaign donors and independent PACs, as well as local, state, and national parties, often depend on this speculation that is based on the polls for where to direct their political ad buying and other campaign assistance — so erroneous polls can have a serious impact on a candidacy, especially for challenges to incumbents who are usually well-funded.

A case in point is the key race for Minnesota governor, pitting incumbent Democrat (DFLer) Tim Walz against Republican physician Scott Jensen. Minnesota has been a reliably blue (liberal) state for some time, and Walz was an early favorite. But polls showed him under 50% and only a few points ahead of Jensen. Then a poll appeared in early September from a major polling group showing Walz leading 51%-33%. Labeled an “outlier” by Jensen partisans, it nevertheless depressed the challenger’s fundraising, already smaller than his incumbent opponent’s, until two new public polls were reported, one showing the race at 48% to 41%, and another showing Walz leading 46% to 43% (within the poll’s margin of error). The latter poll was by Trafalgar Group, considered one of the most accurate of all pollsters, and the race now goes into its final days rated as a toss-up.

The apparent undermeasuring of likely Republican voters in the purported outlier poll above is not, however, an isolated case. Since 2016, and even before, many establishment pollsters have underweighted Republicans in their samples. The 2016 and 2020 presidential cycles were most notable for this, as were the two gubernatorial elections (Virginia and New Jersey) in the 2021 off-year cycle.

This practice seemed to be persisting in the 2022 mid-term cycle when a number of gubernatorial and U.S. Senate polls suddenly were reported with double digit or otherwise large leads for Democratic candidates in several major races — feeding a new narrative that an earlier anticipated red (conservative) wave likely in 2022 was not occurring. After Labor Day, however, the same polls often became considerably tighter or even reversed. Pollsters such as Trafalgar Group have consistently tried to avoid undermeasuring likely Republican voters, and have consistently indicated that competitive races are indeed still competitive.

Another case in point is the Pennsylvania U.S. Senate race, where Democratic Lt. Governor John Fetterman defeated a more moderate figure for his party’s nomination while celebrity physician Mehmet Oz narrowly won the GOP nomination against an establishment Republican. Although Fetterman had suffered a heart attack and stroke during the primary campaign, and was absent from active campaigning, he ran TV ads against Oz while the physician crisscrossed the state, meeting voters and speaking to groups. Because President Biden’s energy policies had made him and his Democratic administration unpopular in the state, especially in western Pennsylvania, this race to fill the seat of retiring Republican Pat Toomey had always been considered a likely toss-up, yet polls after the primary showed Fetterman winning by double digits. But Fetterman’s campaign has now seemed to implode, and new polls now show the race to be very close.

Similar examples include the U.S. Senate races in Ohio, Wisconsin, North Carolina, Florida, Arizona, and Nevada, where many major polls showed Republicans, both incumbents and challengers, doing poorly, but which now report numbers showing these races much more competitive.

In 2010, as a challenger, and in 2016, as an incumbent, polls just before the election showed Republican Ron Johnson trailing his opponent, but he won both races. Similarly, Florida Republican Marco Rubio was reported behind in 2010 and 2016, yet he too won both times.

It is generally conceded that polls of simply adults or even registered voters are likely to be misleading. Only those carefully determined to be likely voters should be included in a good poll — and a sample of 1,000 or more is likewise necessary. Polls that arbitrarily weight their raw data by predicting the percentage turnout of a major party and independent voters risk skewing their results. The same is true of arbitrarily adjusting raw data by gender.

The bottom line is that polls are often more speculative than they purport to be. Because of procedures mentioned above, margins of error are really larger, sometimes much larger, than advertised.

Using a phrase normally applied to other human institutions, polls are something we can’t quite live with, but can’t live without.

In a key cycle such as the 2022 midterms, there are lots of expectations, but no matter what pundits and pollsters say, there remains a true sense of suspense until the ballots are counted.