State polling problematic — again

To this day people still think that the polls had it wrong about Trump in 2016. That’s not really the case. Trump lost the popular vote to Hillary Clinton 48.2 percent to 46.1 percent — well within the margin of error for most national polls (Rassmussen was within one point). The problem was two-fold. First, the media and the Trump-hating pundits forgot that polling margin of error is real. Second, the same crowd forgot that the President is elected state-by-state via the Electoral College — and much of the state-level polling in the key states of Michigan, Pennsylvania and Wisconsin was off the mark.

The current election looks like it could be more of the same, with Trump running unaccountably further behind in the state polls against his national polling. This is cold comfort for Team Trump as he remains behind Biden by an average of 8.3 points in the 538 average and 7.4 points in the RealClearPolitics average. Using the more generous RealClearPolitics average of Biden leading 49.4 percent to 42 percent, Trump is down 4.1 points from 2016 and Biden is up 1.2 points over Clinton. (Note that any incumbent President should be expected to trail his prior performance as there is always a segment of undecided voters that exists in polling but disappears once people have to make a choice in the voting booth.)

Until the Constitution is re-written, predicting the eventual president requires guessing which candidate wins which states — and that means accurate state-level polling matters, which is a significant problem. The national pollsters are only sporadically polling the states, and local polling groups are of uneven quality. Partisan consultants and think tanks are also polling, but their results are suspiciously flattering for their own party.

In a best-case scenario, highly-regarded pollsters would regularly poll battleground states using the same methods and large sample sizes. But the expense and a lack of coordination (or unwillingness to do so) has resulted in the current hodgepodge of polls and results.

Currently, state-level polls present a more dire picture for Trump than the national polls.

Trump’s polling numbers are at least 5 points lower than his 2016 totals in every large swing state. After deleting polls from partisans, Trump is down nearly 7 points in Florida (49.02 percent in 2016 to 42.2 percent 2020 recent polling average) and nearly 6 points in Michigan. His drop-off in Pennsylvania, Wisconsin, and Ohio ranges from 5.28 percent to 5.59 percent.

As bad as the swing state numbers are, the reported numbers from states Trump won are even worse. Trump is down over 13 points in Arkansas, over 10 points in Oklahoma and nearly 10 points in Tennessee. Each of those states feature just one poll — which means the results should be treated cautiously. However, Missouri and Kentucky have seen multiple polls, and Trump lags 7.77 and 6.42 points on average — numbers which would put seemingly safe states for Trump in play, if the polls are accurate.

In all, 538 has Trump vs. Biden polls from 37 states with Trump’s numbers falling by more than 4.6 points in 22 states. Two big states, California and Massachusetts show a deterioration of just a few percentage points. In New York Trump is polling down 4.52 percent.

The upshot of this cavalcade of polling numbers lends some credence to Trump’s complaints that the polls undercount his true strength — but only some credence. In the case of the national polls, there is some legitimate concerns for bias, but any bias may only be a percentage point or two. The frequency of polling — and thus the experienced gained — appears to have helped the national polls address the problem of skepticism and lower willingness to participate by conservatives and Trump supporters.

State-level polls may still need to catch up and address the problems identified by the American Association of Public Opinion Research in their post-mortem report from 2016, which included improper weighting of less educated voters, lack of accounting for an unwillingness to admit a respondent was voting for Trump and generally not modeling turnout correctly.

While it’s unlikely that any poll will be as disastrous as the Michigan State poll that had Clinton up 19 points on election eve (and Libertarian Gary Johnson getting 11 percent!), making accurate Electoral College predictions will depend on the herd of polling groups improving their methodology and not indulging a dislike for the incumbent.

Keith Naughton, Ph.D., co-founder of Silent Majority Strategies, is a public affairs consultant who specialized in Pennsylvania judicial elections. Follow him on Twitter @KNaughton711.