Monday, August 24, 2020

2016 Polls Were So Wrong 2020 Polls Can’t Be Trusted and Must Be Ignored

Actually, polling data provides valuable insight into the feelings, attitudes, and leanings of voters. Too often, however, political candidates, pundits and strategists ignore the “devil in the details” in cross tab polling data and go their own way, then blame the polls when things go wrong.

The common meme in 2020 is that polling can’t be trusted because 2016 polls were so wrong. Corollaries to that meme are that polls were wrong in 2016 and are wrong in 2020 because Trump people lie to pollsters about supporting Trump, and Trump voters simply won’t talk to pollsters. But these oft-repeated arguments ignore the fact that the national and state polls in 2018 accurately reflected the building blue wave for Democrats winning control of the House. Further, 2016 polls missed Hillary Clinton’s national popular-vote margin by only about one percentage point.

Clinton’s national popular-vote edge was 2.1 percentage points (equal to about 2.9 million votes) and Hillary won the national popular ballot contest by 2.86 million votes. There couldn’t have been too many Trump voters who lied or wouldn’t respond to pollsters in 2016 for them to be that close.

In 2016, 45 states and the District of Columbia exactly followed polling in those states. Florida and North Carolina were considered toss-ups, along with Maine's and Nebraska’s 2nd Congressional Districts. The top line polling numbers in Michigan, Pennsylvania, and Wisconsin, where Clinton lost by 77K votes, showed Clinton ahead in each, because pollsters did not properly weight the sample of voters without a college degree when calculating their top line “horse race” numbers. Trump's advantage in those key battleground states wasn't noticed by mainstream media pundits who only look at those top line numbers until late in the race, if at all. (Polling companies now correctly weight the sample of voters without a college degree when calculating the top line “who is winning” headline numbers.)