Revisiting 2020 polling

 As we come up on the 2022 midterms, believing in polling could be hazardous to your health. First of all, midterm polling is really hard. Finding out who is actually going to schlepp themselves to the polls is way harder than in presidential years. Anyway, I looked at the polling data leading in to the 2020 election and the final results. I focused on the swing states. First, I looked at the polling aggregates. Second, I looked at a model if the polls were as wrong as they were in 2016. Then I looked at the actual results.

Back in 2016, the polls significantly undercounted support for Trump. Did they do the same in 2020? Yes. In every swing state, Biden underperformed his polling aggregates. In some states, they were off by a lot.

Let's take Wisconsin. Aggregates had Biden at +10, 2016 Model had him at+4, actual results were +0.6.

Nevada was even stranger, as the Aggregate had Biden +6, but the 2016 Model had him at +8, but he finished at +2.4.

In Minnesota, Pennsylvania and North Carolina, he outperformed the 2016 model. Minnesota's 2016 Model was Biden +4 and he won +7.1; Pennsylvania was +1 and he won +1.2; North Carolina's 2016 Model was -3 and he lost -1.

All of these brings me back to Margin of Error. Polling aggregates are designed to reduce the margins of error, but let's assume that the MoE is the usual 3%. Only in Minnesota, North Carolina and Georgia did Biden come within the 3% margin of error for the aggregates. However, if you use the 2016 Model, Biden was within the MoE nationally and in New Hampshire, Michigan, Arizona, Ohio, Iowa and Texas. This makes Florida, Nevada and Wisconsin the real outliers.

The Million Dollar Question as we enter another season of polling and prognosticating is why Biden (and Clinton) underperformed their polling numbers. Obviously, pollsters either couldn't reach Trump voters or accurately sample Trump voters. Given how embarrassing 2016 was for pollsters, I doubt it's a sampling error so much as they simply could not get Trump voters to pick up the phone.

The Billion Dollar Question is more important. Are those Invisible Trump Voters dedicated to Trump or Trumpism? If they really are in Cult 45 and it is "Trump as Trump" who is critical to them, they might not show up to support RINOs and non-Trumpists in the midterms. 

They did not in 2018.

Indeed, Trumpists have seem to fall into that irregular voter demographic, and hopefully that continues to be the case in 2022. Democrats have escaped the direst predictions for re-districting and the Senate map looks favorable to them, but they will have to turn out their voters and hope the die-hard Trumpists stay home.