Visiting Column #49 -- The Underpolling Problem, 2020 Style

I remember 2016 as if it were only four years ago even though it was ... well, yeah, it actually was four years ago.  But Lordy, have things changed.

In the fall of 2016, my best girl and I got a fair number of polling calls asking about our preference in the upcoming presidential election between Donald Trump and Hillary Clinton.  They came through on our cell phones and they came through on the home phone as well.

They were annoying as heck, but we answered them and dutifully indicated our intent to vote (hint: it wasn't going to be for Hillary).  We watched as the poll numbers continually suggested that the former first lady was going to get her wish and get elected.  And we watched on election night with amazement as the polls showed themselves to be completely wrong -- so wrong, in fact, that Larry Sabato, the respected Virginia political science professor frequently on TV commenting and forecasting results, went out the next day to say that he apparently didn't know politics and would have to start all over.

But that was 2016.  This is 2020, and my best girl and I have taken a completely different tack this year.

It is pretty obvious from the caller ID when you are getting a polling call, and we get two or three every day.  Where in 2016 we would have done what we thought was our civic duty and answered the call accurately and honestly, this year we simply do not answer.  We do not pick up the phone, and we do not share our intentions.

If we happen to answer the phone, because the caller wasn't obvious, we do not answer polls, or we do not answer them accurately.  I don't know who the caller is.  People have literally been killed for supporting the current president, and I'm not sure when the Antifa types start doing fake polls.  What a time, am I right?

I didn't consult with anyone on this practice of no longer answering poll calls accurately, mind you; we just do it.  We decided, and that was it.

So as I contemplate the impact of that, I have to ask the obvious question -- "Who else is doing the same?".  And you should be too, and so should Larry Sabato, and Gallup and Pew and Rasmussen and all the other polling companies out there.

Because polling is simply a data collection procedure, it is evident that if the data is inaccurate, the results will be as well -- "garbage in, garbage out", or "GIGO" as we used to say back when I was learning to program in the latter 1960s.  My old college fraternity brother, Fred Faltin, had a similar take 50 years ago, creating the "Faltin Fudge Factor" -- an amount you added or subtracted from your test results to end up with the outcome you actually wanted.  Ahhhh, MIT.

After 2016, we already know that the data is not completely accurate.  I believe that for the most part there was a 5-10% bias in the polling data toward Hillary Clinton; the actual voting went for President Trump, about that percentage higher than where the polling had the numbers a day earlier.  We know that.  You could look it up, as Casey Stengel would have said.

But this is 2020, and that factor, at least 5%, representing polling undercount, exists today at least as much, if our own feelings about responding to polls are in any way representative.  I've seen some broadcasts discussing the topic, and while they still refer to the usual 3% "margin for error", they're also mentioning, although not quantifying, the polling undercount of Trump voters.

And that undercount is at least another 3% on top of the margin for error, based on 2016.  At least.  I hesitate to extrapolate our own situation too far, but we are two Trump voters who were willing to tell polling callers that in 2016, but are either telling them nothing, or lying and saying that we'll vote for the other guy in 2020.

Do the math.  If you poll 100 people and their actual intent is split 50-50 Trump and Biden, the difference is zero points.  If even one couple like my wife and I tell the caller that we're voting for Uncle Joe, then the reporting is 52-48, and that is a four-point spread, above the margin for error!

I hope you get the idea.  There is the accurate data (i.e., whom people are actually going to vote for), and then there is the 2016 factor, the at-least 5% undercount of Trump voters based on how people felt.  Add to that the 2020 factor, where some Trump voters feel threatened or otherwise choose not to answer the calls (accurately) -- and we don't know what that percentage is.  I respectfully decline to guess.

We have not had debates yet, and Biden has essentially not answered a challenging press corps yet to explain all the questions in his policy and opinions on current events.  There is a lot that can change in the last two months of the campaign.  So the "accurate data" part I mentioned in the previous paragraph still has room to swing a bit.

But there is definitely that third group of voters out there, and the polling companies must be wracking their brains trying to figure out how to apply the "new liar" factor, among the 2020 voters, to their predictions.  That's hard because they don't actually "predict", they have to report. If they count 51-46 in one direction, what else can they report?  But if the election comes and their reported 51-46 ends up 53-47 the other way, they look like idiots!

The Democrats have long since started their effort to stack ballots through their mail-out campaign, so what, we have to ask, are they planning for?  Are they thinking that they'll need ten million phony mail-in votes to ensure Biden wins?  How do they even know how to prepare when Gallup can't provide reliable data?

Oh, it will be such fun.

Copyright 2020 by Robert Sutton
Like what you read here?  There are over 1,000 posts from Bob at www.uberthoughtsUSA.com, and after four years of writing a new one daily, he still posts thoughts once in a while as "visiting columns", no longer the "prolific essayist" he was through 2018, but still around.  Appearance, advertising, sponsorship and interview inquiries cheerfully welcomed at bsutton@alum.mit.edu or on Twitter at @rmosutton