Question: What should a Spirited Reasoner do when it turns out his election prediction missed the mark?
Answer: He should do what all good Spirited Reasoners do: (1) admit his mistake, (2) study the reasons that led to the mistake, and (3) learn lessons that will enable him to reason better in the future.
In last week’s post, the Spirited Reasoner predicted that voters would not be required to wait days or weeks to learn the final outcome of the election because of a massive “Blue Wave” that failed to materialize. That prediction was less than accurate, and I admit it.
Why was my reasoning faulty?
There were four mini-proofs given in my post, some more accurate than others. Following is my best attempt at a self-critique:
- I opined that the massive number of early votes pointed to a sizable Biden victory, based on the likelihood that these votes were being cast by demographic groups among whom Hillary Clinton had underperformed in 2016. Why was this inference wrong? Because it now appears that voters did not vote in lockstep with my expectations about their demographic propensities. (That actually seems like a good thing, the more I think about it.) For example, Cuban-American and Venezuelan-American voters apparently turned out more heavily for Donald Trump in 2020 than they did in 2016. Also, it now appears that first-time voters were not necessarily young voters, and that first-timers who were over the age of 30 voted heavily for Trump.
- I opined that Democrats enjoyed a sizable advantage because early votes were “in the bank.” Why was this inference wrong? Because it failed to appreciate President Trump’s enormous success as an election day motivator. In other words, while Joe Biden did indeed break records in terms of early voter turnout, Donald Trump also broke records in terms of election day turnout. Given the rabid nature of President Trump’s core support, the Spirited Reasoner should have foreseen that the turnout of red voters on election day could turn out to be nearly as massive as the pre-election turnout of blue voters.
- I opined that COVID-19 was attacking the very “blue wall” midwestern states that had cost Hillary Clinton the election in 2016. I gave Democrats a minor advantage for this factor in my blog post. This prediction may have been accurate and might explain why Joe Biden was able to reclaim Wisconsin, Michigan, and Pennsylvania, albeit by the narrowest of margins.
- I observed that polls were showing Joe Biden with a percentage margin substantially in excess of Hillary Clinton’s. Had Joe Biden wound up winning all the states in which polls showed him to be leading, he would have racked up 375 electoral votes and Democrats would have gained control of the Senate. As was the case in 2016, the average of these polls was approximately four percentage points too high nationally, and as many as ten points too high in several states. While it made sense, I think, to assume that highly paid pollsters would have learned some lessons from their mistakes in 2016, it now seems that they did not. Most of these polls were farther off the mark this year than they were in 2016.
Above all, the Spirited Reasoner has learned that the margin of error reported by pollsters should be expanded by at least a factor of two. For example, when a poll states that its margin of error is only 3 percentage points in either direction, we should assume a margin of at least twice that, or +/- 6 percentage points, and maybe +/-9.
And, as we reported in our post dated October 24, 2020 (“Erroneous Margins of Error”), our suspicions should be raised whenever we see a bimodal distribution of polling results; that is, pollsters falling into two distinct camps, one set predicting a high victory margin, the other set predicting a much closer race. In the future, when we notice that phenomenon, we should not make the mistake (as I did) of simply averaging those predictions. (Statisticians warn against averaging results obtained from a bimodal distribution.) Instead, I should have practiced what I was preaching in that week’s post: I should have noted the curious nature of the bimodal distribution, admitted that I had no way of knowing which of the two polling camps was more accurate, and accepted the fact that the election was too volatile to predict.