Polling performance for the 2020 election between Joe Biden and Donald Trump. Relative error is a metric for evaluating the accuracy of poll predictions relative to actual voting outcomes. Relative error is calculated as (observed−expected)/observed, where the "observed" value represents actual election results, and the "expected" value denotes the predicted results from polls. This statistic helps to identify the proportion by which polls under or overestimate support for a candidate. In this figure, a positive relative error indicates that the polls underestimated the candidate's actual support, while a negative value suggests an overestimation.
I admit I am confused that polling data for Trump underestimated his support while still losing. Unless, of course, the polling results at the time showed him losing by a wide margin. That isn't my recollection, either.
At best, It seems to reinforce my impression that polling today is complete garbage.
Your recollection is faulty. Polls showed him losing by an extremely wide margin. He over-performed polling by an average of 8%, while Biden under-performed his polling by an average of around 4%. Yes, he lost, but it wasn't the blow-out that polling suggested it would be. It ended up being a narrow victory for Biden. This result is consistent with the differential that was observed in the 2016 election, where again, the polling was predicting a blow-out Trump loss, and in that election, he won.
Thanks for the explanation. I guess my memory AND polling isn't very reliable.