Since the beginning of the 2020 election season, the chattering class has reminded us of the failure of 2016’s polls to accurately predict Donald Trump’s victory. But, what if 2016’s polls were not as wrong as commonly believed? What does it mean that pollsters are taking precautions to prevent a repeat? A few thoughts:
The 2016 polls were not entirely wrong.
A few days before the election, RCP’s polling average showed HRC leading nationally by 2.3% (and over 3% on election day). Clinton did indeed win the national vote by 2.1%, which was within the margin of error.
In Arizona, Florida, and Georgia, Trump’s margin against Clinton on the Friday before the election, as measured by the FiveThirtyEight “now-cast,” almost exactly matched the final result. No wave of “shy” voters materialized; no polls were suddenly “unskewed.”
Further, less than a week before the election, pollster Nate Silver suggested the popular vote/Electoral College split that resulted was a possibility.
State polls are more important than national polls.
National polls are representative of the entire electorate, but the Electoral College is comprised of state votes not national averages. To get a better forecast of a presidential election, it is important to look state-by-state.
In 2016, Trump won several swing-states by razor thin margins even though Clinton led the polls. Grace Panetta explains why the margin of error data is more helpful than “the polls were wrong.“
In Michigan, Clinton led the polls by 3.6 points on average, according to RealClearPolitics’ averages, while Trump carried the state by a margin of 0.3 points, a 3.9 percentage point discrepancy. Clinton also led the polls in Pennsylvania by 2.1 points while Trump ended up winning by a margin of just 0.7 points, a 2.8 percentage point difference. And in Florida, where polls showed Trump leading by 0.4 points on average, he ended up winning by 1.2 points, a 0.8 percentage point difference.
Some swing states might have already swung.
Several of the swing states that went for President Trump by razor-thin margins in 2016 are consistently polling for Joe Biden, outside the margin of error. These include states the president almost must win in order to be reelected. From FiveThirtyEight:
Using the current “including toss-ups” version of the RCP electoral college map, if Biden wins the three states above that are outside the polling margin of error, he would have 262 of the 270 electoral votes needed to win. Adding Arizona, Minnesota, Ohio, or Georgia to that trio would be ballgame.
Pollsters modified their polling techniques.
Panetta again:
Some of the disparities between the polling margins and actual results in states like Wisconsin came from were state-level polls not adequately weighting, or factoring in, education levels, which ended up obscuring some of Trump’s support from white, non-college-educated voters. Since then, pollsters have reevaluated many of their methods and now factor in education to a greater extent.
Is it possible the polls are off enough for Trump to pull another election surprise? To this observer it does not seem likely. It is, of course, possible he could get a last minute surge reflected in tightened national and state-by-state poll projections, showing a closed gap by Election Day. But, unless such a surge happens, a Biden victory seems more likely.
fides quaerens intellectum