Quick Election Forensics: When Did Forecasts For 2020 White House Race Become Problematic?
Forecasters correctly predicted that Joe Biden will win, but vastly underestimated the strength of support for Donald Trump. A quick post-mortem suggests that the polls and forecasts became too bullish after August 31.
First things first: Forecasters correctly predicted Joe Biden’s victory in the 2020 White House race despite widespread skepticism of polling data after the shock election four years ago that brought Donald Trump to power.
But for the second election in a row, pollsters and forecasters in the US underestimated the true extent of electoral support for Trump. Final forecasts heading into the November 3 vote were far too bullish for Biden, further fuelling public expectations of a landslide victory for the challenger and his fellow Democrats in downstream races.
It will take weeks, if not longer, for experts to fully unpack what went wrong with US electoral polls this time and why. In this article, I’ll take a look at when the forecasts started heading in the wrong direction via a quick post-mortem of the forecasts by data analysis outfit FiveThirtyEight and weekly magazine The Economist.
1. TURNING POINT: AUG 31 2O2O
For the 2020 election, both FiveThirtyEight and The Economist built sophisticated forecasting models that combined state-level and national polling data with economic and population indicators to predict a range of outcomes.
Their final forecasts predicted a thumping victory for Biden: FiveThirtyEight gave him an 89% chance of winning the Electoral College and predicted that the Democrat would win 348 Electoral Votes (EVs). The Economist was even more bullish, predicting that Biden would win 356 EVs and assessed that he had a 96.6% chance of winning.
But their forecasts weren’t always so bullish. A look-back at their models’ output point to August 31 as the point when the projections for a Biden win began rising sharply, particularly for FiveThirtyEight.
In its August 31 forecast, FiveThirtyEight projected Biden to win with 313 EVs — the lowest figure since June 1 as well as the figure closest to what analysts now expect Biden to win eventually — 306 EVs — after the vote count is completed (a handful of states have not been called at the time of writing, including Georgia, which is headed for a recount).
But for reasons unknown, FiveThirtyEight’s projected EV count for Biden began rising noticeably from September 1 onwards, as seen in the chart above. The Economist’s projections followed a broadly similar trend, though it recorded its lowest forecast in that same period — 328 EVs — much earlier on June 4.
Was this caused by a change in their models’ parameters and assumptions? Or was it the result of systemic polling errors? This is unclear for now.
To be fair, both outlets have maintained from the outset that their forecasted EV counts were never meant to be a precise figure. But it is clear from the chart above that the projections had been moving in the wrong direction since August 31, and consistently so for the subsequent months.
This is obviously problematic as you would expect the polling data and projections to give us a more accurate picture as we got closer to the election. But looking back, the opposite was true: the gap between the forecasts and actual results grew larger as we got closer to November 3.
2. FUELLING UNREALISTIC EXPECTATIONS?
Aside from the EV count, both FiveThirtyEight and The Economist also published daily forecasts for Trump and Biden’s respective chances of winning the election. Again, both outlets have been at pains to stress to the public that a low chance of winning is not the same as having no chance of winning.
But the sharp rise in their predictions for Biden’s projected chance of winning, especially from August 31 onwards, likely fueled public expectations of a massive victory for Biden.
Between August 31 and November 3, Biden’s chance of winning the White House went up from 67.4% to 89.2%, according to FiveThirtyEight’s forecasts. The Economist, whose model has been consistently more bullish about Biden, saw its forecast for the Democrat’s winning chance rise from 84.9% to 96.6% in the same period.
Like the forecasts for the EV count, it was clear once actual vote counting began — and the depth of electoral support for Trump became evident — that projections for Biden’s winning chance had been overconfident and persistently so for the final stretch of the campaign.
Defying expectations of a collapse in support at the polls, the beleaguered White House incumbent won over 70 million votes — more than the previous record of 69.5 million held by former US President Barack Obama in the 2008 election, and improved on his 2016 performance in some states. Forecasters and experts have put up nuanced explanations (here and here) of their work, but it is hard to square their predictions of a major decline in Trump’s chance of winning with his actual performance.
Biden has won more than half of the popular vote, with over 74 million votes and counting. The White House race is not decided by the popular vote but rather by the electoral college, a unique system where the successful candidate has to stitch together a winning coalition from various states that would give him or her at least 270 EVs out of a possible total of 538.
3. LOOKING AHEAD
There’s no doubt that political polling in the US has some serious problems, from falling response rates to regional discrepancies in accuracy. But the reasons behind these problems are far from clear, to say nothing of possible fixes needed to prevent another “rubbish-in-rubbish-out” scenario.
So will the US polling industry shrivel and die as a result of this worsening credibility crisis? Far from it. Current trends will only fuel the emergence of more polls and forecasts, not less.
The 24/7 US news media machinery has an insatiable appetite for new information and data. If current polls don’t meet their needs, new ones will emerge to satisfy that market demand. Without changes in demand, the supply of new polls will only go up.
More importantly, the 2020 election points to an even more competitive and complex political landscape in the years ahead. For instance, Trump has defied political conventional wisdom and proved to be highly resilient at the polls despite fumbling the national response to Covid-19 and failing at his role as commander-in-chief.
Reliably red states like Georgia have turned “purple”, with Biden and the Democrats proving to be very competitive in races in the southern state. How best can politicians navigate this complex new environment?
Well, with more polls — and forecasts — of course.
The only safe bet for the 2024 US Presidential Election is that we would be debating — yet again — the accuracy of polling data and forecasts .
