Biden’ lead remains at 4 points in May expert survey

The Pollyvote has completed its second survey of elections experts to forecast the 2020 presidential election. In the survey, conducted from May 15-18, twelve political science professors from a variety of U.S. universities predicted that Joe Biden will win the popular vote by roughly four points.

The mean forecast is that Biden will garner 52.2% of the popular two-party vote, compared to 47.8% for President Trump. Eleven experts expect Biden to win the popular vote, while one expert predicts a Trump victory.

This forecast is nearly identical to last month, when the combined expert forecast was that Biden would receive 52.3% of the two-party vote.

PollyVote thanks the experts who participated in this round, namely

  1. Alexa Bankert (University of Georgia)
  2. Andrea Benjamin (University of Oklahoma)
  3. John Coleman (University of Minnesota)
  4. Scott Blinder (University of Massachusetts Amherst)
  5. George Edwards (Texas A&M University)
  6. Sarah Fulton (Texas A&M University)
  7. Sandy Maisel (Colby College)
  8. David Redlawsk (University of Delaware)
  9. Larry Sabato (University of Virginia)
  10. Mary Stegmaier (University of Missouri)
  11. Michael Tesler (University of California, Irvine)
  12. Charles Walcott (Virginia Tech)

Fair model forecast turns in Biden’s favor due to economic downturn

Ray Fair did not release a forecast update from his presidential vote equation in late April due to the uncertainties regarding the GDP figures for the second and third quarter, which are inputs to his model. Fair wrote on his website:

I did not make an economic forecast with my US model this time because the model has nothing to say about the effects of pandemics. I could try to subjectively constant adjust the estimated equations, but this would only be guessing. So I am going to let the user decide what values of G and P to use.

Ray Fair, April 29, 2020

PollyVote thus decided to use the estimates from the second quarter survey of professional forecasters conducted by the Federal Reserve Bank of Philadelphia. In this survey, published today, the consensus of 42 forecasters was that the Q2 economy will drop at an annual rate of 32.2%, and rebound in Q3 (+10.6%).

Entering these values in Fair’s tool to compute vote predictions yields a forecast of 54.2% of the two-party vote for the Democratic candidate Joe Biden, and 45.8% for President Trump. This is a dramatic change from the previous forecast from January 30, which predicted more or less the opposite result (54.4% Trump vs. 45.6% Biden).

As can be seen in the following chart, this change is also noticeable in the combined PollyVote, which registered an almost 1-point drop in Trump’s predicted two-party vote share.

First PollyVote forecast predicts a tight race

The first PollyVote forecast for the 2020 U.S. presidential election, published in late April, predicts a tight race for the national popular vote.

As of April 26, the combined PollyVote predicts incumbent President Donald Trump to gain 49.1% of the popular two-party vote, compared to 50.9% for former Vice President Joe Biden, with an 80%-uncertainty interval of +/-2.7 percentage points.

Based on these numbers, the estimated probability for Biden to win the popular vote is 67%. In other words, the PollyVote predicts Biden to win the popular vote two out of three times, whereas Trump is predicted to win one out of three times.

Table 1 shows the point forecasts from the PollyVote and the components, for which data are currently available. Forecasts based on both polls (intentions) and expectations see Biden ahead, whereas model-based forecasts and naive forecasts see a very small advantage for Trump.

Intentions (polls) (RCP average)46.753.3
– Prediction markets (7-day IEM average)49.051.0
Expert judgment47.752.3
– Retrospective (1 model, DeSart)54.445.6
– Hybrid (1 model, Fair)47.152.9
Naive forecast (2 models, electoral cycle and 50/50)50.649.4
Table 1: The PollyVote and its components, forecasts from April 26, 2020

It should be noted that available data that go into the PollyVote are still limited. The poll-based component is merely based on the RCP average, prediction markets rely on the small-scale Iowa Electronic Markets, and there is only one model forecast each in the retrospective model and the hybrid model category.

Loyal followers of the PollyVote will note that the PollyVote structure has changed compared to previous elections. We will explain our rationale for this adjustment in a future post. Stay tuned.