Abramowitz Equation: Could it really be this simple?

Professor Alan Abramowitz of the University of Virginia has discovered an incredibly simple way to predict the Presidential election. You don't need to know who might win in the swing states. All you need to know is Donald Trump's net approval rating from Gallup polls. The net approval rating is the approval rating minus the disapproval rating.

The equation was developed based on elections dating back to 1952, and includes only those elections where the incumbent president was running for re-election. Based on past elections, the incumbent president has a definite edge being re-elected. Abramowitz presented two similar equations, one using approval ratings from late June, and the other from late October.

I used Abramowitz data and calculated the model's coefficients using EXCEL. The data used are given at the end of this posting. The resulting equation is:

EV-Trump = 5.65 (NA) + 261.8, where NA = Trump's approval rating minus his disapproval rating in percent.

The equation is nearly identical to Alan Abramowitz. He may have included 11 elections including the 1948 election.

As of September 13, Trump's approval ratings was 42% approval and 56% disapproval, so NA = -14%, according to Gallup. The website, fivethirtyeight.com shows NA = -9.3, and I used their value because they take an average of several polls. Based on NA = -9.3. Trump will win 208 electoral votes, so Biden is the winner with 329 electoral votes.

In the plot below, I have indicated the point on the line for NA = -9.3, and EV = 208.

Point 1: Carter, 1980, Point 2, Bush I, 1992, Point 3, Ford, 1976 and Point 4, Bush II, 2004.

In Table 4 of Abramowitz article, he predicts Trump would win 205 EV's if NA = -10%, and 233 EV's if NA = -5%. I calculate the same values. In the same table, for NA = -10%, I calculate Trump has a 20% chance of winning, while Abramowitz calculates 17%. The winning probability is the probability that Trump's electoral votes will be 270 or higher, given an NA = -10%.

I tried two additional experiments on this data, using NA = -10 for comparison purposes. First, I culled out all elections which were won by more than 400 EV's, as these are elections where the incumbent was very popular (NA > 30%). The second experiment was to reduce the dataset further by removal of Carter loss in 1976 with only 49 EV's. All results are stated with respect ot the predicted winner (Biden).

Results of Cases with Net Approval = -10

This is one of the simpliest methods of prediction the 2020 election. Note, these predicted EV's are close to results using poll based methods.

David Lord

September 28, 2020

Data used (NA values are from Table 6 of Abramowitz article)