This is a follow-up to my previous post, in which I discussed the mechanics and early predictions of a few statistical models I developed to estimate the chances that Donald Trump will be re-elected to the Presidency during the 2020 General Election.
As I wrote in my last post, Trump’s job approval numbers have been pretty dismal the past several months. According to my polling aggregator, which estimates approval ratings based on polls gathered from HuffPost, a full 58% of Americans presently disapprove of the job Trump is doing in the White House. Only 38% approve.
Although that’s obviously bad news for the Trump administration and its current agenda, the good news for Trump personally is that, historically at this point in a president’s first term, job approval numbers are not a very good predictor of re-election.
Let’s take a look at what I mean.
Far in advance of Election Day, approval ratings bear little to no sensible relationship to an incumbent’s chances at re-election
To get a better understanding of just how poorly job approval ratings predict re-election at this point in a president’s first term, I fed some simulated data into my forecasting models, which aim to estimate Trump’s re-election chances based solely on current and historical job approval ratings. You can read a little more about how the models work, and their limitations at this point, in my earlier post.
The simulated data comprised a range of job approval numbers, from 0% to 100%, for between 0 and 1,100 days prior to Election Day. As of the time of my writing this, we are 1,100 days out from Election Day on November 3, 2020.
The graphic below shows estimates of re-election chances, plotted as a function of actual approval ratings, for, at most, 1,100 days prior to Election Day.
The estimates come from two models – (1) a “Bayesian Model,” which compares current and historical job approval ratings to try to determine whether Trump’s latest numbers seem characteristic of a president who will later go on to be re-elected; and (2) a simple “Logistic Regression Model” with two predictor variables, namely approval ratings from Gallup tracking polls and the number of days prior to Election Day each poll was conducted.
As you can see in the graphics above, there is a whole lot of variability in the models’ prediction estimates, particularly for approval ratings that approach 0% or 100%.
According to the logistic regression model, for example, a hypothetical incumbent with just a 20% approval rating has anywhere between a 5% chance at re-election and an 89% chance at re-election.
Similarly, at 1,100 days out from Election Day and with an approval rating of only 38%, the models peg Trump’s current re-election chances at between 81%, according to the Bayesian Model, and 84% according to the Logistic Regression Model.
So, what’s going on here? Are these models just failing spectacularly? No, not really.
The reason for these varied and counterintuitive predictions is because, far in advance of Election Day, approval ratings bear little to no sensible relationship to an incumbent’s chances at re-election. In fact, the models have determined that, for polls conducted very early on in a president’s first term, approval ratings are actually slightly negatively related to re-election chances. This means that, very early on, it is lower approval ratings, rather than higher approval ratings, that signal a better shot at being re-elected.
The graphics below are simplified versions of the ones presented above and show, rather clearly, how the models see the relationship between approval ratings and re-election chances at various points in time prior to Election Day.
But during the final year of a president’s first term, approval ratings become a much more reliable predictor of re-election.
So, it’s clear that approval ratings are not very predictive of re-election early on in a president’s first term, at least not in any way that’s intuitive and sensible.
It’s also clear, however, that as we get closer to Election Day, the nature of the relationship between approval ratings and re-election chances changes considerably, and things start to make a bit more sense.
The next pair of graphics show estimates of re-election chances relative to approval ratings for all days up to and including 300 days prior to Election Day.
At a little under a year out from Election Day, there is considerably less variability in the models’ prediction estimates, compared to the situations depicted above.
At 300 days or fewer prior to Election Day, an incumbent with a 20% approval rating has only between a 5-18% chance at re-election, according to the Logistic Regression Model, and essentially a 0% chance according to the Bayesian Model.
Clearly, this is more reasonable. And what we should take from this more reasonable relationship between approval ratings and re-election chances is that, during the final year of a president’s first term in office, approval ratings evidently become a better, much more reliable predictor of re-election.
So, it’s too early to make predictions today. But where might things go from here, for Trump and for our country? And what sort of predictions might we be able to make in the near future?
At this point in time, Donald Trump is a deeply unpopular president. Of that there really is no question. But as I’ve tried to emphasize, it’s entirely too early to know what the public’s low opinion of Trump will eventually mean for the outcome of the 2020 Election. As described above, approval ratings bear little to no sensible relationship to an incumbent’s re-election chances this far in advance of Election Day.
But given that approval ratings eventually become a pretty good predictor of an incumbent’s chances at re-election, we can speculate a bit about where things might go from here over the course of the next three years.
Below are three possible scenarios.
Scenario #1: Trump’s approval ratings stay the same from today until Election Day on November 3, 2020.
As of the time of my writing this on October 30, 2017, I estimate that President Trump’s approval rating is around 38%.
This is obviously not where team Trump will want things to be on Election Day if they want a good chance at winning re-election. If Trump fails to pull his approval numbers up over the next three years, then he will have a 43% chance of being re-elected, according to my Bayesian Model, and a mere 33% chance according to my Logistic Regression Model.
But as unappealing as this scenario would be to Trump’s supporters, it probably wouldn’t necessarily spell disaster to Trump’s re-election bid. Remember that most presidential forecasting models pegged Trump’s chances of winning in 2016 at 25% or less.
Scenario #2: Between today and Election Day, Trump’s approval numbers increase by about the same amount as for an average incumbent who eventually goes on to win re-election.
When we look at historical approval ratings for former presidents who eventually went on to win re-election – excluding Lyndon B. Johnson, because there is limited polling data for him prior to the 1964 Election, and George W. Bush, because his extremely high approval ratings following the 9/11 terror attacks contributed to a particularly steep decline in approval by the time of his 2004 re-election – we see that the average winning incumbent typically sees only a 1.25 percentage point increase in approval ratings over the course of the 1,100 days leading up to re-election.
If Trump follows this trend – and there’s no guarantee he will – then we can expect his approval rating on Election Day to be somewhere around 40-42%, so obviously not very different from where it is currently. This would translate to about a 56% chance of re-election according to my Bayesian Model and a 41% chance of re-election according to my Logistic Regression Model.
But given that, as of right now, Trump’s approval ratings are exceedingly low – and lower than those of any other president in recent political history at this point in a first term – one could argue that Trump might see a larger than average increase in approval from now to Election Day simply because of some regression toward the mean.
Regression toward the mean is a concept in statistics that refers to the tendency for extreme values to drift, gradually over time, back toward some theoretical average without any intervention whatsoever. In Trump’s case, it’s unclear what this theoretical average might be, but we could again speculate by looking at some historical data for past presidents.
The average approval rating on Election Day for incumbents re-elected to a second term is about 55% (this time including Lyndon B. Johnson and George W. Bush). If Trump’s approval ratings drift up to around this level by Election Day 2020, then his chances of winning a second term will be 81%, according to my Bayesian Model, and 80% according to my Logistic Regression Model.
Scenario #3: Between today and Election Day, Trump’s approval ratings decline even further
In recent political history, only three presidents have failed to be re-elected – Gerald Ford in 1976, Jimmy Carter in 1980, and George H.W. Bush in 1992. Over the course of the final 812 days leading up to Election Day (I chose this cut-off because it’s the earliest point in time for which we having job approval data for all three of these former presidents), the approval ratings for these three losing incumbents decreased, on average, by about 20 percentage points, from an average of 61% to an average of 41%.
Given that Trump’s approval ratings are already extremely low, it seems unlikely that his numbers will sink by quite this much over the next three years. But if, by Election Day, Trump’s approval rating is, in fact, down to only 20%, then his chances of being re-elected will obviously be quite dismal. His chances of being a two-term president will be essentially 0%, according to my Bayesian Model, and only slightly higher at about 5% according to my Logistic Regression Model.
Below are summaries of my models’ predictions for the scenarios I’ve just described.
To confidently predict Trump’s chances of re-election based on approval ratings, we’ll probably have to wait until early January 2020.
So, there you have it – three scenarios outlining what Trump’s chances at re-election might look like on November 3, 2020.
Obviously, only time will tell as to which of these three scenarios will come closest to the truth on Election Day. But if past data on the subject is any indicator, then things might start to get a bit clearer around 300 days prior to the election, when I suspect job approval numbers become a relatively better predictor of election outcomes.
So, although I’ll update my models’ predictions regularly from here on out – and you can view the latest predictions here – we should all stay tuned for early January 2020.
Brian Kurilla is a psychological scientist with a Ph.D. in cognitive psychology. You can follow Brian on Twitter @briankurilla