A few weeks ago, Greg Koger examined Donald Trump's approval rating by state and foresaw a difficult reelection effort for the President. Trump, Greg noted, had negative net approval in 30 states, and he didn't have an easy path for even getting back to his slender victory margins in 2016. His campaign, and his party, would have to expend considerable resources to win even pretty reliably Republican states.
The piece left me curious, though -- how well do approval ratings predict vote shares a year later? Any forecast a year away from an election is going to be imperfect, of course, but approval ratings are asking a fundamentally different question than the one voters will see on their ballots.
One way of examining this is to look at Barack Obama's approval ratings by state a year before he faced reelection. I've done that in the scatterplot below, using Obama's 2011 approval ratings to predict his vote shares in 2012. I've also added a green trend line that indicates the share of the vote he would have received if his approval ratings predicted that perfectly.
First of all, this is a very strong relationship (with an r-squared of .86). But approval ratings tended to understate vote shares somewhat, particularly in the more Democratic states. Obama had an approval rating of 54.5% in California in 2011, but ended up with 61.9% of the two-party vote there in 2012, for example.
We can get a much stronger prediction by controlling for Obama's vote share in 2008. Using both that and the approval ratings to predict Obama's reelection vote share yields an r-squared of .96, with both factors highly statistically significant.
Next, I used the regression formula resulting from that* to predict Trump's vote share in 2020, using his vote share in 2016 and his 2019 approval ratings. I then took those forecasts and plugged them into the map below. The "lean-D" states are those where Trump is predicted to get between 40 and 48 percent of the vote, and the "lean-R" states are those where he's predicted to get between 52 and 60 percent. The tossups are between 48 and 52 percent.
This is notably a much more competitive map than the one Greg posted. It leans Democratic -- Trump would need to hold pretty much all the tossups in order to remain in office -- but neither side has a lock on this one.
Again, as Greg noted, these are just projections based on very early data, and quite a bit (impeachment, recession, etc.) can happen between now and November 2020. It's also quite possible that Trump's approval ratings won't predict his vote share in the same way that Obama's predicted his. But at least from this look, while Trump's task is a difficult one, it is far from insurmountable.
*If you're interested, the regression formula is:
predicted vote share = -9.07 + .83 × previous vote share + .34 × state approval rating
Data available here.
Comments