2020 is finally here, which means this year, we’re finally voting that monster out. But to listen to the conventional wisdom, you’d think that Trump will be all but impossible to defealt. There’s usually some variant of “Incumbents almost always win,” or “with this economy, he should be cruising to reelection.”
In a series of posts, I’m going to make the argument that, unless something changes dramatically, Trump is very likely to lose — and history backs me up.
In a recent post, I outlined that, high quality pollsters have been outstanding at predicting electoral outcomes since 2016 — and even in 2016, the “huge” errors were about only about 3% at the state level. Here’s a 2018 summary:
Today, I’ll present a short numerical history of presidential reelections. Because some data is a pain to collate, and might be of help to others, here are a nice set of resources that you may find useful for other projects. Feel free to share:
I’ll save the last 4 of these for a subsequent post, but for now, I thought you folks might want them all in one place. So let’s start with some basics…
1. By and Large, Presidents aren’t Especially Likely to get Reelected
Start with some stats:
- 39: Thirty-nine individuals have been elected to the Presidency, not counting non-consecutive terms, folks who ascended to the presidency after death or resignation, but didn’t win a term on their own.
- 25: Of those who got elected in the first place, only 25 sought reelection, and got at least as far as winning their primary. Roosevelt ran for reelection three times, winning all three.
- 15: Of the 25 reelection campaigns, only 15 won. That’s 60%, barely more than half. Of the last 6 modern presidents (post-Ford), the ratio is only 67%.
As an illustration of the lack of the power of incumbency, I’d like you to consider the children’s card game, War. Each person throws down a random card from their pile, and whoever has the higher card, wins the pot (ignore ties for this example). You could think of the value on each card as the “quality” of the nominees from each party.
Suppose we tweak the game to model the rules for presidential elections. In our version, imagine that a winner can leave their winning card down for one more hand, but the loser has to put up a new card. Since your winning card is probably above average, you’re more likely than not to win a second time.
In fact, running the odds, you’re about 67% likely to win “reelection,” in this toy model. Further, if you won your election by defeating an incumbent (like in 1980 or 1992) then you’ve beaten someone who’s beaten someone else, and those odds rise to 75%!
Using the toy model, we’d expect that at least 2/3 of incumbents should win reelection (even ignoring those who were simply blackballed by their own party or who saw the writing on the wall, pre-election). In other words, presidential reelection is a little less likely than simply asserting that candidate quality (a number representing “electability” within a district) is some immutable number.
There is, however, one important thing that can be used to predict whether a particular candidate will be reelected: how popular they are.
2. Job Approval is Pretty Much All that Matters
The correlation between job approval and reelection prospects is so obvious that it should be trivial. But you’ll find no shortage of prognosticators who insist that economic factors can and should be used on their own to predict who will win an election. In the next post, I’ll relate historical data to reelection results, and show that while there is a connection, the real issue is that normally some economic metrics predict approval which in turn predicts reelection. If you’ve got a good economy, but a lousy approval, you’re going to lose.
Here is the net approval (approval-disapproval) from the final 2 gallup polls averaged (to reduce noise) for each post-war president, including those like Johnson and Ford who inherited their first terms:
Incidentally, I use Gallup only because they go back so far, not because they’re particularly good pollsters.
Something should immediately pop out at you, even without the guide line: unpopular presidents tend to lose, and the less popular they are, the more they tend to lose by. The sole exception on the chart was “Dewey beats Truman” in 1948, which, like in 2016, saw a polling miss due to a misunderstanding of how to correct for certain segments of the population. In 2016, it was by not including education as a key question (no college white voters were very different than college voters); in 1948, the issue was that only rich folks owned a phone.
Ford also seems a bit of an outlier, but besides the complication that Ford wasn’t elected in the first place, the last gallup poll was in June of 1976. Also worth noting was that while his net approval was above water, his actual approval was 45%, with a lot of undecided voters.
Nevertheless, the correlation between these two measures is 0.87. For those unfamiliar, 1.0 is a perfect predictor, 0.0 is random, and -1.0 is perfectly anti-correlated. This correlation is extremely good.
We could also look to see whether “approval” or “net approval” is the better predictor. For what it’s worth, “approval” has a similar trend, but a lower correlation:
As a quick reminder, Trump is currently sitting at around -10 net approval according to fivethirtyeight. Using my own low noise tracker, the number is -11, with an approval of 42:
In that territory — and remember, he’s near his current highs — there’s no realistic chance that Trump could win if his numbers remain where they are now.
3. Head-to-Head and Approval Converge
Finally, I want to make the case that in a very real sense, the challenger doesn’t matter all that much, at least in predicting reelection. In particular, while head-to-head polling is interesting at this stage, it may be less informative than net approval.
Here, my case is a little weaker, in large part because it’s hard to get quality tracking polls until relatively recent elections. The only 2 reelections with polling data are 2004 and 2012:
Here’s a visual comparison of the 2004 polling:
And 2012:
What you’ll note is that prior to about April of each election year (before the challengers were entirely settled), there’s a bit of a disconnect between the approval polling and the head-to-head polling with the eventual challenger. After April, the two remained consistent with each other to about 2 points, and both were within 3 points of the final electoral margin.
Yes, this is based on only 2 elections, but it still generally supports the whole, “The net approval gives you a great sense of where the election is” theory.
Finally, an important caveat:
Yes, this is a measure of the popular vote, not the electoral college. Absolutely true. But it’s also true that if Trump loses the popular vote by 10 points, he’s not winning the electoral college in any scenario.
Stay tuned for Part 2.