In an earlier post, I noted that President Obama’s approval rating was higher than a model of presidential approval predicted — something that seemed to set him apart from previous presidents. In response, commenters here on FiveThirtyEight and The Monkey Cage (see also Jay Cost) noticed that, for most presidents from Dwight D. Eisenhower until Mr. Obama, the model’s predicted approval rating was consistently lower than the actual approval rating. In this post, I want to do four things:
- Describe why this was true.
- Report results from what I think is a better model.
- Demonstrate that my original point — Mr. Obama is more popular than expected — still holds, but to a lesser extent.
- Extend the analysis by showing that Mr. Obama’s “better-than-expected” popularity is due to his appeal among Democrats and independents, but not Republicans.
First, why did the model seem to underestimate the approval of most presidents? The reason was how the model dealt with this fact: in addition to the systematic factors that influence presidential approval — economic indicators, for example — idiosyncratic factors will also influence approval. If we want to look at presidential approval from 1948 to 2008, then predict what Mr. Obama’s approval “should be,” how should we account for these idiosyncratic factors?
In my earlier post, I did this by including a variable for each presidential administration from Eisenhower to George W. Bush. I did not include a variable for Harry S. Truman, since the model cannot have a set of factors that perfectly predict each other — and given that the presidential approval data start in 1948 and end in 2008, knowing that the president is not Eisenhower up through George W. Bush means that you can perfectly predict that it must be Truman. (See the “dummy variable trap.”) Instead, Truman’s approval rating would be reflected in the model’s intercept — that is, the estimated approval rating when the other variables in the model equal zero and when the president is none of Eisenhower through George W. Bush.
The problem arises because the intercept is part of the prediction of Mr. Obama’s approval ratings. And because Truman was unpopular during his second term — on average, only 39 percent of Americans approved of him — the intercept is deflated. Thus, the model’s predictions for Mr. Obama and most other presidents would tend to be lower than their actual ratings, simply because most presidents have been (on average) more popular than was Truman during his second term. This is what commenters noticed. After further reflection, I believe this feature of the model was a mistake. I thank commenters for making me think harder about this. (And now might be a good time to remind readers, especially those reading via RSS, that the “I” is not Nate Silver but John Sides, a guest blogger.) This is a nice example of how online commentary about articles can generate important insights.
A better model would essentially average the idiosyncratic factors of each president to generate a baseline against which Mr. Obama’s approval could be evaluated. One way to do that is to estimate a fixed-effects model. In this model, the intercept becomes the average of the approval ratings of the presidents besides Mr. Obama, again assuming that any other factor in the model is equal to zero. Using this model, and including the same economic and other factors as in the original model, I can again generate an out-of-sample prediction for each president from Truman through Mr. Obama (click image to enlarge):
)
CLICK TO ENLARGEThis model’s predictions are not consistently greater than actual approval. There is instead a mixture of under- and overestimation. To be sure, the model is not perfect. Because the model includes a “lagged” measure of presidential approval (that is, presidential approval in the previous quarter), the model’s predictions also sometimes lag the president’s actual approval rating by a quarter. This is particularly true when the model does not fully anticipate sharp increases in approval, like the rally effects at the outset of the gulf war, after Sept. 11, etc. Still, the model of approval from 1948 to 2008 explains about 84 percent of the variation in approval ratings. The median discrepancy between actual approval ratings and the model’s predictions is only 3.2 points in absolute value.
So what about Mr. Obama? Here is a graph just of his actual and expected approval ratings.
As in my original post, Mr. Obama is generally more popular than he should be. But his approval rating exceeds the prediction by about 2.7 points, on average, not by the nine points I originally reported. In certain quarters, his approval rating does exceed expectations by a larger amount — for example, by eight points at the beginning of 2011, when gross domestic product grew very slowly and inflation increased somewhat, but his approval rating remained steady. His rating is also six points larger than expected in the first quarter of 2012, which is the most recent quarter in which the economic data is available. Indeed, over the past three quarters, there is a hint that the gap between Mr. Obama’s expected approval and his actual approval is beginning to grow.
Mr. Obama is again more popular than most other presidents, relative to the model’s prediction. If you consider the out-of-sample predictions for all presidents, only one president consistently “beat” the prediction in his first term to an extent greater than Mr. Obama: Ronald Reagan, who was about 3.3 points more popular than the prediction, on average.
Mr. Obama’s better-than-expected popularity arises largely because he is more popular with Democrats and independents, but not Republicans. Using historical presidential approval data broken down by party, I estimated the same model and compared its prediction to Mr. Obama’s actual approval among Democrats, independents and Republicans (click on image to enlarge).
On average, Mr. Obama’s approval is higher than expected among Democrats and independents, but not among Republicans. In fact, his approval is actually slightly lower than expected among Republicans. Among first-term presidents, his performance versus expectations within his own party is exceeded only by Reagan’s performance. In fact, it’s worth contrasting Mr. Obama to other recent Democratic presidents. Bill Clinton performed no better or worse than expected among Democrats. Jimmy Carter performed much worse than expected (seven points). These facts may allay frequent fears (or hopes) during Mr. Obama’s first term that he was “losing the base” or otherwise losing support among important Democratic constituencies. This may have been true among some Democratic or liberal activists — many of whom disagreed with the president on various issues — but it was largely not true among Democrats in the public. Not only does Mr. Obama remain relatively popular among Democrats, but he is also more popular than we might otherwise expect. In the most recent quarter, his approval among Democrats (as well as independents) is about six points higher than expected.
Of course, these results aren’t necessarily surprising. As the political scientist Gary Jacobson has documented, presidential approval ratings are increasingly polarized by party. This has certainly been true of Mr. Obama. But the very forces of partisan polarization — the “red America” and “blue America” — that Mr. Obama deplored as a presidential candidate may have buoyed his approval rating as president.