Brendan Nyhan

Over the weekend, the New York Times reported on Yale economist Ray Fair’s forecast (PDF) that the economy will rebound and President Obama will win a substantial re-election victory:

[Fair’s model forecasts real annualized growth in gross domestic product of 3.69 percent for the first three quarters of 2012. A survey of leading economists by Blue Chip Economic Indicators shows an average forecast of 3.2 percent growth in real G.D.P. in 2012, while the Congressional Budget Office estimates 3.4 percent. Plug either of these estimates into his election algorithm and the result is the same: President Obama wins.

In the quarter that just ended, however, the economy was growing at a rate of just 2 percent. If that sluggish pace continued — or, more ominously, if there were a double-dip recession or a steep plunge in the markets — that forecast would change.

Under those circumstances, regardless of other issues or the identity of President Obama’s opponent, the model shows the president losing.

I would be more cautious. First, I’m reluctant to place much stock in the details of Fair’s model due to his history of making ad hoc changes to the model specification (see Larry Bartels and Douglas Hibbs for more [PDFs]).

In addition, I think Fair’s estimate overstates the likelihood of an Obama victory given what we know today. The Philadelphia Fed survey of professional forecasters revised its forecast of 2012 growth downward last week from 3.6% to 2.9% (somewhat lower than the Blue Chip 3.2% figure or CBO’s 3.4%). If we plug that value into Alan Abramowitz’s simple linear fit of second-quarter GDP in election years and presidential election performance, we find Obama right around where President Bush was in 2004:

Many people forget, but Bush’s re-election victory was one of the narrowest in recent history. Obama is truly on the knife’s edge, especially once you consider the uncertainty around economic performance in 2011 and 2012. Here, for instance, is the estimated range of uncertainty in 2012 GDP growth among forecasters surveyed by the Philadelphia Fed:

What makes the 2012 campaign especially hard to predict is that Sarah Palin has a significant chance to be the GOP nominee (currently estimated at 20.9% on Intrade) despite having what seem to be unprecedented negatives for a serious presidential candidate at this point in the election cycle. Here, for instance, is an update of my previous chart comparing the trajectory of Palin’s favorable/unfavorable ratings to Hillary Clinton at the equivalent stage in the 2008 election cycle:

Obama should lose if the economy is bad, but we don’t have a good precedent for a presidential nominee who enters the primary campaign with an unfavorable rating of 52%. It’s possible that Palin would significantly underperform the forecast, giving Obama a chance in circumstances where he might otherwise face near-certain defeat.

Update 11/23 10:09 PM: I edited a sentence above to clarify that Palin’s negatives seem to be unprecedented for a serious presidential candidate at this stage in the election cycle. Also, Matt Yglesias notes that the Fed governors and reserve bank presidents are more optimistic about GDP growth in 2012 than the sources cited above (“Central tendency”: 3.6%-4.5%, “Range”: 2.6%-4.7%).

Update 11/24 8:40 AM: The Weekly Standard’s Jay Cost raises concerns about the GDP/election results graph above:

False positives are also a possibility. They are why I am suspicious of graphs that show a tight relationship between a single economic factor and electoral outcomes, like this one which tracks election results against Q2 GDP. There are so many economic variables that we could compare to electoral outcomes, which points to a critically important, yet often forgotten, caveat about statistical analysis: if we run enough tests, at some point we will wind up with a false positive, thinking that we’ve hit upon a causal relationship when in fact we haven’t. For instance, suppose your friend gives you a hundred coins and asks you to test to see which, if any, are rigged. You have about a 70 percent chance of getting heads 13 out of 16 times on at least one coin, even if all of the coins are fair. That would be a false positive, i.e. you conclude that the coin is rigged when in fact it isn’t. By the same logic, if we run enough tests between economic variables and election outcomes — sooner or later we are going to “find” something that isn’t actually there.

All of this is why that old phrase “lies, damned lies, and statistics” is so often repeated. We always need to check statistical results against common sense, being sure to strongly favor the latter. For instance, I’d ask: how many people thought to themselves in November 1948, “Hmm…well I like the cut of Tom Dewey’s jib, but boy-oh-boy GDP last May was really strong, so I’m going with Harry Truman!” Probably not that many! That’s not to say that there is no relationship between presidential outcomes and Q2 GDP, in fact I think there is one. Instead, my point is that common sense suggests that the graph is perhaps overstating it, or at least inducing us to oversimplify what is in fact a very complicated relationship.

Cost then shows a plot of “final election outcomes against the Gallup question of ‘who do you trust more to handle the economy,’” arguing that “the relationship is much less fuzzy.”

Two points are worth noting. I disagree with Cost’s suggestion that the relationship between pre-election economic growth and presidential election outcomes is overstated, oversimplified, or a potential false positive. While it’s certainly true that the data set is extremely small and that the mechanics of the relationship between growth and the presidential vote are complicated, the correlation is extremely strong and holds across a range of different measures of growth. See in particular the weighted-average measure of real per-capita income growth in the Bread and Peace model of Douglas Hibbs, which produces a very tight fit to the data (the most significant deviations are well-explained by a measure of military casualties that Hibbs uses as a second predictive variable):

It’s certainly true that you can often produce a tighter fit to the data by using subjective measures of economic satisfaction, approval of the president or his economic performance, etc. instead of raw economic data (see, for example, the Tom Holbrook plot reproduced in this post), but that doesn’t mean the relationship with the economic fundamentals isn’t real.

Update 11/28 3:32 PM: Alfred Cuzán at the University of West Florida emails to note that he has also issued a conditional forecast under different assumptions about the state of the economy using the “fiscal model” (PDF).

[Cross-posted to HuffPost Pollster]

Reminder for those who are interested: I’ll be giving a public talk titled “Barack Obama and the 2010 Election: What Political Science Can Teach Us” tonight at North Carolina State University (3712 Bostian Hall, 7 PM).

For those of you in the Research Triangle area, I’ll be giving a public talk titled “Barack Obama and the 2010 Election: What Political Science Can Teach Us” on Tuesday at North Carolina State University (3712 Bostian Hall, 7 PM).

Eric McGhee, John Sides, and I have a new article up at Boston Review that synthesizes our post-election analysis here and at The Monkey Cage:

Midterm Postmortem

Eric McGhee, Brendan Nyhan, and John Sides

On November 2 the Democrats suffered a historic defeat, losing more than 60 seats in the House of Representatives. It was the worst midterm loss for the president’s party in the post–World War II era. Why did Democrats do so poorly? One week after the election, there is much we do not know, but political-science research and our own analyses of the election results suggest some provisional answers.

Read the whole thing and then send it to friends, family, journalists, etc. who might be interested in a political science take on what happened last Tuesday.

[Cross-posted to HuffPost Pollster]