Monday, August 15, 2011

Using economic projections to make vote projections

When I'm asked whether I expect President Obama to get reelected, I often look around to see if there's an economist nearby to tell me how the economy is going to be doing next year. Well, here are some economists now, in the form of USA Today's survey of 39 economists. Unfortunately for my purposes, they don't forecast real disposable income, which is probably the most reliable predictor of presidential elections, but they do use gross domestic product, which is the next best thing. Their average projection of GDP growth from the fourth quarter this year to the third quarter next year is 2.53 percent. How does that stack up with economic growth and electoral performance in previous years?
Well, if we use this metric and the USA Today forecast, Obama will be facing lower economic growth than what Bush faced in '92 and his son faced in '04. That's real nail-biter territory. The regression line suggests Obama getting 50.9 percent of the vote, although, as the graph shows, that line is being pulled upward a bit by Eisenhower's unexpectedly good performance in 1956. If you take out that case, the Obama forecast is 50.03 percent of the vote. In other words, don't expect to get a lot of sleep on election night next year.

All the usual caveats apply here. The forecast for 2012 economic growth is just a prediction -- it could get substantially better or worse. And this is just a bivariate regression model, which doesn't include things like foreign policy or ideological extremism or campaign quality. And, of course, past performance does not necessarily predict the future.

But this could be a really, really close race.

15 comments:

Andrew Oh-Willeke said...

The regression is really dubious. McCain 08 was not an incumbent in the ordinary sense. Likewise, I do not recall our nation having a President Stevenson or a President Humphrey or President Gore in the late 20th century. The dynamics of open races are fundamentally different from those of reelection campaigns. A reelection campaign is a referrendum on the incumbent, an open race is a choice.

Carter 80 is an outlier that is profoundly influencing the regression line that otherwise shows a much weaker economy to electoral outcome relationship.

Seth Masket said...

Andrew, this regression is in line with a large literature that sees presidential elections -- all of them -- as referenda on the party of the president, not just the incumbent president. And, indeed, this literature shows that, at least from the perspective of the economy, the dynamics of open races aren't really that different from those of reelection campaigns. McCain wasn't an incumbent, but his party suffered because it was the incumbent party during bad economic times. Bush Sr. benefited from the economic expansion of the mid-1980s even though he wasn't the incumbent president.

Anonymous said...

I've noticed several different blogs arguing that the literature seems to back up this kind of approach. Can anyone point me to a reading list or syllabus where I could read this literature first hand?

Seth Masket said...

Anonymous, I'd encourage you to check out the Zaller/Bartels paper on this topic, which gets into the election variable and the debate over RDI vs. GDP. Also, read Hibbs' Bread and Peace Model.

Steve Greene said...

Nice post, Seth. I was surprised that the economists predicted growth that high. It's certainly not great, but I would've expected predictions for much worse. I wonder just how bad Obama's number would be if we actually get that double-dip recession.

Andrew Oh-Willeke said...

What does the r^2 look like on that plot? What percentage of variation is explained by GDP?

Seth Masket said...

It's .37. It goes up to .40 if you weight GDP toward the more recent figures.

Robert said...

Layperson checking in. So is the r^2 a measure of the strength of the correlation? And, more to the point, if so, is there any other measured variable that suggests a stronger correlation than this GDP measure?

Seth Masket said...

Robert, yes, the R-squared measures the strength of the relationship. An R-squared of .40 means that the predictor (GDP growth) explains 40% of the variation in the dependent variable (presidential vote share). RDI usually has a higher R-squared than GDP. Hibbs' model combines RDI with military fatalities to produce an R-squared of .86.

Anonymous said...

did somebody order two "Mountain man mike" pizza's and a salad? i got an order from this address for two mountain man mike's and a salad.

Bart DePalma said...

The regression line suggests Obama getting 50.9 percent of the vote, although, as the graph shows, that line is being pulled upward a bit by Eisenhower's unexpectedly good performance in 1956.

Eisenhower 56 enjoyed full employment.

Indeed, every post WWII president who earned reeleection either enjoyed near full or full employment (less than 6%) or, in the case of Reagan 84, a sharp decrease in unemployment off a previous high (- 3% between 83 and 84).

Carter 80 and Bush 92 suffered from either unemployment rising or plateuing over 7% and both lost badly.

Obama is in a far worse situation with U3 unemployment among those actually looking for work at between 9% and 10%, with long term unemployment among the millions who have given up looking for work reaching levels not seen since the Great Depression.

Think Hoover.

Seth Masket said...

Bart, neither the unemployment level nor the annual change in unemployment predict vote share in the regression model. Unemployment was low for Ike in '56, but it was even lower in '52, '68, and 2000, all years in which the incumbent party's candidate nonetheless lost.

Bart DePalma said...

Seth said...

Bart, neither the unemployment level nor the annual change in unemployment predict vote share in the regression model.

I am unsure how you designed your model, but no President seeking reelection has lost with "full employment." Likewise, no President seeking reelection with less than full employment has won in the modern era unless unemployment sharply declined from a previous high during that term. See FDR 36, FDR 40, FDR 44, Reagan 84.

There are no exceptions, so you may wish to revisit your model and restrict it to reelections and extend its scope past a single year.

Unemployment was low for Ike in '56, but it was even lower in '52, '68, and 2000, all years in which the incumbent party's candidate nonetheless lost.

What your examples show is that open elections with new candidates during "full employment" are not referendums on prior president or his party. I suspect that voters feel free to shop around during peace and prosperity. There is not the same impetus to reward the success of a sitting President.

Seth Masket said...

Bart, we're sort of talking past each other, since my dependent variable is the vote share of the incumbent president's party, and you're just talking about cases in which an incumbent president is up for reelection. That is, you're focusing on a lot fewer cases than I am. You may be right that "no President seeking reelection with less than full employment has won in the modern era unless unemployment sharply declined from a previous high during that term" (depending on how we're defining "full employment"). But given that only two presidents have lost their reelection bids in the past 75+ years, that's really not that informative.

Bart DePalma said...

Seth:

I agree that we do not have enough reelection data points for statistical reliability, but a perfect record of the handful of data points that do exist strongly suggests a correlation between unemployment and presidential reelection.