Since James Carville famously wrote “The Economy, Stupid” in Bill Clinton’s campaign headquarters, the mass public has become increasingly familiar with what is one of the more important “facts” discovered by political scientists: US presidential elections are undoubtedly affected by the state of the economy. The question of which part of the economy matters most, however, still remains unsolved. Nate Silver has two very interesting recent posts on this topic. In one, he makes the claim that the key figure is 150,000 jobs per month: if the US economy performs better than that between now and the elections, Obama has a good chance of being elected. In the second, he subjects a whole host (43 to be precise) of different economic variable to the test of predicting 16 different US presidential election results, and then compares which of these variables performs best; again his payroll job growth variable does quite well. Also recently, The Monkey Cage’s Larry Bartels, also writing in the NY Times, highlights changes in real disposable income (which comes in 10th on Silver’s list), a long time favorite variable of political scientists studying American politics.
Both posts are worth reading, as they continue to provide irrefutable evidence of a link between economic conditions and election results in the US. Moreover, the question of which economic conditions matter most is of both academic and pragmatic interest. Whether it is actually possible to study this in an empirically satisfying matter when limiting oneself to US presidential elections, however, remains an open question. As Silver notes:
When you’re testing 43 different economic indicators over a sample of just 16 elections, the best-performing ones are likely to have been a little lucky. In fact, the relative rank of the economic indicators has historically been very inconsistent: those that perform best over one set of elections do not do much better over the long-term.
In my book on economic voting in post-communist countries in the 1990s, I took a slightly different tack. Rather than try to parse out the effect of different economic variables on election results, I used the best data I had available to try to get a general estimate of how parties would perform when the economic was “good” as opposed to “bad” (see Ch.3 for a description of my methodology). Now I was doing something quite different — trying to use regional variation in economic conditions to predict regional variation in election results one election at a time — but at the end of the day I wonder how much more we can really claim with 16 elections to draw on in the US beyond the (very valuable) observation that better economic conditions help the incumbent, while worse economic conditions hurt the incumbent.
As Silver aptly notes, a variable that does well through 13 elections can suddenly perform poorly in the 14th, and then where are we? Most obviously, I think this probably follows from the fact that an awful lot of Silver’s 43 variable co-vary with one another. If his results showed 3 variables with huge effects and then 40 with no effect, I would think we were on to something quite important. Instead, however, we see a gradual decline across all the variables, suggesting to me at least that after 4 more presidential elections we might still see the same gradual decline but with a reordering of the rankings.
Indeed, the most interesting thing from Silver’s horse race type of analysis is probably the variables that have *no* relationship to vote outcomes. And while these may also be the result of random noise, there is one that is worth mentioning, which is unemployment. I have always been struck by the lack of unemployment measures in most of these US economic voting models, and Silver suggests why this is the case: in his data at least, unemployment has no effect on election results in US presidential elections. (As a side note, while I used a variety of variables in my models in post-communist countries, unemployment is an — if not the most – important driver of the results). Silver does, however, find a very strong effect for change in unemployment between January-September of the election year, which actually overlaps very nicely with Bartels’ basic claim in his piece that voters are myopic, heavily weighting recent economic developments at the expense economic developments from earlier in one’s term. So once again, we may be in a situation where economic variables co-vary with one another (we would expect growth in real disposable income to go up as unemployment comes down), but we can make some interesting observations about the time-frame within which the economy matters.
[h/t to Nadia Hassan]