Nadia Hassan writes:
One debate in political science right now concerns how the economy influences voters. Larry Bartels argues that Q14 and Q15 impact election outcomes the most. Doug Hibbs argues that all 4 years matter, with later growth being more important. Chris Wlezien claims that the first two years don’t influence elections but the second two do.
After 2000, Larry Bartels and John Zaller used Bayesian model averaging (BMA) to assess why election models overestimated the Gore vote in their pre-election forecast. Erikson, Bafumi and Wilson (2002) had some beefs with their arguments and conclusions. But both groups agreed BMA is useful for understanding elections. Do you think BMA could help evaluate these arguments and the uncertainties surrounding them?
My reply:
There are interesting debates on how the economy influences voters and also how presidents influence the economy, and in both cases the question of timing comes up. With regard to the particular question above, I haven’t looked at the details myself, but my guess is that the data at hand would not be enough to decide among these various theories. Ultimately, “N” is not large, and you have to use outside criteria to decide what model to go with. I doubt that Bayesian model averaging will give you much here. But I guess it won’t hurt.
I only see one model here, at least as stated: Bartels and Hibbs are entirely consistent with each other. Wlezien is implausible as paraphrased here (exactly zero is a very unlikely effect size), but any reasonable restatement would match the other two. If all three are qualitatively the same, the only remaining debate is over effect size, Since they’re all running off the same data, the only place hwere multiple models would help is in getting the confidence intervals to be wider, as they probably should be.
I probably should have mentioned this paper by Wlezien just to clarify that it’s not just his hunch or reading of others that lead him to his conclusion. That he found similar patterns among three different indicators is telling, and the analyses seem to be well done. (But, they need a good graph or scatterplot!) They may be, nevertheless, limited by the things you discuss above.
http://www.researchgate.net/publication/271504822_The_Myopic_Voter_The_Economy_and_US_Presidential_Elections
How much is this differential-economic-quarter impact in the grand scheme of all those things that DO impact voting? And what’s the external validity of all this. If, for arguments sake, we do conclude that Q14 and Q15 impact election outcomes the most, is this fairly generalizable across elections? Does it even make sense to search for a generalizable metric of this nature, given the huge variations over cohorts?
If instead of concluding that “Q14+Q15” matter, we conclude that “last two years matter” how qualitatively different is this conclusion in the goal of predicting outcomes? Especially when evaluated in the context of all the other noise that’s out there? Isn’t the difference between “Q14 matters” vs “Last 2 years matter” totally swamped out by the temporal variation between (say) 2012 vs 1992?
Isn’t some of this fine grained structural modelling a bit like arguing about angels and pin-tip dances?
Rahul,I think there are two reasons why scholars are interested in that: theories about democratic accountability and voting and incentives for politicians. If Q8-Q16 matter equally, politicians might have less incentive to get growth over a short period of time. There is also the question of who the voters are and how sound their judgments are.
OTOH Poli. Sci. researchers have an incentive in generating neat causal arguments.
The question is, whether the noisy & ever changing patterns of voting reality do have any such significant temporal difference between the impact of various quarters to an extent where you can draw meaningful conclusions about accountability & incentives.
Color me skeptical.
As an aside, what percent of voter behavior is explained by economic performance anyways?
I’ll tell – Obama. Hilary Clinton won’t be inaugurated till 2017.
You win.