John Sides and Noam Scheiber got in a bit of a twitter fight yesterday about election prediction models, political science, and the 2012 election. John wrote a good piece summarizing his position here; I recommend it highly.
I have two additional points, but it gets lengthy so I'm going to break it up into two posts. This one is about what we mean when we separate "fundamentals" from "campaigns."
John gets to this in comments, but not in the main post: what's generally thought of as the "fundamentals" -- as opposed to the campaign -- should definitely not be thought of as purely economic performance. I think that's especially true on a theoretical level. It's useful for many reasons to separate the campaign from everything else. That "everything else" is going to include political context (is the incumbent running? How long has the incumbent party been in the White House? Is there unified or divided government?) on the one hand, and everything that would go in the category of retrospective voting on the other -- that is, things that would go into a voter's impression of how the president has done. Of that final bit, the economy empirically seems to be the dominant factor, strong enough that you can get fairly close to a useful prediction if you toss out campaign factors and the political context and the rest of the "how are things going" stuff. But then again: you can do better, both in prediction and explanation, if you don't throw that stuff out. Thus we have Hibbs, including casualties from wars; thus we have Abramowitz, including both political context and a presidential approval pre-campaign variable that presumably captures "how are things going" that goes beyond the economy.
I should note, by the way, that for an incumbent president any model that includes approval is going to capture not only how events affected approval, but also how the president's perceived personality and anything else that makes people like or dislikes him affect approval. Given a permanent campaign, one could certainly quibble that these are a type of campaign effect; indeed, even without a permanent campaign, things such as "personality" may wind up created by the incumbent's original campaign. That's all fine, however; what we're trying to separate out are the effects of the (current) campaign, and we're treating how people feel about the president at the beginning of that campaign as separate, even if it was produced by the exact same process. Note too that this means that an incumbent president's personality, then, is part of "fundamentals" while the challenger's personality is part of "campaign," as are all of their other relevant characteristics. Again, you may find that goofy, but there's a logic to it.
(Okay, that's not completely true; if there's something about a president that would affect approval differently than it affects vote choice, maybe it winds up in the "campaign" pot. I think.)
So, again: there are campaign effects and fundamentals. Campaign effects themselves include various things -- issue positions, ads, the candidates, GOTV and other mobilization efforts, and more. Fundamentals include political context stuff and performance stuff, of which the economy has turned out to date to be by far the biggest. The overall finding has been that fundamentals matter more than campaign effects, but that campaign effects are real -- but given that overall campaign effects are of limited (but real!) importance, it's going to be very hard for any specific campaign action, an ad or a debate quip or an issue, to do all that much.
It's even nastier, though, once we get down to brass tacks.
ReplyDelete-Presidential approval: when do you measure it? With a permanent campaign, when is appropriate? I think Abramowitz uses the June measure, which is a pretty defensible choice, as campaigns don't usually get into full swing until the summer (particularly going back in time), and primaries are often not done with before April (again, this looks better further back in time). But, since Prez App is REALLY a "kitchen sink" variable (in that it really is meant to capture everything else going on), its nasty: it's both theoretically appropriate and an obvious econometric trick at the same time. Which brings me to:
-What does a coefficient "mean" in a model? Is it that voters are blindly retrospective (for which there's more than a little evidence for it), and non-partisans are particularly prone to this? Or, are our economic variables really capturing a true campaign effect? That is, when the economy is bad, a challenger pointing that out can score actual rhetorical points with an audience; those attacks land. When the economy is good, challenger attacks don't work, because they're either desperate ("the sky is falling!"), illogical ("think how much better things would be if I was in charge!"), or just not compelling enough ("yeah, you're right, abortion is murder, but I care more about how well the economy is doing, and I like the incumbent on that score").
For my money, I tend to actually be that caricature of political scientists that Scheiber is making fun of (incorrectly, though, as he doesn't seem to be aware that most of the models are predicting a very close election, with Obama ahead but well within the MoE, and therefore, neither a narrow Obama win or narrow Romney win would constitute evidence against the models); I'm relatively agnostic on the "campaigns activate fundamentals" argument, but I find it interesting.