Here is another “not my field” question about the really interesting debate going on on this blog and others, concerning presidential election forecasting models and the question of how much campaigns matter versus the state of economy. Basically, I’ve never understood what all you adepts mean when you argue about whether “campaigns matter.”

We can do simple thought experiments to prove that “campaigns matter,” and matter enormously. For example, suppose that the Republican nominee decides to campaign strongly for changing our system of government to a communist dictatorship. I predict he or she would lose for sure, provided the opponent chose a normal campaign strategy, and for any plausible state of the economy.

That’s an extreme example, but the point is more general: Candidates *choose *campaign strategies and try to choose them optimally, which is to say that they are trying to choose the strategy that maximizes votes (or actually, probability of winning) given the other candidate’s strategy. To the extent that candidates succeed in this, historical data on campaign strategies, the economy, and election outcomes give us no leverage at all for estimating the causal effect of variation in campaign strategies on election performance. Campaigns could matter enormously, in that if any given candidate had chosen differently he or she would have lost a lot of votes. But you won’t be able to estimate the size of this campaign effect from comparing across campaigns and elections.

I guess what you all must have in mind is that candidates *don’t *necessarily chose optimal campaign strategies. They don’t choose obviously bad strategies like advocating for Leninism or monarchy, but for reasons of competence they may fail to identify the best campaign strategy within some window around the best. So then I guess the question you are asking is something like “how much do campaigns matter within the range of likely errors in identifying the best campaign strategy”?

But even here we can’t get an estimate of the size of this effect by comparing across campaigns, since maybe the reasons why you ran a less than optimal campaign are also factors that lower your vote share independent of the campaign (for one thing).

The experiment we’d like to run is to convince candidates to randomly choose from a menu of different campaign strategies, and do this for lots of elections so we could identify average effects. Not going to happen for US presidential campaigns. (Though Leonard Wantchekon famously pulled it off with candidates for a parliamentary election in Benin, and I’ll bet we start seeing some of this for lower level elections in the US in the coming years.)

The points are: (1) campaign strategies obviously matter, in a meaningful sense, and matter a lot, and (2) seems unlikely that you could estimate how much from data on past presidential elections, because candidates choose their strategies in light of all kinds of idiosyncratic, situation-specific information that Nate Silver, Doug Hibbs, Ray Fair, et al will never have.

If this is right (and maybe I’m missing something), then “how predictable are presidential elections based on economic fundamentals?” is an interesting question, but one that doesn’t have much or any bearing on the question of how sensitive is vote share to changes in campaign strategy. Further, the closer that candidates are getting to optimal campaign strategies, the more it will be true that “campaigns matter” in the sense that they could hurt themselves by choosing different strategies, but (obviously) they couldn’t help themselves by choosing differently.

Elections