The most clever insight in the paper is that you can combine data from different prediction markets to estimate an interesting conditional probability -- the probability that a primary candidate will win the general election conditional on winning the nomination. (If p(G) is the probability of winning the general election and p(N) is the probability of winning the nomination (both of which are evident in prediction market contract prices), p(G|N) -- the probability of winning the general election if nominated -- can be calculated as p(G)/p(N).) In the first part of the paper, the authors focus on how individual primaries in the 2008 election affected this conditional probability for each candidate. This is interesting because classic theories in political science posit that primary elections force candidates to take positions that satisfy their partisans but hurt their general election prospects by making it harder for them to appeal to the electoral middle. If that is the case, then ceteris paribus one would expect that the conditional election probabilities would have gone down for Obama and Clinton each time it looked like the primary season would become more drawn out -- which is what happened as results of several of the primaries rolled in.

As it turns out, p(G|N) didn't move much in most primaries; if anything, it went up when the primary season seemed likely to extend longer (e.g. for Obama in New Hampshire). Perhaps this was because of the much talked about positive countervailing factors -- i.e. the extended primary season actually sharpened each candidate's electoral machines and increased their free media exposure. Of course, Malhotra and Snowberg have no way of knowing whether the binding effect of primaries exists and was almost perfectly counterbalanced by these positive factors, or whether none of these factors really mattered very much. [...]

The second part of the paper explicitly considers the problem of assessing how "surprised" the prediction markets were in particular primaries (without explaining why this was not an issue in the first part), and employs a pretty ad hoc means of upweighting effect estimates for the relatively unsurprising contests. Some kind of correction makes sense but it seemed to me that the correction was so important in producing their results that it should be explained more fully in further revisions of the paper.

We have no reason to thank the University of Chicago Press.