The short answer: a little of both.
I’m attending the NYU CESS conference, and the highlight of the day so far (and a contender for the new alpha paper on election monitoring and election day fraud) is this:
We address these questions by studying observers’ effects on two markers of fraud—overvoting (more votes cast than registered voters) and unnaturally high levels of turnout — during Ghana’s 2012 presidential elections. Our randomized saturation experimental design allows us to estimate observers’ causal effects and to identify how political parties strategically respond to observers.
We show that observers significantly reduce overvoting and suspicious turnout at polling stations to which they are deployed.
We also find that political parties successfully relocate fraud from observed to unobserved stations in their historical strongholds, where they enjoy social penetration and political competition is low, whereas they are not able to do so in politically competitive constituencies.
I learned more about elections listening to Susan Hyde’s 10-minute discussion of the paper than I’ve learned in several years of reading. Sadly not easily summarized or online.
Nonetheless, full paper here.