A job market paper I am excited about:
we use a set of weather and school holiday measures as instrumental variables for protest violence, a design made possible by disaggregated, micro-level protest event data constituting the universe of reported French protests from 1980-1995 and including an indicator for property destruction. Our results suggest that protest violence lowers the incidence of obtaining a concession.
From Emiliano Huet-Vaughn at Berkeley.
If you think that groups need to commit small-scale violence is to establish a credible threat of greater disruption, Emiliano’s result is not what you would predict.
I’m a newbie at this stuff, so please forgive if this is a naive question, but why does it not matter that the first-stage r-squared in this paper is less than .07? Can this really be a good instrument with an R2 that low? Is there a good discussion that anyone can direct me to about how to think about R2 in the first stage regression?
Totally fascinating — nice defense of the exclusion restriction in this piece. Only complaint is that the author doesn’t really distinguish between estimands — the IV estimate is a LATE, OLS estimate is an ATE — the incidence of violence due to weather and holidays is a only a fraction of overall protest violence, some of which may be effective in obtaining concessions, who knows. But at least we’ve pinned down a casual estimate for the variation we can explain with weather and holidays!
Though my priors incline me to agree with the premise, I am not sure whether this could be driven by unrelated factors such as concessions being less likely in the Summer (because no one is actually doing anything, the protest is less likely to get public attention, etc.)? Also, could it not be a case that less planned violence (i.e. violence that derives from the fact that the weather is high, and its is not raining) is less likely to be effective than planned violent protests?