May 26, 2011 § Leave a comment
Card and Dahl: Domestic male-on-female violence increases after upset NFL losses.
The basic results of a Poisson regression are in this table. I would have preferred if the model had been specified in terms of categories instead of interactions. This would give you impacts on violence (multiplying their coefficients by 100, so they can be approximately interpreted as percentage effects; positive means more violence) as follows:
(“Expected close” means a point spread of less than four points.)
Clearly the upset loss is far away from the other numbers. They give standard errors for the model specified as interactions; I’d guess that the SEs for the graphed specification are about +/- 4.