>higher abortion rates could be the result of other factors that themselves
>relate to a later decline in crime. Obviously, multi-variate regression
>tries to filter out all those factors and the assumption is that given two
>cities with all factors filtered out, if abortion is the only thing left
>standing, you have proved causality.
Well, if we try to assign partial coefficients to unemployment, poverty and fewer teenagers, their retort seems obvious (though I can't access their paper from here): abortion accounts for the reduction in teenagers as % of pop. (esp. of the type prone to crime), unemployment and poverty. It seems that they will claim that these other 'factors' were relatively or first reduced in those states early to legalize abortion--though I would be surprised if their evidence turns out be very strong in this regard.
Moreover, this study would have no counterfactual import ("Extrapolating our estimates out of sample to a counterfactual in which there were no abortions, crime rates might be 10-20 percent higher than they currently are with abortion") without the operation of a socio economic machine in which those surplus to capital's demands are tracked for poverty, unemployment and criminalization. Yet how does one index such a machine in the regression analysis--much less determine its durability against evolutionary transformation or external modification on the basis of such statistical analysis, though the discovered 'causal' relationship between abortion and reduced crime would have no relative stability or counterfactual import without the durability of such a machine?
At any rate, what kind of social machine must be in place for an allowance of abortion twenty years earlier to 'cause' a reduction in crime today?
yours, rakesh