Mechanisms and Extrapolation in the Abortion-Crime Controversy - Mechanism and Causality in Biology and Economics - page 181

arguing for the selection mechanism is easy to appreciate given that there are no
known records in the USA of women who desired to terminate pregnancy but were
denied legal right to do so. 2 However, there are studies using USA data that bear on
the selection mechanism less directly. For example, Donohue and Levitt cite studies
finding that women who terminate pregnancies are, not surprisingly, significantly
more likely to be in circumstances known to adversely affect the life prospects of
children, such as being an unmarried teenager (Levine et al. 1999 ). Similarly, it is
possible to study overall effects of abortion legalization on the well-being of birth
cohorts. Gruber et al. ( 1999 ) do this and find that post-1973 birth cohorts exhibited a
marked reduction in a number of adverse factors, including single-parent

households and poverty.
In addition to tracing a mechanism from legalized abortion to the 1990s decline
in crime rates, Donohue and Levitt argued for their hypothesis on the basis of
statistical data concerning abortion and crime rates. Donohue and Levitt's statistical
argument turns on three main points. First, since five states in the USA (Alaska,
California, Hawaii, New York, and Washington) legalized abortion around 1970,
Donohue and Levitt's hypothesis predicts that the declines in the early legalizing
states began about 3 years before those in the other states. And since abortion rates
continued to vary between states after Roe v. Wade, Donohue and Levitt's hypoth-
esis predicts that states with higher abortion rates would experience greater
reductions in crime. Secondly, because teenagers are much more likely to commit
property crimes than violent crimes, Donohue and Levitt's hypothesis predicts that
the decline in crime rates would begin with property crime and then spread to more
violent crimes. Finally, given the “age in” impact of abortion legalization, Donohue
and Levitt's hypothesis predicts that the decline in crime would disproportionately
result from cohorts born after legalization. Donohue and Levitt argue that the data
support all three of these predictions. 3
Besides its intimate link to a perennial “hot button” social issue, the above
complex argument makes a fascinating case study in social science methodology.
Numerous philosophers and social scientists have discussed the role of mechanisms
in providing support for causal claims in social science (Elster 1989 ; George and
Bennett 2005 ; Hedstrom and Swedberg 1999 ; Kincaid 1996 ; Little 1992 , 1998 ;
Reiss 2007 ; Steel 2004 ). Two features of the Donohue and Levitt study make it
interesting and relevant in connection to these discussions: (1) the interconnected
role of mechanisms and macro-level statistical data in supporting its central
hypothesis and (2) the role of extrapolation in providing evidence for a mechanism.

2 No doubt such women exist. But since there was never any official process of applying for
permission to terminate a pregnancy in the USA, there is no way to identify which women these
were.
3 Several critics have challenged these statistical arguments (Joyce 2003 ; Foote and Goetz 2008 ).
See Donohue and Levitt ( 2004 , 2008 ) for replies.