Mechanisms and Extrapolation in the Abortion-Crime Controversy - Mechanism and Causality in Biology and Economics - page 182

2 Mechanisms and Scale-Up Models

Tracing a mechanism can strengthen a causal inference in several ways. It allows
for additional tests of the hypothesis and creates further lines of relevant evidence.
In addition, experimental or quasi-experimental studies of important aspects of the
mechanism may be possible even when no such studies are feasible at the macro-
level (Steel 2011 ). These points are illustrated in the Donohue and Levitt, wherein
studies of women who requested but were denied abortion constitute “natural”
experiments of the effects of being born unwanted on child development. But in
order for inquiries concerning mechanisms to support macro-level causal claims,
there must be some reasonably clear connection between the mechanism, which
typically involves individual behaviors and interactions, and the macro-level,
which makes general claims about a population. What I will call a “scale-up
model” is used to forge a link between micro and macro. A scale-up model specifies
how an inference can be made from mechanisms to macro-level phenomenon.
Donohue and Levitt in fact explicitly present a scale-up model, which they describe
as a “back-of-the-envelope” calculation that provides a “crude prediction of the
impact of legalized abortion on crime” ( 2001 , p. 389).
Their approach combines research on how legalized abortion affects the compo-
sition of birth cohorts as judged by four factors—race, teenage motherhood,
unmarried motherhood, and unwantedness—along with research on the impact of
each of those factors on criminality (ibid). The model then breaks down 1990
Census data using the eight possible combinations of the first three factors (appar-
ently assuming that race is either white or black), finding the proportion of the
population in each group. They then use estimates from a study of the impact of Roe
v. Wade on birth rates (Levine et al. 1999 ) to decide what those proportions would
have been if abortion had not been legalized. Next, Donohue and Levitt use
previous research to assign crime rates for each cell (e.g., for children of a white
unmarried teenage mother). Thus, the effects of abortion on crime mediated by the
first three factors can be estimated by summing the proportion-weighted crime rates
for each cell with two different proportion weightings: one based on 1990 Census
data and the other set based on estimates of what these proportions would have been
if abortion had not been legalized. Since unwantedness is not measured in the data,
Donohue and Levitt estimate the number of unwanted births by assuming that 75 %
of unwanted births would be prevented by abortion. 4 They then extrapolate the
result that children born from unwanted pregnancies are about twice as likely to

4 That is, number of abortions

number of unwanted births (if abortion had not been
legalized). Thus, given the number of abortions (for which there is data), the number of unwanted
births in the hypothetical nonlegalization scenario can be estimated. The number of unwanted
births in the actual legalization case would just be this number minus the number of additional
abortions performed due to Roe v. Wade.

75 % ¼