Mechanisms and Extrapolation in the Abortion-Crime Controversy - Mechanism and Causality in Biology and Economics - page 177

been prevented by abortion were it available—are more likely to be born into
circumstances that increase the chance that they will engage in criminal behavior
upon entering early adulthood. One line of evidence cited by Donohue and Levitt
for this hypothesis consists of data from Scandinavia and Eastern Europe, where for
some periods of the twentieth century, women desiring an abortion were required to
receive legal permission. Thus, data concerning the life outcomes of children whose
mothers requested but were denied abortions are directly relevant to the mechanism
proposed by Donohue and Levitt. The use of data from Scandinavia and Eastern
Europe to support a claim about a mechanism in the United States is an example of
an extrapolation , that is, using a causal relationship found in one context as a basis
for an inference about causal relationships in another that may differ in a number of
relevant respects. Extrapolation is essential for imparting significance to scientific
research beyond the confines of the original investigation, yet relatively little
philosophical attention has been devoted to the question of how and under what
circumstances such inferences are justifiable.
In this chapter, I examine the role of extrapolation in Donohue and Levitt's study
from the perspective of some recent attempts to clarify the underlying logic and
principles of such inferences (Pearl and Bareinboim 2011 ; Steel 2008 ). This project
is intended both to explicate the methodology of the Donohue and Levitt study as
well as to advance philosophical understanding of extrapolation. Three main
philosophical themes emerge from the discussion below. First, several different
types of causal claims might be at issue in an extrapolation—including claims about
mechanisms and probabilistic causal effects—and these distinctions matter for
methodology because different conditions may be required to support extrapolation
in each case. Secondly, scientific study of a phenomenon typically generates
evidence at a variety of levels of aggregation, and this has important implications
for extrapolation. The Donohue and Levitt study, for example, discusses data
concerning psychosocial effects of unwantedness on individual children in addition
to comparisons between rates of abortion and crime among different states in the
USA. As explained in the final section, data from the macro-level can provide a
means of indirectly testing assumptions about similarities of the model and target
made at the level of mechanisms. The third and final point follows on the heels of
the second. Like almost all other scientific inferences, extrapolations are normally
components of a complex web of interrelated evidence that must be considered
together in assessing a hypothesis. Thus, to focus on whether extrapolation alone
could have established a conclusion in a real scientific example (see Lafollette and
Shanks 1996 ) would be, more often than not, to miss the point. In real-life cases, the
question can only be whether and to what extent the extrapolation strengthens the
overall body of evidence.
I begin with a synopsis of the Donohue and Levitt study. Next in Sect. 2 ,I
consider the role of multiple levels in the study, in particular, the connection
between the psychosocial mechanism concerning unwantedness and criminal
behavior and the state-level comparisons concerning abortion and crime rates. I
introduce the notion of a “scale-up model” that links a micro-mechanism to a
macro-level statistical

relationship. In Sect. 3 , I present a framework for