Towards the Methodological Turn in the Philosophy of Science - Mechanism and Causality in Biology and Economics - page 13

extrapolation (Steel 2007 ), is to investigate whether methods, hypotheses, and facts
of one field can be applied to another. David Teira and Julian Reiss's chapter
compares research methods of randomization in medical and economic sciences:
randomized clinical trials (RCTs) and randomized field experiments (RFEs),
respectively. Randomized controlled trials have long been regarded as the gold
standard for finding causal relations between interventions and experimental
outcomes. One reason, as Teira and Reiss point out, is that they provide mechanical
objectivity , meaning that randomized trials usually follow rigorous and transparent
rules so that the results are immune to the bias of subjective expert judgment. They,
however, argue that such objectivity is hard to come by. It is because the
participants both RCTs and RFEs could act so strategically to obtain their best
interest from the trial experiment that the supposed invariance of the controlled
environment breaks. Consequently, it is questionable to infer from the evidence the
causal connections between treatments and the results.
Both Hsiang-Ke Chao and Szu-Ting Chen's and Steel's chapters deal with the
issue of extrapolation that was conceptualized by Steel ( 2007 ), and they both deal
with the studies that can be categorized as freakonomics— using economic
principles to (surprisingly) explain a social phenomenon that was first thought to
be out of the realm of economics—popularized by the economist Steve Levitt. Steel
uses John Donohue and Levitt's ( 2001 ) controversial article of the causal relation
between legalized abortion and crime rate in the United States. Because there is no
direct evidence that can be used to check whether the hypothesis is correct, social
scientists support the US case by analyzing results derived from a survey of a
similar case that happened in the Scandinavian and Eastern European areas during
some periods in the twentieth century. But can we legitimately use evidence
obtained in a different time and a different area to support the local case? The
problem of extrapolation is analyzed by applying a mechanism-based approach—
what Steel calls “comparative process tracing.”
Another case where extrapolation could lead to possible explanation is the
“missing women” debate discussed by Chao and Chen, in which a biological
explanation—hepatitis B virus infection—for the abnormal inequality of sex ratio
at birth in Asia is offered by, extrapolated by, and instantiated by the sampling data
in the other area. But the biological explanation is claimed to be rejected by
economists who used Taiwanese population-level data. They find empirically that
cultural factors such as son preference are the cause for the missing women
phenomena. Chao and Chen argue that such empirical study does not necessarily
deny the existence of the underlying biological causal path, since what is observed
is a net causal result. Taking the net causal result as an evidence of ruling out minor
causal paths is equal to treating the underlying mechanism as a nontransparent box.
In this regard, extrapolations regarding evidence in different time and space can be
seen as complementary rather than substitutive.
This echoes our account of the ontology of mechanisms and causal structure: In
search of an explanation for a phenomenon, it is adequate to specify the mechanism
or identify the causal structure that underlies it. Science progresses thus from black
box to grey box and in turn to transparent box, but not the other way around.