everything, therefore he explains nothing. Any strange behaviour, or behavioural change, can be
“explained” by positing a direct preference for the behaviour. Obviously, such explanations are
empty and meaningless unless there are constraints on the set of preferences that one can
reasonably put forward as an explanans.
Where should the constraints imposed on preference-based explanations come from? One
way to discipline such explanations is to apply knowledge about the causal determinants of
preferences. Historically, two obstacles have stood in the way of producing this knowledge.
First, the ability to infer causality from non-experimental data was quite limited, and the
econometric tools that helped overcome some of these limits (e.g. natural experiment
approaches, instrumental variables) have been widely applied only in the last two decades.
Second, economics and the social sciences generally had a poor understanding of the “nature” of
risk, time, and social preferences because the standard approach - expected utility theory,
exponential discounting theory, and the assumption of purely self-interested preferences - failed
to capture important preference patterns. It is not possible to properly measure preference change
if one lacks a full understanding of the fundamental patterns of preferences. What appears as a
change in preferences under the wrong theory may, in fact, reflect a stable preference.3
Independent empirical measures of preferences and of preference change are indispensible
for disciplining preference explanations. At the time when Stigler and Becker wrote their paper,
such preference measures were not available. However, progress in experimental and
econometric methodology over the last 20 years now makes it possible to measure preferences
much more precisely. Today, there is, for example, a relatively large literature on risk (e.g.
Bruhin, Fehr-Duda, and Epper 2010; Starmer 2000), time (e.g. Frederick, Loewenstein,
O’Donoghue 2002), and social preferences (e.g. Fehr and Schmidt 2003, Cooper and Kagel,
forthcoming) that provides constraints on the types of preferences that one can reasonably
invoke. The experimental and econometric tools available today enable researchers to get much
3 The following example illustrates this. Suppose that a subject is motivated by negative reciprocity, i.e. he responds
to hostile acts with hostility (Falk and Fischbacher 2006). In the ultimatum game this means that the subject - if in
the role of a responder - rejects low, unfair, proposer offers because the rejection punishes the proposer for his
unfairness. Suppose now that the responder is put in a different condition, in which the same low offer is generated
by a random mechanism, implying that the proposer is not responsible for the low offer. A responder motivated by
negative reciprocity will not reject this offer because the randomly generated low offer is not a signal of unfair
proposer behavior. If one assumes a wrong social preference model that neglects the true underlying responder
preference, the change in behavior across conditions appears to reflect a change in preferences while in fact, the
change in behavior is fully compatible with a stable preference for negative reciprocity.