[Note for bibliographic reference: Melberg, Hans O. (1997), The assumption of
rational selfishness - How can it be justified? When should it be rejected?, www.oocities.org/hmelberg/papers/971117.htm]
The assumption of rational selfishness
How can it be justified? When should it be rejected?
by Hans O. Melberg
Introduction
In his book The Armchair Economist Steven E. Landsburg writes that when faced with
apparently irrational phenomena like $2.99 pricing (why not $3?), long queues to get
tickets to a concert (why don't they rise the price?), and voting (seldom worth the
trouble) he
"... insists on finding rational explanation, no matter how outlandish, for all of
this apparently irrational behaviour" (1993, p. 11-12).
The topic of this short observation is how this rather dogmatic adherence to the
assumption of rationality can defended or attacked.
The costs and benefits of assuming rationality depends on the general aim of your
study. Three such aims are: to explain, to predict, and to aid policy formation.
Rationality and explanation
Take the first aim - to explain. Clearly, some people would argue, people are not always
rational, so we can explain more if we bring in variables like norms and emotions. Against
this the dogmatic rationalist might argue, first, that norms and emotions are simply not
important enough to be worth out time. In short, the increase in explanatory scope is not
enough to justify the inclusion of more variables (the parsimony argument). This defense,
I think, is weak. There are numerous studies documenting that we quite frequently act in
non-rational ways (see, for instance Kahneman et. al. Judgment under uncertainty).
There is, however, a second and stronger line of defense. While admitting the importance
of norms and passions in human behaviour, one could argue that the costs of bringing in
these variables are not worth the benefits.
To understand this, it is necessary to distinguish between an increase in
explanatory scope (being able to explain more phenomena) and explanatory reliability (the
uncertainty attached to your explanation). By bringing in norms one might increase
explanatory scope, but explanatory reliability might decrease since explanations based on
norms are often difficult to quantify and test empirically. This distinction makes it
clear that in order to justify the inclusion of more variables in addition to rational
selfishness, it is not enough to argue that we can explain more. We also have to
show that the new variables do not decrease the reliability so much that it was not worth
the increase in scope. In short, to justify inclusion of new variables you must show that
explanatory power increase (defined as scope multiplied by reliability; a multiplication
that needs to be weighted according to your preferences for reliability vs. scope). This
is a much more difficult task than simply showing that people are sometimes irrational or
that we can explain more if we bring in norms.
Rationality and prediction
Milton Friedman, in his (in)famous article "The methodology of positive
economics", argues that evidence of irrationality on a micro level does not
invalidate the rationality assumption. Take the following example: You might make a model
to predict the strength and angle of a pool player's shot. The central equations in this
model might be a quite complex mathematical expressions with many variables (e.g. the
strength of your shot depends on the position of all the other balls, your own ball, your
risk-type etc.). Now, the point is that you need not (and do not) claim that the pool
player actually goes through all these complicated calculations in his head before he
makes his shot. He is only acting "as-if" he went through the calculations.
Hence, arguments and proofs to the effect that this is not how the pool player decides how
to shoot, does not impress you. Only the accuracy of your prediction counts, not the
accuracy of the underlying assumptions.
To argue against the rationality assumption on these terms, one would have to develop
alternative models which are better at predicting behaviour. There is however, a slight
problem here, because one might ask whether these alternative models also have to be as
general as the rational maximization assumption. In short, is the criterion the accuracy
of your predictions in a very specific case, or the general accuracy across many instances
in many different fields? Although in doubt, I tend to think that the first is correct: If
I show that an alternative model is better at predicting a specific type of behaviour I
have succeeded in demonstrating the value of non-rational motivation in that case. (There
is probably, once again, a trade-off: generality vs. accurate predictions).
Rationality and Policy formation
Assume that you want to reduce the crime rate. The question is then how you should try to
do this. The answer, in turn, depends on your view of the driving forces of human
behaviour. If you think humans are mainly driven by norms, your answer to the question
"how to reduce crime" might be: "Try to change the attitudes of
people." If you are an economist like Gary S. Becker, who thinks that crime is a
result of cost/benefit considerations on part of the criminal, your answer might be:
"Increase the penalties to make it more unprofitable to commit crimes."
The paradox is that the sociologist might be right in assuming that attitudes and norms
are more important than cost/benefit considerations, yet Becker's recommendation might
still make more sense. The problem is this: Behaviour is probably driven by a mixed set of
forces - partly rational maximization of net benefit, partly non-rational attitudes,
norms, and emotions. Now, while it is easy to change the reward structure, it is very
difficult to change attitudes. We simply do not know how to change norms, and even if we
knew the causal connections it might take so long time as to make the result so uncertain
(in the long run many other variables will also change) that it is not worth the effort.
In sum, even if crime is mainly motivated by norms, for policy purposes might be best to
focus on crime as a rational activity. At least this focus give better policy results
(assuming that people are at least somewhat motivated by cost-benefit considerations).
At this point one should note the implications of the above argument for my first
theme - rationality and explanation. As an illustration consider voting: To explain the
high level of voting you clearly (?) need some kind of non-selfish and possibly
non-rational mechanisms. The benefits of voting are simply not worth the costs. However,
to explain variations in the level of voting, one might find that a rational model works
best. Whether this is just an "as-if" model, or whether it is really true that
deviations are really caused by rational considerations, I do not know. However, the
general point should be clear: Explaining change is different from explaining levels, and
different mechanisms might be used.
Conclusion
If the aim is to explain, one can defend rationality by putting emphasis on parsimony,
reliability, and change. If the aim is to predict, rationality can be defended on grounds
that it produces the most accurate predictions, regardless of whether this is how people
actually make up their minds. Lastly, the focus on incentives and rationality can be
justified from a policy perspective by arguing that it produces a more certain and larger
effect than slow and unreliable efforts to change non-rational motivations.
My own position is not as extreme as demonstrated in the quote from S. Landsburg. His
position amounts to a lexicographic (no tradeoff) and theoretical (empirical evidence of
irrational behaviour do not impress him) preference in favour of rationality (read the
quote in the introduction and note the "no matter how outlandish" part.) On the
contrary, I think it is an empirical argument - to be resolved by detailed empirical
studies. The purpose of the above discussion was to clarify the criteria for this
empirical argument: To find out exactly what we should demand of an explanation based on
non-rational mechanisms, and when it can be said to be better than an explanation based on
rational choice. To give more precise answer, one would have to do empirical studies
demonstrating the costs and benefits of the competing explanations. But this is the topic
of another essay, the present essay had a more limited scope: to clarify the general
criteria - the various categories of costs and benefits.
References
Landsburg, Steven E. (1993), The Armchair Economist, New York: The Free Press
[Note for bibliographic reference: Melberg, Hans O. (1997), The assumption of rational
selfishness - How can it be justified? When should it be rejected?, www.oocities.org/hmelberg/papers/971117.htm]