Sunday, September 05, 2010
Jonah Lehrer recently wrote about the identifiable victim effect. In the piece he talks about the work of Paul Slovic and identifiable victims. I appreciate the work that Slovic has done. I have met and interacted with some of Slovic's co-authors and think the work is great. The idea behind identifiable victim bias is that people respond differently when they see one 'identifiable' victim rather than a bunch of statistics indicating the 'true' depths of poverty. For example, people give £3.50 in the one case and less than half that in the other. Lehrer says the following about the different reactions that people have:
"Of course, this is a deeply irrational reaction. We are much less interested in helping a victim – we only want to help the victim. (This bias is known as the identifiable victim effect, since it suggests that we react much more strongly when the victim can be specified.) Why do we this? Because human charity is ultimately rooted in our compassionate feelings, and not in some rational, utilitarian calculations. We are not Vulcans."But, Lehrer's claim that it's 'irrational' for us to respond in one way to an emotive stimulus and another to an intellectual one is false. Again, this abuses the idea of 'rationality'. What Lehrer is actually saying is that people behave in a way that he interprets as inconsistent, and thus irrational. I would argue, instead, that the behaviour is not inconsistent. That is, in an agent's preference rankings they will (or may, depending on the agent) consistently prefer - and rank higher - cases in which they can identify victims rather than observe the statistics of a problem. This can be a consistent response. It just seems as though Lehrer wants to project a certain kind of value on what an individual's ranking should be and, because he thinks its inconsistent, infer it is irrational. Were it inconsistent for all agents at all times, it would display irrational behaviour.
He then goes on to examine a new paper in which some people are 'less irrational' than others. That is, some people who tend to be 'more analytical' also tend not to 'fall victim to' the identifiable victim bias. OK, we're getting somewhere. What this says to me, instead, is that people who value analysis are going to approach and rank their preferences in such a way that they would like their emotional and analytical selves to be consistent. That is their preference rankings mirror the correlate of tendency to analyse. What Lehrer does not say is that both groups could be rational and satisfying preference functions that are indeed consistent, but that he does not know what those preference sets are. If we accepted this, and measured at the level of the population, we might be able to infer what those preference sets are from the behaviours that people display, rather than labelling one group as rational and the other as not.
The problem here is one of identification. That is, can we identify at the level of the individual ex ante and without additional information what a given individual will rank higher? No, we can't. We need their preferences to be revealed by their behaviour. OK. Once we have data, though, might we be able to list the probabilities with which an individual's preferences might fall into one class or another? Yes. In fact, I'd say that would be progress. Admitting the possibility of agents with heterogeneous preference sets, that is, people who rank things differently, and therefore rank their responses to things like statistics and identifiable victims differently, is a step toward improving the behavioural sciences. The point is the idea of rationality is non-tautologous as long as we have accurate descriptions of the classes of preferences that individuals might have. If I understand the problem accurately, then this is indeed what many researchers are trying to consider when it comes to other-regarding and self-regarding preferences. Some agents seem to display other-regarding preferences. Other agents seem to display self-regarding preferences. Given no knowledge about an individual we cannot say with certainty which kind of preferences they have. But, as more data mounts, we should be able to say that individuals with certain kinds of characteristics are more likely to display a certain set of preferences given a certain set of conditions. We could then provide probabilities with which a given individual about whom we have limited information will display one set of preferences or another. Rationally. Similarly, once we have better and more comprehensive data about how states affect behaviour, then we can improve our understanding of state-based and endogenous preferences (see Bowles's (2006) book Microeconomics: Institutions, Behavior and Evolution for a decent summary of these ideas).
To me, that is more interesting than identifiable victim bias being 'irrational'. I must also leave this post with a caveat: I do not believe people are fully rational. Moreover, I feel studying psychology can improve our understanding of agents in the settings that they are accustomed to and to understanding proximate motives for behaviour (see Greg Mankiw's NYT column from today where he urges students to study psych, along with one or two other subjects). But I also think that inaccurate commentary on what rationality is and means does not help us to improve the behavioural sciences. Moreover, I think a proper understanding of rationality and its applications in the behavioural sciences serves to improve them rather than impoverish them (consistent, for example, with this recent post at Crooked Timber).