Most of classical economic theory is built on the foundation of revealed preference. The guiding principle is that, whatever is going on inside her head, an individual’s choices can be summarized as the optimal choice given a single, coherent, system of preferences. And as long as her choices are consistent with a few basic rationality postulates, axioms, this can be shown mathematically to be true.
Most of modern behavioral economics begins by observing that, oops, these axioms are fairly consistently violated. You might say that economists came to grips with this reality rather late. Indeed, just down the corridor there is a department which owes its very existence to that fact: the marketing department. Marketing research reveals counterexamples to revealed preference such as the attraction effect. Suppose that some people like calling plans with lots of free minutes but high fees (plan A) and others like plans with fewer free minutes but lower fees (plan B). If you add a plan C which is worse on both dimensions than plan A, suddenly everybody likes plan A over plan B because it looks so much better by comparison to plan C.
The compromise effect is another documented violation. Here, we add plan C which has even more free minutes and lower fees than B. Again, everyone starts to prefer B over A but now because B is a compromise between the extreme plans A and C.
Do we throw away all of economic theory becuase this basic foundation is creaking? No, there has been a flurry of research recently that is developing a replacement to revealed preference which posits not a single underlying preference, but a set of preferences and models individual choices as the outcome of some form of bargaining among these multiple motivations. Schizonomics.
Kfir Eliaz and Geoffrey de Clippel have a new paper using this approach which provides a multiple-motivation explanation for the attraction and compromise effects. Add this to papers by Feddersen and Sandroni, Rubinstein and Salant, Ambrus and Rosen, Manzini and Mariotti, and Masatilioglu-Nakajima-Ozbay and one could put together a really nice schizonomics reading list.
Cheap Talk 
4 comments
Comments feed for this article
July 20, 2009 at 7:30 am
AndyfromTucson
I speculate that individuals actually do have a single coherent system of preferences, but its a system of preferences for certain psychological rewards, and unfortunately those psychological rewards have a (currently) murky relationship with events in the physical world (hence the Marketing Department and phone plan C). I suspect that once the relationship between objectively measurable events and mental rewards is better understood it will turn out that most behavior is quite rational and consistent once you accept that humans are seeking a certain optimum flow of mental rewards and not long term objective utility.
July 20, 2009 at 10:05 am
michael webster
This is very interesting, and I look forward to reading the papers. The attraction effect works if C is a close second to A, but clearly worse. The compromise effect also works by introducing A, a clearly superior but unaffordable option.
I would note that the attraction and compromise effect may be rational, if the introduction of the new outcome gets somebody to make a choice that they didn’t regret, and would have made no choice otherwise. Sort of like breaking the dam of action.
The other research project, less well known, is to use choice functions to simply describe the individual’s choices and then to construct set theoretic conditions which are necessary and sufficient conditions for a preference relation to be constructed. Ultimately preferences might simply reduce to choices and choice criterion.
July 21, 2009 at 2:15 pm
rbhui
michael webster: I’m not sure I see the difference between what’s normally done and your “other research project”. Could you elaborate?
July 21, 2009 at 2:51 pm
michael webster
In the foundations of economics, we ask which is basic: choice or perference?
A.K. Sen had a paper in the mid 70’s which answered the question by exploring the following two concepts.
1. Take a choice function, C(S), define some intuitive properties of choice, contraction consistency, property alpha, and expansion consistency property beta, then define a preference relation to be Rxy iff for x,y in S, x is in C(S).
What logical properties, if any does R have? Loosen alpha and beta, what logical properties, if any, does R have?
2. Take a preference relation Rxy, and define C(S,R) to be if x,y in S and Rxy, then x is in C(S,R).
What set theoretic properties does C(S,R) have, given certain logical constraints on R.
Now it turns out, that if R is the usual transitive preference relation, then C(S,R) satisfies alpha and beta, and vice versa.
So for some situations, it doesn’t matter whether you start with choice or preferences.
Unfortunately, it turns out that all you have to do is specify how C works on the pairwise choices to generate the entire choice function.
This is obviously incorrect for the two examples above since context or background matters.
And to require less than alpha will usually entail a failure of path independence, again not surprising in light of the contextual dependence.
Now, you have to come up with a generalization of maximization that satisfies your weakened conditions on the choice function.
If you think path matters, and is dictated by the use of choice criterion along the choice path, then a) not all choices are relevant, but b) you need a concept rational choice criterion to rule out deviant decision paths.
In the attraction case, you might model the situation with n+1 decision criteria, all the ones that A is superior to C on, and the null criteria – cannot choose.
Your decision path is simply: if not null, maximize on n. If null, then flip a coin or don’t choose. The decision path over A,B, and C, will select A, but the decision path over A and B will either be empty or both, depending on how you interpreted the null criteria. The decision path over A and C will select A, and C and B will be treated the same way A and B was.
Property alpha, contraction consistency, is violated by this choice function.
Path independence is violated by this choice function.
WARP is violated, as is binariness.
But, all in all, the choice function seems pretty reasonable and so a more general notion of rationality is required if it is to encompass this choice behavior.
So, you can either move away from the notion of one single preference relation or move away from the notice of a simple choice function. And obviously, there are going to be important points of contact between the two approaches.
I am looking forward to reading this paper in detail.