You often hear economists (proudly) say that they don't take stated preferences seriously, but instead rely on observed preferences. Correct?

To me both are problematic if you want to understand preferences, since people often do things they don't want to do. Example: I prefer to go to bed at 22:00 but I observe that I very often get caught up and don't do this.

Decision making of individuals is complicated and I think everyone should be careful interpreting observerd preferences?

My question is: Can anyone provide further points thinking about this issue and/or articles about the interpretability of observed preferences?

I am aware there exists micro-models with "Hyperbolic discounting" making preferences time-inconsistent. I am looking for some more general discussion and frameworks.

  • $\begingroup$ What is the tag "soft" supposed to be? $\endgroup$ – FooBar May 16 '15 at 17:58
  • $\begingroup$ Was supposed to be 'soft-question'. Maybe I'm wrong: I thought it fitted since the question was not technical and did not have a single correct answer. $\endgroup$ – snoram May 16 '15 at 18:05
  • $\begingroup$ Feels like a meta tag to me: blog.stackoverflow.com/2010/08/the-death-of-meta-tags $\endgroup$ – FooBar May 16 '15 at 18:42
  • $\begingroup$ I don't think it was invented as a meta tag. Anyway I don't know how this site is governed really. Here is some discussion: meta.engineering.stackexchange.com/questions/229/… $\endgroup$ – snoram May 16 '15 at 20:37

In my view, it is useful to distinguish two phenomena:

  • individuals have time-inconsistent preferences (this fits your example). There is a huge literature on that. The common practice is to evaluate welfare from the ex ante perspective, thereby assuming that people have well-defined preferences over the long-run, but that their choices might deviate from it because of immediate impulses. In that case, the true preference can be elicited by asking the individual to choose in advance. Some authors question the ex ante perspective, see for instance "The Welfare Economics of Default Options in 401(k) Plans" by Bernheim, Fradkin & Popov.

  • individuals make mistakes because of cognitive limitations. In this situation, it is very hard (impossible?) to recover the true preferences from the choices. At best, we can have a measure of the extent to which the individual makes consistent decisions. You can have a look for instance at "The Money Pump as a Measure of Revealed Preference Violations" by Echenique, Lee & Shum.

Overall, it is true that economists usually reject stated preferences, and there are many good reasons for that. It does not necessarily imply that we assume that observed choices are perfect; just that we do not have a better alternative to elicit people's preferences. However, some researchers are now studying how we could take advantage of non-choice data: (i) from a neuroscience perspective (many recent papers by Antonio Rangel and Colin Camerer for instance); (ii) using stated preferences, like "Non-Choice Evaluations Predict Behavioral Responses to Changes in Economic Conditions" by Bernheim, Bjorkegren, Naecker & Rangel.


You've walked into a serious philosophical landmine! To begin, this state of disharmony between "actual action" and "right action" is called akrasia. Your bedtime story is one of akrasia, and I suspect Aristotle might tell you your action is morally wrong.

The reasoning why we disregard unobserved preferences is grounded deep in the philosophical history of economics. Economics has adapted the position that we do not inherently care about a person's feelings. We only care what they do about it. Do they actively apologize? Do they feel guilty? Do they adjust their future behavior? Do they offer concessions in some other real, overt, tangible, way?

Economics is following the consequentialist philosophy that defines the discipline, utilitarianism. In this understanding of ethics, a person's invisible preferences are irrelevant, but their actions in the real world are relevant. The consequence matters, not the motive.

Lastly, our goal is to predict future observed behavior. Hopes and dreams are not in accordance with actual action, people are often in a state of akrasia. So why would we study the unobservables? Dreams and hopes remain as immaterial and ultimately pointless in comparison to that which is actually done. I admit there are cases where stated preferences are the topic, but to sell the idea as economics those preferences must eventually be expressed as action. For any paper on preferences, there will be a chain of literature that you can trace back explaining how feelings eventually lead to tangible action. What use is a dream deferred?


Here are two quick references, which attack your question from two sides:

  1. On the behavioral side, it seems that people who make statements about preferences systematically choose differently than people who are forced to interact with the question more extensively before making choices. Specifically, when people people simply make a statement about a choice over risk vs simply "trying the risky choice out," the former display much more risk aversion than the latter. See this paper and the literature cited therein for examples of people displaying different risk aversion when asked vs when forced to interact with a simple simulation of a risky process. That paper itself aims to apply these insights to trying to help people make better investment decisions. So why do people appear to act differently than they say they will? ? There are many possible stories -- one possibility is that thinking is costly, and when you simply answer a question, you aren't actually doing an enormous amount of "thinking calculation" work. However when you must actually make a real choice, the payoff is eminent and you actively mentally chew on the problem more -- and having done more actual mental calculations, you come up with a different answer. This is likely an open research question which would benefit from more applied (experimental) work.

  2. A little further afield: one possibility on the theoretical side is that there may be people who don't fully order their preferences over different baskets of consumption. In simple verbal questions about choices, there may be some easy mis-communication about choices then. However when choices are actually observed, they can still be ordered under a particularly constructed preference ordering, such that they still fulfill the appropriate theoretical requirements to be useful for much of economic theory (eg. building utility functions). See the first question here, and answer here. This is actually a very fun result from basic preference theory. (I'll note that if the consumer agent is careful and truthful with their wording, there shouldn't be a problem between words/actions, as both should be useful in the above-referenced preference orderings. Also, the observer needs to be careful as well in this context.)

If you want to chew on the "thinking is costly so I won't do a lot of it until I need to" approach, Ariel Rubinstein's "Modeling Bounded Rationality" takes an interesting approach. He attacks the question from the perspective of how to actually model costly calculation, from some interesting perspectives (including computational theory from computer science). It certainly isn't the only reference on this topic, but an interesting read and a good place to start. I will warn you that this is not a field that has been widely explored (that I am aware of).

You should also look into the experimental literature cited in the first paper I linked to above. Many interesting topics discussed there.


The idea that revealed preference is preferable to stated preference is pretty common. It makes a lot of sense. That said, there is a lot of value in stated preference. It does offer something that RP does not.

  • It can be used to run estimate demand for products where revealed data does not exist. This is common in health economics to assess the value of an as yet, undeveloped product.

  • Methods such as contingent valuation can be used. It can provide a lot more information about the shape of a distribution.

  • SP allows variation in attributes. (What if this product was $10 cheaper. Would you buy it then?)

  • RP results may reflect market equilibrium. For example, under a specific choice set, individuals may choose a certain bundle. Not much can be gained about the proximity of this to other similar bundles.

“Factors that are most important to consumers will often exhibit the least variation due to the natural forces or market equilibrium. Their importance might therefore be difficult to detect with revealed-preference data” Train (2003)

There is a lot to be gained for stated preference data. Stated Preference data has enormous application in economics and in many cases is a better source to tackle certain problems.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.