What the study found
The authors identify a sufficient condition under which choice data alone can be used to identify consumer preferences even when consumers are not fully informed. They show that an alternative method based on cross derivatives of choice probabilities can recover preferences regardless of the search protocol, and that this approach nests several standard models, including full information.
Why the authors say this matters
The study suggests that this method can address bias in canonical models when consumers do not observe all relevant information. The authors conclude that it can also provide natural tests for full information and can be used to forecast how consumers will respond to additional information.
What the researchers tested
The researchers considered a search model in which the attribute hidden to consumers is observed by the econometrician. They assumed that if a consumer searches good j, she also searches goods that are better than j in terms of the non-hidden component of utility, and that consumers choose the searched good with the highest overall utility. They also verified their approach in a lab experiment involving costly search.
What worked and what didn't
Under the stated conditions and additional mild restrictions, the alternative method succeeds in recovering preferences regardless of the search protocol. The authors note that canonical models will be biased because the value of the hidden attribute will be understated when consumers do not search some goods. The lab experiment is reported to verify that the approach succeeds when consumers engage in costly search.
What to keep in mind
The abstract presents the result as holding under a sufficient condition plus additional mild restrictions, so the scope is conditional. It also says that canonical models are biased in this setting, but the abstract does not describe all limitations of the method or the experiment.
Key points
- The paper says choice data alone can identify consumer preferences under a sufficient condition, even when consumers are not fully informed.
- A cross-derivatives-based method is described as robust to the search protocol and compatible with standard models, including full information.
- Canonical models are said to be biased because the value of a hidden attribute is understated for goods consumers do not search.
- The authors say the method can be used to test for full information and forecast responses to additional information.
- A lab experiment with costly search is reported to verify the approach.
Disclosure
- Research title:
- Preference recovery can be robust to consumer search
- Authors:
- Jason Abaluck, Giovanni Compiani, Fan Zhang
- Publication date:
- 2026-01-07
- DOI:
- 10.1086/740223
- OpenAlex record:
- View
Get the weekly research newsletter
Stay current with peer-reviewed research without reading academic papers — one filtered digest, every Friday.