Estimating Choices When Consumers Don’t See Everything

This paper describes a way to recover consumer preferences from choice records even when buyers do not see all product attributes. The authors identify a set of conditions in which observed choices are enough to learn underlying tastes, even if consumers search selectively and miss some information. They show that standard estimation approaches can understate the importance of attributes that consumers often fail to examine. The proposed alternative uses mathematical derivatives of choice probabilities to recover preferences and works across many search behaviors, including the case where buyers are fully informed. The authors also report a lab experiment where the approach recovered preferences when search was costly.

What the study examined

This work looks at whether it is possible to learn what people value from the choices they make when those choices are not always fully informed. In many settings, shoppers do not see every feature of every product before deciding. The paper frames the situation as a search model: some product attribute is hidden from buyers but observed by the researcher, buyers look at a subset of options, and they pick the item that gives the highest overall benefit among the items they did inspect.

The central question is whether recorded choice data alone can reveal true preferences about product attributes when consumers do not observe everything. The authors lay out a clear set of conditions under which such recovery is possible and propose an estimation approach that aims to remain valid despite incomplete information.

Key findings

• Under a stated set of assumptions, observed choices alone can identify preferences even if buyers do not examine all attributes. One key assumption is that when a buyer searches a given product, she also inspects products that are better on observable features than that product.
• Standard estimation methods can be biased in this environment. Specifically, the importance of the hidden attribute tends to be understated because consumers who never search some goods appear insensitive to variation in that attribute.
• The paper proposes an alternative estimation strategy based on cross derivatives of choice probabilities. This approach succeeds in recovering preferences regardless of the particular search protocol, so it remains valid whether consumers are uninformed or fully informed.
• The proposed approach includes common models as special cases, meaning it generalizes several standard settings while remaining applicable more broadly.
• The authors translate the theoretical idea into practical checks: the method naturally suggests tests that indicate whether consumers are effectively fully informed in the data.
• The approach is evaluated experimentally: a lab experiment shows that the method can recover preferences when search is costly to participants, providing an empirical demonstration of the procedure.

Why it matters

Many applied studies rely on observed choices to infer how much people value product features. When people do not look at everything before choosing, those inferences can be misleading. This research supplies a path to more reliable preference estimates in those situations by giving conditions and a concrete estimation technique that remain robust to selective inspection.

Because the method nests familiar models, it can be used both to reframe existing analyses and to test whether data behave like full-information cases. The lab evidence reported offers support that the technique performs in practice when individuals incur costs to search, suggesting usefulness for empirical work that studies consumer demand or forecasts responses to new information.