Context-dependent choice and evaluation in real-world consumer behavior
Human information processing is naturally limited. To compensate for these limitations, humans rely on contextual information to inform their choices. A classic example of such context-dependence occurs in value-based choice: the relative value of an option depends not only on the option in question but also on the other options in the choice set, or context. While context effects of this sort have been observed primarily in small-scale laboratory studies where choice sets are tightly constrained, it is unknown whether context takes hold of choice “in the wild”. Here, we demonstrate the generality of context-dependent valuation by analyzing a massive real-world restaurant rating dataset (Yelp.com; 4.2 million ratings). We find that Yelp users make fewer ratings-maximizing choices in choice sets with higher overall average ratings. This behavior is quantitatively well-described by a divisive normalization model of choice, wherein the value of available options is scaled to the average of options in a choice set. We follow these analyses up with data from an online experiment, in which we (a) replicate the choice pattern seen in real-world Yelp users and (b) demonstrate that participants’ expectations of an option’s quality are also context dependent, in accordance with the ratings of the options the choice set, even in the absence of explicit choice. The experimental choice data was again well-characterized by a divisive normalization model of valuation. Taken together, we find compelling evidence for context-dependent valuation in behavior, manifesting both in users’ real-world and hypothetical choices and expectations.
Interesting work! I'm wondering how some of the effects might relate to (anticipated) regret. If the choice set contains many high-valued options, and the average ratings are a noisy reflection of the true value of an option, the probability that the highest rated option is indeed best will decrease when other options have a close average rating. I...
Cite this as: