Context-sensitive rationality: Choice by salience

Abstract

We describe a context-sensitive approach to individual choice, in which the explanation is provided by a family of linear orders indexed by all available items. Selection from a menu is then recovered by the classical maximization paradigm, subject to the constraint that the justifying rationale must be indexed by an item of the menu. This approach allows us to pursue two complementary goals: (1) a fine classification of all possible choices into classes of rationality, and (2) a bounded rationality model based on an ordinal notion of salience. Concerning (1), we refine the context-free model of rationalization by multiple rationales, partitioning the class of all choice functions on $n$ items into $n$ classes of rationality. The least rational class is expressive of a moody behavior, which is rare for small $n$, but prevailing for large $n$. Concerning (2), we enrich our framework by a binary relation of salience, which guides the selection process. Upon requiring that all rationales associated to equally salient items coincide, choice is explained by appealing to the unique linear order indexed by a maximally salient item of the menu. Choice by salience is a specification of choice with limited attention. Numerical estimates show the sharp selectivity of this model of bounded rationality.