Anchoring and Adjustment within Categorical Judgments of Saving Schemes: a Psychomimetic Approach

The aim of this paper is to study the interactions between categorization and decision making. While these two fields have mostly evolved separately, we introduce a model for categorization processes that is derived from a general decision-making heuristics. Such a model is able to account for both tasks involved in categorization, i.e. classification and category construction. One of the most important principle that support this work is the bounded rationality as introduced by Simon [Simon, 1955] in order to take into account the human limitations. We designed a connectionist network, called Categ ART, to implement our categorization model. We illustrate our method for an experiment where subjects had to categorize saving schemes. Categ ART produces automatically polynomials that can be interpreted as classification rules based on prototypes. The analysis of these rules allows us to suggest two anchoring and adjustment heuristics that seem to play an important role in our experiment. 1 Categorization theories While decision making and categorization theories have evolved mostly separately in cognitive psychology, we propose to build a bridge between these two fields and claim that it may benefit both. In general, categorization involves two main kind of processes: 1. a category construction process, where categories are constructed, using semantic and/or episodic knowledge and/or experiences, in order to represent and structure some sets of stimuli; 2. a categorical judgment process, sometimes referred as a classification process, which determines the appropriate category for a particular stimulus. Early research in categorization was dominated by the so-called classical view, whereby subjects were supposed to use formal and logical rules that gave a strict definition of the category [Bruner et al., 1956]. The study of more ill-defined categories dramatically revealed the limitations of this approach. The studies conducted by Rosch and her colleagues on natural categories suggested that categorization does not consist of hypothesis testing but rather learning and processing of relevant information [Rosch, 1973]. In particular, Rosch defended the idea