“My own behaviour baffles me. For I find myself not doing what I really want to do but doing what I really loathe.” Saint Paul What behaviour can be explained using the hypothesis that the agent faces temptation but is otherwise a “standard rational agent”? In earlier work, Gul and Pesendorfer (2001) use a set betweenness axiom to restrict the set of preferences considered by Dekel, Lipman and Rustichini (2001) to those explainable via temptation. We argue that set betweenness rules out plausible and interesting forms of temptation including some which may be important in applications. We propose a pair of alternative axioms called DFC, desire for commitment, and AIC, approximate improvements are chosen. DFC characterizes temptation as situations in which given any set of alternatives, the agent prefers committing herself to some particular item from the set rather than leaving herself the flexibility of choosing later. AIC is based on the idea that if adding an option to a menu improves the menu, it is because that option is chosen under some circumstances. From this interpretation, the axiom concludes that if an improvement is worse (as a commitment) than some commitment from the menu, then the best commitment from the improved menu is strictly preferred to facing that menu. We show that these axioms characterize a natural generalization of the Gul-Pesendorfer representation.

Abstract We study how strategic interaction and cooperation are affected by the heterogeneity of cognitive skills of groups of players, over consecutive plays of repeated games with randomly matched opponents using Prisoner’s Dilemma as stage game. We observe overall higher cooperation rates and average final payoffs in integrated treatment groups—where subjects of different IQ levels interact together—than in separated treatment groups. Lower IQ subjects are better off and higher IQ subjects are worse off in integrated groups than in separated groups. Higher IQ subjects adopt harsher strategies when they are pooled with lower IQ subjects than when they play separately. We demonstrate that this outcome should be expected in learning and evolutionary models where higher intelligence subjects exhibit lower frequency of errors in the implementation of strategies. Estimations of errors and strategies in our experimental data are consistent with the model’s assumptions and predictions.

Abstract We provide two characterizations, one axiomatic and the other neuro-computational, of the dependence of choice probabilities on deadlines, within the widely used softmax representation $$\beginalign* p_t\left( a,A\right) =\dfrace^\fracu\left( a\right) λ\left( t\right) +α\left( a\right) \sum_b\in Ae^\fracu\left( b\right) λ\left( t\right) +α\left( b\right) , \endalign*$$ where $p_t\left( a,A\right)$ is the probability that alternative a is selected from the set A of feasible alternatives if t is the time available to decide, λ is a time-dependent noise parameter measuring the unit cost of information, u is a time-independent utility function, and α is an alternative-specific bias that determines the initial choice probabilities (reflecting prior information and memory anchoring). Our axiomatic analysis provides a behavioural foundation of softmax (also known as Multinomial Logit Model when α is constant). Our neuro-computational derivation provides a biologically inspired algorithm that may explain the emergence of softmax in choice behaviour. Jointly, the two approaches provide a thorough understanding of softmaximization in terms of internal causes (neuro-physiological mechanisms) and external effects (testable implications).