A multi-self model of self-punishment
Angelo Enrico Petralia
TL;DR
The paper introduces a multi-self model of choice in which a DM may punish herself by distorting her true preferences through harmful distortions that move the top $i$ alternatives to the bottom in reverse order. It defines Harm($\rhd$) and the degree of self-punishment $sp(c)$, and provides necessary and sufficient axioms and characterizations for estimating $sp(c)$ from observed choices, including weakly and strongly harmful regimes. It shows that weakly harmful behavior captures common biases such as second-best procedures and decoy effects, while strongly harmful behavior, which corresponds to maximal self-punishment, becomes increasingly prevalent as the choice set grows, culminating in an inconsistency characterization. The framework sits within the multi-self literature, offering testable restrictions and partial identification of the underlying preferences, and points to stochastic, context-sensitive, and dynamic extensions for future work.
Abstract
We investigate the choice of a decision maker (DM) who harms herself, by maximizing in each menu some distortion of her true preference, in which the first i alternatives are moved, in reverse order, to the bottom. This pattern has no empirical power, but it allows to define a degree of self-punishment, which measures the extent of the denial of pleasure adopted by the DM. We characterize irrational choices displaying the lowest degree of self-punishment, and we fully identify the preferences that explain the DM's picks by a minimal denial of pleasure. These datasets account for some well known selection biases, such as second-best procedures, and the handicapped avoidance. Necessary and sufficient conditions for the estimation of the degree of self-punishment of a choice are singled out. Moreover the linear orders whose harmful distortions justify choice data are partially elicited. Finally, we offer a simple characterization of the choice behavior that exhibits the highest degree of self-punishment, and we show that this subclass comprises almost all choices.