Self-Control as Value-Based Choice

Abstract

Self-control is often conceived as a battle between "hot" impulsive processes and "cold" deliberative ones. Heeding the angel on one shoulder leads to success; following the demon on the other leads to failure. Self-control feels like a duality. What if that sensation is misleading, and, despite how they feel, self-control decisions are just like any other choice? We argue that self-control is a form of value-based choice wherein options are assigned a subjective value and a decision is made through a dynamic integration process. We articulate how a value-based choice model of self-control can capture its phenomenology and account for relevant behavioral and neuroscientific data. This conceptualization of self-control links divergent scientific approaches, allows for more robust and precise hypothesis testing, and suggests novel pathways to improve self-control.

Figure 1

Value-based choice model of self-control. The cumulative subjective value of each each response option (middle column) is a weighted sum of value inputs based on the option’s attributes (left column). Example attributes for a choice option include primary rewards, effort costs, social acceptance or rejection, and self-consistency and -verification. The subjective value integration is not strictly rational, but instead is modulated by a number of choice “anomalies” such as the tendency to discount delayed gains. Value accumulates dynamically and stochastically across time until a threshold is reached, and attention can influence the accumulation process by altering the relevant attributes. The option with the greatest value when the threshold is reached or time runs out is enacted.

Figure 2

Value accumulation across time for two hypothetical choice options. Action A (solid line) accumulates subjective value rapidly then drops off, whereas Action B (dashed line) accumulates value more slowly but it eventually reaches a greater value. These temporal dynamics could occur either due to randomly-accumulated fluctuations, or due to systematic differences in the nature of A and B (e.g., more abstract versus more concrete attributes). In either case, Action A would tend to be selected (and more quickly) if a low decision threshold were used because it reaches the threshold first, but Action B would be selected (and more slowly) if a higher decision threshold were set. The selected action also depends on the time available for the decision: Action A would tend to be selected if a short limit were imposed. Also, the noise depicted in the lines indicates stochasticity in the valuation process: repetitions of the same choice might result in selection of Action B occasionally, even in a short response window, due to random variation; for the same reason, Action A would sometimes be selected in a long response window.