The Value of Choice

In this study, I investigate whether the out-of-sample predictive accuracy of existing distributional preference models can be improved by appending a "discretion" term which captures the intrinsic value of choosing for others (i.e., valuing control over others as distinct from caring about their outcomes). I propose a specific functional form for the discretion term that distinguishes between positive discretion (options to increase another’s payoff) and negative discretion (options to decrease it), with diminishing marginal utility. I append this term to four distributional preference models: inequality aversion (Fehr & Schmidt, 1999), quasi-maximin (Charness & Rabin, 2002), constant elasticity of substitution (Andreoni & Miller, 2002), and bliss point (Gidseg, 2026). I estimate individual-level parameters using an incentivized experiment with 105 subjects playing a Power Game, in which subjects spend their own earnings to gain control over a partner’s payoff. I then assess predictive accuracy on non-estimation data using the root mean squared error. I find that the discretion term improves spending prediction accuracy for approximately 18–23% of subjects across all four distribution models. Importantly, the discretion term rarely worsens prediction (1–4% of subjects). The estimated discretion parameters show significant heterogeneity across individuals but are not significantly correlated with measured personal characteristics. These findings suggest that incorporating the intrinsic value of decision rights over others into utility models can meaningfully improve 'economists' ability to predict behavior in social contexts.