A Model of Justification

I consider decision-making constrained by considerations of morality, rationality, or other virtues. The decision maker (DM) has a true preference over outcomes, but feels compelled to choose among outcomes that are top-ranked by some preference that he considers “justifiable.” This model unites a broad class of empirical work on distributional preferences, charitable donations, prejudice/discrimination, and corruption/bribery. I provide a behavioral characterization of the model. I also show that the set of justifications can be identified from choice behavior when the true preference is known, and that choice behavior substantially restricts both the true preference and justifications when neither is known. I argue that the justifiability model represents an advancement over existing models of rationalization because the structure it places on possible “rationales” improves tractability, interpretation and identification.

💡 Research Summary

The paper introduces a formal “justifiability” model of decision‑making in which a decision‑maker (DM) possesses a true preference ordering over outcomes but restricts his actual choices to those that are top‑ranked by some “justifiable” preference relation. A justifiable preference is any complete and transitive ordering that the DM believes could be held by a “good and reasonable person.” The core axiom, Irrelevance of Unjustifiable Alternatives, states that any alternative that is not top‑ranked by any justifiable preference is irrelevant: adding or removing it from the menu does not affect the DM’s choice. This captures the intuition that the DM will only act on outcomes that can be plausibly defended to external observers or to his better self.

The first major result (Theorem 3.1) provides necessary and sufficient conditions for observed choice behavior to be consistent with the justifiability model when the true preference ≻ is known. The set M of all justifiable preferences consistent with the data is the maximal collection of complete, transitive relations that satisfy the irrelevance axiom. Proposition 3.1 extends the representation to incorporate any dominance relation (e.g., impartiality, anti‑discrimination, stochastic dominance), showing that the model can accommodate additional normative constraints.

When the true preference has an expected‑utility (EU) representation, Theorem 4.1 shows that the justifiable preferences can also be represented by EU functions. This linear structure yields tractable comparative statics: to compare the moral strictness of two DMs it suffices to examine their choices from binary menus that share one fixed alternative (Corollary 4.2). The EU version also explains risk‑related phenomena observed in experiments by Exley (2016), where subjects’ charitable versus self‑payoff choices reverse depending on which component is risky.

If the true preference is unobserved, Sections 5 and 5.1 present two behavioral characterizations (Corollary 5.1 and Theorem 5.1). They demonstrate that even with a completely unrestricted true preference, the justifiability model imposes strong restrictions on observable behavior; in fact, as few as three alternatives can falsify the model. Theorem 5.1 shows that a cyclic pattern of choices over a three‑element menu is incompatible with any set of justifiable preferences, providing a simple empirical test.

The paper then maps a wide range of empirical studies onto the model. In the classic “moral wiggle‑room” experiment by Snyder et al. (1979), subjects’ cyclic choices over rooms with and without a wheelchair‑bound stranger are explained by a set M that requires any justifiable preference to rank the stranger‑present option above the empty room. In gender discrimination experiments (Norton et al., 2004), M consists of preferences that prioritize qualifications over gender, allowing subjects to claim merit‑based reasoning while still favoring males. Bribery experiments (Gneezy et al., 2016) are modeled by M that is independent of the bribe but varies in humor or risk attitudes, letting subjects rationalize acceptance of a bribe as “genuinely superior.” Allocation experiments (Rodríguez‑Lara & Moreno‑Garrido, 2012; Haisley & Weber, 2010) are captured by M that treats self and other symmetrically but differs on the weighting of earnings versus accuracy or ambiguity attitudes.

Section 6 situates the justifiability model within two strands of literature. The first strand includes two‑tier models (Gul & Pesendorfer, 2005; Manzini & Mariotti, 2007) where a secondary preference breaks ties in a primary one; these correspond to the special case where M is a singleton. The second strand concerns rationalization models (Cherepanov et al., 2013) that impose far fewer structural requirements on “rationales.” By insisting that justifications be complete, transitive, respect dominance, and satisfy independence, the present model achieves stronger predictive power, identification, tractability, and interpretability.

Overall, the paper makes three substantive contributions: (1) it formalizes a broad class of morally or rationally constrained choices, (2) it provides clear axiomatic characterizations and identification results both when the true preference is known and when it is not, and (3) it demonstrates the model’s explanatory breadth across diverse experimental domains. The justifiability framework thus offers a powerful tool for researchers seeking to disentangle genuine preferences from the socially acceptable rationales that shape observed behavior.