The Structure of Moral Motivation

Abstract

Understanding why people act morally requires a precise account of moralmotivation: how moral goals are structured, how they are inferred from behavior and narrative, and how they manifest in action. Dominant theories of moral motivation typically rely on valence-based or unidimensional distinctions (e.g., prosocial vs. antisocial; approach vs. avoidance), which struggle to accommodate mixed, internally conflicting, or context-dependent motives. The present dissertation develops and empirically evaluates a multidimensional framework of moral motivation—the Motivation of Moral Action (MoMA) model—and examines how moral motivations are structured, represented, inferred, and behaviorally expressed. Across integrated lines of investigation, the findings provide convergent support for MoMA's proposed multidimensional structure. First, formal theoretical development and computational simulations demonstrate that MoMA exhibits greater representational adequacy than competing bidimensional models. Simulations spanning behavioral prediction, emotional experience, moral judgment, and character inference show that collapsing moral motivation into simpler valence-based frameworks results in systematic information loss and reduced predictive precision. In contrast, MoMA preserves distinct motivational states that yield empirically differentiable outcomes. Second, large-scale annotation of real-world moral narratives reveals that diverse configurations of moral motivation are reliably represented in natural language. Human annotators consistently distinguish among motivational profiles predicted by MoMA, and frontier language models exhibit partial convergence with human inferences while also revealing systematic divergences. These results suggest that multidimensional moral v motivation is psychologically real, communicatively encoded, and inferentially accessible. Third, the dissertation introduces and validates Moral Approach–Avoidance Tendencies (MAAT), an implicit behavioral paradigm designed to index moral motivation through action tendencies toward morally good, neutral, and bad stimuli. Across studies, moral content did not reduce to a simple valence-driven approach– avoidance pattern. Instead, morally good and morally bad stimuli elicited distinct and sometimes competing action tendencies, consistent with MoMA’s claim that moral motivation cannot be reduced to affective valence alone. Collectively, this work advances a unified and empirically supported account of moral motivation that clarifies its internal structure, demonstrates its representational presence in natural language, and establishes its measurable behavioral expression. By integrating formal modeling, simulation, narrative inference, and implicit measurement, the dissertation provides a foundational framework for studying moral motivation across psychological, computational, and applied contexts.