An agent-based analysis of context-dependent preferences

In "Skating on Thin Ice", Frohlich and Oppenheimer (2006) describe a phenomenon they observed in public goods experiments that is rarely discussed in the literature: individual contributions to the public good are often inconsistent over time, appearing to fluctuate between two distinct contribution levels. Although they conjecture that individuals have complex context-dependent preferences, they did not develop a full specification of the theory. Using an agent-based simulation model, we explore the likelihood of these psychological conjectures, and thereby provide a possible specification of a theory of complex context-dependent preferences. We consider two main theories: first, that inconsistent contributions arise from a deterministic avoidance of exploitation and second, that inconsistent contributions arise from a probabilistic response to exploitation. We show the former theory clearly fails and the latter theory, under specifiable conditions, does produce the observed pattern of contributions. Two simple alternative theories are also considered, that of a highly-stylized "probabilistic guilt" and of goal-oriented but non-utility maximizing behavior (with stable preferences). Both alternatives, under certain conditions, are also able to generate the observed pattern. We develop an analysis of situations in which the predictions of these theories diverge and suggest that one could discriminate between them in laboratory settings. Finally, we consider a possibly fruitful relationship between simulation and experimentation to consider the implications of one's models and conjectures: this research can be seen as one step in an iterative process of theory development, vetting and testing, generating an empirically grounded theory of individual behavior in VCM games.