| This dissertation consists of two essays on social interactions, evolutionary game theory and evolutionary computation. The first essay studies the recidivism of felons on probation in the United States. Using a national sample, I attempt to identify the risk factors for recidivism among Female, Male, Black, White and Hispanic felony probationers. I adopt the Cox proportional hazard model and use Bayesian model averaging for the selection of covariates and subsequent inference. The study extends earlier research on recidivism in three directions. It departs conceptually from most previous studies in considering and identifying social interactions as one of the most significant factors affecting recidivism. Methodologically, I show that Bayesian model averaging leads to better evaluation of risk factors for recidivism, as well as improved risk assessment for potential recidivists. Finally, this study differs from most predecessors in exploiting a national sample rather than data from a single state or a number of counties, and in studying recidivism separately for each gender, ethnicity and race group.; In the second essay, I model the conflict between individual and collective rationality as a Prisoner's dilemma game and develop various algorithms to simulate learning by boundedly rational players. The study restricts the players to strategies representable by Moore machines and assumes that more complex strategies are more costly.; Extensive simulations suggest that cooperative behavior persists when mutants are less frequent than a threshold mutation rate and strategy complexity costs are not too high. When attention is restricted to one and two state Moore machines, the cooperative behavior is supported by nice strategies, i.e., strategies that are never the first to defect. Furthermore, amongst all these nice strategies, the ‘Grim’ turns out to be the most evolutionarily fit.; When the analysis is extended to more complex strategies, specifically to Moore machines with up to sixteen states, on the other hand, the cooperation is still supported mainly by ‘Grim’ if complexity costs are substantial. Otherwise, the surviving strategies begin with a sequence of defections, but eventually achieve cooperation when playing against themselves. Most of their states induce defection, allowing them to exploit naïve cooperators. |
