An Information-Theoretic Analysis of Offending Over the Life Course

Avinash Singh Bhati, The Urban Institute

Several fully- or semi-parameteric methods have been proposed in the literature that purport to mitigate the ill-effects of population heterogeneity on inferences regarding state-dependence when estimating longitudinal or panel data models. Most, if not all, are maximum likelihood based methods. These methods have desirable properties only asymptotically. Moreover, when behavioral parameters may be unstable over time, allowing this instability without mak-ing a-priori assumptions about its functional form can increase the dimentionality of the problem thereby decreasing the efficiency of likelihood based estimators. In this paper a semi-parameteric information-theoretic method is proposed that can be used to recover information about the evolving impacts of time-stable and time-varying characteristics while permitting persisting population heterogeneity without making any a-priori distributional assumptions re-garding the unobserved heterogeneity nor the error structure. The method is applied to a simple dynamic binary choice model using the Cambridge delinquency study and possible extensions are discussed.

(Return to Program Resources)

Updated 05/20/2006