A Grouped Mixed Proportional Hazard Model with Social Interactions: The Passage of the Motorcycle-Helmet-Use Law

Abstract

We develop a mixed proportional hazard model in discrete time when there is cross-sectional duration dependence from social interactions. We model the cross-sectional dependence using the weighted lagged choices of neighbors based on a proper spatial weight matrix, and nonparametrically specify the baseline hazard and the distribution of unobserved heterogeneity. We use EM algorithm to estimate the duration model and derive the observed information matrix for statistical inference. Using the U.S. state-level panel data, we analyze that the state legislation decision on the mandatory motorcycle-helmet-use law is significantly affected by neighboring states’ choices, whereas the fatality rate from motorcycle-related accidents is not so.