We analyze persistence in marijuana consumption utilizing data from the 1997 cohort of the National Longitudinal Survey of Youth (NLSY97). We allow for three sources of persistence: pure state dependence, time invariant unobserved heterogeneity and persistence in idiosyncratic, time-varying shocks. We also consider intensity of consumption based on days of use per month and estimate a dynamic ordered Probit model using simulated Maximum Likelihood. We consider a Polya model that generalizes the more commonly used Markov models. The results show that there is a causal eect of previous use. However, ignoring unobserved heterogeneity and serially correlated shocks signicantly exaggerates the state dependence.