A Non-Experimental Analysis of True State Dependence in Monthly Welfare Participation Sequences

This paper provides an empirical evaluation of true state dependence in welfare participation using unique administrative data from California that is measured at the monthly frequency, which coincides with the welfare eligibility period and so is free of time aggregation bias. The analysis uses first- and second-order dynamic conditional logit models that non-parametrically control for permanent unobserved heterogeneity to test for state dependence in welfare behaviour. The second-order model also absorbs individual-specific first-order Markov chains, and provides a more robust test for state dependence in high frequency data. The results using the first-order model show substantial first-order state dependence in monthly welfare participation. Absorbing heterogeneous first-order effects, the hypothesis of no second-order state dependence is also easily rejected. This suggests that past welfare participation predicts future participation, given unrestricted effects of both the present state and unobserved heterogeneity, and provides substantive evidence of duration dependence at the individual level