A Hazard Model for Welfare Durations with Unobserved Location-Specific Effects

Many papers have investigated how personal characteristics and environmental variables affect welfare durations of unmarried mothers. This paper estimates proportional hazard models for welfare durations that allow for either fixed state or fixed labor market area effects. Conditioning on residence location by fixed effects can limit the impact of three types of potential bias. (1) Estimates of the effects of personal characteristics can be biased owing to the omission of relevant local area variables. (2) Estimates of the impact of state welfare benefit levels are biased because they proxy for other unmeasured attributes of the state, in particular, the entire state welfare system. Conditioning on state fixed effects limits this bias to the extent that we can use time variation within states to estimate the benefit level effect. (3) With state fixed effects, we can better estimate the impact of local conditions, such as unemployment rates, because they also may have been picking up omitted state-level effects. The models are estimated by the Cox partial likelihood method with time-varying covariates. Data come from the 1984 and 1985 panels of the Survey of Income and Program Participation. I find that some personal characteristics (being black or Hispanic, education) have greater impact after controlling for location-specific effects. A Hazard Model for Welfare Durations with Unobserved Location-Specific Effects