Unobserved heterogeneity in event history analysis: A quantile regression approach

Event history data often exhibit significant unobserved heterogeneity caused by latent characteristics of economic agents or by transience of treatment effects. Neglecting such heterogeneity typically leads to biased and inconsistent estimates of the parameters in duration models.; Several attempts to overcome this problem have been explored based on the class of mixed proportional hazards (MPH) models. MPH models specify unobserved heterogeneity via random variables acting multiplicatively on the baseline hazard function, shrinking (expanding) it when the random effects are smaller (greater) than unity. The main alternative to MPH models is the class of accelerated failure time (AFT) models in which the random effects multiply survival times (instead of hazards), allowing expansion (compression) of the lifespan according to the values of the random effects.; The intrinsic reliance on parametric assumptions, along with the restriction of a simplistic location shift for the random effects, make both MPH and AFT classes of models unattractive to survival modeling of heterogeneous populations. For these reasons, we propose quantile regression as a robust alternative to the existing random effects duration models often adopted in event history analysis.; The quantile regression model (Koenker and Bassett (1978)) provides a more flexible analysis of the conditional duration distribution with unobserved heterogeneity, without imposing stringent distributional assumptions on the baseline hazard function, unobserved heterogeneity, or residuals. Such flexibility offers a more accurate assessment of treatment effects over a large range of duration quantiles.; The dissertation is organized as follows: In Chapter 1 we provide a brief survey of the unobserved heterogeneity problem in event history analysis. In Chapter 2 we analyze the estimation and inference procedures of MPH models. In chapter 3 we examine the theoretical properties of random effects duration quantiles, while in Chapter 4 we present a Monte Carlo study of the small sample performance of the aforementioned estimators. In Chapter 5 we provide an empirical application of the discussed methods on what we consider to be the first systematic assessment of the duration of financial analysts' employment spells. Our concluding remarks and suggestions of future research are given in Chapter 6.