Linear discrete time hazard estimation using Stata

Linear fixed-effects estimators (first differences, within transformation) are workhorses of applied econometrics because they straightforwardly allow for eliminating unobserved time-invariant individual heterogeneity that otherwise may cause a bias. However, I show that these popular estimators are biased and inconsistent when applied in a discrete time hazard setting, that is, with the outcome variable being a binary dummy indicating an absorbing state. I suggest an alternative, computationally simple adjusted first-differences estimator. This estimator is shown to be consistent in the considered nonrepeated event setting under the assumption of unobserved time-invariant individual heterogeneity being uncorrelated with the changes in the explanatory variables. Using higher-order differences instead of first differences allows for consistent estimation under weaker assumptions. In this presentation, I introduce the new community-contributed command xtlhazard, which implements the suggested estimation procedure in Stata.