Using National Longitudinal Survey data, the authors estimate proportional hazard models in order to learn whether it is more difficult for employers to identify female nonquitters than male nonquitters. They find that women may be a higher risk than men in the overall sample because they are more likely to be "movers" for unobserved reasons. When the authors focus on a relatively recent birth cohort, however, they find that it is no longer difficult to identify female nonquitters. Unobserved heterogeneity becomes an insignificant factor among women and virtually all determinants of turnover are observable at the time of hire.

Tenure responses in the Panel Study of Income Dynamics (PSID) and the National Longitudinal Surveys are often inconsistent with calendar time. These inconsistencies pose special problems in the PSID because job changes cannot be identified directly, so researchers must infer them from error-ridden tenure data. We use alternative rules for partitioning PSID data into jobs and then estimate several wage and mobility models to assess the sensitivity of parameter estimates to the partitioning method. We also assess the importance of replacing "raw" tenure data with imputed measures that are internally consistent.