Identifying Selection and Duration Dependence Using Elicited Beliefs

Abstract:

Employment separation hazards decline with tenure, but the forces behind this pattern are not fully understood. It may reflect dynamic selection, as low-quality matches end early and surviving jobs are increasingly composed of better matches, or structural duration dependence, as separation risk declines within an ongoing match. We show that, under plausible conditions, the two mechanisms are observationally equivalent based on conventional duration data alone. They differ, however, in how heterogeneity in separation risk changes over tenure. Although this heterogeneity is not typically observable in standard transition data, elicited separation probabilities are informative about it. We use these beliefs, observed in harmonized survey data from 11 euro area countries and the United States, to identify the relative importance of each channel. To do so, we develop a quantitative model in which workers learn about match quality from noisy signals while separation risk may also decline within a match. The model delivers closed-form posterior beliefs over match-specific separation rates, thereby making the distinction between dynamic selection and structural duration dependence analytically transparent.