Duration data models, unemployment benefits and bias

Abstract

Traditional approaches to measurement conditional probabilities of leaving welfare do not deal with unemployment benefits data in a proper way leading thus to biased estimates of unemployment–employment transition probabilities. In fact, these approaches overestimate hazard rates and hence underestimate the expected welfare duration.

Notes

¹  d _i1 equals to 1 when the unemployed exits to a job receiving UI (0 for the rest); d _i2 equals to 1 when the unemployed exhausts UI disappearing from the records for ever (0 for the rest) and d _i3 separates the uncensored and censored durations of recipients who get UA after the UI exhaustion. It takes value 1 when the unemployed exit to a job (0 for the rest).

² Our sample contains information of unemployed with ages between 18–59 years that entered the Spanish UCS during February 1987. We focus our analysis on the unemployed entitled to UI and UA, where some unemployed who exhausted UI receive UA.

³ A null effect is found by Stancanelli (1998) in UK using only UA information.

⁴ Some authors find that time of benefit entitlement affect the behaviour of the unemployed (Gonzalo, 2002, in Spain using only UI data; Arranz and Muro, 2004, in Spain and Juradjda, 2004, in US aggregating together UI and UA) and others that it does not affect (Stancanelli, 1999, for UK aggregating UI and UA data).