Next: About this document ... Up: 6 Handling spell data Previous: 6.4 Counting employment spells
Censoring is a serious problem when we want to look at the determinants of duration. If we exclude censored durations we systematically bias our sample, but if we include them it is also biased as the observed duration is inevitably biased down. Approaches based on life tables get around some of the worst of the problems: the Kaplan-Meier survival curve, for instance, gives us an unbiased estimate of the distribution of durations, controlling for censoring. More sophisticated multivariate hazard-rate models, or `event-history analysis' (Cox regression etc. ) use exactly the same underlying logic.
Under SPSS the survival command gives Kaplan-Meier survival estimates of the survival rate, taking account of censoring.
For instance, for employment spells only, we can get the estimated survival curve, for men and women, and for age groups (you will probably have to match in sex and DOB, and calculate age):
select if empstat = 1. compute agegp = trunc(age/10). surv tables=duration by sex(1,2) /plot (survival) /int=thru 200 by 3 /status=endtype(2). surv tables=duration by agegp(1,6) /plot (survival) /int=thru 200 by 3 /status=endtype(2).