Using a sub-sample of the BHPS matched to administrative data on participation in three State welfare programs, we investigate how response error affects estimation of duration models. While previous work has addressed response error, assumptions of independence are typically made and few authors have addressed false positive and false negative spells. This paper examines those assumptions by estimating probabilities of both errors of omission and errors of commission, as well as models of differences in length when true spells are reported. Utilizing the results from these models, we estimate a model of spell durations correcting for response error. The results indicate that failure to account fully for response error may lead to serious bias.
Chris Bollinger (University of Kentucky, ISER Visitor) Annette Jäckle (ISER)
Date & time:
18 May 2011 12:00 pm - 18 May 2011 13:00 pm
Internal seminars home