Classification error adjustments for gross flow estimates / A. C. Singh and J. N. K. Rao. : CS11-613/91-13E-PDF

"The problem of estimating gross flows from repeated surveys is considered when an individual's response at successive points in time is subject to classification error. A popular method for correcting classification errors is based on the assumption of independent classification errors and it uses interview-reinterview data for estimating error rates. In this paper a generalized model based on 6-response contamination is proposed which includes the model of independent classification errors as a special case. Using interview-reinterview data, this model can provide a range for bias adjustments for each flow as ɛ moves away from 1 (the value corresponding to independent classification errors model) to a lower bound E bounded away from zero so that a sensitivity analysis to the assumption of indepdendent classification errors can be made. Some numerical examples based on the Canadian Labour Force Survey are presented. It is seen that biases for some cells are fairly stable as ɛ varies hut, for others, they show monotonic upward trends in magnitude as ɛ increases. However, for a wide range of ɛ values (between .5 and 1), the assumption of independent classification errors seems fairly robust. Moreover, the biases for flow differences corresponding to symmetric cells seem to be quite insensitive as ɛ varies. Chi-square tests of symmetry and quasi-symmetry for an adjusted flow table are also presented"--Abstract.

Lien permanent pour cette publication :
publications.gc.ca/pub?id=9.837227&sl=1

• Format MARC XML
• Format MARC HTML

Demander des formats alternatifs