Ac. Singh et Jnk. Rao, ON THE ADJUSTMENT OF GROSS FLOW ESTIMATES FOR CLASSIFICATION ERROR WITH APPLICATION TO DATA FROM THE CANADIAN LABOR-FORCE SURVEY, Journal of the American Statistical Association, 90(430), 1995, pp. 478-488
Gross flows represent transition counts between a finite number of sta
tes for individuals in a population from one point in time to the next
. Such flows are important for researchers and policy analysts; for ex
ample, gross labour flows for understanding labour market dynamics. Un
fortunately, the observed flows are typically subject to classificatio
n error. As a result, the problem of adjusting observed flows for clas
sification error has received considerable attention. Currently, three
methods for adjustment of classification error are available. All the
se methods use the key assumption of independent classification errors
(ICE) in conjunction with interview-reinterview data. We first give m
odifications to two of the methods that ensure that margins of the adj
usted flow table agree with the published ''stocks'' without requiring
a final margin adjustment. We then propose epsilon-response contamina
tion models and procedures for studying the robustness of ICE under di
fferent scenarios of departures from ICE applicable for all the method
s. Our empirical results, based on data from the Canadian Labour Force
Survey, show that for many scenarios the ICE assumption is fairly rob
ust. There are, however, situations where this is not the case. We thu
s suggest that users apply the proposed procedures to check the robust
ness of ICE under scenarios relevant for their own applications. Final
ly, we provide valid chi-squared tests for modeling flow tables adjust
ed for classification error under ICE so that the adjusted flows could
be further smoothed under the model. These tests are also illustrated
using data from the Canadian Labour Force Survey.