Effect of Movers on Triple System Estimation
Sally W. Thurston and Alan M. Zaslavsky
KEY WORDS: alternative list, ethnographers, mover rate, correlation bias, census coverage estimates, administrative list, dual system estimation
The Census Bureau’s use of ‘capture-recapture’ methodology to estimate census coverage relies on the assumption that being in the census is independent of being in the post enumeration survey. One method of checking this assumption, and of measuring bias due to correlation between the two sources, is to use a third source of names and addresses. This makes it possible to expand the 2 x 2 table underlying the dual system estimate into a 2 x 2 x 2 table in which only one of the 8 cells is unknown. Following earlier work by Zaslavsky and Wolfgang, this paper focuses on two estimates of the unknown cell: ‘ratio r1' and ‘ratio r2'.
One challenge of triple-system estimation is proper cross-classification of cases by inclusion or exclusion in each of the three sources. Misclassification may bias population estimates. Movers and non-movers may have different coverage rates in each source, so the accuracy of estimates is likely to be improved by considering movers and non-movers separately.
Alternative lists may be formed from administrative lists, or compiled by ethnographers working in small areas. During the 1990 census, ethnographers compiled independent lists of names and addresses in 29 sites; in four of these, post-enumeration survey data were also collected. Preliminary data from one site (in rural North Carolina) suggest that dual system estimation may understate the true population size.
The authors conclude that proper consideration of movers when using triple system estimation may lead both to more accurate population estimates than is possible using a dual system estimate, and to a way to measure the correlation bias in the DSE. Ethnographers’ lists are potentially more accurate than administrative lists, because they are more up-to-date. To be used in triple-system estimation, it is important that ethnographic lists be accurate at the time of the post-enumeration survey and at the time of the census, and that they define mover status reliably.
For the ethnographer’s account of the site, see: Patricia B. Lerch (1992) Coverage Differences in the Census of a Rural Minority Community in North Carolina: The Little Branch Area of the Waccamaw Sioux Tribe. Ethnographic Evaluation of the 1990 Decennial Census Report # 20 (EV-92/20).
CITATION: Citation: Sally W. Thurston and Alan M. Zaslavsky. (1992) “Effect of Movers on Triple System Estimation.” Proceedings of the Survey Research Methods Section (American Statistical Association): 176-181.