Comment Uncertainty in Demographic Analysis

Abstract

The article by Robinson, Ahmed, Das Gupta, and Woodrow (RAGW hereafter) gives the basic logic of Demographic Analysis (DA) along with the primary data sources used to estimate components of change in population. RAGW give almost all of the population estimates and associated undercount rates available from the DA method. These include coverage estimates by demographic subgroup (with very coarse age groupings), as well as patterns in undercount rates across several censuses. The main new feature of DA for 1990 is the provision of bounds on the DA estimates that are given probabilistic interpretation. In our opinion there are four main reasons for considering DA seriously in evaluating census coverage. First, the method is cheap, especially in comparison with Dual-Systems methods based on the Post-Enumeration Study (PES). Second, the basic idea of DA is sound, deriving from the fundamental equation of demographic accounting. Information from a variety of data bases (e.g., Medicare enrollment, immigration, and vital statistics) is used, and demographic and administrative information from the “past” is incorporated. Third, DA estimates are independent of estimates obtained from the PES with Dual-Systems methods, so an important check on the latter can be made. To many observers outside the courtroom, discrepancies between the two sets of estimates seem modest, but it is worthwhile to try to explain why slight differences arise. Finally, the DA method does not change much from census to census, so reasonably valid historical comparisons can be made. (If corrections are made, then it is a simple affair to recalculate DA estimates for previous censuses.) Inter-census patterns that exhibit unusual sex ratios or cohort patterns lead to refinements in the estimates. An example that RAGW cite is the finding of an inconsistent cohort “effect” for Blacks over the previous three censuses, which led to modification of estimates of completeness of coverage in birth registration for Blacks in 1940 (and to related proportional adjustments in adjacent cohorts). It is more difficult to compare undercount patterns derived from Dual-Systems methods across censuses, because the PES and the estimators used with it have changed appreciably over the last four censuses. A summary of the historical development of the DA method appears in Himes and Clogg (1992), which acknowledges the important role of Coale (1 955) in establishing