The Accuracy and Coverage Evaluation (A.C.E.) Survey relies on dual system estimation to estimate coverage in Census 2000. In making these estimates, the A.C.E. must account for missing data; these missing data are in the form of non-interviews and non-response. Various procedures in the A.C.E. make adjustments in the dual-system estimation process to take these missing data into account.
We wanted to examine the spread of national dual system estimates (DSE) when alternative missing data procedures were used on the same A.C.E. data. The resulting range would give us an indication of how sensitive the DSEs were to changes in one or more of the missing data procedures. We selected seven alternative procedures for our analysis: two non-interview adjustment alternatives, four non-response adjustment alternatives, and one alternative covering both adjustments.
Our analysis includes: an examination of ranges and standard deviations of the DSEs, for both the entire set of alternative procedures and for specific subsets; graphical analyses; a regression analysis; a study of how the individual alternative procedures effected the DSEs.
|