1
Chapter 7: Assessing the Estimates
Mary H. Mulry
The evaluations of the A.C.E. Revision II estimates may be divided into two categories. One
category contains the evaluations that focus on individual error components. The other group
consists of comparisons of the relative error between the Census and the A.C.E. Revision II
estimator.
This section provides a brief description of the evaluation studies. The component errors
examined by separate studies are sampling error, error from imputation model selection, error
due to using inmovers to estimate outmovers in PES-C, synthetic error, error in the identification
of the census duplicates as determined by administrative records, error in the identification of
computer duplicates as determined by a clerical review, error from inconsistent poststratification
variables, and potential error arising from the automated coding of some cases, called the at-risk
coding, in the Revision Sample. The comparisons of relative error between the Census and the
A.C.E. Revision II estimator include a comparison with Demographic Analysis, the construction
of confidence intervals that account for bias as well as random error, and loss function analyses.
Also in this category is an examination of the consistency of the estimates of coverage error
measured by the A.C.E. Revision II estimator and the Housing Unit Coverage Study (HUCS).
Although an adjustment for correlation bias is included in the A.C.E. Revision II estimates, no
evaluations address the error in the level of correlation bias or the model used to distribute it
across poststrata. The reason is that examining alternative models only would consider
differences in models. Those differences would reflect the variations in how the several models
correct the original DSEs for correlation biases, but would not reflect the presence or absence of
correlation bias in these corrected DSEs.
7.1 Sampling error
Sampling error gives rise to random error, quantified by sampling variance. The sampling
variance is present in any estimate