Summary: GAO discussed the status of the Census Bureau's 1990 Post Enumeration Survey (PES) to determine the accuracy of the census. GAO noted that: (1) PES estimated a net undercount of 2.1 percent, or approximately 5.3 million persons, and the net undercount based on demographic analysis was about 1.8 percent, or approximately 4.7 million persons; (2) the gross number of errors was much higher than what was shown by PES data demographic analysis, since some people were missed by the census, while others were counted more than once; (3) the 1990 census was the first census not to reduce the net undercount over the preceeding census and had the highest racial differential undercount since the Bureau began estimating coverage; (4) the major delay encountered in reaching an adjustment decision was in the statistical process used to reduce sampling variability, known as smoothing; (5) the restrictive time schedule for an adjustment may have jeopardized the quality of PES data and left insufficient time for analysis and interpretation of results; (6) the results of the Bureau's evaluation projects on missing data indicated that the model used to impute such data was reasonable when compared to alternative models and simulations, but this did not necessarily mean that all concerns about PES quality were resolved and the census could be adjusted with confidence; and (7) since there were people who were not counted by the census or PES, correlation bias in PES estimates of population tended to be biased downward from the true population that would be calculated if all persons were counted.
