Any survey’s usefulness lies in part on what its answer is to the question “Are you actually getting information from everyone you want to?” Missing people who should be in your survey can cause serious biases in the results, if you disproportionately miss persons that differ on variables of interest from those you did measure. Both nonresponse and undercoverage can be significant contributors to this “missingness” problem, and since their effects of biasing the survey are similar, we would like to measure their combined impact. Undercoverage is often measured separately, as with a standard coverage ratio. In this paper, we propose a new measure called “sample completeness” which will examine the effect of undercoverage and nonresponse together on the 2000 American Community Survey (ACS) national test (the Census 2000 Supplementary Survey (C2SS)), and we will compare its sample completeness to that of the 1990 census long form. The ACS is a monthly survey which will provide data comparable to that from the decennial census’s sample (or long form) questionnaire and will replace the long form for the 2010 census.

We will briefly discuss the sampling methodology for the C2SS and the 1990 long form and our reasons for using data from the 1990 census instead of Census 2000, how to calculate the sample completeness ratio, some important differences between the long form and C2SS data, and a comparison of the sample completeness ratios for various demographic groups.