All of these results are subject to sampling and nonsampling error and should be interpreted cautiously. Nonetheless, assuming that the estimates are roughly accurate, it appears that the 2000 MAF may have had a negligible percent net undercount of occupied units. A small net coverage error, however, masked large numbers of gross errors of omission and erroneous inclusion—as many as 2.3 million erroneously enumerated occupied units (an estimated 0.7 million units that should have been deleted in the special unduplication operation, but were not, and 1.6 million erroneously enumerated units estimated from the A.C.E.), and as many as 2.7 million omitted units. The omitted units included not only 1.4 million units that were never on the MAF but also 0.4 million units that were incorrectly dropped from the MAF before Census Day, 0.6 million units that were incorrectly deleted from the MAF on the basis of such operations as nonresponse follow-up, and an estimated 0.3 million units that were incorrectly deleted from the MAF in the special unduplication operation (Vitrano et al., 2003b:Table 33). Without the special unduplication operation, even with its problems, the gross errors in the MAF would have been larger yet.17
Among the erroneously enumerated occupied housing units identified in the Housing Unit Coverage Study were census units in A.C.E. block clusters that were incorrectly geocoded to a block in a ring of surrounding blocks; such geocoding errors were estimated as 0.4 percent of total housing units (Barrett et al., 2001:Table 10). A study that looked for geocoding errors for census housing units in individual A.C.E. blocks by searching the entire tract and one ring of surrounding census tracts estimated that 4.8 percent of total housing units were misgeocoded (Ruhnke, 2003:iv). Most or all of these geocoding errors, however, would not contribute to gross errors for larger geographic areas (e.g., towns, cities, counties, states).
We do not know whether the errors in the MAF contributed more or less to population coverage errors than did omissions and erroneous inclusions of people within correctly identified housing units. We also know little about the variability in the accuracy of the