The maps reveal a surprising level of structure and suggest how localized clustering of similar return rates can be. There is a strong concordance between 1990 and 2000 rates in the Plains and Mountain states. In contrast, the large majority of the tracts experiencing 20 percent or more dips in return rate lie in the eastern United States, beginning in central Texas. Large clusters of lower return rates are evident in various parts of the map, perhaps most strikingly in central Indiana and in Brooklyn, New York (Figure B-3), and abound throughout Kentucky, Tennessee, and the Carolinas. At the other extreme, as suggested by our regression models, the tracts that experienced higher mail return rates are concentrated in the Pacific division (particularly around Los Angeles and the extended Bay Area) and also in New England. Within the nation’s largest cities (save, perhaps, Los Angeles), similar return rates are concentrated in portions of the city; a zoom-in on Chicago highlights a cluster of tracts with lower return rates on the west and south sides of the city (data not shown), while the striking cluster of tracts with lower rates in Brooklyn contrasts markedly with other portions of New York City (see Figure B-3).
These geographic effects are consistent with the broad geographic dummies (by division) incorporated in our regression models. Furthermore, these extreme-value tracts have higher hard-to-count scores (with half of the values lying between the first quartile 22 and third quartile 58, with median 39) than do tracts whose change in return rates was less than 20 percent (first quartile 9 and third quartile 51, with median 27). But the effects do not suggest any other obvious explanatory factors. There does not, for instance, appear to be an unban/suburban/rural divide at work, nor do areas of either growth or decline in return rate appear to correspond with areas experiencing greater population growth over the 1990s and for which new construction might be a major part of the address list. Again, our analysis here is limited by the available data; we look forward to the Census Bureau’s evaluations of factors influencing mail return rates.
We did not find that changes in mail return rates explain the reductions in net undercount rates shown in the A.C.E. In fact, our analysis found very similar mail return patterns between the 1990 and 2000 censuses. The patterns in each census were explained by much the same variables, and the available demographic and socioeconomic characteristics for tracts did not explain changes between 1990 and 2000. However, census tracts experiencing unusually large increases or decreases in mail return rates did show a tendency to cluster geographically. Further investigation of these clusters and of local operations and outreach activities in these areas would be useful to identify possible problems and successes to consider for 2010 census planning.