This study compared 140 estimates and 14 derived measures (mostly medians) from the 2005 ACS with similar estimates and derived measures from the 2005 NYCHVS at the New York City level. Of the 140 estimates, 61 estimates or 44 percent tested as statistically different. Of the 14 derived measures, 8 measures or 57 percent tested as different. This outcome may seem to imply that the two surveys are far apart in their view of New York City's housing and households, but that conclusion would overstate the case. The relatively large sample sizes for both surveys contribute to the large number of estimates found to be "statistically different," but only 26 of these estimates (not including derived measures) differed by 1 or more percentage points and only 10 differed by 2 or more percentage points.
Differences between estimates from the 2005 ACS and the 2005 NYCHVS were expected. However, the level of differences should provide confidence in the quality of the estimates for users of both surveys. This study pointed out the fundamental differences in data collection methods, residence rules, questionnaire design and question wording, and other areas, and offered theories, explanations, etc. about how they may have accounted for some portion of the observed differences in the estimates compared. Because of the interdependencies between the different methods, the relative effect of these methodological differences cannot be determined.
The limited scope of this study invites further work. Areas of further exploration might include comparing the estimates from the NYCVHS with the different ACS modes of collection to determine what role this plays in the comparisons; comparing estimates at the borough level; and, for characteristics common to both occupied and vacant units, comparing estimates by occupancy and vacancy status.
The overall conclusion reached by this study is that, for the characteristics examined here, the ACS and NYCHVS both provide acceptable estimates; where differences in estimates do exist, it is important for the data user to understand that the two surveys employ different methodologies, procedures, and processes that may affect survey estimates. In this way, users can better determine which survey's estimates of a particular characteristic are most appropriate for their needs.