by Calculated Risk on 4/18/2012 03:42:00 PM

CR note: This is an important topic on trying to understand the number of excess vacant housing units in the US. Unfortunately the various surveys do not match up with the decennial Census data. It appears the vacancy rates in the HVS survey are way too high - yet this is the data most analysts use to estimate the excess number of vacant housing units! In other words, most reported estimates are way too high. The good news is the Census Bureau is trying to understand why ...

From economist Tom Lawler (Lawler identified this issue and pushed for this review):

The Census Bureau posted the following paper presented at the January 2012 meeting of the Federal Committee on Statistical Methodology, and folks interested in the topic should read it.

"Evaluation of Gross Vacancy Rates From the 2010 Census Versus Current Surveys: Early Findings from Comparisons with the 2010 Census and the 2010 ACS 1-Year Estimates" by Arthur R Cresce, Ph. D., Assistant Division Chief for Housing Characteristics, Social, Economic and Housing Statistics Division,
U.S. Census Bureau, SEHSD Working Paper Number 2012-07

Here is an excerpt of the purpose of the paper.

"This paper is part of a larger effort to understand why there are differences in the level of occupied and vacant housing units among the 2010 Census, the 2010 American Community Survey (ACS), the Current Population Survey/Housing Vacancy Survey1 (HVS), and the American Housing Survey2 (AHS). The specific focus of this paper is to provide a snapshot of research completed to date on factors that might explain differences in the level of vacant and occupied housing units between the 2010 Census and 2010 American Community Survey (ACS). Thus, this paper is not intended to answer all questions or issues concerning these differences. The 2010 ACS 1-year estimate for the gross vacancy rate (GVR) was 13.1 percent compared to 11.4 percent for the 2010 Census. We expect to produce a more comprehensive report on the 2010 Census – ACS differences in 2012 with additional reports to address the differences between the 2010 Census, the HVS and AHS. The goals of these reports are: 1) to understand better why these totals differ and 2) to address particular factors, where possible, that might lead to more consistent results across data collection efforts in the future.”

As noted above, this paper focuses on the decennial Census gross vacancy rates and the 2010 ACS gross vacancy rates. The paper notes, however, that Census analysts are also focusing on decennial Census vacancy rates vs. the HVS vacancy rates, as the below excerpt indicates.

“We plan to produce a series of reports in 2012 that will provide a more in depth analysis of potential factors that could explain the reasons for these differences, not only between the ACS and the census, but also among the ACS, the census and the Housing Vacancy Survey. From these reports, we hope to draw conclusions that will enable us, where possible, to take specific actions that could help provide more consistent results between the ACS and the census and, in general, among all our current surveys.”

Here are some summary conclusions from the paper.

“1. Although the census and the ACS have different reference periods and different residence rules, we do not believe differences in the reference period and residence rules were major contributors to the overall difference in the gross vacancy rates. However, problems can arise when implementing reference periods combined with residence rules. In the 2010 census, vacant housing units were enumerated in either Nonresponse Followup (NRFU) or in Vacant Delete Check (VDC) which was at least two months after Census Day. This enumeration of the Census Day reference date can make the determination of occupancy status problematic. FRs in the ACS and census enumerators can also misunderstand or misapply a usual residence or current residence rule.

“2. Response categories for occupancy status and vacancy status are similar between the ACS and the 2010 Census, but the way the questions are asked are different. It is not clear, though, if this played a role in explaining some of the differences in classification of housing units.

“3. The 2010 ACS sample was not drawn from the 2010 Census, which may help to explain at least a portion of the difference between the ACS and census GVRs.

“4. Large differences in the reporting of “Other” vacancy status and a possible connection between difficulty in obtaining a response (as measured by percent CAPI in the ACS and “hard-to-count” scores in the census) and differences in the GVR may provide some clues to understanding these differences.

“5. The census implemented coverage improvement procedures, such as special methods to review and confirm the status of housing, which are unique to the census and are not implemented in the ACS. The VDC operation in 2010 resulted in a net decrease of about 537 thousand vacant units.

“6. It was clear from debriefings with interviewers that they faced a very difficult task, Despite common procedures, differences in interpretation of what is an occupied unit can occur, especially in hard to count areas and, in general, in areas experiencing large numbers of foreclosures. Determining the occupancy status of a unit is especially hard in some areas when no household members can be contacted and neighbors are unwilling to provide information. “

Net, the paper suggests that the aggregate ACS vacancy rates for 2010 were probably “too high,” though by how much varied significantly by area/region.

Since the Housing Vacancy Survey vacancy rates were well above the ACS vacancy rates, the implication is that the HVS vacancy rates are substantially overstated. However, why the HVS vacancy rates are way too high is still being investigated.

Census currently plans to release the HVS for the first quarter of 2012 at the end of April, though it is not clear why!