Non-Elderly Disabled Category 2 Housing Choice Voucher Program: An Implementation and Impact Analysis. A. Overview of Analytical Approach and Data Sources


We used a difference-in-difference analytical method to estimate how likely nursing home residents eligible for NED2 vouchers in treatment areas were to move to the community before and during the period of voucher availability, as compared to people residing in comparison areas. Data availability (described below) led us to focus on a subset of PHA areas that received and leased the highest number of NED2 vouchers in 2011. For each PHA analyzed, we first identified within-state comparison areas that aligned on several characteristics, including rental housing market and community transition rates among nursing facility residents, before HUD allocated the vouchers to PHAs in January 2011.

Second, within each treatment and comparison area, we identified two samples of NED individuals who were eligible for the NED2 vouchers either: (1) before vouchers were available (the pre-intervention period); or (2) during the period of voucher use (the intervention period). The pre-intervention period was structured to cover the same duration as the post-intervention period, which was truncated based on data availability. Among our sample of NED2 users, the first voucher was used in April 2011, and data are available through the end of 2011. Accordingly, for the purposes of our analysis, the intervention period begins on April 1, 2011, and runs through December 31, 2011. The pre-intervention period covers the same time frame as the intervention period one year before voucher use began: April 1, 2010, through December 31, 2010.

Third, we identified a sample of non-elderly individuals in the treatment and comparison areas who were likely to use NED2 vouchers. Restricting the sample to those most likely to use vouchers was critical to detecting the program's effects. The nursing home population in the study areas contained a large proportion of individuals who were extremely unlikely candidates for voucher use, according to an empirical analysis of the characteristics of those who did not receive vouchers such as people receiving hospice care.22 If we were to include the entire population of NED2-eligible nursing home residents in our samples, it would have been nearly impossible to detect a statistically significant impact on transitions since the number of available vouchers (no more than 100 in a PHA) was only a small fraction of the number of residents. Consequently, we constructed a more targeted, smaller sample of people likely to use a voucher to increase the chance of identifying an impact.

Last, we calculated the impact of the availability of NED2 vouchers on the probability of making a transition from a nursing facility to a community-based setting in selected areas using a difference-in-difference estimator. This approach controls for time-varying factors that have common effects across treatment and comparison areas between the pre-intervention and intervention periods, as well as for fixed differences between the treatment and comparison groups that exist in the absence of voucher availability. We used within-state comparison groups to control for state-level policies, regulations, and changes in the economic environment. Differences may have existed in policies and service systems within states that affected institutional care and community living, but we assumed these were constant across the pre-intervention and intervention periods and therefore factored out in the estimation method. We used regression-adjusted difference-in-difference estimates to control for any observable differences between the characteristics of individuals in the treatment and comparison samples.

Issues related to data availability and the pace at which NED2 vouchers were used in each community influenced the time frame of the analysis, selection of communities, and approach used to estimate the impact of the vouchers on community transition rates. Our primary data sources included: (1) HUD administrative data; and (2) MDS data derived from assessments of nursing home residents.

HUD Administrative Data. Information used to identify 2011 NED2 voucher users came from administrative data provided by HUD's Office of Policy Development and Research. The data were collected from two sources: (1) reporting forms used by PHAs participating in the HUD Moving to Work program; and (2) data entries in the PIC for all other PHAs. In both datasets, NED2 voucher users were flagged as program participants.

HUD administrative data contained fewer voucher users than reported by PHAs to the TAC in December 2011 (Table III.1). Several explanations are possible. First, HUD administrative data may have been incomplete. HUD staff speculated there may have been delays in the entry of NED2 data in some PHAs, and other PHAs may have provided incorrect codes for identifying NED2 voucher users. Second, TAC data were reported during interviews and may have been inaccurate. Limitations in available data, combined with the limited time frame for completing this study, restricted our analysis to 2011 voucher users identified in the HUD data, or about one-third of NED2 users who leased vouchers in 2011.

TABLE III.1. Voucher Use by December 2011
State PHA Area   Vouchers Funded     Vouchers Leased  
by 12/11
Vouchers Leased
by 12/11
  (HUD Administrative Data)  
PHA Areas Included in Quantitative Analysis
Maryland Baltimore City 40 26 28
Baltimore County   50 20 13
Ohio Cincinnati 100 42 26
Washington Snohomish 50 37 32
Tacoma 100 44 40
PHA Areas Excluded from Quantitative Analysis
California Pasadena 40 8 5
Orange County 50 9 9
Georgia Decatur 35 4 5
Massachusetts   Lynn 35 2 0
New Jersey Statewide 100 36 11
Ohio Lucas 60 4 9
Texas Austin 36 14 11
Washington Longview 35 17 0
SOURCES: Information on the number of vouchers funded was provided by HUD. Voucher distribution data came from two sources: (1) TAC; and (2) HUD administrative data.

Minimum Data Set (MDS). This data set contains clinical assessment information for all patients in Medicaid-certified and Medicare-certified nursing homes and provides detailed information on the demographic characteristics, residential status (length of stay, discharge date, discharge status), functional status, and health conditions of nursing home residents. One unique element in the MDS is a query to residents about their desire to be discharged to the community; we used residents' answers to help to narrow both treatment and comparison samples to include only those with an interest in and some expectation of returning to the community. The most recent MDS data available for this study covered the period through the end of 2011; data for 2012 are not yet available.23

To identify all residents who resided in nursing facilities during our pre-intervention and intervention periods, we drew data from MDS version 2.0, which has data on people who resided in nursing facilities from the start of the pre-intervention period (April 2010) through September 2010, and from MDS version 3.0, which has this information for October 2010 through December 2011.24

The MDS data had two limitations for this study. First, they did not contain information on the geographic location of each institution, which was necessary to identify individuals residing in a treatment or comparison area. We compensated for this by linking the facility's National Provider Identifier (NPI), which is available in the MDS, to the NPI Registry maintained by CMS, which includes unique identifiers for health care providers assigned using the National Plan and Provider Enumeration System (NPPES) and corresponding zip codes for providers.25 The second limitation of the MDS data was the inclusion of only those NED2 voucher recipients residing in nursing facilities, and thus the exclusion of those residing in other institutions--for example, intermediate care facilities for individuals with intellectual disabilities (ICFs-ID)--who may have been eligible for NED2 vouchers. This was a minor limitation, however, as 94 percent of identified 2011 NED2 voucher users were matched to the MDS and thus known to reside in nursing facilities before making transitions to the community.

Selection of PHA Areas Included in the Analysis. Although the NED2 voucher program became effective in February 2011, only 36 percent of the total available vouchers in the 13 areas covered by the PHAs that received at least 35 vouchers each were leased by December 2011. Because low voucher use would have made it difficult to detect statistically significant impacts, we limited the analysis to the five treatment areas in which, according to HUD data, the highest number of vouchers had been issued by the end of 2011: Baltimore County, Baltimore City, Cincinnati, Snohomish County, and Tacoma (Table III.1).26

View full report


"Cat2Housing.pdf" (pdf, 1.44Mb)

Note: Documents in PDF format require the Adobe Acrobat Reader®. If you experience problems with PDF documents, please download the latest version of the Reader®