Appendix C: Changes in the 1997 American Housing Survey and Their Impacts on Estimates of Worst Case Needs

• It is the first time that the survey was conducted using Computer Assisted Personal Interviewing (CAPI). CAPI incorporated a wide range of improvements that had been identified during 2 years of research and testing. Skip patterns involving complex branching and calculations that would not have been appropriate with a paper questionnaire became simple with an automated instrument for the respondent and field representative. It was almost impossible for field representatives to skip appropriate questions. Online editing features reduced errors at the point of data collection. These changes should improve the quality of the data. In addition, the timeliness of data delivery will improve in the future. Nevertheless, responses to questions are likely to be affected in unknown ways by these changes, possibly making results for 1997 less comparable with earlier results, even when exactly the same questions were asked.

• The Census Bureau changed computer platforms from a UNISYS mainframe to a UNIX workstation and changed its processing software/language to SAS. These changes mean that all the software for processing the data, editing the data, allocating missing values, recoding and transforming variables, preparing tables, and estimating variances had to be rewritten.

Appendices A and C of the Census Bureau's published volume in the 1997 AHS (American Housing Survey for the United States, 1997, Current Housing Reports H150/97) provide more detail on all changes.

This report's estimates of worst case needs were also directly affected by changes in the questions used to identify households receiving rental assistance. In AHS data prior to 1997, households were counted as receiving Federal housing assistance if they answered "yes" to one of the following questions: "Is the building owned by a public housing authority? Does the Federal Government pay some of the cost of the unit? Do the people living here have to report the household's income to someone every year so they can set the rent?" Although the number and characteristics of households responding affirmatively to these questions were generally consistent with program data, detailed examination revealed that households often do not report their assistance status correctly. (See Duane T. McGough, Characteristics of HUD-Assisted Renters and Their Units in 1998, May 1997.) In particular, many of the households responding that "a State or local government pay[s] some of the cost of the unit" were found to participate in Federal assistance programs, suggesting that the past practice in worst case reports of counting such households as unassisted was misleading.

Based on research using focus groups of assisted households, a different battery of questions was developed to better identify receipt of housing assistance. Accordingly, the order and content of the AHS questions about housing assistance were changed. The questions now used to identify assisted households are: "As part of your rental agreement, do you need to answer questions about your income whenever your lease is up for renewal? (If so,) to whom do you report your income? Do you pay a lower rent because the government is paying part of the cost of the unit? Is the building owned by a public housing authority?"

As noted in the body of the report, this new battery of questions identifies 5.6 million households as receiving rental assistance, 400,000 more than the total of units with Federal, State, or local assistance reported in 1995. This total is consistent with the sum of all households actually assisted by Federal rental assistance programs run by HUD and the Department of Agriculture, low-income renters occupying units assisted by the Department of Treasury's low-income housing tax credit (LIHTC), and State and local programs. It is unlikely, however, that the number is comparable to the 1995 estimate, or that the total number assisted actually grew by 400,000 over the 1995-97 period.

Because worst case needs also, by definition, include households with severe physical problems, the 1997 estimates are also affected by changes in questions and processing regarding plumbing and other physical problems. Many questions were changed in wording or sequence to allow respondents to concentrate sequentially on whether the problem occurred in their house, met the (more precise) definition, and occurred within the time period (such as the previous 3 months) specified. The revised questionnaire counts 1.8 million households as having severe physical problems in 1997. Although this estimate should be more accurate for 1997 than previous estimates were, it can not be directly compared with the 1995 count of 2 million. Specifically, it is unlikely that this accurately implies that the total number of units with severe physical problems fell by 200,000 between 1995 and 1997.

Income is another important component of the worst case needs determination. Some changes occurred in the collection of income data in 1997. Although the text of the questions remained the same, some persons-those age 21 or older who were related to the reference person and who did not contribute toward household expenses—were no longer asked separately about wage versus nonwage income. This change contributed to a drop in reporting nonwage income from 74 percent of households with one or more nonwage sources in 1995 to 64 percent in 1997. An accompanying upturn in those reporting wage income produced a median household income somewhat higher than the 1995 figure. However, the count of households with income less than $5,000 rose by 540,000 between 1995 and 1997. The Census Bureau estimates that about 60 percent of the increase occurred in households reporting zero or negative household income. Although households with zero or negative income are by definition not included as households with severe rent burden, they nevertheless are counted as having extremely low incomes, and thus affect the measures of affordable units per household in the report.

To summarize, because of the improvements in AHS questions and processing, HUD believes that the 1997 estimates of worst case needs more accurately count households with severe physical problems and households not receiving assistance than did estimates from earlier surveys, and thus provide a more reliable point-in-time estimate of worst case needs. Because of the major differences in questions identifying assistance and other differences in processing, however, the Census Bureau recommends against using differences between the 1995 and 1997 worst case needs estimates alone as indicative of trends.