The data for our study come from the first three interviews or ``waves'' of the
HRS, a
nationally representative longitudinal survey of 7,700 households whose heads
were between the ages of 51 and 61 at the time of the first interview in
1992 and 1993. Each adult member of the household was interviewed
separately, yielding a total of 12,652 individual records. Waves two and three
were phone interviews conducted in 1994/95 and 1996/97, respectively, using
computer assisted telephone interviewing (CATI), which enabled interviewers to
condition their questions on information provided in previous interviews.
Deaths and sample attrition reduced the sample to 11,596 individuals in
wave two of the survey and 10,970 individuals in wave three.
From our perspective the HRS has several advantages over the alternative sources of data previously used to analyze the DI award process such as the SIPP data (e.g., Lahiri et al. (1995) or Hu et al. (1997)). The HRS is a long panel focusing on older individuals nearing retirement age, with separate survey sections devoted to health, disability, and employment that contain numerous questions on objective and subjective indicators of health status, questions about disabilities, their dates of onset, beginning and ending dates of employment spells, questions about employer accommodations to disability, and questions about applications and appeals for DI benefits. In wave one respondents were asked about the date they first applied for DI benefits and whether or not they were awarded benefits. A respondent who was awarded benefits was asked the month and year they first started receiving them. If a respondent was not awarded benefits, they were asked whether or not they ``appealed or applied again later''; if the answer was positive then the respondent was further asked about the month and year of their last application or appeal. In subsequent waves individuals who reported receiving DI benefits in the previous wave were asked if they were still receiving DI benefits, and if not, the month and year at which they stopped receiving benefits. Those who were not receiving benefits at the time of the interview were asked in subsequent waves whether or not they applied for DI benefits since the previous survey, whether or not they were awarded benefits then, and whether or not they appealed or applied again if they were denied. A respondent was further asked about the month and year of the last application or appeal if they appealed or applied again.
There are several limitations of the HRS data for studying the DI award process. First, unlike the SIPP data, there is no match to the SSA Master Beneficiary Record, so we are unable to verify individuals' self-reported information on dates of application and appeal for SSDI and SSI benefits. Second, the HRS did not distinguish between SSI and SSDI applications, appeals, and awards, since all questions combined the two programs into a single category. Third, due to the limited level of detail in the HRS questions on the DI appeal process and the fact that relatively few cases reached higher levels of appeal, we collapsed the multi-stage appeal process into a single stage. Finally, the HRS did not include appropriate follow-up questions that would allow us to determine whether the DI applications or appeals reported in previous surveys were awarded, denied or pending, resulting in potential censoring of information on appeals and reapplications. Fortunately we were able to rectify some of these censoring problems using other information in the HRS survey. For example the income section of wave two of the HRS included a question about whether respondents received Social Security income, and if so, the type of Social Security Income (SSI/SSDI benefits, retirement, etc.) and the date at which the respondent began to receive those benefits. We used this information to determine that certain previously pending DI cases resulted in awards. However in some cases we could not use other survey information to resolve ambiguities when there was no information on the outcomes of previously reported pending DI applications (i.e., in waves two or three). Since only a small fraction of cases are pending for more than 24 months, we classified an ambiguous pending DI application or appeal as a denial if we could determine that it had been pending for more than 24 months.
Despite the problems in the HRS questionnaire design, we found that by imposing reasonable strategies for resolving ambiguous cases and imputing missing dates of application, appeals, and awards (described in more detail in Appendix A), we feel confident about the quality of the resulting data set. To the extent that we were able to verify, our estimates of the mean durations between various events such as application and appeal, and appeal and first receipt of benefits, lined up fairly closely with independent information from SSA on the delays at various stages in the DI process (some of which are included in Figure 2.2).
Other important variables for our analysis are monthly and annual indicators summarizing the respondent's employment history. These variables could be important factors in DI determinations since they provide evidence of an applicant's ability to engage in substantial gainful activity. In particular, any evidence of employment subsequent to the reported date of disability onset or the filing of an application for DI benefits could be grounds for immediate rejection at the first stage ``SGA screen'' in Figure 2.1. We constructed employment histories using information on beginning and ending dates of employment spells in the employment section of the HRS. In particular we calculated, for each individual in every year between 1991 and 1996 annual hours worked and annual earnings. Monthly employment indicators for each month between January, 1989 and December, 1996 were also calculated. We employed a battery of consistency checks to validate the extensive number of calculations necessary to translate reported dates of beginning and leaving previously held jobs and ``intermediate jobs'' held between successive survey waves to determine the time path of employment down to the finest possible time period allowed by the survey questions (i.e., monthly). In addition to the work history data we also constructed a number of wealth variables, the most important being net worth, housing wealth, and non-housing assets.
Finally, we had to address issues of time aggregation in modeling the DI award process. Although, individual decisions about when to apply for disability are made in continuous time, we only observe individuals at successive surveys which are approximately two years apart. Obviously the wider the window of time over which we observe any given person, the higher the likelihood that they will submit an application or appeal. For this reason it is inappropriate to model the DI application decision in a static context, treating each individual as constituting a single observation as has been done in most of the previous literature. Instead we want to exploit the panel nature of the HRS, where a single individual can yield many observations on decisions to apply or not apply. The number of ``application'' or ``no application'' observations that one can obtain from a single individual is in some respects arbitrary since it depends on how finely we discretize time. However the finer we discretize time, the more information we need on other covariates which affect application and appeal decisions like health status. In the HRS we only observe an individual's health status at points in time that are roughly two years apart. We decided that a reasonable compromise would be to estimate a model that discretized time into two year intervals. This yields a maximum of three ``person-period'' observations for an individual who responded to each of the three waves of the HRS, provided the outcome of a previous application or appeal was not still pending. Since application, appeal, and award decisions do not coincide with the interview dates, we used data from the survey wave closest to the date of the application, appeal, or award as representing the state of the individual at that time. However if the closest wave was previous to the date of application, appeal, or award, and if the person did not report being disabled at that wave, then we used the information from the subsequent wave.