This paper provides an exploratory empirical analysis of the Social Security Administration's (SSA) disability application, appeal, and award process using a sample of 12,652 men and women from the first three waves of the Health and Retirement Survey (HRS). We estimate separate discrete choice models of the individual's decision to apply for benefits and the government's decision to award benefits. The disability application and award process is not a one shot game: applicants whose initial applications are rejected are eligible to appeal within 60 days of a denial, and they may submit any number of new applications. Our results indicate that it may be worthwhile to appeal or reapply after an initial rejection. While the probability of an award for an initial application by one of 54 state-based Social Security Disability Determination Services (DDS) is 50% in our sample, the ``ultimate award rate'' increases to 72% once we include individuals who appealed or re-applied following initial rejections. Nevertheless, there is a significant delay cost to appealing an initial rejection. The mean duration between application and award for ``first stage'' awardees is 4.6 months compared to a mean delay of 14.7 months for those who received benefits after one or more stages of appeal. These findings suggest the importance of modeling dynamics and uncertainty in order to obtain an accurate understanding of the DI award process.
One of the practical motivations for conducting a more detailed investigation that explicitly accounts for the dynamics of the appeal process is the fact that in recent years the SSA system has been overwhelmed by the large numbers of cases of DDS denials that have been appealed to SSA's Administrative Law Judges (ALJ). The number of appeals grew from 225,000 in 1986 to 498,000 in 1996 (U.S. General Accounting Office (GAO), 1997). Much of the growth in appeals is a result of the surge in overall disability applications, awards, and benefit payments. These payments have increased from $32 billion in 1988 to $75 billion in 1996, a seemingly unsustainable growth rate of 10.7% per year. The huge increase in applications has created a backlog of nearly a half million cases and large increases in processing delays. For example, the average processing time for appealed cases increased from about 10 months in 1994 to over one year in 1996. To resolve these problems the SSA initiated a comprehensive ``disability process redesign'' intended to simplify and streamline the current evaluation process used by the DDS. However, we need to have good understanding of the behavioral impacts of delay in the application and appeal process in order to determine whether attempts to streamline and speed up the application and appeal process could inadvertently lead to significant additional growth in applications and appeals.
To our knowledge, the only other attempt to explicitly model the
complete disability application, appeal, and award process comes from
Riphahn and Kreider (1997). They used retrospective information from the first wave of
the HRS to model the individuals' decisions to apply for DI benefits over
the period 1986 to 1991. However, since the main goal of the study was to
approximate the full returns to applying for DI benefits, the various stages
of the process were not modeled. Consequently, similar to previous models
of the DI application decision by Halpern and Hausman (1986), there is no
explicit consideration of the substantial delays involved in appealing an
initial rejection and the type of self-selection that these delays might
induce. The analyses of Lahiri et al. (1995) and Hu
et al. (1997) provide more insight into the initial (``first stage'')
acceptance decision made by the DDS, but ignored the possibility of
appeal.
Due to the limitations of the reduced-form models employed in this paper, we are unable to generate predictions of behavioral responses to changes in DI program parameters (such as the proposed disability process redesign), or identify the relative importance of policy and ``macro'' level causes of the recent rapid growth in the DI rolls. Nevertheless, our analysis does shed considerable light on the ``micro'' determinants of application and award decisions under the current regime. Our goal in this paper is to use relatively simple and flexible reduced-form econometric methods to uncover the most important predictors of application, appeal, and award decisions. Our probability estimates can be viewed as approximations or ``projections'' of decision rules used by individuals and the government onto the fairly rich set of information provided in the HRS.
The results in this paper are based entirely on self-reported data on
health status, employment status, income, and dates of application, appeal, and
award of DI benefits. It is important to interpret our
results with caution when it comes to making causal inferences about
the impact of health on the propensity to apply for disability since
the self-reported health and disability status variables used in our
analysis may be subject to measurement error and endogeneity problems.
Measurement error can arise from the usual sorts of survey recording errors,
misreporting due to memory or other cognitive limitations of respondents,
and from a variety of other problems associated with using respondents'
subjective self-assessments of health and disability as measures of
``true'' health or disability status. There is also significant disagreement in
the literature about the endogeneity of
self-reported health and disability measures due to various types of
intentional or unintentional misreporting on the part of respondents
such as ``rationalization bias'', where respondents use health
problems as a convenient excuse to explain their non-participation.
Since we do not attempt to deal with potential measurement error and
endogeneity problems in this paper, we acknowledge that
some of the relationships we discover could be subject to complicated biases.
However we do not believe our principal findings are simply a
reflection of ``spurious causality''. Instead the evidence we
have seen so far confirms our initial working hypothesis that to
a first approximation respondents report all information truthfully
and accurately. Space constraints prevent us from presenting this
evidence here: we defer an analysis of the magnitude
of endogeneity and
measurement error biases resulting from self-reported disability status
to a subsequent paper.
The rest of the paper is organized as follows. Section 2 provides a brief background on the U.S. Social Security Disability Insurance program (SSDI) and the Supplemental Security Income Disability Insurance program (SSI). Section 3 describes the HRS data set and some of the compromises imposed by data limitations. Section 4 presents the empirical results from nonparametric estimation of the distribution of delays at various stages of the application process and estimation results for our four- stage model of DI process. Section 5 offers a summary and conclusions.