Errors in the Social Security Disability Award Process

10/18/2011
Featured in print Bulletin on Aging & Health

The multistage process for determining eligibility for Social Security Disability Insurance (DI) benefits has come under scrutiny for the length of time the process can take - 1153 days to move through the entire appeals process, according to a recent Social Security Administration (SSA) analysis - and for inconsistencies that suggest a potentially high rate of errors. One inconsistency is the high reversal rate during the appeals process - for example, administrative law judges, who represent the second level of appeal, award benefits in 59% of cases. Another inconsistency is the variation in the award rates across states - from a high of 65% in New Hampshire to a low of 31% in Texas in 2000 - and over time - from a high of 52% in 1998 to a low of 29% in 1982.

The SSA has been working on a long-term strategy to address these issues since the mid-1990s. As SSA Commissioner Barnhart said in remarks before the House Social Security Subcommittee in September 2003, "claimants and their families expect and deserve fair, accurate, consistent, and timely decisions."

Despite these concerns, the actual error rate in the DI award process is unknown. The most recent studies on the question date from the 1960s, and thus may not reflect the current situation. In "How Large Are the Classification Errors in the Social Security Disability Award Process?" (NBER Working Paper 10219), Hugo Benitez-Silva, Moshe Buchinsky, and John Rust conduct an audit of the DI award process for a recent sample of applicants and construct an alternative screening mechanism that may have a lower error rate.

The authors use data from the Health and Retirement Study (HRS) for a sample of individuals who applied for DI between 1992 and 1996. The authors first compare self-reported disability status to the outcome of the DI application. Self-reported disability status is based on the respondent reporting that they have "an impairment or health condition that prevents them from working entirely." Working under the assumption that self-reported disability status is equivalent to true disability, the authors find large classification errors in the awards process - 58% of those who are denied benefits are truly disabled, while 22% of those who are awarded benefits are not truly disabled.

These estimates rest on the assumption that self-reports of disability are truthful and accurate, a subject the authors turn to next. They point out that 18% of respondents who receive DI benefits say that they could work, throwing doubt on the theory that people exaggerate health problems to justify benefit receipt. The authors also show that respondents have a similar definition of disability as the SSA, as the self-reported disability rate and DI award rate are similar for people with a given health condition, such as cancer. Finally, the authors recompute the error rate under a less restrictive assumption, that both self-reported disability status and the SSA award decision measure true disability imperfectly but without systematic bias, and find very similar results.

The authors also examine each stage of the award process to see where errors are most likely to occur. One interesting finding is that there is a high degree of self-screening by applicants - persons who report that they are disabled are much more likely to apply for DI benefits (47% of disabled persons apply vs.1% of non-disabled) and to appeal an unsuccessful initial determination (73% vs. 47%). The authors suggest that processing delays may discourage non-disabled persons from applying and appealing, as the loss of wages during the award process represents a real cost for applicants who are able to work. While there had been some concern that administrative law judges might be too lenient in awarding benefits, the authors find that the judges' decisions reduce the probability of rejecting a truly disabled person by ten percentage points, without increasing the probability of awarding benefits to a non-disabled person.

Finally, the authors design a new statistical screening rule for DI applicants. To use this rule, one would collect data on an applicant's health conditions and level of functioning, feed this data into a model created by the authors (based on observed relationships between health inputs and self-reported disability status in the HRS) to obtain a predicted probability of disability, and make an initial award to applicants with a sufficiently high probability. The authors estimate that if this screening rule replaced the first stage of DI award process, the probability of awarding benefits to non-disabled applicants in that stage would fall from 29% to 18%, while the probability of denying benefits to disabled applicants would fall from 67% to 53%.

The authors caution that there are a number of practical obstacles to implementing such a rule, such as the possibility that applicants may distort reports of their health characteristics to game the system, and that the DI award process can never be completely computerized. Nonetheless, the authors believe that their method may prove useful in helping to redesign the DI award process. They add that some of the changes recently proposed by Commissioner Barnhart, such as providing quick decisions in clear-cut cases and establishing a team of medical experts at each Regional Office, may encourage the implementation of a procedure such as the one the authors describe.