For agencies not in OOS status, we find that in a bivariate context, limited VR service availability (a lower percentage of non-beneficiary applicants to the same agency in the same month that receive an IPE and a longer mean duration from application to IPE, a measure of access delay) is associated with higher SSA program application and receipt. After controlling for several factors that might partially explain the bivariate relationships, these relationships remain. Although other factors might account for the latter results, they are consistent with the hypothesis that more timely access to VR services for those applicants not already receiving SSA program benefits reduces the likelihood that they will apply and receive those benefits in the next 48 months. The results are substantially stronger for SSDI than for SSI. The weaker results for SSI might reflect the inclusion of many VR applicants who do not meet the SSI means test in the sample for the SSI analysis.
Of course, the results do not prove that greater access to VR services for non-beneficiary applicants would reduce SSA program application and allowance receipt. Our statistical methods and focus on never in OOS agencies tries to isolate variations in service availability that are plausibly exogenous to individual factors, but our results nonetheless remain associations.
Parallel analyses for agencies that were in OOS during all or part of our sample period found much weaker and less consistent evidence of a negative relationship between our access measures and SSA program outcomes. It seems likely that the way these agencies provide priority to applicants with the most significant disabilities undermines the validity of our access measures for those most likely to apply for SSA benefits.
One caveat to the magnitude of our estimates is that the time period over which we were able to follow individuals necessarily led to right-censoring on the outcome measures at month 48 after the VR applications. Censoring might be especially important for benefit receipt. If waiting for VR services induces VR applicants to apply for SSA benefits, it seems likely that SSA applications would be filed only some months after VR application. Receipt of SSA benefits may not occur until much later. During the study period, initial SSDI decisions took approximately 6 months, and many awards were made only after the appeal of an initial denial—a process that can take years (Lindner and Burdick 2013). Though not presented, our findings were qualitatively robust to alternative specifications, including both shorter and longer time frames of application and benefit receipt and consideration of an outcome measure that accounted for application to either SSDI or SSI (rather than considering each outcome separately).
There are several limitations to the analysis. First, we did not consider two groups of VR applicants that might be deterred from leaving the labor force and entering SSDI by timely access to VR services: those who had applied for SSDI and then applied for VR services while their SSDI applications were pending, and those who were not disability insured at the time they applied VR services but became disability insured in the near future, perhaps with the help of VR services. Both groups are problematic to identify but could be very important. The first group is important because the SSDI determination process is lengthy, so some might receive VR services and return to work while waiting for a decision from SSA. The second group is important because VR services received by those who are not disability insured might be instrumental to gaining the work experience they need to become disability insured. It seems likely that the effect of VR service availability for these groups would be in the same direction as for the disability-insured VR applicants included in our sample, but smaller.
Second, we have focused on just one aspect of how resource constraints affect VR services: timely availability. Resource constraints can also affect SSA outcomes by their effects on other aspects of service quality, but no suitable measures of quality are available. It might be that our two measures of service availability are highly correlated with other dimensions of quality that are affected by resource availability, in which case the estimated coefficients reflect effects of resource constraints on SSA outcomes via effects on unobserved dimensions of service quality and overstate the impact of the timely availability of VR services per se.
Third, our model controls for only characteristics and the environment at the time of VR application. The broader environment for workers with disabilities could have changed in the years after VR application in ways that could have affected the decision to apply for federal disability benefits. For example, during the time period of our study, states were enacting programs such as Medicaid Buy-In and making broader systems change with the help of Medicaid Infrastructure Grant funding, which aimed to promote employment. Perhaps more importantly, the 48-month window of the last annual applicant cohort (2005) includes the recession that started in the last quarter of 2007. These factors might well be important for some VR applicants, but they would have to be correlated with the service availability measures even after accounting for our control variables to affect the estimated coefficients for the service availability measures; we have no reason to think that they are.
To illustrate the magnitude of the point estimates for agencies never in OOS, we considered how modest exogenous changes in our access measures would change SSDI application and receipt if we assume that the estimated coefficients are unbiased estimates of the effect of such exogenous changes over the range of the variable considered. The specific change we simulate is an increase of IPEat in each month in which the value is below the 90th percentile of the observed distribution to the 90th percentile, and a reduction in the mean waiting time in months that are above the 10th percentile to the 10th percentile. The selection of these percentiles was arbitrary, but can be considered to reflect attainable levels of service availability as they have, in fact, been observed in a substantial share of sample months. The exercise also assumes that the effects of the two changes are additive—an assumption that is built into the regression model, but not necessarily true. Given our assumptions, these changes would together yield 9,300 fewer applicants to SSDI in the following 48 months, a reduction of 11.0 percent, and 6,900 fewer recipients, a reduction of 24 percent. As these predictions are for four annual applicant cohorts, the mean reductions per annual cohort over the sample period are 2,325 applicants and 1,725 fewer recipients. These predictions are only suggestive because of the limitations that are inherent to our analysis.
Even if we assume the coefficients are unbiased estimates of the additive effects of exogenous changes to the two access measures, the estimated effects of these substantial changes in access on SSA outcomes would be modest relative to the number of annual SSDI applications and awards. From 2002 to 2005, SSA averaged around 2 million applications for SSDI each year, of whom approximately 900,000 eventually received benefits (Social Security Administration 2013a). It should be noted that the agencies not in OOS include only about half of all agencies, and we do not have a plausible prediction for how expanding access to their non-beneficiary applicants would affect SSA outcomes. Even if we could double or triple the projected impact for the non-OOS agencies, however, the impact would still be small relative to the scale of all applicants and awards.
Yet, if VR services can divert even a few individuals from applying for federal benefits, cost savings could be substantial. Using administrative data from all four programs, Riley and Rupp (2012) estimated the present value of federal expenditures on SSDI, SSI, Medicare, and Medicaid for the average new adult SSDI or SSI beneficiary in 2000 to be $123,000 (adjusted for inflation to 2011 dollars) through 2006 (six to seven years after award)14. As savings would continue to accrue over time, it is not hard to see that relatively few cases diverted from benefits by VR services would lead to savings rivaling the payments that SSA makes to VR agencies for providing services to existing beneficiaries under its cost reimbursement payment system (which was $73 million in 2011) (SSA 2013b).
It is also possible that substantial savings could be achieved by changing the regulations and incentives under which VR agencies operate in a manner that would cause them to give greater priority to those non-beneficiary applicants who are most likely to apply for and enter SSDI or SSI. Current regulations and incentives favor serving existing beneficiaries, and even helping non-beneficiaries obtain benefits. The federal requirement to give priority to those with the most significant disabilities means that most SSA beneficiary applicants for VR services receive the highest priority. Further, SSA will pay the VR agency for serving beneficiary applicants if the applicant becomes competitively employed for nine months at the completion of VR service receipt, even if the beneficiary continues to receive benefits. The state government may also benefit when a VR applicant becomes an SSDI beneficiary, especially if the state is paying for some of the applicant’s health care under Medicaid or through other state programs. That is because the new beneficiary will eventually be entitled to Medicare, a federal program. Our estimates reflect a system that was not designed to deter entry into SSDI or SSI.
While our findings are not sufficiently strong to support a major overhaul in federal funding and regulations for VR services, they do suggest the need to develop and test approaches to determine whether VR agencies could help keep their applicants, or potential applicants, in the labor force and out of SSDI, and whether such efforts could pay for themselves through reduced benefits and, more broadly, greater productivity. For example, VR agencies might be able to reduce SSDI entry by reaching out to workers who are experiencing the onset of disability, perhaps via their employers. Delivering services to such workers before they lose their jobs might be a more effective way to keep them in the labor force. VR agencies rarely do this however. Alabama’s Retaining a Valued Employee program is an interesting exception that has been gaining interest from other state agencies in recent years. At present, however, federal regulations and funding do not encourage such innovations, and any such innovations are not being systematically evaluated.