Employment Patterns Before Applying for Disability Insurance

Introduction

Many Social Security Disability Insurance (DI) policy proposals feature early-intervention and worker-retention objectives. If workers with disabilities are provided with adequate supports, they may be diverted from applying for DI benefits. To be effective, these proposals should identify the types of people who could benefit most from such proposals, as DI applicants have varied backgrounds and characteristics (Thompkins and others 2014). Casting too broad a net might misplace resources on individuals who are not able to remain in or return to the labor force, or on those who already have adequate access to supports. Casting too narrow a net might miss people who could benefit from employment supports, which would limit the potential returns both for at-risk individuals and for the program.

This article examines the employment patterns and demographic characteristics of DI applicants in the period before application. Such information can help inform various policy proposals involving early intervention, worker retention, and program diversion by identifying how various groups of applicants could be better targeted and by assessing the potential reach of such proposals. We rely on Survey of Income and Program Participation (SIPP) data matched to Social Security Administration (SSA) records to answer questions about the employment, demographic, and other characteristics of DI beneficiaries before they apply for DI, with particular emphasis on their detailed employment patterns, their participation in non-DI public programs, and their coverage under selected types of insurance. For brevity, in this article, we use “program participation” to refer broadly to receipt of benefits provided by non-DI public programs or private insurance.

Using four distinct preapplication observation periods ranging from 6 to 24 months, we find that about half of DI applicants were consistently employed until they applied or until they experienced a single definitive work cessation before application. The other half, all of whom met DI's overall work-history requirements, either did not work at all or had intermittent employment in the preapplication period. DI allowance rates, program-participation patterns, and demographic characteristics differed between those two halves of the observed applicant population.

Our findings contribute to the literature by identifying types of people, in terms of work histories and DI award probabilities, who might be likely candidates for early-intervention or worker-retention initiatives. DI proposals could target at-risk workers for supports either through their employers or through the public programs in which they participate. Evidence on the employment patterns of people who are likely to apply for DI can help policymakers identify potential target populations, tailor program changes to subgroups that may follow very different paths to DI, and more generally, ensure that program changes are successful and cost-effective.

Background

A substantial body of literature addresses policy proposals that aim to support employment retention before workers apply for DI, including several SSA studies (such as Kearney and others 2005/2006). Providing supports to workers while they still have jobs—at the time when they encounter a potentially disabling health condition or their existing health condition worsens—is widely seen as a better way to promote independence than waiting until they apply for DI to provide such supports (Autor and Duggan 2010; McCrery and Pomeroy 2016). Advocates of early-intervention initiatives cite the economic advantages to all parties involved—workers, employers, communities, and state and federal agencies—of keeping people with disabilities in the workforce when possible. These initiatives are informed by increasing evidence suggesting that some DI beneficiaries could work if given appropriate supports (Autor and Duggan 2003; Black, Daniel, and Sanders 2002; von Wachter, Song, and Manchester 2011). This evidence is not overwhelming, however, as other studies have found that the level of retained employment might be relatively low (for example, Maestas, Mullen, and Strand 2013). Evidence from SSA demonstrations (such as Frey and others 2011 and Gubits and others 2014) also shows that although some DI beneficiaries can work with targeted services and supports, few attain earnings levels sufficient to cease benefits.

What most proposals have in common is a need to identify workers who are likely to apply for and receive DI benefits before they actually do so—ideally, while they are still in the labor force. A related objective is to identify those potential applicants who would be most effectively served by a particular policy intervention, as well as the best time to intervene, especially given variations in the timing of earnings declines by age, sex, and disability type in the period before an individual applies for DI (Costa 2017). Successful targeting of potential applicants is crucial to the effectiveness of any proposed policy. Previous work has highlighted several characteristics that might provide a starting point for identifying such people.

Early-Intervention Approaches to Worker Retention

People rarely exit the DI rolls once they begin receiving benefits (Liebman and Smalligan 2013; Liu and Stapleton 2010), and many disability researchers and policymakers have concluded that the most effective intervention occurs before a person applies for benefits—preferably, while he or she is still working (Stapleton, Mann, and Song 2015). This conclusion adds an unanticipated dimension to policy considerations because the DI system was created to support people who can no longer work, rather than those who remain employed (Burkhauser and Daly 2011). Most early-intervention approaches, although designed to provide better supports to workers to divert them from needing benefits, target either employers or public programs.

Employer-focused proposals. These efforts would either mandate private (or hybrid) short-term disability insurance to cover employees who acquire a disabling condition or use an experience-rating approach in administering the disability portion of the Old-Age, Survivors, and Disability Insurance (OASDI) payroll tax. Mandatory short-term disability insurance would provide employers and employees with worker-retention supports (such as income replacement, vocational rehabilitation, and workplace accommodations) for up to 2 years. Eligibility for DI benefits would begin only at the end of the 2-year period. Depending on their size, employers would pay insurance premiums that vary based on the tendency of their employees to file claims, which would provide incentives for firms to retain their workers with disabilities (Autor and Duggan 2010). An alternative policy proposal would shift the disability portion of the OASDI payroll tax to an experience-rating system (Burkhauser and Daly 2011). Experience rating is already used in determining workers' compensation and unemployment insurance (UI) employer contribution amounts, for similar reasons. Employers with large numbers of employees claiming disability benefits would face higher tax rates, which would encourage them to lower disability claims by finding ways to keep workers with disabilities on the job.

Employer-focused approaches offer many advantages. For example, because employers are best positioned to observe their employees' work performances, policies can establish incentives or laws for small employers to provide accommodations (such as those mandated for employers of 15 or more workers by the Americans with Disabilities Act) that help workers stay in the labor force.¹ Employers and employees both might benefit from improved rehabilitation and vocational supports that encourage employees with work limitations to maintain employment. Taking a more longitudinal perspective, employers might also institute measures that can delay or prevent the onset of health conditions that could lead to work-limiting disabilities, such as promoting ergonomic work environments. An important downside, however, is that placing another burden on employers in the form of short-term disability insurance premiums or experience ratings might increase existing incentives to avoid hiring or retaining people who are at greater risk for disability.

Public program–focused proposals. These efforts could encourage program changes, either through SSA and the disability determination process or through the collaboration of state-level organizations, to provide more workplace supports to people with disabilities. Providing individualized supports and wage subsidies to potential DI applicants might encourage them to stay in the workforce, alleviating the need to apply (for example, Liebman and Smalligan 2013). More ambitious proposals would aim to identify individuals as they experience the onset or worsening of medical conditions that threaten their ability to remain employed and potentially qualify them for DI. Targeted individuals would receive appropriate supports that might enable them to remain in work (McCrery and Pomeroy 2016, Chapter 3). Expedited DI application and processing for individuals who are not capable of work could be included in either proposal.

Another public-program policy change would switch funding for state disability determination services from the discretionary to the mandatory side of the budget, making public disability programs more akin to Temporary Assistance for Needy Families (TANF), Medicaid, and the Supplemental Nutrition Assistance Program (SNAP) (Liebman and Smalligan 2013). This change would provide more resources for purposes such as reducing backlogs, performing continuing disability reviews, and collecting evidence that leads to better disability determinations early in the application process. If improved administration at the state level results in more appropriate benefit allowances or continuing disability decisions, marginal applicants might opt not to expend the effort to apply. On the other hand, faster decisions might reduce the opportunity cost of applying for DI. One important caveat of this approach is that if expenditures for administrative costs were mandatory rather than discretionary, there would be no effective limit on them.

Alternatively, government policymakers could offer or expand evidence-based early-intervention services to their own workforce or to the general population through incremental or targeted policies (Stapleton, Mann, and Song 2015). Those services could be delivered either as part of the package of benefits given to state and federal employees or, more broadly, as a coordinated part of the services already delivered by state-level labor and education agencies (such as American Job Centers and state vocational rehabilitation agencies) to a state's employers and employees. Such approaches would build on the experiences of states pursuing similar initiatives (Ben-Shalom and others 2017). The latter policy option might be bolstered by recent changes at the state and federal levels in the provision of such services resulting from the Workforce Innovation and Opportunity Act of 2014.

Research Questions

The proposals summarized above raise questions about who could be targeted for each type of initiative. DI applicants who meet the work-history requirements for DI eligibility provide a useful sample for considering these questions, as they have recent labor-force attachment and have health conditions that are serious enough to warrant an application for the economic supports that DI provides. DI applicants are not a homogeneous population (Honeycutt 2004; Lahiri, Song, and Wixon 2008; Lindner 2013; Livermore, Stapleton, and Claypool 2010). Some have strong work histories; others have sporadic or no labor-force attachment in the period before DI application. Applicants may also have widely varying levels of prior involvement with various types of public programs.

This analysis answers four research questions:

1. What are DI applicants' preapplication employment patterns?
2. How do employment patterns differ over varying preapplication time periods?
3. What demographic, job, and non-DI program participation characteristics are associated with each employment pattern?
4. How do these characteristics and patterns relate to the likelihood of DI allowance?

The temporal relation between DI application and prior labor-force and non-DI program participation can provide insights into how best to reach potential DI applicants and into identifying which applicants are more or less likely to be affected by various proposals.

Data

The analysis relies on a pooled sample from the 1996, 2001, and 2004 panels of the SIPP.² The SIPP is nationally representative of households in each panel's initial year, with its sample weighted to reflect the civilian noninstitutionalized population aged 15 or older.

We used Social Security administrative files linked to SIPP data to identify people who applied for DI, along with their application dates, their receipt of DI benefits and Supplemental Security Income (SSI), and the outcome of their application at the initial or reconsideration levels. We only considered applicants who met the earnings requirement for DI eligibility (meaning they had a qualifying work history and were fully insured for the program) and received a medical disability determination. Not all SIPP data could be matched to SSA records: Some SIPP respondents did not provide Social Security numbers, some respondents opted out of having their data matched to federal records (beginning in 2004), and some SIPP information (such as Social Security number, name, sex, and date of birth) that respondents provided did not match the administrative data (McNabb and others 2009). The match rates were 84 percent for the 1996 panel, 60 percent for the 2001 panel, and 79 percent for the 2004 panel. Using the Social Security administrative data, we excluded people who had already received DI benefits as of the first SIPP wave from our analysis sample. The statistics presented here could therefore be biased if nonmatched respondents differ systematically by DI receipt or application status from matched respondents; we did not calculate new weights based on the sample exclusions.

We restricted our sample to DI applicants aged 25–55 with matched data whose first survey response occurred in wave 1 of each SIPP panel (as opposed to including those who joined a panel after wave 1). We excluded applicants younger than 25 because they are less likely to qualify for DI and more likely to be enrolled in school.³ We excluded applicants older than 55 to avoid tracking sample members who might qualify for early retirement benefits during our observation periods.

We tracked preapplication employment patterns to categorize and compare individuals according to various characteristics. We first established four observation periods, respectively consisting of the final 6, 12, 18, and 24 months before the month of DI application. The observation-period subsamples overlap in that each person included in the 24-month subsample is also included in the larger subsamples for each successively shorter observation period. That is, the subsample for the 6-month observation period is larger than the 24-month subsample because more applicants had 6 months of preapplication SIPP data than had 24 months of preapplication data. Second, for each observation period, we categorize applicants based on their employment data for all observed months before the month of DI application. The four mutually exclusive categories we consider are (1) consistently employed: employed in all months of the observation period; (2) ceased employment: consistent or intermittent employment that ended, without subsequent resumption, during the observation period; (3) intermittent employment: employed in some months but not others, with no single or definitive work cessation in the observation period; and (4) not employed: no employment in the entire observation period. A given individual's employment category might differ from a shorter observation period to a longer one. For example, an individual observed for 12 months who worked steadily until 8 months before DI application and then ceased employment would be classified as “ceased employment” in the 12-month period but as “not employed” in the 6-month period.

In addition to employment histories, we analyzed the following characteristics of the DI applicants in our sample:

• Demographic characteristics, including sex, age, race, marital status, presence of children in the household, poverty status of the household, and education. This information came from the earliest SIPP observation for each individual.
• Program participation, including receipt of Medicaid, SNAP, TANF, UI, and workers' compensation benefits. Under this category, we also analyzed private health insurance and employer- and self-provided disability insurance coverage. We used SIPP data to assess these measures across all months of an observation period; for some applicants, participation or coverage status varied from one time period to another.
• Job characteristics, including binary indicators for full-time status, ever being laid off, ever having two or more jobs (moonlighting), and union membership; industry division (services, goods-producing or other); and industry sector (private for-profit, public, nonprofit or self-employed). These data were based on the person's earliest employment experience in the SIPP observation period.⁴
• DI claim outcome. From the Social Security administrative data, we identified whether a DI claim was allowed at the initial or reconsideration levels of review. We did not consider the outcome of applications at the hearing or appeal levels, which account for about 1 in 3 ultimately successful applications.⁵ We focus on the potential for early-intervention programs to offer an alternative to pursuing DI benefits.

Methodology

This analysis incorporated both descriptive statistics and regression models, with separate estimations for each observation period. We weighted the data using the wave-1 weights for each SIPP panel, and we applied the SIPP-recommended adjustment factors to our variance estimates to account for the survey's complex sampling design.

For the descriptive statistics, we started with the number of applicants in each employment category for each observation period, along with their DI allowance rates. Then, we stratified applicants in the 6-month period (the largest sample) by employment category to compare their demographic, program participation, and job characteristics (described above).

Next, we assessed the relationship between the preapplication employment category and each of the observable applicant characteristics, holding other characteristics constant. To do so, we used multinomial logistic regression models to estimate the sample members' preapplication employment categories; that is, the dependent variable for each of these regressions was the employment category. We ran two models four times each, once for each observation period. The first model included all DI applicants and only those characteristics that were observed for all of them. The second model used the subset of DI applicants who reported working at any point during the observation period and included job characteristics as predictors. In the results section, we report the average marginal effect of each predictor on the probability of belonging to a given employment category.⁶

Finally, we used logistic regression models to estimate DI allowance as a function of employment category and non-DI program participation during the preapplication period. We again ran two models four times each, with the first model including all applicants and the second including only those employed at some point during the relevant period. Both models also analyzed job characteristics. Again, we report the average marginal effects.

Results

In this section we present findings from our descriptive and regression analyses.

Descriptive Analyses

The sample distribution across the four employment categories varies based on the length of the observation period (Table 1 and Chart 1). In the 24-month period, only 13 percent of our sample was not employed, and only 22 percent was employed the whole time. This distribution shifts in shorter observation periods. In the 6-month preapplication period, one-third (33 percent) of the sample was not employed, and more than one-fourth (28 percent) was consistently employed.

To explore the changing distribution among employment categories over time, we further analyzed the subsample of applicants for whom we were able to observe the full 24-month preapplication employment history. We tracked that subgroup's employment-category patterns across each of the four observation periods. Additionally, for the consistently employed and not-employed categories, we distinguished the individuals who met the category definition in the 24-month period from those who met the definition only in one of the shorter periods (Table 2 and Chart 2). An individual who met the definition for ceased employment in the 24-month period could have been classified only as ceased employment or not employed in shorter periods. By contrast, an individual defined as intermittently employed in the 24-month period could have an employment history that meets the definition for any of the four employment categories in shorter periods. The majority of individuals classified as intermittently employed in the 24-month period were classified as either intermittently employed or consistently employed in the 6-month period, indicating that for some workers, intermittent work is the norm. In addition, a large proportion of those who worked consistently until a definitive cessation experienced that cessation within 6 months of applying for DI.

The patterns shown in Charts 1 and 2 suggest that we can consider DI applicants as being one of two types. The Type 1 applicant works consistently up to or shortly before the point of application. Such a person can be considered to have a strong attachment to the labor force. Policy proposals such as those described earlier are generally geared toward this type of worker in that they presume a substantial existing relationship with an employer that continues until or almost until the point of DI application. The Type 2 applicant has a weaker attachment to the labor force, working only intermittently or not at all for long periods (up to and possibly exceeding 24 months) before applying for DI. Early-intervention efforts that rely on identifying people with current or recent work attachments are likely to miss persons in this group. Because about half of the applicants in our sample were Type 2, policy proposals that tacitly focus on Type 1 applicants overlook a substantial target population.

Table 3 presents summary statistics for our 6-month subsample, the largest of the four, broken down by preapplication employment category.⁷ Here again, we see differences between Type 1 and Type 2 applicants, with the former more likely to have higher educational attainment, more likely to have employer-provided health insurance, and less likely to have relied on public programs. Applicants who were consistently employed were more likely to be male, white, have a child in the household, have income above the poverty level, and have a college education than were applicants in other employment categories. In addition, compared with those in other employment categories, applicants who were intermittently employed were more likely to be black, to be never married, and to have a high school diploma or equivalent as their highest education level, while applicants who were not employed had a higher proportion with less than a high school education. Consistently employed applicants had the highest average annual household income; applicants who reported being not employed had the lowest annual household income.

Regarding program participation, applicants with intermittent or no employment were more likely to be receiving Medicaid, SNAP, or TANF benefits. Applicants who were consistently employed or had ceased employment were more likely to have had private health insurance. People with intermittent or no employment were more likely to report receiving UI or workers' compensation.

Job characteristics did not vary widely across the three applicable employment categories. Applicants who were consistently employed were slightly less likely to work in the private sector or to have moonlighted in a given month and more likely to work in the public sector.

Another way Type 1 and Type 2 applicants differed is in the likelihood of DI allowance. Tables 1 and 2 show the allowance rates for the full sample and the 24-month subsample, respectively, across employment categories.⁸ In all time periods, applications of persons who were employed or who ceased employment after working consistently were much more likely to be allowed than were those filed by persons who worked only intermittently or not at all (a difference of more than 10 percentage points in any time period).

Regression Analyses

In the previous section, we identified two types of DI applicants, characterized primarily by their labor force attachment in the preapplication period. In this section, we summarize the findings from logistic regression models that predict preapplication employment category. We first consider the association of demographic, program participation, and job characteristics with the likelihood of an individual's falling into a given employment category for a given observation period. We then use employment categories (along with program participation and job characteristics) as predictors of DI allowance.

Predicting employment category. Many of the age, marital-status, and educational-attainment categories were consistently significant predictors of employment category (Table 4). For instance, being older than 40 reduced the chances of being intermittently employed. Unmarried applicants were more likely than married ones to be consistently employed and less likely to be not employed. Relative to college graduates, applicants with lower educational attainment were less likely to be consistently employed.

Within the Type 1 employment categories, receipt of self-funded private disability insurance, UI, or workers' compensation benefits was associated with lower probability of consistent employment and higher probability of employment cessation. This pattern generally held across the four observation periods. The explanation seems clear: These programs are designed to help established workers who lose their jobs because of a newly disabling condition or, in the case of UI, an involuntary layoff for any reason.

Among Type 2 applicants, program participation is weakly associated with whether a person was employed intermittently or not at all. There is some evidence that receipt of UI benefits correlates positively with intermittent employment and negatively with not being employed, but only in the longer observation periods. In the 6-month period, receipt of UI benefits is actually associated with increased odds of being not employed, likely because most job losses that precipitated UI benefits occurred more than 6 months before DI application. UI may be associated with not being employed in part because of a pattern wherein people lose their jobs, collect UI benefits for the typical duration of 6 months, then apply for DI either while still receiving UI benefits or directly after they are exhausted (Lindner 2016). Another factor might be that DI requires a 5-month waiting period between the established onset date and initial benefit eligibility.

We also observed significant associations between employment category and use of programs such as private health insurance, SNAP, and TANF. We do not believe the model identifies a causal relationship, but we think this association likely reflects the effect of employment status on program participation rather than the other way around. For example, we observed that having private health insurance was positively associated with consistent employment and ceased employment, and was negatively associated with not being employed; we observed the opposite for SNAP and TANF. People who are consistently employed often have health insurance through their employer and earn too much to qualify for means-tested programs such as SNAP and TANF.

Our analysis of the relationship between job characteristics and preapplication employment category necessarily excluded applicants in the not-employed category, as they had no job characteristics to observe. Controlling for demographic characteristics, we found that job characteristics were generally weakly and inconsistently related to preapplication employment categories (Table 5). Union membership was associated with a higher probability of employment cessation and a lower probability of intermittent employment. However, this result emerged only in the two longest observation periods. Not surprisingly, having ever been laid off is positively correlated with intermittent employment and negatively correlated with consistent employment. (Being laid off and subsequently finding another job meets our definition of intermittent employment.)

Predicting DI allowance. With our second set of regressions, we found that the preapplication employment categories serve as useful predictors of DI allowance at the initial or reconsideration levels after controlling for individual characteristics (Table 6). The difference between the probability of allowance for applicants who were consistently employed and the probability for applicants who ceased employment was not statistically significant. By contrast, applicants who were intermittently or not employed were less likely to be allowed, with the effect most significant in shorter observation periods.

Age was the only demographic characteristic for which we found statistically significant results. Relative to applicants in their 20s, older applicants were more likely to be allowed. This finding is not surprising, as most disabled-worker beneficiaries are aged 50 or older and DI eligibility rules consider age in the last step of the determination process. With few exceptions, we found that participation in a specific non-DI program was not significantly associated with DI allowance. The most noteworthy finding is that receipt of workers' compensation was negatively associated with DI allowance, with statistically significant estimates in the 6- and 12-month preapplication periods.

We found similar results when we limited the sample to applicants who reported some employment during the observation period (Table 7). Again, intermittent employment was associated with a significantly reduced likelihood of DI allowance, and employment cessation was not associated with a probability of allowance that differed from that for people who were consistently employed. We again found that older applicants were more likely to be allowed. We did not find any evidence that the job characteristics we analyzed were associated with the probability of DI allowance.

Conclusion

This study uses Social Security administrative data to examine patterns of employment and non-DI program participation in the months leading up to DI application. People follow different preapplication paths, and a given individual's path may indicate the likelihood of his or her application being allowed at the initial or reconsideration level. About half of DI applicants worked consistently either to the point of application or shortly before application, with a cessation and no subsequent resumption. We call these individuals Type 1 applicants. They are characterized by stable employment in well-paying jobs, often with benefits such as private health insurance. Applicants from this group had a higher likelihood of DI allowance.

The other half—the Type 2 applicants—either had been out of the workforce for a long time (many for at least 24 months) or had intermittent work histories. Members of this group were less likely to receive DI benefits than were the Type 1 applicants, and they tended to rely more on means-tested and social insurance programs (such as UI and workers' compensation) for support.

Based on our results, early-intervention or return-to-work programs that focus on DI applicants with more recent attachments to the workforce (Type 1) are likely to fail to target about half of the individuals who eventually apply. The question, therefore, is whether policy proposals can capture Type 2 applicants while those applicants, even without a long-term attachment to an employer, still consider themselves to be in the labor force. Type 1 applicants likely have better human capital and skills to build upon as they attempt to return to—or maintain—their employment, so early interventions that provide high-quality and timely medical and rehabilitative services, accommodations, and assistive technologies may help them to use those skills, potentially with the same employer. However, their higher rate of DI allowance may indicate that Type 1 applicants have impairments that clearly inhibit their ability to work at substantial levels, in which case interventions may be less likely to succeed.

Type 2 applicants typically have comparatively limited human capital and skills, as well as lower income and fewer resources—characteristics which, when combined with medical problems, make it difficult for them to find and maintain good jobs. Given their economic situations, their opportunity costs of applying for DI might be lower than those of Type 1 applicants, and their lower DI allowance rates might indicate less severely disabling conditions on average. To succeed, efforts to help these applicants should identify them either when they are still working (with early-intervention services that address the full array of issues that prevent them from holding better jobs) or after they have left the labor market (with services that help them to reconnect with employers). Identifying such people before they apply may require outreach via health care providers, administrators of other programs in which they may participate, and the media.

Because Type 2 applicants do not have secure labor force attachment in the preapplication period, employer-focused proposals might have less reach than do broader systemic approaches that improve supports for those seeking DI benefits or that focus on work capacity. If an intermittent work history is symptomatic of a disability that could be managed with appropriate supports, then providing ongoing and condition-specific supports might be logical policy objectives.

It is hard to know whether the return on investment for early-intervention services that target Type 2 applicants is higher or lower than that for services that target Type 1 applicants. On one hand, Type 1 workers' longstanding attachment to the workforce—and potentially to a particular employer—might make it easier to retain them in the workplace, as they may need only timely access to rehabilitation services, workplace accommodations, or supportive technology to remain productive. On the other hand, Type 1 applicants may already have access to such services through their employers, and to the extent that their higher allowance rate reflects impairments that more clearly meet SSA's disability definition, focusing efforts on these people may not offer the greatest return on investment for early-intervention programs. Type 2 applicants likely have lower human capital, are harder to target, and may require a broader array of services (including ongoing support) to stay employed, but they may be less likely than Type 1 applicants to already have access to services that would keep them in the workforce. Furthermore, the benefits of enhancing the capacity for independence among Type 2 applicants would include not just diversion from DI, but a potential decrease in reliance on the other public programs that these individuals turn to for support. Even if investments in Type 2 applicants ultimately provide a lower return, efforts to target them could likely be justified on equity grounds because of their low income, frequency of experiencing poverty, and other potential barriers to employment.

Although this information adds to our understanding of DI applicants, important unknown factors remain. SIPP information on health and disability characteristics is incomplete, and this analysis would have benefited from having additional information to allow the consideration of health status over time, condition type, and the timing of the health-condition onset that precipitated the DI application. Future research could explore these relationships, as well as the reasons for possible denial and whether they differ between Type 1 and Type 2 applicants.

We imposed further data limitations as well. Specifically, we did not include more recent SIPP waves, and we assessed application outcomes only at the initial and reconsideration levels. In addition, even after we pooled the three SIPP panels, the number of people whom we observed applying for DI benefits was small, particularly for some of the characteristics in which we were most interested. The sample was further restricted in that not all SIPP respondents could be matched to SSA records. The small sample sizes, particularly for the 18- and 24-month subgroups, provided less precision for our results than we would like, and may explain why we found that job characteristics are, for the most part, not predictive of preapplication employment categories or DI awards.