Abstract
Introduction
The benefits of employment for physical and psychological health is well-documented for people with and without disabilities (Bishop, Chapin, & Miller, 2008; Dutta, Gervey, Chan, Chou, & Ditchman, 2008). Individuals with disabilities who are employed have lower rates of alcohol use, depression, and anxiety and are found to have higher levels of self-esteem and quality of life (Dutta et al., 2008). However, individuals with disabilities have significantly lower rates of employment compared to the general population, with an estimated 32 percent of working-age adults with disabilities employed compared to 73 percent of people without disabilities from 2010 to 2012 (U.S. Department of Labor, 2013). Recognizing the importance of employment to the individual, the state-federal vocational rehabilitation (VR) program plays a critical role in reducing vocational disparities, but only 60 percent of VR consumers successfully obtain employment. Despite poor employment outcomes for people with disabilities overall, evidence suggests that subgroups—defined by type of impairment, educational attainment, or demographic characteristics—have more robust employment outcomes (Chan et al., 2016).
Furthermore, evidence suggests that competitive employment provides the greatest benefit on a broad range of rehabilitation outcomes for people with disabilities. As a result, we undertook the current study to utilize competitive employment as the outcome measure since it represents the ideal VR case closure. Findings can then be more directly applied to the state-federal VR programs in enacting the mission of integrated employment outlined by the Workforce Innovation and Opportunity Act (WIOA) of 2014 and Rehabilitation Services Administration (RSA).
In this study we specifically examine differences in type of employment defined as competitive versus not. While most studies of VR compare closures with or without employment, recent policy changes have emphasized the importance of employment in integrated settings. First, in 2001 RSA updated the definition of employment outcome for the state-federal VR program to mean work in an integrated setting (Rehabilitation Services Administration, 2001). More recently, in 2014 WIOA amended Title IV of the Rehabilitation Act of 1973 to emphasize competitive integrated employment at competitive wages as a major initiative (U.S. Department of Education, 2014).
The purpose of this study is to examine employment outcomes for 17 impairment types included in the RSA-911. We hypothesize that differences evident among types of impairment will vary across personal and demographic characteristics including age, gender, educational attainment, and race/ethnicity. A better understanding of how impairment type interacts with personal and demographic factors can enable a greater understanding of facilitators and barriers to vocational success. Moreover, this improved understanding can be used to help direct state-federal VR policy and intervention strategies to improve vocational outcomes.
We find significant differences between type of employment and impairment type, gender, age, and education. In contrast to existing research examining employment alone, we find competitive employment is lowest among those who are blind or visually impaired. Competitive employment is also lower for those with mobility, orthopedic, or mental impairments; women; older clients; and those with lower levels of educational attainment. Notably, interaction effects revealed that the differences across demographic groups vary by type of impairment.
Literature review
Impairment type
Impairment type has been found to be an important predictor of employment outcomes in the state-federal VR program. While the RSA-911 contains 19 impairments, research on employment outcomes often categorize disability type into three broad categories including: (1) sensory or communicative disabilities (e.g., visual or hearing impairments); (2) mental impairments (e.g., serious mental illness); and (3) physical disabilities (e.g., spinal cord injury). These studies have found that individuals with sensory or communicative disabilities typically have better employment outcomes, followed by individuals with physical disabilities, and then by individuals with mental impairments (Rosenthal, Chan, Wong, Kundu, & Dutta, 2006). Other studies have focused exclusively on a specific impairment population, leaving little research examining differences across the many specific impairments within the RSA-911. Our intention with the current study is to systematically study employment outcomes across 17 impairments included in the RSA-911. We are not able to include two of the 19 impairments due to limited sample size.
Gender
The role of gender in employment outcomes of those with disabilities is mixed in the literature and not fully understood. It is clear that women are typically employed at a lower rate when compared to men. However, many studies have failed to demonstrate a difference in successful case closure in the state-federal VR program by gender. While other studies have found that despite having no impact on the rate of employment, men have significantly higher earnings than women (Ireys, Gimm, & Liu, 2009).
Among people with disabilities, one recent study found, while men with disabilities are employed at higher rates than women, the employment gap between individuals with disabilities and those without disabilities is smaller for women (35 percentage points) than for men (37 percentage points) (Sevak, Houtenville, Brucker, & O’Neill, 2015). Those authors suggest that while there is no theoretical explanation for this difference, it is possible that men with disabilities face greater obstacles to employment. Further research is needed to understand the interaction of gender, disability type, and other personal and demographic factors.
Race/ethnicity
The rehabilitation literature has found considerable differences between blacks and whites in VR eligibility, service provision, and employment outcomes. However, interpretation of these finding is complex, as the relationship between race and employment outcomes varies. In a seminal analysis examining race and VR outcomes, researchers found African Americans had lower rates of receipt of VR services, training, job placement, and wages (Atkins et al., 1980). Contemporary analyses continue to find evidence of such disparities, showing that whites are more likely to be deemed eligible for VR services (Wilson, 2002; Wilson, Alston, Harley, & Mitchell, 2002), that African Americans receive less vocational training or college and university training (Boutin & Wilson, 2012), and that African Americans are less likely to be placed in competitive employment (Capella, 2002; Feist-Price, 1995; Rosenthal et al., 2006; Rosenthal, Ferrin, & Wilson, 2005; Wilson, 1999).
Despite being the fastest growing segment of the U.S. population, Latinos have not been the focus of many studies of VR. One study found that Hispanic clients were deemed eligible to receive services at a higher rate than non-Hispanics (including whites, African Americans, Asians, and Native Americans) (Wilson & Senices, 2005). Capella (2002) reported that Hispanics were 1.77 times more likely to be closed with competitive employment compared to whites. One recent study found that Latinos had a smaller disability employment gap than non-Latinos (Sevak et al., 2015). That study also identified a smaller gap among Asians.
Some studies have also noted racial differences by specific health conditions. In an analysis of RSA-911 data, African Americans with a sensory or communicative disability were 20 percent less likely to obtain employment through VR, while Native Americans with a physical or mental impairment were 51 percent and 50 percent less likely, respectively (Dutta et al., 2008).
Education
While individuals with disabilities have lower rates of educational attainment compared to the general population (Gilmore & Bose, 2005), employment outcomes are better among people with disabilities who have higher levels of education than among those with less education, and the employment gap progressively declines with higher levels of educational attainment (Sevak et al., 2015). One study found that individuals with disabilities in their twenties and thirties had a 23 percent greater chance of obtaining employment if they completed high school (Loprest & Maag, 2007), while another estimated that people with disabilities have a 50 percent greater chance of employment when completing college (Gilmore & Bose, 2005). Higher educational attainment is also associated with higher earnings for people with disabilities (Beveridge & Fabian, 2007). Evidence also suggests that the support from VR program for higher education can improve employment outcomes and job quality and is one of the strongest predictors of earnings (Boutin & Wilson, 2009; Manyibe, 2008; Migliore, Timmons, Butterworth, & Lugas, 2012). The positive benefit of college and university training has been demonstrated across a variety of disability populations, including individuals with psychiatric disabilities (Boutin & Accordino, 2011) and those with chronic arthritis (Mamboleo et al., 2015). These findings suggest that the deployment of education and training through VR might well work as an intervention to improve employment outcomes for people with disabilities.
Age
An aging population, combined with extensions in the retirement age for receipt of full Social Security benefits, will likely increase the need for VR services for older adults. However, the relationship of age and VR employment outcomes is unclear.
Several studies have found that, among individuals with chronic health conditions, older age was associated with lower employment ratings (Ipsen, 2006). One recent study found that the disability employment gap was the largest during middle age (Sevak et al., 2015), typically the highest earning years in an individual’s career. Authors also noted that decline for the workforce started at age 30 for people with disabilities, compared to age 50 for the general population (Sevak et al., 2015). In contrast, one study focused on VR clients with a sensory or communicative disability found that those over the age of 65 were three times more likely to successfully obtain competitive employment compared to individuals between 16 and 34. However, the older VR clients were also more likely to be employed at application to VR and were seeking services to maintain employment (Dutta et al., 2008). An important caveat is that many in the general population are exiting the workforce at this age, and that the employment gap gets smaller as a result.
Methods
Data
We used case service data from the Rehabilitation Services Administration Case Service Report (RSA-911) from 2010 to 2013 for the analysis. The RSA-911 details all cases closed by the state-federal VR system in a fiscal year. The dataset includes information about each case, including demographic information, primary and secondary disability, employment outcomes, and VR service utilization. The RSA-911 provides an opportunity to assess the efficacy of employment services provided by the state-federal VR program and the context from which each individual utilizes VR services, facilitating a greater understanding of factors that might enhance successful employment outcomes.
Within states, vocational rehabilitation (VR) agencies are administratively organized in two ways: one general agency serving all individuals with disabilities or two separate agencies, one for the blind and one for everyone else. For this analysis, we used data from all three kinds of agencies but nested individual observations by state and impairment.
In addition to the data provided in the RSA-911, we included a variety of state-level variables that could be related to employment outcomes. These include the yearly state unemployment rate (from the Bureau of Labor Statistics); whether the agency was in order of selection, which means some individuals were placed on a wait list (from RSA-113 reports); and the amount of federal funds allocated to the states for VR per person with disability. The funds provided per capita variable was calculated by dividing federal funds provided by RSA (from RSA-113 reports) to each state by the number of people with disabilities in the state (from the American Community Survey). We used single regression imputation (McKnight, 2007) to impute values when federal funds data were incomplete.
Sample
Because our study is focused on the quality of employment outcomes, defined as having competitive employment or not, the sample was limited to individuals who were employed at closure. The sample was further limited to individuals who lived in the United States, were between the ages of 18 to 65, were not employed at application, had no previous closures, and had complete data on all of the variables used in the models. Due to small within group sample sizes, the 266 individuals who were deaf and blind and the 1,018 individuals who had other hearing impairments were dropped from the study. The final sample size was 354,414.
The sample (see Table 1 Descriptive Statistics) was 63 percent white, 24 percent African American, and 10 percent Hispanic (any race). Only 1 percent of the sample was Asian, and 2 percent of the sample was of other races or multiracial. The scarcity of individuals in those two smaller race/ethnicity groups made it very difficult to obtain good inferences about them in the statistical model used later in this paper. The average age of the clients was 35 years old. The most frequently observed educational outcome was a high school diploma (37 percent) followed closely by not finishing high school (32 percent). Only 7 percent of the sample finished college. The remaining 23% had some college or post high school educational experience. The most frequent impairment category was cognitive (29 percent).
Models
We estimated a sequence of five nested logistic regression models using STATA v 14 (StataCorp, 2014). In these models the outcome variable was lack of competitive employment. Since 96 percent of the people in the study obtained competitive employment, only 4 percent of the sample did not.
In order to have a baseline for the subsequent models, the null model, which only includes the intercept, was fit. The first nontrivial model only fit the state variable as a predictor. The second model added yearly state characteristics such as unemployment rate, order of selection, and the federal funding per capita. The third model added the primary impairment variable. The fourth model added educational and demographic variables. The fifth model added interactions between an individual’s primary impairment and his or her educational and demographic characteristics. Based on fit statistics, the fifth model was chosen to be included in this article.
While our study focuses on differences by individual characteristics and we were not primarily interested in state variables, we did control for the state variables described above (the unemployment rate, the mean order of selection from 2010 to 2013, and the federal grant per person with disability) to isolate differences by impairment and demographics. We also controlled for individual receipt of benefits from Social Security disability benefit programs (SSDI, SSI) because it has been shown to be closely related to VR employment outcomes (Nord & Nye-Lengerman, 2015; Schaller, Yang, Ji, & Zuna, 2013). Due to space constraints, we do not present estimates for the state variables.
Results
The final model contained 81 variables and 158 interacted variables. Given the large sample of 354,414, the traditional p-value of 0.05 could lead to many false positives. In order to mitigate the potential for type 1 error, a p-value of 0.01 was used as a threshold to determine statistical significance. The large number of variables prohibits the reporting of each estimate, so we largely limit our presentation of results to those that are statistically significant. While we discuss the main effects of variables in the text, we note that their interpretation is limited and do not display them in tables. We do display tables and figures of the estimated interaction effects between impairment, educational attainment, and race/ethnicity.
We discuss the estimates as relative risks for ease of interpretation. In models where one of the two outcomes is very small, the odds ratios can be interpreted, approximately, as the relative risk of lack of competitive employment of the group relative to the reference group (Agresti, 2013). The reference group used for the analysis was white men from Alabama who had hearing loss and communicated with speech and did not graduate from high school.
Main effects
Because the model contains a series of interactions, it is difficult to interpret the coefficients on the individual variables, which are often called main effects. They represent the level difference between the reference category and a group, whereas the coefficients on the interacted variables represent the additional differences between the reference category and a group. To estimate the full difference in risk between groups requires multiplying the odds ratios for each group and its related interactions. We do this and discuss those estimates in the next section.
Based on a comparison of the noninteracted variables for type of impairment, we find that blindness is associated with the highest relative risk for lack of competitive employment (θ= 7.73, p < 0.001), relative to the reference impairment group. Visual impairment is associated with the next highest relative risk of lack of competitive employment (θ= 4.61, p < 0.001). Mobility orthopedic impairment is associated with the third highest relative risk of lack of competitive employment (θ= 2.97, p < 0.001), followed by other orthopedic impairments (θ= 2.68, p < 0.01). Cognitive impairments and other mental impairments were also at a higher risk for lack of competitive employment (θ= 2.45, p < 0.001 and θ= 2.59, p < 0.002). Lastly, other physical impairments and psychosocial impairments were at higher risk for lack of competitive employment (θ= 2.27, p < 0.012; θ= 2.02, p < 0.015).
The odds ratios for noninteracted demographic variables also revealed significant differences. In gender, females had a greater risk for lack of competitive employment compared to males (θ= 2.32, p < 0.001). Greater age was also associated with a higher risk of lack of competitive employment (θ= 1.05, p < 0.001). Hispanic ethnicity was associated with half the relative risk (θ= 0.44, p < 0.001) of non-Hispanic, white race/ethnicity.
Higher educational achievement was associated with lower relative risk of lack of competitive employment. High school completion was associated with half the relative risk (θ= 0.53, p < 0.001) of competitive employment relative to those who had not completed high school. Having an associate’s degree or some college education is associated with four times the likelihood of competitive employment (θ= 0.27, p < 0.001), and having a college degree is associated with six times the likelihood of competitive employment (θ= 0.17, p < 0.001).
Interaction effects
As described above, the model included interactions of impairment with gender, age, race/ethnicity, and education. We first describe findings on interactions between education level and impairment. In general these interactions revealed that the lower relative risk associated with higher educational attainment was not as dramatic for most impairment groups relative to the reference group of hearing loss with auditory communication. For example, we estimate odds ratios greater than one for educational attainment indicators interacted with other visual (θ HS = 1.78, p < 0.003; θ SC = 2.66, p < 0.001; θ C = 2.82, p < 0.001) and mobility and manipulation impairments (θ HS = 1.99, p < 0.002; θ SC = 3.59, p < 0.001; θ C = 6.09, p < 0.001). Additionally, having some postsecondary education but not having completed a bachelor’s degree is associated with increased relative risk of lack of competitive employment for people with the following impairments: blindness (θ= 1.78, p < 0.003), deafness with visual communication (θ= 1.980, p < 0.008), mobility and orthopedic (θ= 1.97, p < 0.002), respiratory (θ= 3.69, p < 0.003), psychosocial (θ= 2.98, p < 0.001), and other mental problems (θ= 2.69, p < 0.001). Lastly, compared to those with a bachelor’s degree in the reference group of hearing loss with auditory communication, risks of non-competitive outcomes were higher for those with a bachelor’s degree deafness with visual communication (θ= 3.11, p < 0.002), hearing loss with visual communication (θ= 3.98, p < 0.01), communicative (θ= 5.05, p < 0.006), mobility and orthopedic (θ= 3.18, p < 0.001), other orthopedic (θ= 2.72, p < 0.006), general physical (θ= 2.45, p < 0.011), psychosocial (θ= 2.53, p < 0.001), and other mental (θ= 3.05, p < 0.001) impairments.
To facilitate interpretation of the differences in relative risk by impairment and educational attainment, in Table 2 we combine the relative risk estimated on the noninteracted education variables and the education variables. These risks can be interpreted relative to the risk for the reference group hearing loss with auditory communication with educational attainment less than high school completion.
The difference in relative risk from one column to the next can be interpreted as the change in relative risk by educational attainment. For people with other visual impairments, an increase in education from completing high school to having some college reduces the relative risk of lack of competitive employment by 19 percent, and an increase in education from having some college to completing college reduces the relative risk by 37 percent. People whose primary disability is in the mobility and orthopedic category enjoy a relative risk decrease of 3 percent if they graduate from college instead of having only some postsecondary education. Individuals with mobility and manipulation impairments benefit from a 7 percent relative risk reduction if they possess some postsecondary education but did not earn a bachelor’s degree. Oddly, persons with a bachelor’s degree in this impairment category have a 1.7 percent increase in relative risk compared to people who had only some postsecondary education. People with a respiratory impairment who possess some postsecondary education less than a bachelor’s degree have 46 percent reduction in relative risk compared to people who only graduated high school. Persons with a bachelor’s degree in the impairment categories general physical, psychosocial, and other mental have a 37 percent, 27 percent, or 32 percent reduction in relative risk compared to people who had some postsecondary but not a bachelor’s degree.
Interactions between age and impairment were statistically significant for the following impairments: deafness with visual communication (θ= 0.97, p < 0.002), communicative (θ= 0.96, p < 0.01), mobility and orthopedic (θ= 0.96, p < 0.001), mobility and manipulation (θ= 0.97, p < 0.001), other orthopedic (θ= 0.96, p < 0.001), general physical (θ= 0.97, p < 0.00), other physical (θ= 0.96, p < 0.001), cognitive (θ= 0.97, p < 0.001), psychosocial (θ= 0.956, p < 0.001), and other mental (θ= 0.950, p < 0.001). These estimates indicate that the increase in relative risk of lack of competitive employment with increased age is smaller for these groups. At first glance, the interaction effects may seem trivial since they are all just slightly less than one. In reality they are not at all trivial. Since age is a continuous predictor, the product of its effect interacting with any variable grows exponentially. The cumulative age effects graph (see Fig 1) shows different impairment groups experience different degrees of relative risk of lack of competitive employment as they age. For example, the relative risk of lack of competitive employment hardly changes over time for those with psychosocial and other mental impairments. Those with deafness who communicate visually, however, have almost a fourfold increase in the relative risk of lack of competitive employment as they age. The other five impairment groups fall somewhere in between.
While the impairment interactions were most notable by age and education, some additional interactions were notable. Hispanics with cognitive impairments were over three times as likely to not be competitively employed as non-Hispanics if they had hearing loss with primarily auditory communication (θ= 3.270, p < 0.001). Women had a reduced relative risk of lack of competitive employment compared to men who had a hearing loss whose primary mode of communication was auditory in five impairment categories. These were general physical (θ= 0.66, p < 0.01), other physical (θ= 0.64, p < 0.003), mobility orthopedic (θ= 0.65, p < 0.005), mobility and manipulation (θ= 0.591, p < 0.001), and cognitive (θθ= 0.50, p < 0.001).
While the estimated relative risks for any variable or interaction may seem large, the predicted probabilities of noncompetitive closure implied by the model, when taking into account all estimates, were quite similar. It is important to note two things about predicted probabilities: (1) small absolute changes in predicted probabilities can be very meaningful if all of the predicted probabilities are small; and (2) they are generated by averaging all the variables in the model. In this sample, the overall probability of lack of competitive employment was 4 percent, so we expect to see the predicted probabilities being close to 4 percent. All of the predicted probabilities by any impairment or demographic were between 2 percent and 6 percent except those for blindness and other visual impairments, which were 22 percent and 18 percent respectively (not shown). This is not surprising since the approximate relative risks for those impairments were much higher than for the other impairments.
Summary and discussion
We first summarize the results for main effects and interactions and then discuss some of the implications for policy and practice. The relative risk of competitive employment varied across impairment groups. Women were more likely to not be competively employed than men, though this relationship was attenuated, but not totally eliminated, for a number of impairment groups. Hispanics were more likely to be competively employed than whites, which is consistent with the literature (Capella, 2002); however, the relationship was opposite for Hispanics with cognitve impairments. Increasing levels of education contributed as much as a sixfold decrease in the risk of not being competively employed, which is consistent with previous research (Boutin & Accordino, 2011; Loprest & Maag, 2007; Mamboleo et al., 2015; Sevak et al., 2015). However, these positive effects were somewhat attenuated when we looked at education interacting with the impairments. Consistent with prior research (e.g., Giesen & Cavenaugh, 2013; Ipsen, 2006), we found that increasing age is a significant predictor of poor employment outcomes, but this effect is substantively smaller for some impairment groups.
Of particular interest in the findings by impairment group is that the blind and those with other visual impairments have a greater risk of not being competitively employed. This finding shows that not all VR clients with sensory or communicative impairments have similar employment outcomes and suggests that combining these groups for analysis runs the risk of losing important details (Rosenthal et al., 2006).
Women being at greater risk of not becoming competitively employed may be a consequence of powerful social role expectations despite changes in favor of more equality in the professional and domestic lives of men and women (Kvam, Eide, & Vik, 2013). Vocational rehabilitation professionals as well as family and friends can also have an influence on women with disabilities not becoming employed by focusing on family participation instead of employment (Kvam et al., 2013). Researchers (Doren, Gau, & Lindstrom, 2011) have recommended that school and agency personnel who are providing employment and career services to yong women with disabilities ensure that these services are free of stereotypical perceptions of womens’ roles related to work and family. More specifically, they recommend that school and agency staff engage gender-neutral career assessments and career-related learning experiences and that young women with disablilties be exposed to female role models with diabilities who are employed in high-skill, high-wage jobs to expand career aspirations.
We found that higher levels of educational attainment were associated with significantly reduced risk of not being competitively employed. This was true for 12 impairment groups, and the risk reduction ranged from 3 percent to 46 percent. The good news is that rates of postsecondary education access are increasing for students with disabilities, though rates of completion remain relatively low when compared to students without disabilities (Madaus, Grigal, & Hughes, 2014). While our evidence is correlational, it is consistent with research that suggests relationships between students and academic advisors and social integration in nonacademic activities while in postsecondary educational programs may be instrumental for students with disabilities persisting to degree completion (Koch, Mamiseishvili, & Higgins, 2014). These findings point to the importance of vocational rehabilitation staff working closely with college disability service staff to focus on more than accommodations. There is also longitudinal research pointing to antecedent secondary educational factors related to completion of postsecondary education; these factors include students with disabilities self-advocating, participating in extracurricular activities, and having opportunities to develop vocational skills through work-study or paid employment (Achola, 2014). Evidence such as this provides direction to state vocational rehabilitation agencies as they take on additional responsibilities under WIOA devoting 15 percent of their budgets to working with special educational students to ensure a meaningful transition from high school to postsecondary education and ultimately competitive employment.
We found a more nuanced story for aging than is presented in previous studies (e.g., Giesen & Cavenaugh, 2013; Ipsen, 2006) when we look at the interaction of age with impairment. We saw great variation in the degree to which aging is associated with poorer employment outcomes across impairment groups. This finding is consistent with the more general observation that older workers are remaining in the labor force for longer. It is estimated that by 2050 22 percent of the U.S. population will be over the age of 60 and this aging population will increase the demand for VR services considering two factors: (1) many older adults are wanting or needing to work well beyond the traditional retirement age; and (2) aging brings with it related chronic illness and disability conditions that can benefit from rehabilitation services to enhance productive capacity. Moreover, employment for older adults is a protective factor for health and wellness, improves psychosocial well-being, and life satisfaction (Momtaz, Hamid, Haron, & Bagat, 2016; Wahrendorf, 2015).
This trend toward older workers staying employed presents opportunities for vocational rehabilitation staff to support older workers as they approach retirement age. For example, among older workers who want to stay in their current jobs either full time or part time for psychological or financial reasons, some without disability histories will develop age-related impairments and need services to stay employed. In addition, there are individuals who have had disabilities for some time and who acquire additional age-related impairments and need services to keep their jobs. There may also be older workers with disabilities approaching retirement that want to leave their current employment situation and shift to another field of work and look to vocational rehabilitation practitioners to assist them with career exploration and securing meaningful employment. Regardless of the older consumer’s employment aspirations, vocational rehabilitation counselors need to appreciate that movement into retirement is fraught with issues related to identity disruption, decision anxiety, loss of self-assurance, disruption of social supports, and existential anxiety (Osborne, 2012). Hershenson (2014) even recommends that counselors develop an individual retirement plan in addition to an individual plan for employment for older vocational rehabilitation consumers.
Conflict of interest
The authors have no conflict of interest to report.
Footnotes
Acknowledgments
This project was funded by the National Institute on Disability, Independent Living, and Rehabilitation Research (U.S. Department of Health and Human Services) Rehabilitation and Research Training Center on Individual Characteristics, under cooperative agreement 90RT5017-01-01. The findings and conclusions are those of the authors and do not represent the policy of HHS or NIDILRR. The authors retain sole responsibility for any errors or omissions.