Return-to-Work Outcome Rates of African American Versus White Veterans Served by State Vocational Rehabilitation Agencies

Abstract

The purpose of this study was to identify disparities in successful return-to-work outcome rates based on race, gender, and level of educational attainment at closure among veterans with a signed Individualized Plan for Employment (IPE). A randomized split-half cross-model validation research design was used to develop and test a series of logistic regression models for goodness of fit across two samples (i.e., screening and calibration) of case records (N = 11,337) obtained from the national Fiscal Year (FY) 2013 Rehabilitation Services Administration (RSA)-911 database. The final predictive multinomial logistic regression model indicated that (a) the odds of White veterans successfully returning to work were nearly 1½ times the odds of African American veterans returning to work and (b) African American female veterans had the lowest probability for successfully returning to work. Moreover, findings indicated that African American veterans’ successful return-to-work rates in 5 of the 10 RSA regions were below the national benchmark. Recommendations for policy development and future research directions are presented.

Keywords

African American veterans state vocational rehabilitation agencies RSA-911 data assessment and cross-validation research methods minority access and outcome rates

Over the past 50 years, millions of U.S. Armed Forces men and women have participated in wartime conflicts to include the Global War on Terror (GWOT [Iraq and Afghanistan Theaters of Operation; post–September 11, 2001, to present]), the Persian Gulf War (also referred to as the Gulf War [August 2, 1990, to February 28, 1991] and the Vietnam War [August 7, 1964, to April 30, 1975]). There are an estimated 22,328,000 veterans residing in the United States (Frain, Bishop, & Bethel, 2010; National Center for Veterans Analysis and Statistics, 2013). Of this total, approximately 10.8% are African American, while White, American Indian, Asian, Hispanic, and individuals from the Other racial category make up 80%, 0.6%, 1.3%, 6%, and 1.4%, respectively, of the universe of veterans. An estimated 15.4% of GWOT, 16.9% of Persian War veterans, and 8.9% of Vietnam War era veterans are African Americans. African American female veterans make up about 20.1% of all women veterans. Remarkably, nearly 1 in every 3, or 33%, of veterans who served in the Persian Gulf War I and GWOT were African American (National Center for Veterans Analysis and Statistics, 2013).

Thousands of these veterans have returned home with more than honor, having sustained service-connected physical and/or mental disabilities (Feist-Price & Khanna, 2011; Moore, Johnson, & Uchegbu, 2011). For instance, as noted in the U.S. National Institute on Disability and Rehabilitation Research (NIDRR) 2010–2014 Long Range Plan, 29,978 service personnel were wounded while participating in Operation Iraqi Freedom from March 2003 through April 2008 (NIDRR, 2009). According to the Rand Corporation, approximately 40% of these veterans sustained mild traumatic brain injury (TBI) or post-traumatic stress disorder (PTSD), whereas others returned with physical and sensory impairments (Tanielian et al., 2008). As many as 25% of GWOT veterans experience hidden disabilities such as PTSD, TBI, and depression, while others have returned with physical and sensory impairments (Tanielian et al., 2008). Other disabilities sustained historically in “war time” theaters of operation include spinal cord injuries (SCIs), amputations, vision loss, substance dependence, limitations owing to orthopedic injuries, as well as disfiguring burns (Church, 2009; National Research Council, 2008). The GWOT’s protracted urban warfare has led to an unexpectedly high number of wounded veterans—in no small part because of the advances in both combat medicine and protective armor—and has led to surprising survival rates (Glasser, 2005). These medical and technological advances have resulted in higher numbers of African American veterans with disabilities (Moore et al., 2011).

Approximately 1 of every 5 (22.6%) African American veterans possess a documented service connected disability, and they also experience higher rates of health conditions such as diabetes, heart disease, acquired immune deficiency syndrome, and strokes in comparison with non-Latino Whites (National Center for Veterans Analysis and Statistics, 2013). Due to functional limitations, they also more often report the need for independent living (IL) services (Sheehan, Hummer, Moore, & Butler, 2012). Using the 2010 National Veterans Survey data, Sheehan et al. (2012) found that African American veterans were almost twice as likely to need assistance with activities of daily living (ADL; for example, assistance with personal hygiene, eating, transferring to bed/chair, using toilet) and 1.67 times more likely to need assistance with instrumental activities of daily living (IADL; for example, assistance with cooking, managing money, household chores) than White veterans, respectively.

The reintegration to occupational functioning and prevention of job loss are perhaps the most important aspects of success for veterans with disabilities (Bell, Boland, Dudgeon, & Johnson, 2013; Frain et al., 2010; London, Heflin, & Wilmoth, 2011; Moran, Schmidt, & Burker, 2013). According to Moran et al. (2013), veterans perceive delayed career as one of the most undesirable experiences in transitioning to the civilian workforce. As African American veterans transition to civilian life, they require a myriad of vocational rehabilitation (VR) services to assist them in returning to work. The origins of authorized VR services to veterans can be traced back to the 1918 Soldier’s Rehabilitation Act. The Rehabilitation Act of 1973 subsequently mandated that services be delivered to veterans (Alston, Lewis, & Loggis, 2014) in a manner consistent with their strengths, resources, abilities, interests, and concerns, along with informed choice (Fleming, Del Valle, Kim, & Leahy, 2013). There are 80 individual or combined state vocational rehabilitation agencies (SVRAs) across the United States. SVRAs are eligibility programs in which individuals must meet specific criteria to receive services that assist them to secure integrated competitive employment. Veterans determined to be eligible for services can possess a service-connected or a non-service-connected disability.

Public Policy Context

Section 21 Legislative Mandate

Section 21 of the Rehabilitation Act amendments documented patterns of inequitable treatment of persons from traditionally underserved populations in all junctures of the VR process. Congress found that persons of color (a) possessed higher rates of disability, (b) were underrepresented in the public VR system, and (c) were less likely to achieve positive employment outcomes when compared with Whites (Lewis, Shamburger, Head, Armstrong, & West, 2007). This legislative mandate context is important because information garnered from the current study could inform the state-of-the-science on improving outcomes for the target population.

Post-911 Veterans Educational Assistance Act of 2008

This act, also known as the New Government Issue (G.I.) Bill, is an effort to pay for veterans’ college expenses similar to the extent mandated in the original G.I. Bill after World War II (Madaus, Miller, & Vance, 2009). Veterans with disabilities covered by the New G.I. Bill are eligible to receive the full amount of tuition and fees charged by a college or a university, not to exceed the most expensive in-state public institution (Grossman, 2009). This legislation provides an important context for this study as the educational attainment level at closure is assessed as a predictor of rehabilitation success.

Methodological Concerns—Rehabilitation Services Administration (RSA) 911 Database

Over the past several decades, researchers (Atkins et al., 1980; Feist-Price, 1995; Moore, 2001a, 2001c, 2002a; Moore et al., 2009; Wilson, 2000) have conducted various Rehabilitation Services Administration (RSA)-911 data–driven studies to assess the relationship among race/ethnicity, VR access, and rehabilitation outcomes. Findings regarding the impact of race/ethnicity on VR experiences have been mixed, however; consequently, there has been little consensus among researchers about the significance of race/ethnicity as a correlate of differential outcomes. Several of these studies have addressed the same research questions, but nearly all have used a multitude of different research methodologies. These studies have differed with respect to the time period of collection, state versus national data, differing sample populations (heterogeneous vs. homogeneous), sample sizes, and statistical analyses (Moore et al., 2009).

Understanding ongoing African American VR customer employment outcome problems has been hampered by methodological differences among studies that have ultimately affected the consistency of outcome findings (Moore et al., 2009). Although researchers have conducted RSA-911 data–driven studies to determine the importance of race/ethnicity as a predictor, relatively few of these studies have utilized random assignment as a technique to control for plausible threats to internal validity (e.g., education, previous work experience, geographic locale). To control many of the extraneous variables that would destroy the study if not controlled, researchers often randomly assign participants to various comparison groups of the investigation (Huck & Cormier, 1996). Unlike many previous RSA-911 data–driven studies, in the current study we attempted to control for such plausible threats to internal validity by using a randomized split-half cross-model validation research design. This technique allows for case records to be randomly assigned to two different samples (screening and calibration). The screening sample is used for logistic regression model development, while the calibration sample is used to validate the best fitting model by using mean square error (MSE), which measures the accuracy of probability estimates. More specifically, a series of logistic regressions tested the effects of veterans’ demographic variables (i.e., race, gender, and level of educational attainment at closure) on the return-to-work criterion. The best fitting model was verified using the aforementioned cross-validation procedures. This data reduction process showed that cross-validation methods reduce variance in the models. The employment of cross-validation procedures for analyzing RSA-911 data can be an important exploratory tool in developing a contemporary and accurate description of VR experiences, which is vital to the development of a national agenda and strategies to achieve full participation for African American veterans with disabilities.

Purpose of the Study

Although Section 21 emphasizes the need to increase VR participation and outcomes among African American veterans, to date, relatively little attention has been paid to evaluating their VR service access and successful return-to-work outcome rates. Far too often, VR research has failed to address African American veterans’ unique needs. Consequently, national benchmark findings on access patterns and outcomes are relatively non-existent. Moore (2001b) encouraged SVRAs to examine their own RSA-911 data to assess service delivery patterns for underserved populations and compare them with national benchmarks generated from the national RSA-911 database. This ex-post-facto inquiry could contribute to filling this apparent research gap. The generated national profile could address questions that policy makers and other key stakeholders need answered regarding African American veterans’ VR experiences as they consider future funding priorities, how to redirect funding resources, and which new research and service initiatives to fund.

The purpose of this analysis was to identify disparities in successful return-to-work outcome rates based on race, gender, and level of educational attainment at closure among veterans with a signed Individualized Plan for Employment (IPE). The generated national profile was broken out by SVRA and RSA region. We compared successful return-to-work outcome rates between African American and White veterans across SVRAs, the 10 RSA regions, and the national benchmark or average. The following research questions were addressed:

Research Question 1: What is the national profile of VR service access and return-to-work outcomes for African American and White veterans with a signed IPE?

Research Question 2: Are race, gender, and level of educational attainment at closure significantly associated with successful return-to-work outcomes for veterans?

Method

Sample

The overall sample for this study consisted of 11,337 VR consumers who were served by the 56 state and territorial VR agencies across the nation during Fiscal Year (FY) 2013 (October 1, 2012–September 30, 2013), and were (a) reported as being a veteran (veteran status = 1), (b) identified as White (race code = 100000) or African American (race code = 010000), and (c) reported as having a developed and signed IPE; that is, closed Status 26 (successfully rehabilitated) or 28 (not successfully rehabilitated). No cases representing consumers identifying themselves as multiracial were selected for analysis. Of these 11,337 veterans, 3,120 (27.5%) were African American, 8,217 (72.5%) were White. Overall, males accounted for 9,899 (87.3%) of participants, while there were 1,438 (12.7%) females in the study sample. A total of 2,639 African American male veterans (23.3%) and 481 African American female veterans (4.2%) were included in the sample. A plurality of these veterans (n = 4,675 or 41.24%) possessed a high school diploma/equivalency or less, while veterans with some post-secondary education (n = 2,824 or 24.91%), an associate’s degree (n = 2,244 or 19.79%), a bachelor’s degree (n = 1,202 or 10.60%), or a master’s degree or higher comprised of the residual of the sample. The total sample was utilized to generate the profile of VR service access and successful return-to-work outcomes.

The employment of a single regression analysis, absent a cross-validation procedure, is oftentimes problematic for accurately predicting the analysis to the population (Bellini, Neath, & Bolton, 1995). As noted by Kromrey and Hines (1996), regression weights that are developed in one sample and applied to a new sample will almost always yield a smaller explained variance coefficient of effect size. Pedhazur (1982) recommended the splitting of one sample into two samples, in the absence of multiple samples, to address effect size variation issues. Moore (2002b) and Moore, Alston, Donnell, and Hollis (2003) demonstrated the use of a split-half cross-validation research design. Essentially, they used randomized block design procedures that included the splitting of an overall sample of RSA-911 case records into two approximately equal samples (screening and calibration) and running independent logistic regressions on each subsample to test significance. This study is intended to extend these previous studies by validating effects across models. Steyerberg et al. (2001) showed that the cross-validation algorithm for logistic regression has better performance with small bias and root mean square error (RMSE). The randomized split-half cross-model validation research paradigm as shown in Figure 1 is constructed to minimize MSE and will use the smallest MSE to select the best fitting predictive model. Model cross-validation techniques can better detect and prevent overfitting (Hsu, Chang, & Lin, 2003).

Figure 1.

Research paradigm for randomized split-half cross-model validation research design.

Data Collection

The national FY 2013 RSA-911 database (N = 589,402) was used in this analysis. It is important to note that this database does not differentiate between veterans by “wartime” era. For example, the “veteran” variable in the database does not distinguish between a Wounded Warrior, a Vietnam War–era veteran, Persian Gulf War veteran, or non-war-time veteran. The “veteran” variable only indicates whether the consumer was a veteran (code = 1) or not a veteran (code = 0). The two data categories for the criterion included Statuses 26 (i.e., exited with an employment outcome) and 28. The RSA data in the type of closure categories labeled 4 and 5 were combined to reflect Status 28, which indicates that a veteran was not successful in returning to work. The category labeled 4 (“exited without an employment outcome, after receiving services”) included Statuses 14, 16, 18, and 20. The category labeled 5 (“Exited without an employment outcome, after a signed IPE, but before receiving services”) included Status 12 only.

Data Analysis

Descriptive and multivariate statistics were utilized to analyze data. Access frequencies and return-to-work percentage rates were generated, compared, and reported for the two comparison groups. Next, logistic regression analyses were conducted for the screening, calibration, and final predictive model to evaluate the relationship between race (i.e., White vs. African American), gender, level of educational attainment at closure, and type of closure. The Statistical Analysis System (SAS), desktop version 9.3, was used in these calculations (SAS Institute, 2011).

Results

Key Observations—Profile

Several observed differences across state/territorial VR agencies, RSA regions, and the nation exist and could have future implications for both African American veterans with disabilities and the state–federal VR program that serves them. Many of these differences can be observed in Table 1 and do not require additional explanation. As such, we will discuss only a select number of key observations.

Table 1.

National RSA-911 Data on African American Veterans’ Access and Return-to-Work Outcomes.

State/territory (VR) agency	Agency code	AA and W vets access comparison after IPE signed–f		AA vets RTW successful (n = 1,316)		W vets RTW successful (n = 4,265)
State/territory (VR) agency	Agency code	AA	W	f	%	f	%
RSA Region I
Connecticut (B)	064	0	2	0	0.00^a,b	1	50.00
Connecticut (G)	008	14	87	7	50.00^b	56	64.37
Maine (B)	078	0	11	0	0.00^a,b	8	72.73
Maine (G)	022	0	60	0	0.00^a,b	29	48.33
Massachusetts (B)	080	0	7	0	0.00^a,b	5	71.43
Massachusetts (G)	024	8	62	5	62.50^b	40	64.52
New Hampshire (G)	032	1	57	1	100.00	34	59.65
Rhode Island (G)	044	1	10	0	0.00^a,b	5	50.00
Vermont (B)	106	0	9	0	0.00^a,b	7	77.78
Vermont (G)	050	3	144	1	33.33^a,b	77	53.47
Regional subtotal		27	449	14	51.85^b	262	58.35
RSA Region II
New Jersey (B)	089	3	8	0	0.00^a,b	4	50.00
New Jersey (G)	033	23	19	8	34.78^a,b	11	57.89
New York (B)	091	4	12	2	50.00	6	50.00
New York (G)	035	154	317	92	59.74	189	59.62
Puerto Rico (G)	043	4	37	2	50.00^b	19	51.35
Regional subtotal		188	393	104	55.32^b	229	58.27
RSA Region III
Delaware (G)	009	22	32	11	50.00	14	43.75
District of Columbia (G)	010	37	5	11	29.73^a,b	3	60.00
Maryland (G)	023	66	49	28	42.42^a,b	21	42.86
Pennsylvania (G)	042	199	657	79	39.70^a,b	356	54.19
Virginia (B)	107	4	3	0	0.00^a,b	3	100.00
Virginia (G)	051	54	91	20	37.04^a,b	45	49.45
West Virginia (G)	054	13	330	9	69.23^b	263	79.70
Regional subtotal		395	1,167	158	40.00^a,b	705	60.41
RSA Region IV
Alabama (G)	001	73	52	53	72.60	26	50.00
Florida (B)	067	9	38	2	22.22^a,b	23	60.53
Florida (G)	011	132	250	41	31.06^a,b	79	31.60
Georgia (G)	012	115	94	26	22.61^a,b	34	36.17
Kentucky (B)	076	4	10	3	75.00^b	8	80.00
Kentucky (G)	020	43	186	20	46.51^a,b	108	58.06
Mississippi (G)	027	41	86	22	53.65^b	74	86.05
North Carolina (B)	092	1	4	1	100.00	4	100.00
North Carolina (G)	036	248	253	87	35.08^a,b	89	35.18
South Carolina (B)	101	2	6	1	50.00	2	33.33
South Carolina (G)	045	296	237	182	61.48^b	155	65.40
Tennessee (G)	047	33	90	18	54.55	42	46.67
Regional subtotal		997	1,306	456	45.74^a,b	644	49.31
RSA Region V
Illinois (G)	016	114	190	46	40.35^a,b	98	51.58
Indiana (G)	017	43	319	16	37.21^a,b	185	57.99
Michigan (B)	081	1	5	1	100.00	3	60.00
Michigan (G)	025	185	274	100	54.05^b	162	59.12
Minnesota (B)	082	1	2	0	0.00^a	0	0.00
Minnesota (G)	026	17	128	4	23.53^a,b	66	51.56
Ohio (G)	039	97	237	20	20.62^a,b	92	38.82
Wisconsin (G)	055	40	254	18	45.00^a,b	124	48.82
Regional subtotal		498	1,409	205	41.16^a,b	730	51.81
RSA Region VI
Arkansas (B)	061	3	16	3	100.00	11	68.75
Arkansas (G)	005	13	110	7	53.85^b	75	68.18
Louisiana (G)	021	183	171	28	15.30^a,b	31	18.13
New Mexico (B)	090	0	4	0	0.00^a,b	0	0.00
New Mexico (G)	034	8	69	4	50.00	26	37.68
Oklahoma (G)	040	37	125	12	32.43^a,b	63	50.40
Texas (B)	104	13	46	9	69.23^b	37	80.43
Texas (G)	048	260	686	128	49.23^b	414	60.35
Regional subtotal		517	1,227	191	36.94^a,b	657	53.55
RSA Region VII
Iowa (G)	018	10	126	6	60.00	57	45.24
Kansas (G)	019	32	114	11	34.38^a,b	49	42.98
Missouri (B)	084	1	7	1	100.00	7	100.00
Missouri (G)	026	69	250	42	60.87	148	59.20
Nebraska (B)	086	0	1	0	0.00^a,b	1	100.00
Nebraska (G)	030	12	102	5	41.67^a,b	55	53.92
Regional subtotal		124	600	65	60.00	317	52.83
RSA Region VIII
Colorado (G)	007	13	156	7	53.85^b	88	56.41
Montana (G)	029	1	32	0	0.00^a,b	12	37.50
North Dakota (G)	037	2	17	0	0.00^a,b	6	35.29
South Dakota (B)	102	0	4	0	0.00^a,b	3	75.00
South Dakota (G)	046	1	26	1	100.00	18	69.23
Utah (G)	049	12	228	8	66.67	119	52.19
Wyoming (G)	056	3	79	0	0.00^a,b	35	44.30
Regional subtotal		32	542	16	50.00^b	281	51.85
RSA Region IX
Arizona (G)	004	10	80	2	20.00^a,b	28	35.00
California (G)	006	304	654	89	29.28^a	175	26.76
Hawaii (G)	014	1	2	0	0.00^a	0	0.00
Nevada (G)	031	3	19	1	33.33^a,b	9	47.37
Northern Marianas (G)	038	0	2	0	0.00^a,b	2	100.00
Regional subtotal		318	757	92	28.93^a	214	28.27
RSA Region X
Alaska (G)	002	15	59	10	66.67	37	62.71
Idaho (B)	071	1	5	1	100.00	3	60.00
Idaho (G)	015	2	144	2	100.00	87	60.42
Oregon (B)	097	0	4	0	0.00^a,b	4	100.00
Oregon (G)	040	5	145	1	20.00^a,b	91	62.76
Washington (B)	109	1	6	1	100.00	3	50.00
Washington (G)	053	0	4	0	0.00^a,b	1	25.00
Regional subtotal		24	367	15	62.50	226	61.58
National totals		3,120	8,217	1,316	42.18	4,265	51.90

Note. National benchmark/average for successful RTW rate—all veterans = 49.22%. RSA = Rehabilitation Services Administration; VR = vocational rehabilitation; AA = African American; W = White; IPE = Individualized Plan for Employment; f = frequency; RTW = return to work; B = blind; G = general or combined.

AA veteran return-to-work percentages below national return-to-work averages (49.22%). ^bAA veteran return-to-work percentages below White veteran return-to-work percentages.

First, as reflected in Figure 2, 3 of the 10 (30%) RSA regions had successful African American veteran return-to-work outcome rates that were above the same types of rates for White veterans. Second, nationally African American veterans with signed IPEs were less likely to return to work successfully than White veterans with signed IPEs. As shown in Table 1, we found that 42.18% of African American veterans with a signed IPE across the nation successfully returned to work when compared with 51.90% of White veterans with a signed IPE.

Figure 2.

National benchmark and RSA regional successful return-to-work outcomes (Status 26 only) among African American and White veterans.

Figure 3.

Predicted probability of successful return to work.

This finding represents a remarkable 9.73% national disparity (almost 10 percentage points) in successful return-to-work outcome rates among African American veteran VR consumers. Third, the national benchmark or national average for successful return-to-work outcomes for the study population was calculated to be 49.22%. Another striking observation is that 5 of the 10 RSA regions had successful return-to-work outcome rates for African American veterans with IPEs that were below this national benchmark. As shown in Figure 2, the five RSA regions with outcome rates for African American veterans below the national benchmark included III, IV, V, VI, and IX.

RSA Region X had the highest African American veteran successful return-to-work outcome rate (61.58%), while RSA Regions III, VI, and IX possessed the lowest rates: 40%, 36.94%, and 28.93%, respectively. These RSA regions appear to reflect the greatest need area for research initiatives aimed at addressing barriers to successful return-to-work outcomes among African American veterans.

Consumer Characteristics and Return-to-Work Outcomes

The association between select characteristics (i.e., race, gender, education level at closure status) was tested using multinomial logistic regression and verified using randomized split-half cross-model validation procedures (Hastie et al., 2009). The procedures are reflected broadly in Figure 2. The algorithm and detailed description of the procedure used to build, test, and cross-validate the models follow.

Procedure 1

Data for the overall sample of case records (N = 11,337) were randomly divided into two samples: screening (n = 5,648 or 49.8%) and calibration (n = 5,689 or 50.42%). A new variable was added and the rand (UNIFORM) function in SAS version 9.3 was used to generate random numbers from the uniform distribution (0, 1; SAS Institute, 2011). If random numbers were less than 0.5, the data or case record was assigned to the screening sample. If random numbers were greater than or equal to 0.5, data or case records were assigned to the calibration sample.

Procedure 2

Multinomial logistic regression is a form of statistical modeling and is appropriate for analyzing categorical outcome variables (Hosmer & Lemeshow, 2000). Data were analyzed using the PROC LOGISTIC procedure of SAS (Allison, 1999; SAS Institute, 2011). The data description of the three explanatory variables and the dependent variable is provided in Table 2.

Table 2.

Descriptions and Codes for Variables.

Variable	Definition	Codes
Y	Dependent variable	26 = successfully return to work28 = not successfully return to work
Gender	Dummy variable	1 = male2 = female
Race	Dummy variable	1 = African American2 = White
Edu	Level of educational attainment at closure	1 = high school diploma/equivalency or less2 = some post-secondary education3 = associate’s degree/vocational technical certificate4 = bachelor’s degree5 = master’s degree or higher

A series of multinomial logistic regression models (1) were conducted on the screening sample to find the best fitting model and to estimate the vector $\hat{β}$ .

\begin{array}{l} \log i t (P (Y = closed status 26 | X_{1}, \dots, X_{p})) \\ = β_{0} + β_{1} X_{1} + \dots + β_{p} X_{p} + ε, \end{array}

where β_i (i = 1, 2,…, p) are the coefficients estimated using maximum-likelihood estimation, X_i (i = 1, 2, …, p) are explanatory variables, and ε is error. A predicted logit was obtained from the solved logistic regression equation by substituting the explanatory variables’ value into the sample estimate of the logistic regression equation:

\log i t (P) = {\hat{β}}_{0} + {\hat{β}}_{1} X_{1} + \dots + {\hat{β}}_{p} X_{p} .

The predicted probability is given by

\hat{P} = \frac{\exp (\log i t)}{1 + \exp (\log i t)} .

This value that represents veterans’ successful return to work serves as the binomial distribution of Y at values of X. This research design increases the efficiency of internal validation procedures for the predictive logistic regression model (Taylor, Ankerst, & Andridge, 2008).

Plausible models were tested for goodness of fit using Wald chi-square measures, and the best fitting model under the screening sample was selected and applied to the calibration sample to calculate logistic probability. The SAS procedure PROC LOGISTIC: CLASS statement was used to test the null hypothesis and develop the three models in the screening sample tested for goodness of fit to the calibration sample. The following steps were used to generate the results shown in Table 3 under the screening sample.

Table 3.

Multinomial Logistic Regression Models Selection for Both Samples.

Model	Model effects	Log-likelihood ratio	Wald χ²	df	p value
Screening sample
1	Intercept, race	49.78		1	<.0001
2	Intercept, race, education	112.14	61.20	4	<.0001
3	Intercept, race, education, and gender	117.99	5.84	1	.015
Calibration sample^a
1	Intercept, race	36.47		1	<.0001
2	Intercept, race, education	75.85	39.10	4	<.0001
3	Intercept, race, education, and gender	80.53	4.66	1	.0307

Reversed to fit the screening sample.

Step 1: To develop Model 1, the following null hypothesis H₀ was tested: There is no significant difference in successful return-to-work outcome rates between White and African American veterans. The likelihood ratio test yielded statistical significance, χ² = 49.31, df = 1, p < .05.

Step 2: To develop Model 2, we entered the level of education attainment at closure variable into Model 1. The following null hypothesis was tested: Model 1 (the reduced model) is an adequate model. The alternative hypothesis H₁ was that Model 2 (full model) is an adequate model. A Wald chi-square test, χ² = 61.20, df = 4, p < .05, was sufficient to reject the null hypothesis. Consequently, Model 2 was a better fit to the data.

Step 3: To develop Model 3, gender was added to Model 2. The following null hypothesis was tested: Model 2 (reduced model) is an adequate model. The alternative hypothesis was that Model 3 (full model) was an adequate model. Wald chi-square test, χ² = 5.84, df = 1, p < .05, yielded significant confidence to reject the null hypothesis, which indicated that Model 3 made the best fit to the data when compared with Model 2. Overall, Model 3 made the best fit in comparison with Models 1 and 2.

Procedure 3

MSE was used to measure the model’s accuracy. In general, low values of MSE indicate an accurate estimate. MSEs were calculated using the following formula:

MSE = \sum_{i = 1}^{k} \frac{{(P_{i} - \hat{P_{i}})}^{2}}{k},

where $P_{i} - \hat{P_{i}}$ is the difference between observed probabilities and average predicted probabilities of being coded as Status 26 on Y for the i-th ordered category, the value of k is the number of the ordered subgroups (in this case, k = 20; Pohlman & Leitner, 2003).

In essence, each characteristic (i.e., race, gender, and level of educational attainment at closure) in the calibration sample was multiplied by its corresponding model coefficient to calculate logistic probability and MSE. The MSE (0.00587) for the calibration sample was calculated by Equation 4 and indicated that the model was a good fit to the calibration sample because they were small. The PROC INMODEL was used to calculate the predicted probability and the PROC SQL or DATA procedure calculated the MSE.

Procedure 4

We reversed the role of the two samples, and repeated Procedures 2 and 3. Thus, the calibration sample as shown in Table 3 was used to develop and test models for goodness of fit to the screening sample. We then calculated the MSE (0.00145) for the screening sample. The smaller MSE (i.e., screening sample = 0.00145) rather than the larger MSE (i.e., calibration sample = 0.00587) was used to select the final predictive model. The best fitting model shown in Table 4 was used to analyze the data to address Research Question 1.

Table 4.

Parameter Estimates: Calibration Sample Best Fitting Model—Final Predictive Model.

Closed status	B	SE	Wald χ²	df	Significance	Exp(B)
Intercept	−.142	.167	.728	1	.394
Race = 1 vs. race = 2	.337	.060	31.740	1	.000	1.40
Gender = 1 vs. gender = 2	.172	.080	4.668	1	.031	1.19
Edu = 1 vs. Edu = 5	−.311	.154	4.071	1	.044	0.73
Edu = 2 vs. Edu = 5	−.552	.157	12.345	1	.000	0.58
Edu = 3 vs. Edu = 5	−.153	.160	.910	1	.340	0.86
Edu = 4 vs. Edu = 5	−.077	.169	.210	1	.647	0.93

The odds ratios (ORs) or effect sizes for the final predictive model are shown in Table 4 and provide the estimated coefficients that predict successful return-to-work outcomes. The coefficients (B) were the log odds of the event occurring (i.e., change in the log odds associated with one-unit change in the independent variable). Logistic regression results indicated that race (OR = 1.40; 95% confidence interval [CI] = [1.24, 1.54], p < .05,) and gender (OR = 1.19; CI = [1.02, 1.39], p < .05) were significant predictors of successful return to work. All things being equal, the OR coefficient indicated that the odds of a White veteran achieving a successful return-to-work outcome is 1.40 times greater than that for an African American veteran. Similarly, the odds of a male veteran achieving a successful return-to-work outcome was 1.19 times greater than that for a female veteran. In addition, results were significant for the level of educational attainment and return-to-work success; high school diploma/equivalency or less versus master’s degree or higher (OR = .73; CI = [0.54, 0.99], p < .05), and some post-secondary education versus master’s degree or higher gender (OR = .58; CI = [0.42, 0.78], p < .05). In short, the odds of a veteran who had a master’s degree or higher successfully returning to work was 0.73 times greater than that for a veteran who had a high school diploma/equivalency or less. Similarly, the odds of a veteran with a master’s degree or higher successfully returning to work was 0.58 times greater than that for a veteran with some post-secondary education (no degree or certificate).

The predicted probability for race, gender, and educational attainment at closure is presented in Figure 3. The results indicate that White male veterans had the highest probability of successful return to work, followed by White females, African American males, and then African American females. For all subgroups, veterans who had some post-secondary education had the lowest probability of successful return to work, followed by those with a high school diploma or less, associate degree, bachelor’s degree, and master’s degree or higher.

Discussion

The purpose of this study was to identify disparities in successful return-to-work outcome rates based on race, gender, and level of educational attainment at closure among veterans with a signed IPE. The results point out that veterans who are African American, females, and individuals with less education are significantly less likely to achieve return-to-work success when compared with their counterparts. Several plausible reasons exist for these findings. First, perhaps they can be explained by transferrable skill gaps among African American and female veterans. Transferable skills have been described as a major challenge for returning veterans seeking employment in civilian life because many veterans are simply unaware of what jobs their military skills can be transferred to in civilian life (Hillesheim, Sprong, Dallas, Upton, & Musgrave, 2013). In a survey of 84 veterans, Hillesheim et al. (2013) found that 56% of respondents indicated that they were not sure what civilian occupations related to their work history obtained while in the military. This uncertainty about transferable skills could contribute to African American and women veterans’ insecurity about making college major and career choices. According to Faberman and Foster (2013), military experiences transfer to only a few select civilian work tasks and occupations. Thus, the lack of transferable skills can result in high unemployment rates for African American and female veterans. Occupational skills acquired during military service in wartimes are far less transferable to the civilian labor market (Faberman & Foster, 2013).

Second, mental health issues in the form of PTSD (Graf, Miller, Feist, & Freeman, 2011; Moran et al., 2013), major depression, and generalized anxiety disorder (Moran et al., 2013) may limit return-to-work opportunities among African American and female veterans. African American veterans experience PTSD and other mental health issues at higher rates than Whites because they are far more likely to serve on the front line of defense (Dohrenwend et al., 2007). However, their symptoms often remain undiagnosed (Schwartz, Bradley, & Sexton, 2005). PTSD presents an “extensive challenge” to veterans of color, in that lack of employment has been recognized as a form of trauma that manifests as PTSD (Atkins, 2011). Women veterans represent an underexplored population within the mental health context (Atkins). Frain et al. (2010) pointed out that in 2007, between 12% and 23% of women were exposed to trauma in combat areas. Mental health issues can greatly impede African American veterans’ participation in employment and career success (Boutin, 2011; Burke, Degeneffe, & Olney, 2009; Leddy, Stefanovics, & Rosenheck, 2014), and female veterans need the expertise of VR to successfully transition into employment (Atkins, 2011). In addition, mental illness has been associated with slower veterans’ progress in the VR system (O’Connor et al., 2011).

Third, level of educational attainment at closure was found to explain return-to-work differences in general, and it is also a possible contributor to race and gender-based differences on the criterion. Various financial and guidance resources are available to support the educational aspirations of veterans with disabilities. For instance, many African American and female veterans with disabilities are eligible for financial assistance through the New G.I. Bill to pursue education or training programs at accredited colleges or universities or accredited non-college degree-granting institutions. Financial assistance is provided for tuition, fees, books, supplies, and housing (Disabled American Veterans, 2012). These veterans may also be eligible to receive financial assistance through the Veterans Administration Vocational Rehabilitation and Employment (VR&E) program (Bell et al., 2013). Moreover, they can receive financial assistance for education through SVRAs (Boutin, 2011). Perhaps greater outreach efforts are needed to inform them about financial support opportunities that could lead to higher educational level attainment.

The conflicts of war result in a myriad of disabilities that present as barriers to educational attainments and future employment success. Although some of these barriers may be internal for African American and female veterans, other challenges impeding their success can be more external. For example, the limited ability of post-secondary faculty and staff to effectively work with veterans with disabilities in the academic setting has been reported as a challenge (Burnett & Segoria, 2009; Glover-Graf, Miller, & Freeman, 2010). Post-secondary institutions often lack necessary knowledge about the needs of students with disabilities (Lindstrom, Flannery, Benz, Olszewski, & Slovic, 2009). Thus, collaboration between SVRAs and higher learning institutions could benefit veteran consumers, in that counselors could provide faculty and staff with disability-related information regarding available options and resources. Educational training is important to veterans’ transition back to civilian life and can significantly increase their employment opportunities and financial independence (Bell et al., 2013; Lindstrom et al., 2009; Shackelford, 2009). For example, Singh and Noah (2013) found that higher levels of education positively correlated with higher vocational success and work attitudes. Research has consistently demonstrated that veterans with disabilities who receive college training are more likely to become employed and earn higher wages when compared with those who do not receive these services (Boutin, 2011; Lindstrom et al., 2009).

Recommendations and Policy Considerations

These findings indicate that there remains a need for SVRAs to review and retailor, where needed, their outreach policies and practices to include new and promising strategies to more effectively serve African American and female veterans. The development of such strategies could prove pivotal for enhancing their seamless access to VR service programs, where their employment transition needs can be addressed. In SVRAs where African American veterans’ successful return-to-work rates were lower than the national benchmark, administrators and policy makers should consider developing state plan goals aimed at increasing positive outcome rates. The establishment of such goals could serve as a framework to help effectively address disparate successful return-to-work rates among these veterans.

SVRA rehabilitation counselors should also make greater efforts to identify African American and female veterans with an interest and aptitude for participating in retraining and educational programs. Figure 3 shows that African American female veterans have by far lower probabilities for successful return to work than African American males, White females, and White Males. There is clearly a need to offer African American female veterans additional retraining opportunities (e.g., associate degree/vocational technical degree programs, and bachelor’s and master’s degree programs) that will develop skill sets that are in high demand in business and industry. These actions could ultimately help close success gaps.

NIDRR, RSA, Veteran Affairs (VA), Social Security Administration, and other entities across the federal disability agency landscape should consider developing future national research and/or service initiatives aimed at addressing employment barriers experienced by veterans from traditionally underserved racial and ethnic groups and communities (i.e., African American, Latino, Native American, Asian). The establishment of multiple comprehensive research programs triangulating data through multiple data collection techniques could help improve our understanding of such barriers, while extensively addressing the Section 21 mandate.

Conclusions: Advancing the State-of-the-Science

These findings on the role that race, gender, and level of educational attainment at closure play in determining return-to-work outcomes denote the need for future studies aimed at increasing our understanding of these phenomena. Research gaps continue to exist in the current state-of-the-science, and relatively little is known about African American and female veterans with disabilities’ VR access, return-to-work success, and the existing and emerging barriers that prevent them from returning to work. Current research gaps can be attributed to the quantity and, more importantly, the quality of related research. In terms of quantity, very few studies have comprehensively examined VR access and return-to-work rates for these target groups. In terms of quality, RSA-911 data analyses present several challenges that can render results that must be heavily scrutinized. As a result, policy makers and service providers do not always infuse related findings into policy and initiative formulation or practice, respectively.

There is a continuing and serious need for scaled-up RSA-911 methodological models (e.g., randomized split-half cross-model validation research design) and non-archival multi-method and mixed-method research in this area to fill in the gaps as to what is known and what works in VR for increasing successful return-to-work rates for this population. One fertile area for examination is SVRA and VA VR&E coservice practices. Such practices can be described as collaborative interagency partnerships in which resources are pooled and services are coordinated to maximize the benefits to veterans with disabilities. Do best-practice coservice models moderate the negative impacts of race and gender on successful return-to-work outcomes? Preliminary qualitative research to include the convening of focus groups and Delphi panels could prove useful for discovering African American veterans’ perceptions on barriers to employment, and best practice coservice strategies could help veterans of color overcome such obstacles. Subsequently, evidence-based research would be needed in the form of multiple randomized trials to identify promising practices for addressing these barriers. Determining and understanding these issues could be extremely beneficial to increasing the successful return-to-work rates for African American and female veterans with disabilities.

Limitations

This study was limited in two specific ways. First, the results represent an analysis of RSA-911 data for FY 2013 and reflect a “snap shot” of VR experiences. A more complete and perhaps valid picture of these experiences could be gained from the analysis of RSA-911 databases across different periods. Additional research is undoubtedly needed to confirm the current findings and to generate other questions pertinent to consumer satisfaction and needs worthy of study and investigation. Future disability and rehabilitation researchers should consider using multiple RSA-911 databases to assess whether these findings are in fact isolated to FY2006 or a trend that reflects systematic concerns. Second, information contained in the RSA-911 database is not impervious to human error, and thus this data source could contain erroneous information entered by counselors. Nonetheless, as a broad generalization these conclusions may be helpful to national policy makers and SVRA leaders in their efforts to enhance VR access and successful return-to-work rates among African American and female veterans.

Footnotes

Authors’ Note

The contents do not necessarily represent the policy of the Department of Education, and readers should not assume endorsement by the Federal Government.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported under a grant from the Department of Education, NIDRR Grant H133B130023.

References

Allison

(1999). Logistic regression using SAS® system: Theory and application. Cary, NC: SAS Institute.

Alston

Lewis

Loggis

(2014). Assistive technology and veterans with severe disabilities: Examining the relationship among race, personal factors, medical support, income support, and use. Medical Care, 52, 17–24.

Atkins

B. J.

(2011). African American veterans, post-traumatic stress disorder (PTSD) and employment. In Moore

C. L.

Johnson

J. E.

Washington

A. L.

(Eds.), Improving vocational rehabilitation access and return to work and career outcomes among African American Wounded Warriors, Gulf War and Vietnam War era veterans with disabilities (pp. 45–70). Oklahoma City, OK: Department of Rehabilitation Counseling and Disability Studies, Langston University.

Atkins

B. J.

Wright

G. N.

Humphreys

Provitt

Bolton

Cooper

(1980). Three views: Vocational rehabilitation of Blacks. The statement, the response, the comment. Journal of Rehabilitation, 46, 40–49.

Bell

G. L.

Boland

E. A.

Dudgeon

Johnson

(2013). The post-9/11 GI Bill: Insights from veterans using Department of Veterans Affairs educational benefits. Rehabilitation Research, Policy, and Education, 27, 246–260. doi:10.1891/2168-6653.27.4.246

Bellini

Neath

Bolton

(1995). A comparison of linear multiple regression and a simplified approach in the prediction of rehabilitation outcomes. Rehabilitation Counseling Bulletin, 39, 151–160.

Boutin

D. L.

(2011). Effective vocational rehabilitation services for military veterans. Journal of Applied Rehabilitation Counseling, 42(2), 24–32.

Burke

H. S.

Degeneffe

C. E.

Olney

M. F.

(2009). A new disability for rehabilitation counselors: Iraq War veterans with traumatic brain injury and post-traumatic stress disorder. Journal of Rehabilitation, 75(3), 5–14.

Burnett

S. E.

Segoria

(2009). Collaboration for military transition students from combat to college: It takes a community. Journal of Postsecondary Education and Disability, 22(1), 53–58.

10.

Church

T. E.

(2009). Returning veterans on campus with war related injuries and the long road back home. Journal of Postsecondary Education and Disability, 22(1), 43–52.

11.

Disabled American Veterans . (2012). Post-9/11 GI Bill: General information. Retrieved from http://www.benefits.va.gov/BENEFITS/factsheets/education/Post-911_General_info.pdf

12.

Dohrenwend

B. P.

Turner

J. B.

Turse

N. A.

Adams

B. G.

Koenen

K. C.

Marshall

(2007). Continuing controversy over the psychological risks of Vietnam for US veterans. Journal of Traumatic Stress, 20, 449–465.

13.

Faberman

Foster

(2013). Unemployment among recent veterans during the Great Recession. Economic Perspectives, 37(1), 1–13.

14.

Feist-Price

(1995). African Americans with disabilities and equity in vocational rehabilitation services: One state’s review. Rehabilitation Counseling Bulletin, 39, 119–129.

15.

Feist-Price

Khanna

(2011). Improving vocational rehabilitation access and employment success for African American homeless veterans with disabilities. In Moore

C. L.

Johnson

J. E.

Washington

A. L.

(Eds.), Improving vocational rehabilitation access and return to work and career outcomes among African American Wounded Warriors, Gulf War and Vietnam War era veterans with disabilities (pp. 26–44). Oklahoma City, OK: Department of Rehabilitation Counseling and Disability Studies, Langston University.

16.

Fleming

A. R.

Del Valle

Kim

Leahy

M. J.

(2013). Best practice models of effective vocational rehabilitation service delivery in the public rehabilitation program: A review and synthesis of the empirical literature. Rehabilitation Counseling Bulletin, 56, 146–159.

17.

Frain

Bishop

Bethel

(2010). A roadmap for rehabilitation counseling to serve military veterans with disabilities. Journal of Rehabilitation, 76(1), 13–21.

18.

Glasser

R. J.

(2005, July). A war of disabilities. Harper’s Magazine, pp. 59–62.

19.

Glover-Graf

N. M.

Miller

Freeman

(2010). Accommodating veterans with posttraumatic stress disorder symptoms in the academic setting. Rehabilitation Education, 24, 43–56.

20.

Graf

N. M.

Miller

Feist

Freeman

(2011). Returning veterans’ adjustment concerns: Family views. Journal of Applied Rehabilitation Counseling, 42(2), 13–23.

21.

Grossman

P. D.

(2009). Foreword with a challenge: Leading our campuses away from the perfect storm. Journal of Postsecondary Education and Disability, 22(1), 4–9.

22.

Hastie

Tibshirani

Friedman

Hastie

Friedman

Tibshirani

(2009). The elements of statistical learning (Vol. 2). New York, NY: Springer.

23.

Hillesheim

C. G.

Sprong

M. E.

Dallas

Upton

Musgrave

(2013). Falling through the cracks: A screening assessment for student veterans with educational barriers. VEWAA Journal, 40(1), 23–31.

24.

Hosmer

D. W.

Jr. Lemeshow

(2000). Applied logistic regression (2nd ed.). New York, NY: Wiley.

25.

Hsu

C.-W.

Chang

C.-C.

Lin

C.-J.

(2003). A practical guide to support vector classification (Technical Report). Taipei, Taiwan: Department of Computer Science and Information Engineering, National Taiwan University.

26.

Huck

S. W.

Cormier

W. H.

(1996). Research statistics and research. New York, NY: HarperCollins.

27.

Kromrey

J. D.

Hines

C. V.

(1996). Estimating the coefficient of cross-validity in multiple regression: A comparison of analytical and empirical methods. The Journal of Experimental Education, 64, 240–266. doi:10.1080/00220973.1996.9943806

28.

Leddy

Stefanovics

Rosenheck

(2014). Health and well-being of homeless veterans participating in transitional and supported employment: Six-month outcomes. Journal of Rehabilitation Research and Development, 51, 161–175. doi:10.1682/JRRD.2013.01.0011

29.

Lewis

A. N.

Shamburger

Head

Armstrong

A. J.

West

S. L.

(2007). Section 21 of the 1992 Rehabilitation Act amendments and diversity articles. Journal of Vocational Rehabilitation, 26(2), 89–96.

30.

Lindstrom

L. E.

Flannery

K. B.

Benz

M. R.

Olszewski

Slovic

(2009). Building employment training partnerships between vocational rehabilitation and community colleges. Rehabilitation Counseling Bulletin, 52, 189–201.

31.

London

A. S.

Heflin

C. M.

Wilmoth

J. M.

(2011). Work-related disability, veteran status, and poverty: Implications for family well-being. Journal of Poverty, 15, 330–349. doi:10.1080/10875549.2011.589259

32.

Madaus

J. W.

Miller

W. K.

Vance

M. L.

(2009). Veterans with disabilities in postsecondary education. Journal of Postsecondary Education and Disability, 22(1), 10–17.

33.

Moore

C. L.

(2001a). Disparities in closure success rates for African Americans with mental retardation: An ex-post-facto research design. Journal of Applied Rehabilitation Counseling, 32(2), 31–39.

34.

Moore

C. L.

(2001b). Disparities in job placement outcomes among deaf, late-deafened, and hard-of-hearing consumers. Rehabilitation Counseling Bulletin, 44, 144–150.

35.

Moore

C. L.

(2001c). Racial and ethnic members of under-represented groups with hearing loss and VR services: Explaining the disparity in closure success rates. Journal of Applied Rehabilitation Counseling, 32(1), 15–23.

36.

Moore

C. L.

(2002a). Comparative competitive employment levels for Latinos and non-Latinos without 12 years of education. Journal of Applied Rehabilitation Counseling, 33(1), 13–18.

37.

Moore

C. L.

(2002b). Relationship of customer characteristics and service provision to income of successfully rehabilitated individuals who are deaf. Rehabilitation Counseling Bulletin, 45, 233–239.

38.

Moore

C. L.

Alston

Donnell

Hollis

(2003). Correlates of rehabilitation success among African American and Caucasian SSDI recipients with mild mental retardation. Journal of Applied Rehabilitation Counseling, 34(3), 25–32.

39.

Moore

C. L.

Ferrin

M. J.

Haysbert

Brown

Cooper

Deibel

. . . Cantrell

(2009). Employment outcome rates of African Americans versus White consumers of vocational rehabilitation services: A meta-analysis. Journal of Applied Rehabilitation Counseling, 40(3), 3–10.

40.

Moore

C. L.

Johnson

Uchegbu

(2011). Barriers to VR service access and return to work outcomes among African American veterans: The need for evidence-based research and service strategies. Oklahoma City, OK: Department of Rehabilitation Counseling and Disability Studies, Langston University.

41.

Moran

Schmidt

Burker

E. J.

(2013). Posttraumatic growth and posttraumatic stress disorder in veterans. Journal of Rehabilitation, 79(2), 34–43.

42.

National Center for Veterans Analysis and Statistics. (2013). Minority veterans: 2011. Retrieved from http://www.va.gov/VETDATA/docs/SpecialReports/Minority_Veterans_2011.pdf

43.

National Institute on Disability and Rehabilitation Research. (2009). Notice of proposed long-range plan for Fiscal Years 2010-2014. Federal Register, 74, 2564–2569.

44.

National Research Council. (2008). Gulf War and health: Volume 7. Long-term consequences of traumatic brain injury. Retrieved from http://www.nap.edu/openbook.php?record_id=12436

45.

O’Connor

M. K.

Mueller

Van Ormer

Drake

Penk

Rosenheck

. . . Drebing

(2011). Cognitive impairment as barrier to engagement in vocational services among veterans with severe mental illness. Journal of Rehabilitation Research & Development, 48, 597–608. doi:10.1682/JRRS.2010.06.0117

46.

Pedhazur

E. J.

(1982). Multiple regression in behavioral research (2nd ed.). Fort Worth, TX: Harcourt Brace Jovanovich.

47.

Pohlman

J. T.

Leitner

D. W.

(2003). A comparison of ordinary least squares and logistic regression. The Ohio Journal of Science, 103, 118–125.

48.

SAS Institute. (2011). SAS/STAT 9.3 user’s guide: Survey data analysis (book excerpt). Cary, NC: Author.

49.

Schwartz

A. C.

Bradley

R. L.

Sexton

(2005). Posttraumatic stress disorder among African Americans in an inner city mental health clinic. Psychiatric Services, 56(2). Retrieved from http://ps.psychiatryonline.org

50.

Shackelford

A. L.

(2009). Documenting the needs of student veterans with disabilities: Intersection roadblocks, solutions, and legal realities. Journal of Postsecondary Education and Disability, 22(1), 36–42.

51.

Sheehan

C. M.

Hummer

A. R.

Moore

B. L.

Butler

J. C.

(2012). Duty, honor, country, disparity: Race/ethnicity difference in self-reported health and disability among veterans. Austin: Department of Sociology and Population Research Center, University of Texas at Austin.

52.

Singh

C. K.

Noah

B. V.

(2013) Impact of vocational counseling and education on work attitudes and job placement of recovering substance abusers: Implications for veterans. Journal of Military and Government Counseling, 1(3), 152–169.

53.

Steyerberg

E. W.

Harrell

F. E.

Jr. Borsboom

G. J.

Eijkemans

Vergouwe

Habbema

J. D. F.

(2001). Internal validation of predictive models: Efficiency of some procedures for logistic regression analysis. Journal of Clinical Epidemiology, 54, 774–781.

54.

Tanielian

Jaycox

L. H.

Schell

T. L.

Marshall

G. N.

Burnam

M. A.

Eibner

. . . Vaiana

M. E.

(2008). Invisible wounds: Mental health and cognitive care needs of America’s returning veterans. Santa Monica, CA: Rand Corp.

55.

Taylor

J. M.

Ankerst

D. P.

Andridge

R. R.

(2008). Validation of biomarker-based risk prediction models. Clinical Cancer Research, 14, 5977–5983.

56.

Wilson

K. B.

(2000). Predicting vocational rehabilitation acceptance based on race, education, work status, and source of support at application. Rehabilitation Counseling Bulletin, 43, 97–105.