Predicting earnings of vocational rehabilitation clients with visual impairments. (Predicting Earnings of Clients with Visual Impairments).

Being-able to predict outcomes for rehabilitation clients is valuable, as indicated by the many researchers who have addressed outcome prediction since the 1960s (Bellante, 1972; Bellini, Neath, & Bolton, 1995; Cook & Rosenberg, 1994; DeMann, 1963; Giesen, Graves, Schmitt, Lamb, Cook, Capps, & Boyet, 1985; Miller & Barillas, 1967; Moore, Flowers, & Taylor, 2000; Schwab & DiNitto, 1993; Weikel & Johnson, 1974; Worrall & Vandergoot, 1980). Several potential uses for this type of prediction have been mentioned. These uses include: (a) identifying persons who would be unlikely to have successful outcomes in order to provide them extra attention and help that might ensure successful rehabilitation (Bellini, Neath, & Bolton, 1995; DeMann, 1963; Worrall & Vandergoot, 1980), (b) demonstrating difficulty of cases in order to justify budget requests, and (c) using as an additional measure of counselor performance (Miller & Barillas, 1967).

Pre-service variables have been investigated as to their potential for explaining outcomes of clients served by state-federal Vocational Rehabilitation (VR) programs (Bellante, 1972; Bellini, Neath, & Bolton, 1995; DeMann, 1963; Weikel & Johnson, 1974; Worrall & Vandergoot, 1980). Demographic variables have also been used to predict acceptance for VR services (Harrison & Wayne, 1986). More recently, service variables, usually in addition to demographics, have been investigated for their contribution to the prediction of VR clients' outcomes (Cook & Rosenberg, 1994; Moore, Flowers, & Taylor, 2000; Schwab & DiNitto, 1993). These investigations using service variables have been limited to specific disability populations, such as substance abusers, persons with psychiatric diagnoses, and persons with mental retardation.

Additionally, one investigation focused on predicting employment outcomes for persons with visual impairments (Giesen, et al., 1985). However, this study did not address the issue of type of agency that provided the VR services, as subjects were selected from four agencies that served only visually impaired individuals. Each state has a VR agency that serves persons with disabilities. In addition to these 51 agencies (50 states plus the District of Columbia), 25 states also have separate agencies that serve only persons with visual impairments. Most persons with visual impairments are served by separate agencies in the states that have them. In the other 25 states, services for clients with visual impairments are provided by the one existing state agency, referred to as a combined agency. Although the question has been debated, very little evidence exists regarding whether separate agencies have better outcomes for consumers than combined agencies (Cavenaugh & Pierce, 1998). This question is relevant, as some persons and groups involved in rehabilitation oppose separate agencies, while rehabilitation consumers, rehabilitation practitioners, and leaders in the blindness community strongly believe that separate agencies are critical to ensuring quality services and good outcomes (Cavenaugh & Pierce, 1998). Cavenaugh and Pierce's review of the literature found only three studies that empirically addressed the question of whether differences exist in outcomes of persons served by separate versus combined agencies. Because the findings of these studies were mixed, the question remains as to whether separate or combined agencies produce better outcomes for clients.

Cavenaugh and Pierce addressed this question in their 1998 study by evaluating the Rehabilitation Services Administration's individual-level data from fiscal year 1989 (referred to as RSA-911 data). The major drawback of this important study was that the authors reported only descriptive statistics (means, proportions, and percentages) for the variables they evaluated, as they considered the closed cases for 1989 to be population data. However, the closed cases from a single year can also be considered a sample of the population of persons with visual impairments who have ever been served by the federal-state VR system. The present study takes this alternative view.

This study used multiple regression analysis to predict weekly earnings of successfully closed VR clients with visual impairments and included type of agency as a variable in the analysis. Including type of agency allowed a determination of whether the type of agency (separate or combined) from which the client received services significantly affected earnings. One reason weekly earnings was chosen as the outcome variable is that, compared to other disability groups, persons with visual impairments who are closed successfully (i.e., status 26 closures) are much more likely to be closed in noncompetitive employment. The only other disability group for which this is true is persons with mental retardation (Hayward, 1998). Although persons may be closed successfully, their earnings are a better measure of quality Of closure than simply being classified as status 26. States may have a high percentage of 26 closures, but low average wages of their rehabilitated clients, calling into question the quality of the closures (Capella, 2001). For the general population of persons closed in status 26 during fiscal year 1997, the percentage closed as homemakers was only 2.84%, while this percentage was 41.04% for persons with visual impairments (Capella, 2000). This substantial difference could not be determined by comparing successful versus unsuccessful closure rates. However, evaluating earnings at closure can provide this important information about quality of successful closures for persons with visual impairments.

The Rehabilitation Services Administration (RSA) is currently investigating several questions concerning VR outcomes with a longitudinal study that Congress mandated in the 1992 Rehabilitation Act. Two of the questions concern economic outcomes for VR clients and characteristics of individuals with disabilities that affect their outcomes (Hayward, 1998). The present study addressed these two questions, specifically for persons with visual impairments. The purpose of this study was to determine what variables account for, or explain, differences in earnings of clients with visual impairments served by combined versus separate agencies. The two research questions investigated were: (a) Can education level, age, number of services received, cost of case services, and length of time a case is open predict earnings of successfully closed VR clients with visual impairments?, and (b) Does the type of agency (separate versus combined) that provides services to visually impaired clients affect their earnings at successful closure?

Method

The data used in the analysis were obtained from the RSA-911 database for fiscal year 1997. This database includes information about each person whose case was closed during that fiscal year. Data from all state agencies and the District of Columbia were used. The sample used for the study was limited to individuals who had been closed successfully (status 26 closure) with major disability codes that indicated visual impairment as their primary disability. All successfully closed clients with visual impairments were included in the sample, regardless of severity of their visual impairments. The original sample size was N=16,927; however, because 657 observations had missing values for at least one of the variables, the sample size used for the analysis was N=16,270. The original sample consisted of 9,497 clients served by separate agencies and 7,430 clients served by combined agencies, while the sample that was used for the analysis consisted of 9,196 clients served by separate agencies and 7,074 clients served by combined agencies.

Multiple linear regression was the statistical method used for the analysis. The dependent variable was weekly earnings after successful case closure. The variables chosen as independent variables were education level, age, cost of case services, number of services received, months the case was open, and type of agency (separate versus combined). These variables were selected based on prior research. From a group of pre-service variables, Bellante (1972) found education and age to be two of the best predictors of Florida VR client earnings. When predicting successful versus unsuccessful closure, DeMann (1963) also found education and age to be significant variables. Therefore, these two demographic variables were included in the model. Giesen et al. (1985) found that expenditures for specific services were a significant predictor in their study of persons with visual impairments. Because information about cost of specific services was not available, cost for the entire case was used as an alternative. Schawb and DiNitto (1993) also used this case expenditure variable, along with types of services, number of services received, and length of service as variables in their study. They demonstrated that differences on these variables existed among successfully and unsuccessfully rehabilitated substance abuse clients. Therefore, total number of services received and months the case was open were also included as independent variables. Finally, type of agency was included in the model, as some prior research has indicated it has an effect on client earnings (National Accreditation Council for Agencies Serving the Blind and Visually Handicapped [NAC], 1997).

Type of agency is a categorical variable and was therefore dummy coded, with separate agencies coded a "1" and combined agencies coded a "0." As five of the variables were continuous and one was categorical, the interaction between the categorical and continuous variables had to be evaluated as part of the regression model (Pedhazur, 1997). Five product vectors were created by multiplying agency type by each continuous variable. These product vectors were included in the model, to be tested for significant interactions between the continuous and categorical variables.

The order of entry of the variables into the model was also largely determined by prior research. As education level and age had the strongest evidence for predictive ability, they were entered first in the model. The other variables included in the model were related to VR service. Number of services received, cost of the case, and length of time case was open (measured in months) were entered next. As no evidence about strength of the variables for prediction of earnings was available, they were entered in the order of suspected significance to the model. The final independent variable was type of agency, which was placed fifth due to mixed findings in empirical research about significant differences in outcomes based on where the client received services. It, along with the five product vectors, was entered into the equation last. Because of the extremely large sample size used for the analysis, a alpha level of p < .0001 was established to determine significance.

Results

Means, standard deviations, and Pearson correlations for all continuous variables are presented in Table 1. The amount of variance explained by the model presented above was [R.sup.2] = .26. The F-test associated with it was significant: F(11, 16,258) = 520.92, p < .0001, adjusted [R.sup.2] = .26. The first step was to test the interactions of the categorical and continuous variables. This was accomplished by using a partial F-test to test the additional amount of variance explained by the product vectors. The total additional amount of variance explained by the interaction terms was .000768. Obviously, this amount is not practically significant in any way. The F-value associated with it was F(5, 16,258) = 4.38, p = .0005, which is nonsignificant at the established p < .0001 level. Therefore, the interaction terms were dropped from the model. The next step of the process was to determine if type of agency contributed significantly to the model by testing its unique variance. This step is equivalent to testing for differences between the effects of receiving services from separate versus combined agencies. The F-value associated with type of agency was not significant, F(1, 16,258) = 1.53, p = .2159. Therefore, there were not significant differences in earnings between the groups of clients served by separate versus combined agencies, and one regression line can be used. (See Table 2 for standardized regression coefficients, t-values, and the unique contribution of each independent variable.)

Of the remaining five independent variables, the standardized regression coefficient and the unique contribution of each one was next investigated to determine which variables were the most significant predictors of earnings. Age was found to be the most significant predictor, explaining 13.05% of the variance in earnings on its own t(1) = -53.54, p < .0001. Education followed in unique contribution, explaining 5.10% of the variance, t(1, 16,258) = 33.47, p < .0001. The only other variable found to explain a substantial amount of variance was cost of case services, which explained 1.68% after the other variables had been taken into account, t(1) = 19.18, p < .0001. Number of services and months the case was open were also found to be statistically significant at the p < .0001 level (see Table 2); however, they did not explain a meaningful amount of unique variance in weekly earnings (.08% and .37% respectively).

Discussion

The multiple regression analysis used in this study identified three of six variables that contribute statistically significant and meaningful amounts of variance to the explanation of weekly earnings in visually impaired VR clients. These variables are age, education, and cost of case services. Contrary to expectations, education was not the best predictor. Older age was predictive of lower earnings, while a higher education level was predictive of higher earnings, as would be expected. The relationship between case cost and earnings at closure was that as more money was spent on a client's case, earnings were predicted to be higher. Two variables, number of services received and months case was open, were found to be statistically significant but they did not contribute a meaningful amount of unique variance. One reason for this may be the relatively high intercorrelations between these variables, age, and case cost. It is interesting to note that these variables were both negatively related to earnings in the regression model. In other words, with the other variables being held constant, as number of services and months in VR increased, earnings were predicted to decrease. This result may be indicative of more challenging cases (e.g., clients with the most severe visual deficits) needing more services and taking longer to serve, but resulting in poorer outcomes in terms of earnings. It may be that the more challenging cases are frequently closed in noncompetitive employment. Type of agency (separate versus combined) was not a statistically significant predictor of earnings, nor were there any significant interactions between type of agency and the five continuous variables.

The results of this research are unique in that age was found to be a significant predictor of earnings for this group. In this sample, its unique variance was 13%, or approximately half of the total variance explained. When Bellante (1972) studied the effects of pre-service variables on the earnings of a general population of VR clients in Florida, he found that age explained 2.61% of the variance. Other studies have determined age to be a significant predictor of outcomes or to be significantly related to outcomes, but not in as great a magnitude as found in this investigation. In this sample, education also predicted a large amount of variance in earnings (5.10% as compared to 3.57% in Bellante's study).

Age as a significant predictor of earnings in VR clients with visual impairments is a relevant finding of this study. This is relevant because it has not been found to be as significant of a predictor in other studies with general populations or in studies with specific populations using successful versus unsuccessful rehabilitation as the outcome variable (Bellante, 1972; Cook & Rosenberg, 1994; Schwab & DiNitto, 1993). One explanation for the importance of age in this study may be the fact that, as people age, they are much more likely to experience visual impairments. Therefore, this population is older than many of the other populations served by VR. In fact, the average age of the VR clients in this study was 53.6 years (SD = 18.86), while the average age of the general population of person closed by VR during fiscal year 1997 was 35.23 (SD = 11.96). Incidentally, the average age for clients in Cavenaugh and Pierce's (1998) study, which evaluated fiscal year 1989 closures of visually impaired clients, was 46.4 for consumers of separate agencies and 42.8 for consumers of combined agencies. Although the usual goal of VR services is employment, the "vocational" goal of homemaker is still accepted. Many of these older VR clients with visual impairments may not be attempting to obtain employment, but may be closed successfully in the status of homemaker (which would mean no earnings), as opposed to competitive or noncompetitive employment. In contrast to these findings, Giesen et al. (1985) did not find age at application to be an important predictor of type of employment outcomes for persons with visual impairments. Why age was found to be such a significant predictor of weekly earnings after successful closure in the clients with visual impairments in this sample is a relevant question to consider and should be further investigated.

Perhaps the most relevant finding of this study is that type of agency did not significantly predict earnings of VR clients. This is similar to Kirchner and Peterson's finding reported in their 1982 study, but is in contrast to what the National Accreditation Council for Agencies Serving the Blind and Visually Impaired (NAC) stated in its 1997 report. Using the 1994 RSA-911 database, NAC found that separate agencies reported higher average weekly earnings for successful closures than combined agencies did (NAC, 1997). However, the results of the present study do not indicate any differences in weekly earnings of clients with visual impairments based on the type of agency that served them. If there are not differences in VR clients' outcomes based on type of agency, one would wonder whether there is justification for continued funding of separate agencies. It is interesting to note that none of the interactions between agency type and the other independent variables were significant either, indicating that the effect of a client's (a) age, (b) education, (c) ease expenditures, (d) number of services received, and (e) months case was open on earnings was the same regardless of which type of agency provided services.

Whether the amount of money spent on clients contributes to their earnings at closure is a significant question. The results presented here indicate that, for persons with visual impairments, amount of money spent on the person's case is positively associated with earnings at closure. Although this variable contributed a small amount of unique variance (1.68%), its correlation with earnings was higher than expected (r = .24). This finding could help VR agencies justify spending larger amounts of money on services for clients, when the services are truly needed. Surprisingly, number of services received only contributed .08% additional variance to the model and had the lowest correlation with earnings (r = .08). The length of time a case was open was also not found to be a practically significant contributor to prediction of earnings (unique variance = .37%).

In summary, age was found to be a highly significant predictor of earnings for this sample of VR clients with visual impairments, explaining 13% of the variance on its own. Education level was also a significant predictor, explaining 5.1% of variance in earnings. Amount of money spent on a person's case was the only other significant predictor, however, its unique variance was substantially lower (1.68%). Also of importance was the fact that type of agency that served the VR client was not a significant predictor of earnings. Based on these findings, a model that could be expected to explain approximately 25.5% of the variance in weekly earnings of VR clients with visual impairments, regardless of which type of agency provides services, would include the client's age, education level, and case expenditures. It would be interesting to conduct a similar analysis with RSA-911 data from other years, to determine whether these variables consistently explain such a substantial amount of variance in earnings. It would also be of interest to determine whether the differences noted in client age for FY 1989 and FY 1997 were part of a trend of older clients with visual impairments being served by VR or merely an anomaly.

Table 1
Means, Standard Deviations, and Correlations * of
Continuous Variables

Variable       M          SD       1      2

Earnings      164.58     205.13   1.00

Education      11.81       3.00   0.28   1.00

Age            53.61      18.86   -.43   -.10

Services        4.92       2.75    .08    .09

Case cost   4,581.00   7,911.00    .24    .07

Months         24.98      24.66    .12    .01

Variable       3      4      5      6

Earnings

Education

Age            1.00

Services       -.13   1.00

Case cost      -.30    .37   1.00

Months         -.31    .30   -.51   1.00

* Note: Due to the large sample size, all correlations
above .02 are significant at the p < .0001 level.
Table 2
Standardized Regression Coefficients, t-Values, and Unique
Contributions of Independent Variables

Variable        Regression    t-value       Unique
                Coefficient              Contribution

Education           .23        33.47 *       .0510

Age                -.39       -53.54 *       .1305

Case cost           .16        19.18 *       .0168

Services           -.03        -4.17 *       .0008

Months             -.07        -9.00 *       .0037

Agency type         .01        -1.24         .0001

* p < .0001

References

Bellante, D. M. (1972). A multivariate analysis of a vocational rehabilitation program. The Journal of Human Resources, 7(2), 226-241.

Bellini, J., Neath, J., & Bolton, B. (1995). A comparison of linear multiple regression and a simplified approach in the prediction of rehabilitation outcomes. Rehabilitation Counseling Bulletin, 39(2), 151-160.

Capella, M. E. (2000). [Frequency of VR clients closed in FY 1997 as homemakers]. Unpublished data.

Capella, M. E. (2001). Vocational rehabilitation programs: Relationships among measures of effectiveness and state-to-state comparisons. Journal of Rehabilitation Administration, 25(1), 19-28.

Cavenaugh, B. S., & Pierce, S. J. (1998). Characteristics, services, and outcomes of rehabilitation consumers who are blind or visually impaired. Mississippi State, Mississippi: Mississippi State University, Rehabilitation Research and Training Center on Blindness and Low Vision.

Cook, J. A., & Rosenberg, H. (1994). Predicting community employment among persons with psychiatric disability: A logistic regression analysis. Journal of Rehabilitation Administration, 18(1), 6-21.

DeMann, M. M. (1963). A predictive study of rehabilitation counseling outcomes. Journal of Counseling Psychology, 10(4), 340-343.

Giesen, J. M., Graves, W. H., Schmitt, S., Lamb, A. M., Cook, D., Capps, C., & Boyet, K. (1985, January). Predicting work status outcomes of blind/severely visually impaired clients of state rehabilitation agencies (Technical Report). Mississippi State, Mississippi: Mississippi State University, Rehabilitation Research and Training Center on Blindness and Low Vision.

Kirchner, C., & Peterson, R. (1982). Effects of state agency structure on VR services of blind and visually impaired persons: Part II. Journal of Visual Impairment and Blindness, 76(2), 73-78.

Harrison, D. K., & Wayne, B. (1986). Sex equity in accessibility to rehabilitation services. Rehabilitation Counseling Bulletin, 30, 116-119.

Hayward, B. J. (1998, August). Third interim report: Characteristics and outcomes of former VR consumers with an employment outcome. Research Triangle Park: Research Triangle Institute.

Miller, L. A., & Barillas, M. G. (1967). Using weighted 26-closures as a more adequate measure of counselor and agency effort in rehabilitation. Rehabilitation Counseling Bulletin, 11, 117-121.

Moore, C. L., Flowers, C. R., & Taylor, D. (2000). Vocational rehabilitation services: Indicators of successful rehabilitation for persons with mental retardation. Journal of Applied Rehabilitation Counseling, 31(2), 36-40.

National Accreditation Council for Agencies Serving the Blind and Visually Handicapped. (1997). Outcomes achieved by consumers with vision loss served by specialized and general state agencies, FY 1994. New York: Author.

Pedhazur, E. J. (1997). Multiple regression in behavioral research: Explanation and prediction (3rd ed.). Fort Worth: Harcourt Brace College.

Schwab, A. J., & DiNitto, D. M. (1993). Factors related to the successful vocational rehabilitation of substance abusers. Journal of Applied Rehabilitation Counseling, 24(3), 11-20.

Weikel, W. J., & Johnson, R. H. (1974). A simple model for identifying probable failures in a rehabilitation program. Journal of Applied Rehabilitation Counseling, 5, 106-111.

Worrall, J. D., & Vandergoot, D. (1980). Indicators of nonsuccess for early vocational rehabilitation intervention. Rehabilitation Counseling Bulletin, 23(4), 282-290.

Michele E. Capella, 346 N. West Ave., 302C WAAX, University of Arkansas, Fayetteville, AR 72701. Email: [e-mail].