Does Work Incentive Benefits Counseling Improve Employment Outcomes for Those with Serious Disabilities? Preliminary Evidence for the “Work Oriented” from Two Demonstration Projects Barry S. Delin, Ellie A. Hartman and Christopher W. Sell Stout Vocational Rehabilitation Institute University of Wisconsin – Stout APPAM Research Conference Boston, MA November 5, 2010 The authors thank staff at Pathways to Independence, Office of Independence and Employment, Wisconsin Department of Health Services for their cooperation and support. This work was supported by the Centers for Medicare and Medicaid Services, Medicaid Infrastructure Grant – CFDA No. 93.768 and Social Security Administration Contract No. SS00-05-60008 The descriptions and interpretations in this paper are solely those of the authors. 1 I. Introduction Work incentive benefits counseling has been characterized as an essential service for helping persons with severe disabilities to either obtain employment or to improve outcomes gained through employment. Though there is a range of definitions for work incentive benefits counseling, we think that one offered by Golden et al. remains the best starting point. 1 Work incentive benefits counseling refers to services and supports aimed at (1) promoting and sustaining successful employment outcomes and (2) doing so in ways that allow those receiving the services and supports to make informed choices about their level of work activity. Key functional tasks include assessing the consumer’s situation and goals, identifying available options, and, based on the consumer’s preferences, tracking and managing benefits and/or work activity to help consumers achieve their goals. Nevertheless, claims about the effectiveness of work incentive benefits counseling services rest heavily on anecdotal reports from those either receiving or providing such services or on appeals to a largely implicit intervention theory. However, there are several studies that provide quantitative evidence of the positive impact of work incentive benefits counseling. This paper is intended as a contribution to this developing literature. The claim that work incentive benefits counseling is an essential service is really an assertion about a particular context: participation in public benefit programs in the United States where eligibility is contingent upon establishing that one is sufficiently disabled. Thus, advocates for the “necessity” of work incentive benefits counseling services focus on the two Social Security disability programs, that is, Social Security Disability Insurance (SSDI) and Supplemental Security Income (SSI), and the public health care programs (Medicaid and Medicare) where eligibility can arise from meeting Social Security disability criteria. Perhaps the most important of the criteria is inability to work at the Substantial Gainful Activity (SGA) level because of one’s disabling condition. SGA is defined by having the capacity to work at any job that will result (in 2010) in monthly earnings of $1000 or more. 2 For those participating in the SSDI and/or SSI programs, such work activity has the potential to result in loss of eligibility for these income support programs and for any others that rely on SSA disability criteria. Even when eligibility is not threatened, program rules can result in very large reductions of income when beneficiaries increase their earnings. Thus, work incentive benefits counseling involves identifying ways to best ameliorate tradeoffs between having higher earnings and maintaining program eligibility and/or benefit levels. It is the focus on ameliorating the substantially negative interaction between increases in earnings and disability program rules that most differentiates work incentive benefits counseling from other types of benefits counseling, particularly the type that 1 Golden, T. O’Mara, S., Ferrell, C. and Sheldon, J. (2000) “A Theoretical Construct for Benefits Planning and Assistance” Ithaca, NY: Cornell University, Program on Employment and Disability. Retrieved August 9, 2010 from http://www.ilr.cornell.edu/extension/files/download/theoretical%20construct.pdf 2 SGA is indexed to the Consumer Price Index for Urban Consumers (CPI-U). There is a somewhat higher SGA level for those who qualify for disability because they are blind. 2 concentrates on helping individuals gain eligibility for and/or maximize benefits from public programs. Henceforth, whenever the term “benefits counseling” appears, it refers solely to work incentive benefits counseling. Purposes of Study Recently, we completed an evaluation of a pilot of a SSDI benefit offset provision, a type of work incentive that reduces cash benefits more slowly than increases in earnings. The project was called the Wisconsin SSDI Employment Pilot or, for short, the SSDI-EP. 3 Those assigned to the treatment group had the opportunity to make use of an offset provision after completion of their Trial Work Period. When applied, the offset feature resulted in only one dollar of reduction in the SSDI benefit for every two dollars of earnings over SGA. However, study participants, irrespective of assignment, were promised access to benefits counseling services. This commitment was substantially fulfilled. 4 The pilot was intended to provide the Social Security Administration with information that would inform the design of a national demonstration of a SSDI benefit offset. The pilot also provided a largely unanticipated opportunity to examine the role of benefits counseling in promoting better employment outcomes. In brief, there were statistically significant increases in earnings, the probability of employment, the probability of having earnings at or above the SGA level, and imputed gross income associated with larger dosages of benefits counseling services. Moreover, statistically significant gains in employment outcomes appeared at surprisingly modest dosage levels; that is, at as little as four hours of reported services over a two year period. 5 Data for one of these outcomes (earnings) is presented in figure 1. 3 Delin, B., Hartman, E., Sell, C., and Reither, A. (2010) Testing a SSDI Benefit Offset: An Evaluation of the Wisconsin SSDI Employment Pilot Menomonie, WI: University of Wisconsin – Stout Vocational Rehabilitation Institute. 4 Through the end of the initial calendar quarter following the quarter of study entry, those in the treatment and control group received essentially the same amounts of benefits counseling services. Services were front loaded in the sense that the average participant would have received about half of their total benefits counseling services in the first six months of the two year period included in the analysis. Thereafter, those assigned to the treatment group tended to receive greater service hours. See Delin, et al., 2010. pp. 104-06. 5 Results for between subject differences were significant for all employment outcomes at a 0.05 p-value. MANOVA techniques were utilized because the technique supports time series analyses with limited sample sizes. Delin, et al. (2010) pp. 223-24. More detailed descriptions of model results for benefits counseling dosage are interspersed through pp. 221-58. 3 Figure 1: Quarterly UI Earnings by Benefits Counseling Dosage Categories, All SSDI-EP Participants UI Earnings by Benefits Counseling Dosage Categories, All SSDI-EP Participants 2000.00 1800.00 1600.00 Quarterly UI Earnings 1400.00 1200.00 BC Hrs = 8.0 or > BC Hrs. = 4.0 to 8.0 BC Hrs. = 0.1 to 3.9 BC Hrs. = 0 1000.00 800.00 600.00 400.00 200.00 0.00 q4pre q3pre q2pre q1pre q0 q1 q2 q3 q4 q5 q6 q7 q8 Quarter Relative to Entry Quarter Source: Wisconsin Unemployment Insurance and SSDI-EP encounter data Note: Constant Dollars (CPI-U, August 2005 adjusted to 100) Consequently, we thought that further analysis of the data from the SSDI-EP was warranted. Moreover, we had access to a deidentified data set from an earlier demonstration that was similar to the SSDI-EP on numerous dimensions. This project, Wisconsin Pathways to Independence (WI SPI) was one of the twelve projects the Social Security Administration funded as part of the State Partnership Initiative. Given the strong resemblances between the SSDI-EP and WI SPI and the availability of comparable data about service delivery and employment outcomes, it seemed reasonable to conduct analyses of the impact benefits counseling had in each effort. Additionally, there was enough similarity between the projects to explore whether the samples could be combined for some analyses. In particular, given that for both studies data were collected about the amount and timing of service provision, it would be possible to do go beyond previous studies that had been limited to simply identifying an association between receiving any benefits counseling and differences in employment outcomes. In this paper, we begin to explore the relationship between dosage (the hours of service delivered) and employment outcomes. We seek to answer two general questions, though we may not be able to generalize our answers beyond our samples to the universe of “working age” persons in the United States living with severe disabilities. These questions are: 4 • • What is the impact of work incentives benefits counseling on employment outcomes (such as mean earnings and the probability of employment). What is the influence of prior employment outcomes on the dosage of work incentive benefits counseling services received? We see these questions as closely related. We note that in most of the existing quantitative studies about the effects of benefits counseling there is a strong increase in measured employment outcomes in the time periods immediately following the start of the intervention. By contrast, results in later time periods are quite mixed, with generally either more modest increases or actual declines. This suggests the possibility that those who received benefits counseling were already primed to seek employment or, if already employed, to increase their earnings. Nonetheless, this observation does not pre-judge whether these individuals would have attained the same level of employment and earnings in the absence of benefits counseling. Additionally, we examine how the relationship between the dosage of benefits counseling and employment outcomes varies for subgroups based on employment status prior to entering either the WI SPI or the SSDI-EP. As already noted, we are interested in the effects of dosage for those: • • Seeking to enter the labor market either after long absence or for the first time Already employed at relatively high earnings, but seeking to increase their earnings Literature Review There are relatively few studies that focus directly on the impact of work incentive benefits counseling on employment outcomes. 6 In addition to limiting consideration of dosage to an essentially dichotomous condition (received/not received), we are not aware of any study that includes data about the quality of benefits counseling services received at the individual level. 7 The two most widely cited papers about the impact of benefits counseling services use data for vocational rehabilitation (VR) consumers in Vermont. Both studies estimate earnings for individuals in Social Security disability programs who received benefits counseling through Vermont’s State Partnership Initiative (SPI) project over a period starting two years before the receipt of benefits counseling services and ending two years thereafter. These individuals were volunteers. Their employment outcomes 6 We exclude analyses where the effects of benefits counseling are co-mingled with other services as, in particular, was the case for many of the projects operated during the State Partnership Initiative. Examples of such work include Shea, J. and Ekstrom, S. (2004) California’s Individual Self-Sufficiency Planning (ISSP) Project: Final Evaluation Report of a State Partnership Initiative (SPI) Demonstration Project. Napa, CA: Allen, Shea & Associates. Retrieved August 20, 2010 from http://www.migrats.org/uploads/CASPIReport.doc and Delin, B., Reither, A., Drew, J., and Hanes, P. (2004) Final Project Report: Wisconsin Pathways to Independence. Menomonie, WI: University of Wisconsin – Stout Vocational Rehabilitation Institute. 7 Some studies do include information about the quality implementation of benefits counseling at the project level. In fairness, this study suffers from the same limitation. 5 were compared to those for two matched groups of VR consumers who had not received benefits counseling. One study focused on outcomes for consumers across the “full” range of disabling conditions; the second analysis was restricted to subgroups of individuals with psychiatric disabilities. 8 In both analyses, those receiving benefits counseling services exhibited much stronger earnings performance over the eight post-intervention quarters than they had in the pre-intervention quarters. More importantly, their earnings trends in the postintervention period were much better than those observed for either of the comparison groups who had not received benefits counseling services. In the analysis including “all” disability types, mean quarterly earnings for the intervention group increased to approximately $1100 at the end of the post-intervention period after varying in a range between $530 and $660 during the pre-intervention period. These gains were far larger in both absolute and percentage terms than those for the two comparisons groups. However, both papers show that gains for those in the intervention group were disproportionately concentrated in the quarters immediately following the start of the intervention. More than 50% of the increase in mean earnings occurred in the first full quarter subsequent to study entry. In the case of the “all disability types” sample, earnings gains continued, albeit at a slower pace. However, the psychiatric disabilities subgroup attained their peak earnings in the third quarter and then began to experience a decline. 9 The strong increase in outcomes in the quarter following initiation of services raises an important question that is not examined in the Vermont papers. Are the outcomes driven primarily by benefits counseling or does the service operate principally as an effective means for individuals with severe disabilities to achieve goals they had already embraced? The answer to this question is potentially important for deciding for whom and especially when currently limited service capacity should be used and thus is one area we explore in this paper. A recent study utilizing Connecticut data examined the impact of benefits counseling on earnings and employment for a sample of approximately 5700 consumers divided into three groups: closed vocational rehabilitation cases who did not receive benefits counseling services, closed VR cases who did and consumers who received benefits counseling, but had not been served by the VR agency in the relevant period. 10 8 Tremblay, T., Smith, J., Xie, H., and Drake, R. (2004) “The Impact of Specialized Benefits Counseling Services on Social Security Administration Disability Beneficiaries in Vermont,” Journal of Rehabilitation, 70 (2) pp. 5-11; Tremblay, T., Smith, J., Xie, H., and Drake, R. (2006) “Effects of Benefits Counseling Services on Employment Outcomes for People With Psychiatric Disabilities,” Psychiatric Services, 57 (6) pp. 816-21. 9 This result is not explicitly identified in either paper, but can be discerned by looking at the figures provided in the articles. For example, line sighting the intervention groups’ mean earnings in figure 1 of Tremblay et al. (2004) one notices that there is an approximately $300 increase between the final pre-intervention quarter and the first post-intervention quarter. This constitutes about 60% of the roughly $500 growth between the final pre-intervention and post-intervention quarters. The proportion of peak earnings growth for the intervention group examined in Tremblay et al. (2006) appears to be greater (roughly 75%). See figure 1 in that paper. 10 The benefits counseling only group received services through the Social Security Administration funded WIPA program or its predecessor, the BPAO. 6 Everyone in the sample received benefits through at least one Social Security disability program. However, unlike in Vermont, those getting benefits counseling did not explicitly volunteer for a Social Security sponsored demonstration project. Samples were identified administratively. Thus, the Connecticut study has the virtue of examining a disability program population more likely to be representative of SSDI beneficiaries and SSI recipients than either the Vermont or, for that matter, the Wisconsin based studies. 11 The findings from Connecticut were broadly similar to those reported from Vermont, but exhibited some important differences. In most of the comparisons of average earnings or the proportions having non-zero earnings, the two groups getting benefits counseling performed better than the group that did not. Among individuals receiving benefits counseling services, those who had received VR services as well as benefits counseling exhibited slightly better outcomes. As in the Vermont, gains were concentrated in the period immediately following the start of the intervention. However, the scale and persistence of gains was substantially less than those associated with the Vermont SPI project. The authors suggested that the likely explanation for the more modest outcomes were differences in sample characteristics related to project goals and eligibility requirements. In addition to reviewing the literature about the quantitative relationship between benefits counseling and employment outcomes, we were interested in reviewing literature that identified reasons why benefits counseling would be expected to support better employment outcomes. We did not find a fully articulated intervention theory that might guide both further research and efforts to improve service delivery. Still, many of the building blocks that could be used for developing a formal intervention theory appear to be in place. Multiple observers claim that benefits counseling works because of the quality of information provided. Good information supports informed decision making by consumers; i.e., allowing them to pursue their goals however defined. 12 One observer hypothesized that good information about available work incentives and when they might be useful given a consumer’s specific goals and circumstances is particularly important. That is, the positive impact of benefits counseling might result from increased use of appropriate work incentives. 13 Indeed, discussions of practice domains in the literature posit that benefits counseling involves the application of information to individual needs, preferences, and circumstances. Consequently, benefits counseling may also be effective to the extent it includes an interactive process between the consumer and the professional that allows identification and clarification of consumer needs and goals. 14 11 Gruman, C., Shugrue, N., Kellett, K., Robison, J., and Porter, A. (2010) Medicaid Infrastructure Grant: The Impact of Benefits Counseling and Vocational Rehabilitation on Earnings Farmington, CT: University of Connecticut Health Center. 12 Golden, et al., (2000) p. 2. Also see Kregel, J., (2009) “Work incentive planning and assistance: Assisting beneficiaries to obtain employment and reduce dependence on SSA benefits,” Journal of Vocational Rehabilitation 31 (1), pp.1-9. 13 Personal communication from Susan O’Mara, Director of the WIPA National Training Center, Virginia Commonwealth University, August 25, 2010. 7 Descriptions of effective benefits counseling also stress the need for trust between the consumer and the professional. This is especially so when the consumer faces potential loss in eligibility, the level of benefits received, or in time and peace of mind if there are disputes with government entities over the benefits received or rescinded due to work activity. 15 Trust is said to arise from consumer confidence in the benefits professional’s competence and commitment to act in the consumer’s interest. However, it is possible that trust has other dimensions, for instance in reducing fears that work activity will result in harm. 16 It is also possible that exposure to benefits counseling may increase consumers’ capacities to plan or to increase their confidence in their abilities to make and implement decisions, thereby increasing their willingness to risk employment or increasing earnings. We think that a work group convened by the MIG-RATS (Medicaid Infrastructure Grant Research Assistances to States) made significant progress toward specifying a nuanced intervention theory. This work group created a visual representation of the likely relationships between benefits counseling and both proximate and longer term outcomes, identifying a chain of influences and in some cases the expected direction (positive or negative) of outcomes. The work group also specified partial models that researchers wanting to concentrate on a limited part of the overall model and/or facing data limitations might use. 17 However, the work group never developed justifications for why the model is ordered as it is. By not doing so, the group produced something less than an intervention model. Nonetheless, we have found the visual diagrams useful for helping to organize our thinking. Our analysis has certainly been influenced by what the MIG-Rats work group termed the “Abbreviated Employment and Income Model.” Note that this model omits intermediary conditions such as “understanding work incentives,” “barriers to work,” and “informed decision making” that, in the full model, are located between benefits counseling and the other elements in the abbreviated model shown in figure 2. Given our emphasis on the direct effects of dosage on employment outcomes, we think the use of the abbreviated model is appropriate. 14 Golden et al., (2000) pp. 2-5. and Lui, J., Chan, F., Lin, C., Anderson, C. and Peterson, M. (2010) “Roles and functions of benefits counseling specialists: A multi-trait analysis,” Journal of Vocational Rehabilitation 32 (3), pp. 163-73 (especially p. 171). 15 Kregel (2009) p. 2, Golden et al. (2000) p.2. and Lui, J. (2009) “An Emerging Profession Disability Benefits Specialists, “ Rehabilitation Counselors and Educators Journal 3 (2) pp. 48-54 (especially p. 52) 16 17 Personal communication from Susan O’Mara, August 25, 2010 This work group met in 2007-08. This discussion is based on a document distributed within the MIG-RATS work group in September 2008, “Guidelines for Using Benefits Counseling Outcomes Models.” The work group did not meet after this date. Though there was some external distribution of this product, it must be considered a draft. 8 Figure 2: MIG-RATS Abbreviated Employment and Earnings Model 18 Abbreviated Employment and Income Model + #1 Benefits Counseling #4a Use of Work Incentives + + #5b Cash Benefit Program Participation +/? ? -/? #4b Employment #6b Total Disposable Income ? + #5c Earnings from Work Benefits Counseling and the WI SPI and SSDI-EP Projects Both the WI SPI and the SSDI-EP were largely operated and designed by the Pathways to Independence unit housed in what is now the Wisconsin Department of Health Services. 19 Both projects were sponsored by the Social Security Administration with the purpose of encouraging those on disability benefits to “return to work.” Both projects were organized in a similar fashion with roughly twenty community based (“provider”) agencies enrolling and serving participants and the Pathways office providing project co-ordination, including providing or arranging for the provision of training and technical assistance. WI SPI operated between July 1999 and September 2004. The SSDI-EP started in summer 2005 and officially continued through the end of 2008. 20 Participants in both projects were volunteers. However there were differences in eligibility requirements, the most important of which was, that while WI SPI participants could be attached to either or both Social Security disability programs, eligibility in the SSDI-EP was restricted to 18 “Guidelines for Using Benefits Counseling Outcomes Models” (2008) pp. 5-6. 19 Wisconsin Division of Vocational Rehabilitation had a significant role in the design and management of WI SPI. SSA mandated some of the critical features of the SSDI-EP (e.g., the offset feature and the basic eligibility rules) but left decisions about project structure and service delivery to Pathways. 20 Some members of the SSDI-EP treatment group retained their ability to use the offset provision after this date, necessitating residual capacity to track their work activity and to provide benefits counseling as needed. 9 SSDI beneficiaries who received benefits based on their own work histories and had no attachment to the SSI program. Finally, SSDI-EP participants were randomly assigned to a treatment and control group, whereas WI SPI participants were not. Work incentive benefits counseling was a critical service element in both projects. Participants were expected to have access to as much benefits counseling as they needed to pursue their employment goals. 21 In WI SPI, benefits counseling was supposed to be fully integrated into a consumer centered vocational planning process. In the SSDI-EP, integration was encouraged, but not required. In both projects, training and technical assistance for benefits counseling came from a common source and the continuities in content and rigor were far stronger than the differences. 22 Counselors’ initial training required approximately fifteen days and all active counselors received follow-up training on a quarterly basis. Benefits reviews produced by inexperienced benefits counselors were subject to review by staff at the training entity. Nonetheless, there were important differences between the projects in participants’ receipt of benefits counseling services, a fact that complicates comparisons between the projects. Those in the WI SPI sample received an average of thirty-nine hours of benefits counseling in the nine quarters we look at, compared to slightly over eight for the SSDI-EP sample. Though benefits counseling delivery was front loaded in both projects, those in WI SPI received almost three times as much in the enrollment quarter and that which immediately followed (Q0-Q1). In the WI SPI only about 4% of participants had no reported hours of benefits counseling over the Q0-Q8 period compared to 21% during the SSDI-EP. By contrast, it is likely that only a handful of WI SPI participants received benefits counseling prior to entering the project, while about a third of SSDI-EP participants reported having received some prior service. 23 21 However, the provider agencies were in the strongest position to make decisions about what constituted the upper limits of service provision for individual cases. 22 During WI SPI, Employment Resources Incorporated (ERI) provided training and technical assistance, generally in the tradition developed at Cornell University. As an outgrowth of the SPI project, Pathways helped to organize a permanent training and TA capacity housed at a consortium called the Wisconsin Disability Benefits Network (WDBN). WDBN was housed at ERI and, with some modification, continued what ERI had developed. 23 In 1999, the SSA sponsored BPAO program had not yet begun. Within Wisconsin only four organizations offered any benefits counseling services. However three of these became provider agencies for the project. It is reasonable to suppose that some small number of WI SPI participants had received services prior to project entry. By contrast, by the start of the SSDI-EP, there were many more sources of benefits counseling services in Wisconsin. It is not surprising that a third of participants had some exposure to the service. It is quite probable that our figure is an underestimate as it only captures reports of benefits counseling at the provider agency where a participant enrolled or imputes such receipt for those who had participated in WI SPI (about 4% of SSDI-EP participants). However, as many participants enrolled at provider agencies where they had already received services, we believe that the error is not large. 10 Table 1: Receipt of Benefits Counseling Services, WI SPI and SSDI-EP Participants WI SPI (n=443) SSDI-EP (n=468) Benefits Counseling, 38.7 Mean Hours, Q0 through Q8 Benefits Counseling, Mean 11.1 Hours, Q0-Q1 (“initial”) % Receiving No Benefits 4.3% Counseling Services % with Reported Benefits 0% Counseling Prior to Study Entry Source: WI SPI and SSDI-EP Encounter Data 8.2 4.1 21.2% 35.3% We think it likely that these differences in service provision were rooted in differences in how service provision was funded. During the first years of the WI SPI project, provider agencies received a capitation meant to cover costs for each participant that was negotiated between each agency and project management. Management wanted the capitations to be generous enough to encourage provider agencies to remain in the project and to fully implement the service approach. At the least, the capitation rates were generous enough to provide for what, in comparison to the SSDI-EP, can be viewed as generous service provision. 24 There was no specific funding source for service provision in the SSDI-EP. Provider agencies were expected to find funding to meet each participant’s service needs. In practice, Pathways was always available to pay for benefits counseling services, though in the first years of the project not all agencies took full advantage of this option. 25 Samples Our analyses are limited to the WI SPI and SSDI-EP participants (433 and 468, respectively) with nine quarters of both project service and Wisconsin Unemployment Insurance (UI) earnings data starting with the calendar quarter in which enrollment occurred. In addition, we have access to UI data for all these participants for the four quarters prior to the enrollment quarter (Q-4 through Q-1) to look at employment 24 We have already noted that the mean amount of benefits counseling services for a WI SPI participant was almost five times greater than for a SSDI-EP participant. Over the same nine quarter period a WI SPI participant received fifty-nine hours of vocationally related services compared to a little over nine hours, on average for a SSDI-EP participant (almost a 7:1 ratio). This does not include VR funded services provided by other vendors (including other units within the provider agency). Given the differences between the project samples in the likelihood of being a VR consumer, we would expect differences to again favor WI SPI participants. Nonetheless, other factors than how the projects were organized and funded probably had a role in motivating observed differences. For example, when there is a larger proportion of participants classified as having a cognitive/developmental disability (as in WI SPI) there is likely to be more utilization of some high intensity services, especially those associate with supported employment. 25 Failure of participants to receive services were more often associated with other factors such as attrition of benefits specialists, training delays, and, quite likely, participants’ failure or reluctance to ask for services. 11 outcome trends in the year preceding entry to either WI SPI or SSDI-EP. Given that participants in both projects entered over fairly long periods of calendar time, analyses are always referenced to the calendar quarter of project entry. 26 Table 2 provides information about selected demographic, experiential and program participation characteristics for the two samples. Many of the differences between the WI SPI and the SSDI-EP samples reflect the inclusion of SSI recipients and of SSDI beneficiaries who got benefits based on another’s earnings history in the WI SPI project. Even so, a large proportion of the WI SPI sample (72%) had some form of SSDI eligibility prior to study entry. Thus, these individuals were subject to losing all of the cash benefit associated with that program should their earnings exceed SGA. Advocates for work incentive benefits counseling would contend that this possibility would increase the potential need for and value of benefits counseling services. 26 The WI SPI sample is limited to participants who entered in the first twenty-five months of the project. There were some differences between these and later enrollees. In particular, early enrollees were more likely to be employed at study entry. The SSDI-EP sample is drawn from the full fifteen month enrollment period. 12 Table 2: Selected Demographic, Experiential, and Program Participation Characteristics of WI SPI and SSDI-EP Participants WI SPI (n=443) SSDI-EP (n=468) Age Mean 38.2 45.2 Median 39.0 46.0 Gender Female 37.4% 44.7% Male 62.6% 55.3% Race/Ethnicity Non-White 22.3% 12.8% White 77.7% 87.2% Hispanic 3.6% 3.2% (includes non-white & white) Education Less than High School 16.5% 5.6% Completion High School Completion 30.2% 27.1% (including GED) More than High School 53.3% 67.3% Completion Disability Categories 27 Cognitive/Developmental 23.3% 7.4% Disabilities Affective/Mental Health 27.8% 36.3% Physical (including Sensory 49.0% 53.5% & HIV) Other --2.7% 28 SSA Disability Program SSDI (including concurrent) 72.2% 100% SSI (only) 27.8% 0% Medicaid Buy-in Participation, Q0-Q8 Yes 12.6% 51.1% No 87.4% 48.9% Source: WI SPI and SSDI-EP Encounter and Survey Data, Administrative Data from WI Department of Health Services, WI Division of Vocational Rehabilitation, and the Social Security Administration Note: All percentages calculated based on available cases 27 The process of assigning cases to disability categories differed between studies. For WI SPI, RSA 911 disability codes were recoded into the three categories by the evaluators and were supplemented by survey data. For SSDI-EP, provider agency staff made the assignments based on instructions from project management at Pathways. 28 The information in table 1 reflects program status prior to WI SPI enrollment. Any period of inclusion in the SSDI program in the two years prior to enrollment resulted in a coding of SSDI. 13 We have already identified some differences between the WI SPI and SSDI-EP in project eligibility criteria. These are not insignificant. A project restricted to SSDI beneficiaries will probably have a higher proportion of participants with prior work experience and higher levels of human capital development. Moreover, a project restricted not only to SSDI beneficiaries, but also excluding those getting concurrent SSI benefits or SSDI benefits based on a parent’s earnings record, will generally have a smaller proportion of individuals with severe cognitive disabilities. The SSDI-EP sample exhibits these tendencies, relative to the WI SPI sample. Most importantly, despite some differences in eligibility requirements, both WI SPI and SSDI-EP participants must be characterized as unusually “work-oriented” relative to the typical working-age participants in Social Security disability programs. We borrow this concept from Gina Livermore, who defines the category based on having work goals or expectations. 29 Livermore found that such individuals were far more likely than others in Social Security disability programs to report engaging in work related activity such as job training, searching for work, or being gainfully employed. Though Livermore’s definition leaves some ambiguity as to whether the concept is primarily attitudinal or behavioral, we would lean to the latter formulation. That is statements or measures of goals or intent only indicate “work-orientation” if their presence is strongly indicative of behavior, either current or in the very near future. Table 3 provides information about WI SPI and SSDI-EP participants’ employment histories, both since entering benefit status and, particularly, in the year prior to entering one of the projects. Though volunteering for one of these projects could, by itself, be viewed as an indicator of a strong orientation toward work, we think the data documenting a relatively high level of workforce participation and earnings before enrollment provides even stronger evidence. According to data from the 2004 National Beneficiary Survey approximately 9% of those in Social Security disability programs reported employment at the time of the survey, with about 13% reporting employment during the previous year. 30 By contrast, in the calendar quarter immediately prior to project entry, 35% of the WI SPI sample was employed, as indicated by having positive earnings in Wisconsin UI records. The comparable figure for those in the SSDI-EP sample was 44%. Another indicator of unusually strong commitment to employment can be seen in the proportions of our samples who reported working since going on either SSDI or SSI benefits: 67% for the 29 Livermore, G. (2008) “Disability Policy Research Brief Number 08-01: Earnings and Work Expectations of Social Security Disability Beneficiaries,” Washington, DC: Center for Studying Disability Policy, Mathematica Policy Research, Inc. p. 2 and Livermore, G. (2009) “Disability Policy Research Brief Number 09-05: Work-Oriented Social Security Disability Beneficiaries: Characteristics and Employment-Related Activities.” Washington, DC: Center for Studying Disability Policy, Mathematica Policy Research, Inc. p. 2. 30 Livermore. (2008) p. 2. In point of fact, participants in the WI SPI and SSDI-EP appear to have had, at project entry, higher employment rates than those in the work-oriented subgroup that Livermore identified from the 2004 National Beneficiary Survey (21% at the time of interview, 29% at some point during the prior year). See Livermore. (2009) p. 2. 14 WI SPI group and 81% for those entering the SSDI-EP. 31 However, as might be expected for a SSDI only population, average earnings prior to enrollment and human capital measures like educational attainment (see table 2) were somewhat higher in the SSDI-EP participant sample than in the WI SPI sample. Table 3: Selected Employment History Characteristics of WI SPI and SSDI-EP Participants WI SPI (n=443) SSDI-EP (n=468) 67.0% 80.8% Self-Reported Employment between SSA Disability Program and Study Entry 32 34.8% 43.6% UI Employment in Quarter Prior to Study Entry (Q-1) 49.2% 55.1% UI Employment in Any Quarter in the Year Prior to Study Entry $608 $894 Mean UI Earnings in Quarter Prior to Study Entry (Q-1) 22.6% 27.6% UI Earnings =/>$1200 in at least Two Quarters in the Year Prior to Study Entry Source: WI SPI and SSDI-EP Encounter and Survey Data, Administrative Data from WI Department of Workforce Development Note: All percentages calculated based on available cases Note: Monetary amounts deflated using CPI-U (August 2005=100) It also appears that the samples had, on average, somewhat higher earnings in the period leading up to study entry than another group of employed persons with severe disabilities in Wisconsin. On average, Wisconsin Medicaid Buy-in participants who would have been eligible for the SSDI-EP earned about $400 per quarter, considerably lower than the means for the quarter prior to enrollment (Q-1) for either the WI SPI sample ($608) or the SSDI-EP sample ($894). 33 Additionally, the proportion of WI SPI and SSDI31 The UI employment rate for the WI SPI comparison group in the quarter prior to their nominal entry date (40%) was also unusually high for a group of persons receiving Social Security disability benefits. 32 If there were positive earnings reported in WI Unemployment Insurance records for any quarter in the year prior to study entry, a self-report of non-employment was changed to a report of employment. 33 The Wisconsin Buy-in, with temporary exceptions, requires employment (though payment can be through “in kind” services). The quarterly earnings data for two samples of Buy-in participants was tracked, respectively, from 2005-07 and 2006-08. All amounts are inflation adjusted using the CPI-U with August 2005 set as the 100 index value. See Delin, B. and Hartman, E. (2010). “Characteristics of Joint Participants in the Wisconsin Medicaid Buy-in and the SSDI Benefit Offset Pilot: A Comparison with the Population of SSDI Beneficiaries Enrolled in the Wisconsin Medicaid Buy-in.” Pittsburgh PA: MIG/DMIE Employment Summit, April 2010. pp.16-19. 15 EP participants who earned well above the mean values was substantial. Roughly a quarter of both samples had reported earnings of at least $1200 in at least two of the four quarters preceding study entry. The two samples are also unusual in that earnings growth begins well before enrollment with no evidence of any reduction in employment effort on the expectation of entering a program aimed at improving employment outcomes. The trend lines for mean UI earnings can be seen in figure 3. Figure 3: Mean Quarterly UI Earnings Q-4 through Q8, WI SPI and SSDI-EP Samples Mean Quarterly UI Earnings for WI SPI and SSDI-EP Participants 2000 1800 1600 Quarterly UI Earnings 1400 1200 SSDI-EP WI SPI 1000 800 600 400 200 0 q4pre q3pre q2pre q1pre q0 q1 q2 q3 q4 q5 q6 q7 q8 Quarter Relative to Entry Quarter Source: Administrative Data from WI Department of Workforce Development Note: Monetary amounts deflated using CPI-U (August 2005=100) Employment rate trends (not shown) were generally similar, though understandably less pronounced. These results are notable in that they show that a large proportion of employment outcome gains, especially for the SSDI-EP sample, took place before program entry and thus was unrelated to benefits counseling received during the projects. This is another reason for our interest in seeking to identify factors that increase use of benefits counseling services as well as seeking further understanding of (apparent) dosage effects. It is important to note that the observed differences in mean earnings largely reflect differences in UI employment rates across groups. However, the average quarterly earnings for those employed according to UI records were about 25% higher for SSDI-EP participants at project entry. That difference grew considerably to approximately 65% higher by Q8. 16 Data Sources and Key Variables Data used in this paper come from multiple sources: government agency records, participant surveys, and encounter data collected by each project’s provider agencies. Most of the data elements used were collected in identical or highly comparable ways during the two studies, but we identify significant exceptions. Information about employment and earnings are from Wisconsin Unemployment Insurance records. Though considered accurate within their limitations, there are important exclusions that result in underestimates of earnings and employment rates. The most important exclusions, for our purposes, are self-employment and employment by firms located outside Wisconsin. Additionally, we use UI earnings data to construct a proxy for individual income. This proxy is the quarterly sum of UI earnings and Social Security disability program payments. Though the proxy measure has some important shortcomings, it allows us to explore the relationship between benefits counseling and income. 34 Indeed, an important reason that benefits counseling may be important is that marginal increases in earnings can sometimes leave beneficiaries with far less income. Both UI earnings and the income proxy are inflation adjusted using the CPI-U (August 2005 =100). Information about the amounts of benefits counseling services and other project services were obtained from monthly reports from the community agencies where WI SPI and SSDI-EP participants were enrolled. Service provision was measured in hours and reports were not limited to time spent with participants. Monthly reports were aggregated on a quarterly basis. Definitions for benefits counseling and vocational service categories were highly consistent across the two studies. Nonetheless, the raw quarterly benefits counseling dosage variable raised a difficult issue. Though the magnitude of change in employment outcomes was similar across the two studies, the mean dosage provided to WI SPI participants over the Q0Q8 period was about five times larger than what was provided to SSDI-EP participants. Given this, though we hypothesized that an hour of benefits counseling would be associated with some gain in earnings and perhaps other employment outcomes, it was clear that the estimates would have very different magnitudes. Therefore, we decided to reconceptualize dosage differences in a relational framework. Given relatively easy access to benefits counseling services, what was the effect of differences in dosage relative to the mean for the project? For our estimates of the impact of benefits counseling on employment outcomes, we created a categorical variable based on the standard deviation of Q0-Q8 dosage in each of the studies. Because the distributions include more cases below the mean than above it, we used a half rather than full standard deviation score to create the following dosage categories. The largest dosage category contains those cases that received an amount of service more than one half standard deviation above the mean dosage for the project. A middle group is defined by dosages within one half standard deviation of the 34 Most notably, it ignores income produced by other family or household members as well as excludes some potentially important sources of individual income. Against this, it must be noted that large proportions of study participants resided in households where they were the sole adult. 17 mean. The lowest dosage group includes those participants, who relative to others in their project, received a dosage less than the amount that was one half a standard deviation below the mean. This final group includes those participants who received no benefits counseling services in their Q0-Q8 periods. Details about these dosage groups are available in table 4. When estimating the impacts of employment related factors, particularly those occurring prior to study entry on the amount of benefits counseling received, we use the actual dosage amounts as the dependent variable. Table 4: Benefits Counseling Dosage Categories WI SPI SSDI-EP Mean Hours of Benefits 38.7 8.2 Counseling, Q0-Q8 ½ Standard Deviation 21.1 6.2 % in High Dosage category 19.0% 20.5% % in Medium Dosage 49.0% 38.2% Category % in Low Dosage Category 32.1% 41.2% Source: WI SPI and SSDI-EP Encounter Data Note: Dosage Categories Relative to One Half of a Standard Deviation Above or Below Mean Dosage In addition to our overall estimates for the WI SPI and SSDI-EP samples, we are also interested in looking at two subgroups within the larger samples: (1) those seeking to enter the labor market either after a long absence or for the first time and (2) those already employed with relatively high earnings, but seeking to further increase their earnings. We define the first group as those individuals who did not report any employment between going on disability benefits and enrolling in the study. In the SSDIEP this data was collected as part of the enrollment process. For WI SPI this information was collected by survey. Unfortunately, this resulted in missing data for about 9% of the sample. We defined the relatively high earnings prior to study entry subgroup as those with quarterly earnings of at least $1200 (inflation adjusted) for at least two calendar quarters in the year prior to the quarter of project enrollment. Given our relatively small sample sizes, we use these subgroups as control or independent variables in our models, rather than running separate models for each subgroup. 35 The other control variables used in the models are split into two types: those available, though not always in fully equivalent form, for both the WI SPI and SSDI-EP and those applicable or available for only one of the studies. Variables of the first type include age, gender, disability type, education, and vocational service hours over the Q0-Q8 period. 36 Variables exclusive to the WI SPI study are SSA program type and use of the SSI waiver. Variables exclusive to the SSDI-EP study include prior use of benefits counseling, the Medicaid Buy-in, study group assignment, TWP completion over the Q035 As these two variables are necessarily strongly correlated, they are never used together in the same model. 36 Differences in the source data used to construct the disability type variable have already been discussed. Additionally, there are two variables in this group where there are a noteworthy proportion of cases without information. Disability type information is missing for 5.3% of the SSDI-EP sample. The education variable is missing for 1.4% of the WI SPI sample. 18 Q8 period, and actual use of the SSDI benefit offset provision. 37 These study specific variables are used mainly in the outcomes models that examine the mediating effects of work incentives. Methods Completing this paper required performing two rather different types of analytical tasks. The first and more important task was to discern the impact of different dosages of benefits counseling on employment outcomes over the nine quarter analysis period. The aim was to identify trends over the Q0-Q8 time period as well as to estimate and explain results at the end of that period. The second task was to account for the amount of benefits counseling received by participants based on their characteristics, particularly employment outcomes during the period leading to project entry. Different methods were chosen to complete each of these tasks. Specific choices reflect data limitations, most notably limits on the number of available cases and, secondarily, differences in the magnitude of benefits counseling service delivery between the WI SPI and SSDI-EP projects. We estimate dosage impacts for both the combined samples and separately for each project. We think both approaches have utility in what is in many respects an exploratory study. The first approach maximizes sample sizes and is defensible given the close similarity between the projects in both the character of the benefits counseling provided and overall project structure and goals. Nonetheless, WI SPI and the SSDI-EP had important differences in their eligibility requirements, use of random assignment, work incentive features, and, above all, in the magnitude of service provision. Though defining dosage categories relative to the mean is intended to support useful comparisons across the two projects, one cannot wish away the possibility that dosage hours have a direct impact on employment outcomes. Modeling techniques were chosen based on sample size, though in all cases our goal was to compare group trends across time. In all cases, we limited these analyses to the Q0-Q8 period as the goal is to assess impacts for benefits counseling delivered through the projects. For the analyses including participants from both projects we use a mixed linear regression model to estimate the impact of three benefits counseling dosage categories. For the smaller WI SPI and SSDI-EP samples, we use repeated measures MANOVA (Multivariate Analysis of Variance) to estimate impacts. This technique has the advantage of allowing us to run time series models with multiple control variables despite relatively small sample sizes. Both types of techniques support estimation of differences across time (within participant) and between groups. While readers can be expected to be familiar with the interpretation of regression statistics, it may be useful to provide some particulars about repeated measures 37 By definition, there are no SSI recipients in the SSDI-EP sample. Similarly, the study assignment variable does not apply to WI SPI. Though it is possible that a handful of WI SPI participants received benefits counseling before that project, it would not have been a large enough proportion to include in our regression models. Similarly, as only 12% of the WI SPI used the Medicaid Buy-in in their Q0-Q8 participation period, we chose not to use that variable in estimating outcomes for that study. Finally, though it appears that some information about TWP completion had been collected during WI SPI, for whatever reason it was not retained in the deidentified data set from that project. 19 MANOVA. The significance of a variable for between subject comparisons is a straightforward probability value. The significance of a variable for within subject comparisons is the probability value for the wilks’ lambda statistic. In both cases, we use the standard .05 level to denote statistical significance. The effect size of the variable (i.e. the amount of variation explained) is estimated by the partial eta squared for both within subject and between subject effects. This is estimated separately for each effect type. MANOVA involves some disadvantages relative to regression techniques. Most notably, independent variables have to be categorical. 38 As a consequence, some of the information available when a variable is in continuous form is lost. An additional challenge is that MANOVA does not produce a direct equivalent to the beta coefficients available from regression analyses. Though it is still possible to identify the rate of change over a particular time period, this needs to be separately calculated using the categorical (marginal) means. In the process of conducting our analyses, we discovered circumstances when an independent variable such as benefits counseling dosage was not significant though the categories were seemingly associated with substantial differences in employment outcomes over time. This suggested the possibility of there being important mediator variables. We explore this possibility using standard OLS linear regression techniques, though we restrict these analyses to the use of work incentives. Finally, we also use linear regression techniques to explore the role that preenrollment conditions had on the total hours of benefits counseling services provided through the projects over the Q0-Q8 period. These models also include a range of demographic and disability control variables and, for the SSDI-EP sample, information about the receipt of benefits counseling prior to enrollment. Regressions are performed stepwise and final models are limited to variables that were statistically significant at the .05 level. Findings: Benefits Counseling Impacts All analyses of the impact of the amount of benefits counseling service over the Q0-Q8 period utilize a categorical measure of dosage relative to the mean dosage in the project any given case participated in. This is done to adjust for the considerable difference in the mean dosage values between WI SPI and the SSDI-EP. Dosage is standardized into three categories, based on whether the amount of benefits counseling provided was within one half standard deviation of the project mean. Models for both the combined sample and those for each of the two projects use a common set of control variables. Conceptually, the most important of these is a dichotomous variable intended to distinguish participants based on having relatively strong employment outcomes in the year prior to entering either WI SPI or the SSDI-EP. Inclusion in the stronger outcome group is based on having two calendar quarters of (inflation-adjusted) UI quarterly earnings of at least $1,200 during this period. Our 38 MANOVA allows multiple independent variables. While independent variables must be entered into the model in categorical form, other covariates can be entered as continuous variables. 20 expectation is that strong employment outcomes prior to entering a “return to work” program would be positively correlated with post-entry outcome levels. Other control variables include gender, age, disability type, and education. Excepting age, all of these variables are categorical. Disability type variables include 1) physical/HIV/sensory disabilities, 2) cognitive/developmental disability, 3) affective/mental disability, and 4) others. Education attainment is measured in three categories: 1) having less than a high school diploma/GED, 2) high school (or GED) completion, and 3) having any post-secondary education. Three employment outcomes are examined in both the combined and separate sample analyses: mean quarterly earnings, quarterly employment rates, and the quarterly mean for the individual income proxy variable. We highlight the results for earnings relative to the other outcomes. Our reasons are as follows: Employment, in itself, has essentially no bearing on continued eligibility for Social Security disability programs. While benefits counseling may encourage employment, the primary purpose of the service is to enable consumers to earn more without loss of eligibility and/or reduction in income. Though, given this, income could be seen as more important, from a consumer’s perspective, than earnings, the inherent limitations of our income variable suggest it would be imprudent to focus over much on those estimates. Combined Sample Results We use a mixed linear regression modeling approach to analyze the effects of the predictor variables (benefits counseling, pre-enrollment earnings, and the four demographic variables) on earnings, employment, and income. The following specification is for the earnings model; except for the outcome variable, the employment and income models have identical specifications. Earnings = β 0 + β 1 (Quarter ) + β 2 (HalfSDBCCa tegory ) + β 3 (Pr eEarningsC ategpry ) + β 4 (Quarter * BenefitsCo unseling ) + β 5 (Quarter * Pr eEarnings ) + β 6 (Quarter * BenefitsCo unseling * Pr eEarnings ) + β 7 (Gender ) + β 8 ( Age) + β 9 ( Education ) + β 10 (Cognitive / DevelopmentalDisabil ity ) + β 11 ( Affective / MentalDisa bility ) + β12 (OtherDisability ) 39 Beta coefficients should be interpreted as unit changes in amounts for both the earnings and income models and as changes in percentage points for the employment model. As these models are linear, the changes associated with any quarter are an average of changes across the Q0-Q8 period and thus may oversimplify actual trends. Graphs of the estimates for these models are available in Appendix A. We begin our discussion of findings for these models by looking at earnings. This earnings model looks at the influence of the predictor variables on earnings from the enrollment quarter (Q0) up through eight quarters following enrollment. Benefits counseling did not have a significant influence on earnings at the enrollment quarter, but did influence changes in earnings over time. As expected, those who had at 39 The physical/sensory/HIV disability category serves as a reference category for the other three. 21 least two quarters of earnings of at least $1200 during the four quarter pre-enrollment period did have significantly higher earnings during the enrollment quarter than participants who did not earn as much during the four quarters prior to the enrollment quarter. Gender, age, and disability type also had an overall influence on earnings, but, unexpectedly, education did not. 40 Table five provides the p-values and dollar estimates of the impacts of each of these predictor variables. Table 5: Estimates of Effects on UI Mean Earnings for Combined Samples Estimates of Effects in Mixed Linear Earnings Model (n = 911) Parameter Beta Std. Error P-Value Intercept 564.3 178.90 0.0016 Quarter -14.1 22.69 0.5348 Benefits Counseling 96.7 57.58 0.0931 PreEarnings 1700.5 229.41 < 0.0001 Quarter * Benefits Counseling 34.3 11.42 0.0027 Quarter * PreEarnings 11.3 45.02 0.8016 Benefits Counseling * PreEarnings 48.8 110.90 0.6602 Quarter * Benefits Counseling * PreEarnings -41.9 21.87 0.0557 Gender (reference = female) 97.9 49.64 0.0487 Age -10.5 2.29 < 0.0001 Education 54.4 34.67 0.1165 Cognitive/Developmental Disability -404.8 69.69 < 0.0001 Affective/Mental Disability 156.0 52.94 0.0032 Other Disability -1197.4 211.72 < 0.0001 One way to interpret these results is to visualize the influence of each predictor variable on earnings. To do this we need a starting place. For simplicity sake, let’s start with individuals who are female, have a physical/HIV/sensory disability, are age seventeen, and had less than two of four pre-enrollment quarterly earnings of at least $1200. By adding the beta coefficient of each of these variables to the intercept value one can calculate the Q0 mean earnings estimate. These individuals had predicted enrollment quarterly earnings of $386. Utilizing the coefficients in table 5, one can estimate changes from our initial example reflecting differences in participant characteristics and time in the project. For individuals with pre-enrollment earnings of at least $1200 in at least two quarters, their enrollment quarter earnings are predicted to be $1701 more, adding up to $2086 in baseline (Q0) earnings. If a participant was male, enrollment quarter earnings are predicted to be $98 higher. For every one year increase in participant age at enrollment, enrollment quarter earnings are predicted to be $11 less. Enrollment quarter earnings are predicted to be $405 less for people with cognitive/developmental disabilities, $156 40 Though we do not explore this anomaly in this paper, we speculate that this result is a reflection of the relatively low employment rates for participants classified as having a physical disability. While participants classified as having physical disabilities tended to have greater educational attainment and higher earnings when employed than other participants, mean earnings in these projects appear to be strongly influenced by employment rates. Indeed, the expected benefits of educational attainment on earnings appear in the income models, where greater educational attainment and the higher earnings associated with it increased the probability of SSDI eligibility and, when participating in that program, the size of the cash benefit. 22 more for people with affective/mental disabilities, and $1198 less for people with other types of disabilities relative to that for persons in the physical/HIV/sensory disability category. Most pertinent to our analysis, the amount of benefits counseling positively influences earnings growth over the post-enrollment period. The variable of interest is the interaction between the dosage category and time (Quarter*Benefits Counseling) available. For those individuals who received a benefits counseling dosage that was less an amount one half a standard deviation below the mean for their project, estimated mean earnings increased $34.30 each post-enrollment quarter. Individuals who received an “average” amount of benefits counseling (within ½ a SD of their project mean) increased their earnings by an estimated $68.60 per quarter. The largest increases in earnings are found for those individuals who received dosages more than one half standard deviation above their project mean over Q0-Q8. Their earnings increase an estimated $102.90 per post-enrollment quarter. Consequently, those in the medium dosage group gain an estimated $274 in quarter eight earnings relative to the low dosage group. Those in the high dosage group gain about $549 relative to the low dosage group. When differences in total earnings are calculated for the full period the differences are even greater. The estimated relative difference in cumulative earnings over the Q0-Q8 period between the low and medium dosage groups is nearly $1235. The comparable difference between the low and high dosage group earnings is about $2470. In the employment model, the predictor variables exhibit a similar pattern of influence on the outcome as they did for earnings. Nevertheless, there are key differences. Education has a statistically significant but negative impact on Q0 employment rates. Employment rates did not vary between those with physical/HIV/sensory, cognitive/developmental, and other disabilities, although those with an affective/mental disability did have higher employment rates to match their higher earnings found in the previous model. Finally, there is an overall predicted decrease in employment during the post-enrollment quarters. However, this downward trend in employment rates is actually offset for those individuals in the middle and high benefits counseling dosage categories. Members of both these groups display gains in their estimated employment rates. The following table provides the p-values and beta estimates for the mixed linear employment model. 23 Table 6: Estimates of Effects on UI Employment Rates for Combined Samples Estimates of Effects in Mixed Linear Employment Model (n = 911) Parameter Beta Std. Error P-Value Intercept 0.4831 0.0462 < 0.0001 Quarter -0.0146 0.0059 0.0137 Benefits Counseling 0.0107 0.0152 0.4802 PreEarnings 0.4602 0.0605 < 0.0001 Quarter * Benefits Counseling 0.0130 0.0030 < 0.0001 Quarter * PreEarnings 0.0065 0.0117 0.5771 Benefits Counseling * PreEarnings 0.0299 0.0292 0.3066 Quarter * Benefits Counseling * PreEarnings -0.0143 0.0057 0.0124 Gender (reference = female) -0.0471 0.0126 0.0002 Age -0.0035 0.0006 < 0.0001 Education -0.0316 0.0088 0.0004 Cognitive/Developmental Disability 0.0217 0.0178 0.2223 Affective/Mental Disability 0.1757 0.0135 < 0.0001 Other Disability 0.0797 0.0538 0.1386 Like the mixed linear earnings model, the mixed linear employment model estimates employment rates based on adding the coefficients for the relevant predictor variables to the intercept value. For a starting point, let us assume a participant is female, seventeen years old, has a less than high school education, a disability type other than affective/mental, and did not have earnings of at least $1200 for at least two of the four pre-enrollment quarters. The estimated employment rate at the enrollment quarter for participants who matched this description is 38.3%. Had we identified a group that was identical, except for having earnings of a least $1200 in at least two of the four pre-enrollment quarters, there would have been a fortysix percentage point increase in the Q0 employment rate to 84.3%. By contrast, if the only difference in participant characteristics had been being male rather than female, the estimated employment rate for the enrollment quarter would have been nearly five points lower. The estimated employment rate was also 0.4 percentage points less for every one year older a participant was at enrollment, but a substantial 17.6 percentage points more for participants with an affective/mental disability (relative to the reference physical disability group). Somewhat surprising were lower estimated employment rates for individuals with more educational attainment, about three percentage points lower at project entry for individuals with a high school education and over six percentage points lower for individuals with at least some post-secondary education. Similar to the earnings model, the employment model predicted a larger growth in employment rates during the post-enrollment quarters if the participant were included in either the medium or high benefits counseling dosage groups for their project. Those who received a dosage within one half standard deviation of their project mean exhibited an estimated 1.1 percentage point gain in employment rates each quarter, whereas those who received a dosage at least one half standard deviation above their project mean had an estimated quarterly increase of almost two and a quarter percentage points in their group’s employment rate. Those who received dosages of benefits counseling less than one half standard deviation below their project mean faced an estimated decline in employment rates of 0.2 percentage points per calendar quarter 24 over the Q0-Q8 period. Thus by Q8, the regression estimated employment rate for the high dosage group had grown almost twenty percentage points relative to the estimated rate for the low dosage group. The final model utilizing the samples from both projects estimates the income proxy variable. Overall results for this model are again very similar to those for the earnings model. Still, there are differences of note. First, those with affective/mental disabilities had income levels at enrollment that were not significantly different from those for the physical/HIV/sensory disabilities group. Recall that in the earnings model the beta coefficient for the affective/mental disability group was significantly higher than that imputed for the “PD” reference group. Second, education did predict differences in income, whereas it did not predict differences in earnings. Unlike in the other two models, the higher an individual’s educational attainment category, the higher her estimated income. Third, in the income model, older participants have higher estimated incomes. By contrast, the earnings model predicts that older participants will have lower earnings than younger ones. Finally, as one would expect, income estimates were higher than earnings estimates. The predicted estimates and p-values for these estimates are displayed in table 7. Table 7: Estimates of Effects on UI Mean Income for Combined Samples Estimates of Effects in Mixed Linear Income Model (n = 911) Parameter Beta Std. Error P-Value Intercept 1425.7 191.39 < 0.0001 Quarter -26.4 22.72 0.2456 Benefits Counseling 52.2 60.52 0.3883 PreEarnings 1456.9 241.60 < 0.0001 Quarter * Benefits Counseling 36.7 11.54 0.0015 Quarter * PreEarnings 4.3 45.30 0.9252 Benefits Counseling * PreEarnings 111.6 116.70 0.3392 Quarter * Benefits Counseling * PreEarnings -36.3 22.04 0.0995 Gender (reference = female) 364.5 53.30 < 0.0001 Age 15.1 2.46 < 0.0001 Education 314.8 37.17 < 0.0001 Cognitive/Developmental Disability -680.1 74.63 < 0.0001 Affective/Mental Disability -1.1 56.76 0.9844 Other Disability -997.4 227.58 < 0.0001 As in the earnings model, these beta estimates can be interpreted as changes in dollar amounts. For participants who are female, have either physical/HIV/sensory or mental disability, were age seventeen at enrollment, and had earnings of less than $1200 in three or more of their pre-enrollment quarters, the Q0 estimate for mean income is just over $1997. For participants with at least two pre-enrollment quarters with at least $1200 in earnings, estimated income for the enrollment quarter was $3454; that is, $1457 above that for participants with our “reference” characteristics. Being male increased the Q0 estimate by $365 compared to participants who were otherwise the same except being female. Each additional year of age at enrollment was associated with an additional $15 increase in estimated income. Education also increased enrollment quarter income, with an estimated addition of $315 for those participants who ended their education with a high school diploma (or equivalent) and $630 for those who 25 had any education beyond that. Individuals with cognitive disabilities had an estimated $680 less in enrollment quarter income compared to the average for the physical/HIV/sensory disability reference group. Like quarterly mean earnings and employment rates, the rate at which income estimates changed over the eight post-enrollment quarters depended on the amount of benefits counseling the participant received, relative to others in their project, during this period. For those individuals who received dosages of benefits counseling that did not even reach within a half standard deviation of their project mean, income growth appears to have averaged less than $37 per quarter. Quarterly growth was higher for individuals who received an “average” amount of benefits counseling (within one half standard deviation of their project mean); the rate of increase is estimated at $73 per quarter. The highest rate of increase was exhibited by those individuals who received at least one half standard deviation of service over their project mean. Their incomes grew at an estimated $110 per quarter. Single Sample Results The mixed linear models for the combined samples provide support for the hypothesis that higher dosages of benefits counseling services result in better employment outcomes. This influence remains even when participant earnings in the four quarters prior to the enrollment quarter are taken into account. Nonetheless, there was a large difference in the typical amount of service provided to participants in the two projects. We now examine the WI SPI and SSDI-EP samples independently to assess whether there are comparable relationships between relative dosage and the employment outcomes of interest. Additionally, we look at whether the “pre-earnings” and demographic control variables exhibit similar patterns of influence as observed for the combined samples regression models. As discussed in the methods section, we perform these analyses using repeated measures MANOVA because of the smaller sample sizes. Although we had to sacrifice the beta estimates, we are still able to look at relationships both across time (within participants) and across (between) participants. MANOVA statistics provide both significance levels and measures of variance (partial eta-squared) explained by each independent variable. Though pertinent results are discussed in the text below, tables providing a fuller set of MANOVA statistics are available in Appendix B. We also provide a series of figures displaying the estimated mean values of an employment outcome associated with each of the three benefits counseling dosage categories when the effects of the other predictor variables are accounted for. We supplement the figures with a series of tables containing estimates of the change in an employment outcome over the Q0-Q8 study period for each of the MANOVA models displayed below. These tables are found in Appendix C. In both the repeated measures MANOVA earnings models for the WI SPI and SSDI-EP samples, there is a significant between subject relationship between benefits counseling and earnings, though movement between dosage categories accounts for only 2% of the variance in earnings. By contrast, pre-enrollment earnings accounts for about 12% of the variance in earnings. It is of note that none of the demographic variables are significant predictors of earnings for the WI SPI sample, but gender and age are significant predictors of earnings for those participating in the SSDI-EP. In this model, gender accounts for 4% of the within subject variance in earnings and age 26 accounts for 1% of the between subject variance. The following two graphs (figures 4 and 5) provide visual representations of the relationship between the benefits counseling received and post-enrollment earnings for, respectively, the WI SPI and SSDI-EP projects. Figure 4: Estimated Quarterly Mean UI Earnings (Inflation Adjusted) by Benefits Counseling Dosage Category, Q0-Q8 for WI SPI Sample Estimated Quarterly Earnings Means during the WI SPI Study Half SD BC Category Less than 1/2 SD of BC Mean (n = 139) Within 1/2 SD of BC Mean (n = 214) Greater than 1/2 SD of BC Mean (n = 84) 3000 Estimated Earnings Means 2750 2500 2250 2000 1750 1500 1250 1000 750 500 250 0 0 1 2 3 4 5 6 Quarter Compared to Enrollment 7 8 27 Figure 5: Estimated Quarterly Mean UI Earnings (Inflation Adjusted) by Benefits Counseling Dosage Category, Q0-Q8 for SSDI Sample Estimated Quarterly Earnings Means during the WI SSDI EP Study Half SD BC Category Less than 1/2 SD of BC Mean (n = 177) Within 1/2 SD of BC Mean (n = 172) Greater than 1/2 SD of BC Mean (n = 94) 3000 Estimated Earnings Means 2750 2500 2250 2000 1750 1500 1250 1000 750 500 250 0 0 1 2 3 4 5 6 7 8 Quarter Compared to Enrollment There is a little less consistency in the relationship between the benefits counseling dosage categories and either the employment rate or income outcomes. In the employment models, this relationship is not statistically significant for the WI SPI sample, but is for the SSDI-EP sample. The opposite is true for the income models. This time the relationship is significant for the WI SPI participants, but not for those in the SSDI-EP. When the relationship is significant, the between subjects effects of benefits counseling continues to explain about 2% of the variance in employment outcomes. Meanwhile, the dichotomous pre-enrollment earnings variable explains between 20% and 23% of the variance in employment rates and either 7% or 8% of the variance in the income variable. In both employment rate models, education accounts for about 4% of the variance in within subject employment rates. Other demographic variables were significant for only one of the samples. Disability type accounts for about 3% of the between subject variance in the employment rate for those enrolled in WI SPI, whereas, for SSDI-EP participants, age proves a significant factor accounting for about 2% of the between subject variance in rates. The pattern of relationships between the demographic control variables and estimated income shows even greater disparity than that observed for the employment rate models. In the income model for the WI SPI sample, age and disability category each accounts for approximately 1% of between subject variance. For those in the SSDI-EP, gender accounts for a statistically significant amount of both the within and between subject variance (4% and 1%, respectively). 28 Finally, education explained about 2% of the between subject variance in the income models for both samples. The following four figures provide visual representations of the relationships between the benefits counseling dosage categories and both employment rates and mean income. Separate graphs are provided for each combination of project and employment outcome. Figure 6: Estimated Quarterly UI Employment Rates by Benefits Counseling Dosage Category, Q0-Q8 for WI SPI Sample Estimated Quarterly Employment Rates during the WI SPI Study HalfSDBCSplit Less than 1/2 SD of BC Mean (n = 139) Within 1/2 SD of BC Mean (n = 214) Greater than 1/2 SD of BC Mean (n = 84) Estimated Employment Rates 1 0.8 0.6 0.4 0.2 0 0 1 2 3 4 5 6 Quarter Compared to Enrollment 7 8 29 Figure 7: Estimated Quarterly UI Employment Rates by Benefits Counseling Dosage Category, Q0-Q8 for SSDI-EP Sample Estimated Quarterly Employment Rates during the WI SSDI EP Study HalfSDBCSplit Less than 1/2 SD of BC Mean (n =177) Within 1/2 SD of BC Mean (n =172) Greater than 1/2 SD of BC Mean (n = 94) Estimated Employment Rates 1 0.8 0.6 0.4 0.2 0 0 1 2 3 4 5 6 Quarter Compared to Enrollment 7 8 30 Figure 8: Estimated Quarterly Mean Income (Inflation Adjusted) by Benefits Counseling Dosage Category, Q0-Q8 for WI SPI sample Estimated Quarterly Income Means during WI SPI Study 5000 HalfSDBCSplit Less than 1/2 SD of BC Mean (n = 139) Within 1/2 SD of BC Mean (n = 214) Greater than 1/2 SD of BC Mean (n = 84) 4500 Estimated Income Means 4000 3500 3000 2500 2000 1500 1000 500 0 0 1 2 3 4 5 6 Quarter Compared to Enrollment 7 8 31 Figure 9: Estimated Quarterly Mean Income (Inflation Adjusted) by Benefits Counseling Dosage Category, Q0-Q8 for SSDI-EP sample Estimated Quarterly Income Means during the WI SSDI EP Study 5000 HalfSDBCSplit Less than 1/2 SD of BC Mean (n = 177) Within 1/2 SD of BC Mean (n = 172) Greater than 1/2 SD of BC Mean (n = 94) 4500 Estimated Income Means 4000 3500 3000 2500 2000 1500 1000 500 0 0 1 2 3 4 5 6 7 8 Quarter Compared to Enrollment Single Sample Results for Participant Subgroups Reflecting Selected Pre-Enrollment Employment Outcomes The next analyses focus on two subgroups within the WI SPI and SSDI-EP samples that were chosen to exemplify either relatively strong or weak employment outcomes prior to entering one of the projects. These analyses focus exclusively on earnings. Additionally, these analyses utilize the same demographic control variables used in the MANOVA models already presented. In point of fact, these results are nothing more than the estimated means for those participants in each sample who also meet the subgroup criterion. The first subgroup is composed of persons with relatively strong employment outcomes. Inclusion in this group is defined by having at least two quarters in the four quarters prior to the calendar quarter of project entry with UI earnings of at least $1200. 41 The estimates for mean earnings trends for the WI SPI and SSDI-EP participants with relatively high earnings in the “pre-enrollment” period are shown in figures 10 and 11. 41 This is the higher value of the “pre-earnings” variable used in the full sample MANOVA analyses. Thus, the MANOVA models are identical to those used to estimate quarterly mean earnings for the full WI SPI and SSDI-EP samples. 32 In the models for both samples, participants identified as having relatively high earnings before enrollment are also significantly more likely to have higher earnings during the post-enrollment period than those who did not meet this criterion. This difference in assignment to one of the pre-enrollment earnings categories accounted for about 12% of the variance in post-enrollment earnings. To determine whether the relationship between benefits counseling and earnings is different for these higher earners, the p-value of the interaction between benefits counseling and pre-enrollment earnings is examined. For both studies, this interaction is not statistically significant (p > 0.05). Thus, the relationship between benefits counseling dosage and earnings is not significantly different for high earners in the year before enrollment compared to the overall relationship for all the participants in each study. The following two graphs provide a visual representation of this relationship for the high earners group. The trends displayed can be directly compared to the ones shown for the full WI SPI and SSDI-EP samples in, respectively, figures 4 and 5. Figure 10: Estimated Quarterly Mean UI Earnings (Inflation Adjusted) for High Earners in Pre-Enrollment Period by Benefits Counseling Dosage Category, Q0-Q8 for WI SPI Sample Estimated Quarterly Earnings Means during WI SPI Study For Participants with At Least Two Pre-Enrollment Quarters with at Least $1200 HalfSDBCSplit Less than 1/2 SD of BC Mean (n = 30) Within 1/2 SD of BC Mean (n = 40) Greater than 1/2 SD of BC Mean (n = 26) 3000 Estimated Earnings Means 2750 2500 2250 2000 1750 1500 1250 1000 750 500 250 0 0 1 2 3 4 5 6 Quarter Compared to Enrollment 7 8 33 Figure 11: Estimated Quarterly Mean UI Earnings (Inflation Adjusted) for High Earners in Pre-Enrollment Period by Benefits Counseling Dosage Category, Q0-Q8 for SSDI-EP Sample Estimated Quarterly Earnings Means during the WI SSDI EP Study for Participants with At Least Two Pre-Enrollment Quarters with at Least $1200 HalfSDBCSplit Less than 1/2 SD of BC Mean (n = 41) Within 1/2 SD of BC Mean (n = 54) Greater than 1/2 SD of BC Mean (n =33) 3000 Estimated Earnings Means 2750 2500 2250 2000 1750 1500 1250 1000 750 500 250 0 0 1 2 3 4 5 6 7 8 Quarter Compared to Enrollment In fact, it is hard not to notice the visual overlap in earnings across the different dosage categories of benefits counseling that participants in the higher pre-enrollment earners group received, especially in the results for the SSDI-EP sample. For that sample, the overlap is especially apparent for participants who received a dosage within one half standard deviation of the project mean and those who received higher dosages. When looking at the relationship between benefits counseling and earnings for all SSDIEP participants (figure 5), there is still some, but not nearly as much, overlap. This raises the question of whether the relationship between benefits counseling dosage and earnings may be stronger for those with lower earnings in the year prior to project entry. The earnings trends exhibited in figure 12 are consistent with this interpretation. There appears to be more differentiation between those in different dosage categories. Receiving higher dosages is associated with higher earnings levels (though less clearly with earnings growth). What one does not see for lower pre-enrollment earners group is a decreasing trend in estimated earning for those who received “below average’ amounts of benefits counseling (i.e., less than ½ SD below the project mean). By contrast, the decreasing trend was present, beginning in Q3, for high pre-enrollment earners who received dosages that did not even reach one half standard deviation beneath the project mean. Still, our overall assessment is that, even in the SSDI-EP, the amount of benefits counseling received positively influenced post-enrollment earnings for participants included in both the low and high pre-enrollment earnings groups. 34 Figure 12: Estimated Quarterly Mean UI Earnings (Inflation Adjusted) for Low Earners in Pre-Enrollment Period by Benefits Counseling Dosage Category, Q0-Q8 for SSDI-EP Sample Estimated Quarterly Earnings Means during the SSDI EP Study for Participants with Less than Two Pre-Enrollment Quarters with at Least $1200 HalfSDBCSplit Less than 1/2 SD of BC Mean (n = 136) Within 1/2 SD of BC Mean (n = 118) Greater than 1/2 SD of BC Mean (n = 61) 3000 Estimated Earnings Means 2750 2500 2250 2000 1750 1500 1250 1000 750 500 250 0 0 1 2 3 4 5 6 7 8 Quarter Compared to Enrollment The second subgroup of interest is composed of participants who had not participated in the labor market between entering benefit status and enrolling in either WI SPI or the SSDI-EP. Inclusion in this group is based largely on participant self-report, though in a few cases persons were removed from this group based on having UI earnings in one of the four pre-enrollment quarters. The MANOVA earnings model is identical to the one used for the high pre-enrollment earners subgroup with one exception. The “pre-earners” variable is replaced by a dichotomous indicator of reported employment (EmpSinceDis) at any point in time between entering a Social Security disability program and entry to either WI SPI or the SSDI-EP. 42 Estimated mean earnings for this subgroup are displayed in figures 13 and 14. For WI SPI participants who did not report a span of employment after entering benefit status, benefits counseling dosage is significantly related to earnings in the Q0-Q8 period. The dosage category accounts for about 2% of the between subject variance. However, this relationship is not significant for SSDI-EP participants, with a between subject p-value of roughly 0.2. However, having no reported employment after entering Social Security benefit status is both significantly and negatively related to earnings. Moreover, the variable accounts for the largest amount of between subject variance in these earnings models, ranging from 4% in the SSDI-EP model to 8% in the WI SPI model. No demographic variables are significantly related to earnings in the model for the WI SPI sample, but gender does account for around 4% in the within subject 42 The pre-enrollment earnings variable is removed from this model because of a concern that including both pre-enrollment employment related variables would cancel each other’s effects. 35 earnings variance and age accounts for almost 1% of the between subject variance in the model results for the SSDI-EP sample. Figure 13: Estimated Quarterly Mean UI Earnings (Inflation Adjusted) for Those without Employment after Entering Benefit Status by Benefits Counseling Dosage Category, Q0Q8 for WI SPI Sample Estimated Quarterly Earnings Means during the WI SPI Study for Participants who at Enrollment Had Not Been Employed since Receiving Social Security Benefits HalfSDBCSplit Less than 1/2 SD of BC Mean (n = 40) Within 1/2 SD of BC Mean (n = 70) Greater than 1/2 SD of BC Mean (n = 23) 3000 Estimated Earnings Means 2750 2500 2250 2000 1750 1500 1250 1000 750 500 250 0 0 1 2 3 4 5 6 Quarter Compared to Enrollment 7 8 36 Figure 14: Estimated Quarterly Mean UI Earnings (Inflation Adjusted) for Those without Employment after Entering Benefit Status by Benefits Counseling Dosage Category, Q0Q8 for SSDI-EP Sample Estimated Quarterly Earnings Means during the WI SSDI EP Study for Participants who at Enrollment Had Not Been Employed since Receiving Social Security Benefits HalfSDBCSplit Less than 1/2 SD of BC Mean (n = 43) Within 1/2 SD of BC Mean (n = 27) Greater than 1/2 SD of BC Mean (n = 12) 3000 Estimated Earnings Means 2750 2500 2250 2000 1750 1500 1250 1000 750 500 250 0 0 1 2 3 4 5 6 7 8 Quarter Compared to Enrollment To determine whether the relationship between benefits counseling dosage is different for those individuals who were not employed since receiving Social Security benefits than for other WI SPI and SSDI-EP participants, the interaction between benefits counseling and having employment since receiving social security benefits was examined. No statistically significant interaction exists in the models for either sample. In other words, there is no evidence that the relationship between benefits counseling and earnings was any different for those who were not employed since receiving Social Security disability benefits and enrolling in one of the studies. Finally, we compared the estimated earnings trends for those who had reported having been employed since entering benefit status to those for the subgroup who had not reported any span of employment since going on benefits. Though we do not display results for those participants, in general the earnings differences related to different amounts of benefits counseling is more pronounced. Findings: Mediating Effects of Work Incentives As suggested by the MIG-RATS Abbreviated Employment and Earnings Model (see figure 2) one explanation for the positive relationship between benefits counseling and earnings is the increased use of work incentives. Specifically, receiving greater amounts of benefits counseling may increase the likelihood that an individual in a Social Security disability programs will use available work incentives. In turn, this increased 37 use of work incentives is hypothesized to increase earnings and (hopefully) income. In other words, work incentives may mediate the relationship between benefits counseling and earnings. The following diagram demonstrates this relationship. Figure 15: Work Incentive Use as a Mediator between Benefits Counseling Use and Earnings Use of Work Incentives (Mediator) b a Benefits Counseling (Independent Variable) c Earnings (Dependent Variable) In order to determine whether the use of work incentives actually mediates the relationship between benefits counseling and earnings, four things must be true. First, there must be a positive relationship between benefits counseling and the use of work incentives (a in the diagram). Second, there must be a positive relationship between the use of work incentives and earnings (b). Third, there must be a positive relationship between benefits counseling and earnings (c). Finally, when both benefits counseling the use of work incentives are included as predictors of earnings in a single model, the relationship between benefits counseling and earnings should be reduced. 43 The presence of the first three relationships (a, b, and c) is first examined using Pearson correlations. If three significant and positive relationships are identified, then a two step linear regression model is used to determine whether the use of work incentives reduces the relationship between benefits counseling and earnings. In this two step model, the first model includes only benefits counseling and pre-enrollment earnings as the predictors of post-enrollment earnings, whereas the second model included benefits counseling, pre-enrollment earnings, and the use of work incentives as the predictors of later earnings. Again, the benefits counseling dosage variable has three ordinal categories based on whether dosage is within one half standard deviation of the mean dosage for the project in which the participant was enrolled. The variable used to indicate the level of pre-enrollment earnings also remains the same, dividing participants based on having earnings of at least $1200 in at least two of the four preenrollment quarters. However, the earnings variable is different. Because the use of a work incentive did not necessarily coincide with either project enrollment or the full Q0Q8 study period, we designate quarter eight UI earnings as the dependent variable. These analyses were limited due to differences between the projects in eligibility requirements, available work incentives, and data availability. To some extent we rely on proxy variables to indicate the availability, though not necessarily the use, of a given 43 Howell, D.C. (2002) Statistical Methods for Psychology: Fifth Edition. Pacific Grove, CA: Duxbury pp. 547-77. 38 type of work incentive. As a consequence, analyses are conducted separately for the WI SPI and the SSDI-EP samples. For the SSDI-EP, the work incentives examined include trial work period (TWP) completion during the Q0-Q8 post-enrollment period, Medicaid Buy-in participation during the same period, whether the study participant was randomly assigned to the treatment group (thus having the opportunity to use the benefit offset being tested), and actual offset use. For the WI SPI study, the work incentive examined includes the SSI waiver. 44 Although not a work incentive, we also examine whether SSDI eligibility (including having concurrent SSI eligibility) mediates the relationship between benefits counseling and earnings. Having access to SSDI benefits also gives one access to specific work incentives such as the TWP and EPE. Only TWP completion and the closely related use of the SSDI benefit offset provision mediates the relationship between benefits counseling and earnings. Though the TWP work incentive was available to WI SPI participants who were SSDI beneficiaries, the relationships described below can only be ascribed to the SSDI-EP participant sample. 45 There is a significant and positive correlation between the amount of benefits counseling and TWP completion (r = 0.145; p = 0.001). There is also a significant correlation between TWP completion and earnings (r = 0.357; p < 0.001). Additionally, the third condition of having a significant correlation between the amount of benefits counseling received and earnings (r = 0.167; p < 0.001) is also achieved. Thus the conditions for using a linear regression model to confirm the mediating effect are met. When both the amount of benefits counseling and TWP completion variables are included in the same linear regression model as predictors of quarter eight earnings, the relationship between benefits counseling and earnings is reduced, with the p-value increasing from 0.002 to 0.034. The following table compares the results for the models omitting and including the TWP completion variable. Table 8: Linear Model Comparison to Check if TWP Completion Mediates the Relationship between Benefits Counseling and Earnings Model Beta Std. Error P-Value 1 (Constant) 230.8 245.61 0.3479 Half SD BC 388.2 125.98 0.0022 Pre-Enrollment Earnings 1197.0 213.71 < 0.0001 2 (Constant) 85.4 231.4 0.7123 Half SD BC 254.2 119.52 0.0340 Pre-Enrollment Earnings 1157.4 200.78 < 0.0001 TWP Completion Post-Enrollment 1687.2 212.34 < 0.001 Study WI SSDI EP Participant; n = 468; Dependent Variable = Quarter 8 Earnings 44 The SSI waiver did not go into effect until approximately a year and a half following the first enrollments to the project. Though eligible participants had to explicitly sign-up for this waiver, the standard 1619 provision would have applied as a matter of course. The SSI waiver involved a decreased loss in the SSI benefit for each extra dollar of earnings compared to the existing 1619 provision. As application thresholds were identical, it is uncertain how much of the SSI waiver’s effect on earnings can be directly attributed to the waiver. 45 Though TWP completion information was collected during WI SPI, it is not included in the deidentified data set for that project. 39 Adding TWP completion to the SSDI-EP sample model also increases the amount of variance in Q8 that is accounted for. The r 2 value for the model without the TWP completion variable is 0.089. Thus 9% of the earnings variance is accounted for. By contrast, the r 2 value for the model including TWP completion is 0.198 (accounting for 20% of variance in earnings). This means that, for SSDI-EP participants, the amount of benefits counseling an individual received during the study predicted TWP completion during the Q0-Q8 post-enrollment period. In turn, TWP completion during this period predicted Q8 earnings. In other words, TWP completion during the post-enrollment period mediates the relationship between the categorical amount of benefits counseling a participant received and Q8 earnings. We also examined the mediating effect of benefit offset usage by SSDI-EP participants (see table 9). While offset use required TWP completion and earnings above SGA for the months it was applied, it is also true that the date of TWP completion could have occurred after Q8 and the offset might not have been applied to the SSDI cash benefit in any of the months contained in the Q0-Q8 period. Still, we observe a significant relationship between the amount of benefits counseling received and offset use (r = 0.179; p < 0.001). There is a significant relationship between offset use and earnings (r = 0.370; p < 0.001). There is a significant relationship between the amount of benefits counseling received and earnings (r = 0.167; p < 0.001). 46 Finally, including offset use along with the amount of benefits counseling received into the linear regression earnings model reduced the relationship between benefits counseling and earnings, with the p-value increasing from 0.002 to 0.042. 47 Table 9: Linear Model Comparison to Check if Benefit Offset Use Mediates the Relationship between Benefits Counseling and Earnings Model Beta Std. Error P-Value 1 (Constant) 230.8 245.61 0.3479 Half SD BC 388.2 125.98 0.0022 Pre-Enrollment Earnings 1197.0 213.71 < 0.0001 2 (Constant) 309.3 232.99 0.1850 Half SD BC 246.2 120.94 0.0424 Pre-Enrollment Earnings 935.2 205.64 < 0.0001 SSDI Benefit Offset Use 2112.1 287.95 < 0.001 Study WI SSDI EP Participant; n = 468; Dependent Variable = Quarter 8 Earnings Just as including TWP use increases the amount of variance in earnings that a model accounts for, so too does inclusion of offset use. When offset use is not included in the model, the r 2 value was 0.089, accounting for 9% of the variance in earnings. When offset use was included in the model, the r 2 value increases to 0.184, now accounting for 18% of the variance. Like TWP completion, benefit offset use mediates the relationship between the amount of benefits counseling a participant received and Q8 earnings. 46 47 As with TWP completion, we start the analysis of mediating effects with bivariate correlations. Also note the reduction in the size of the beta coefficient for the dosage category from 388 to 246. 40 Findings: Determinants of Benefits Counseling Dosage In our brief review of studies about the impact of benefits counseling we noted that the growth in employment outcomes tends to be concentrated in the period shortly after the service was initiated. While it is possible this reflects how the intervention works, it is also possible that such rapid gains reflect conditions already in place when an individual first receives the service. We use data from our samples to explore this question, though we stress that caution should be exercised in applying our findings beyond the immediate context of these demonstration projects. In these analyses, the total hours of benefits counseling provided over the Q0-Q8 period (SumBCQ0Q8) is the dependent variable. We use OLS linear regression to estimate total dosage as we are not interested in identifying dosage trends within the Q0-Q8 period. 48 Our analysis concentrates on employment outcomes prior to entry to either the WI SPI or the SSDI-EP as we hypothesized that persons already having relatively good employment outcomes would be well positioned to use benefits counseling to further improve their outcomes. To do this, we use the same variables we use to identify either participants with unusually strong earnings prior to study entry or those who reported having no employment after going on Social Security benefits. These variables are labeled, respectively, as PreEarn1200Cat2 and EmpSinceDis. Additionally, we ran models using our income proxy for the calendar quarter immediately prior to study entry (Incqn1), as we thought it possible that participants with higher incomes might feel a greater need for services such as benefits counseling as they approached earnings or income thresholds that might threaten continued eligibility for disability programs. 49 It is important to remember that relatively high income does not necessarily indicate relatively high earnings. For SSDI beneficiaries, it often reflects a higher than average benefit level. All of the models estimating total hours of benefits counseling include the following control variables: age, gender, disability type and education. Models using the same set of control variables irrespective of project are referred to as “common.” Additionally, we ran models with variables that were specific to one of the studies. For the SSDI-EP sample we added variables that indicated whether a participant had received benefits counseling before entering the project (PreStudyBC) and which study group the participant had been assigned to (ssdieptx). For WI SPI, we added the variable (SSDI) that distinguished participants who had some form of SSDI eligibility from those who were SSI only. Table 10 summarizes the results for both the common and study specific models after non-significant variables were removed. 48 Benefits counseling tended to be front loaded in both projects (see table 1). For the SSDI-EP, quarterly means for the period after Q1 are too small to support analysis. Additionally, in the case of the SSDI-EP, benefits counselors reported that a substantial proportion of later benefits counseling was associated with problems in processing post TWP work reviews and offset adjustments for those in the treatment group. Delin et al. (2010) pp. 116-18 and 120-24. 49 We examined the impacts of several other indicators of employment related outcomes prior to project enrollment. Results proved highly consistent with those presented here. 41 Table10: Selected Regression Results for Predictors of Total Benefits Counseling Hours Q0-Q8 WI SPI SSDI-EP Variables Beta Variables Beta Significant at .05 Significant at .05 Level Level 18.242 “PreEarn1200Cat2” 3.246 Common Model “PreEarn1200Cat2” for “CognitiveDD” 14.503 “CognitiveDD” -6.538 “PreEarn1200Cat2 “AffectiveMH” -14.131 Common Model for “EmpSinceDis” “CognitiveDD” “AffectiveMH” 16.340 -10.344 “EmpSinceDis” “CognitiveDD” 3.024 -6.225 Common Model for “Incqn1” “Incqn1” 0.004 (per $1 increase) 17.367 -11.110 “Incqn1” “CognitiveDD” “Age” 0.001 (per $1 increase) -5.997 -0.139 18.242 14.503 -14.131 “PreEarn1200Cat2” “CognitiveDD” “PreStudyBC” 3.073 -6.602 3.982 18.772 -10.363 -9.730 “EmpSinceDis” “CognitiveDD” “PreStudyBC” 3.219 -5.787 4.066 “CognitiveDD” “AffectiveMH” Study Specific “PreEarn1200Cat2” Model for “CognitiveDD” “PreEarn1200Cat2 “AffectiveMH” Study Specific Model for “EmpSinceDis” “CognitiveDD” “AffectiveMH” “SSDI” 0.004 “CognitiveDD” -6.093 (per $1 increase) “CognitiveDD” 17.367 “PreStudyBC” 3.970 “AffectiveMH” -11.110 “Age” -0.127 Source: WI SPI and SSDI-EP Encounter Data, Administrative Data from WI Department of Workforce Development and the Social Security Administration Study Specific Model for “Incqn1” “Incqn1” The common models suggest that having relatively high earnings for at least half the quarters in the year prior to study entry substantially increases the amount of benefits counseling one receives. The beta coefficient for WI SPI indicates that on average these higher earning participants would receive about eighteen more hours of benefits counseling than other participants. This increment of service is about 47% of the mean service hours for the full WI SPI sample. Though the beta coefficient for SSDI-EP “common” model is much smaller in absolute size, it implies a considerable proportion of additional benefits counseling (39%) relative to the average dosage. The comparable betas for the study specific models are either identical or marginally different. By contrast, the EmpSinceDis variable is statistically significant for the SSDI-EP sample, but not for the WI SPI sample. For SSDI-EP participants the results are very similar to those for the high earners group. By contrast, the regression coefficients for the WI SPI models never reach a p-value of 0.1 and are excluded from the model. We 42 do not understand why a variable that indicates whether there had been any employment history is non-significant for the WI SPI sample, though the control variables we discuss below may provide some insight into the matter. The results for the income variable are mixed, but generally indicate that those with higher income before study entry received more benefits counseling over the Q0-Q8 period. Results for both of the WI SPI model and for the common model for the SSDI-EP are statistically significant with betas of .004 for the WI SPI models and .001 for the SSDI-EP model. 50 Though these coefficients appear to be very small, it is because they measure the estimated increase of benefits counseling dosage associated with a one dollar increase in the income variable. For example, a beta of .001 implies that if a SSDIEP entrant had a Q-1 income $1,000 greater than another study entrant, the former would be expected to be provided with one more hour of benefits counseling over the Q0-Q8 period; i.e., about 12% of the sample average. Among our control variables, one or more of the disability type variables prove statistically significant in every model. Results for these variables are relative to those of the reference category of physical disability. While in all models being categorized as having a cognitive/developmental disability has a statistically significant effect on benefits counseling dosage, the effects do not run in a consistent direction. In WI SPI, this disability category is associated with receiving substantially more benefits counseling than those in the reference category. By contrast, in the SSDI-EP, those in the cognitive/developmental disability category typically received substantially less benefits counseling than those in other disability groups. We have not followed up on this finding, though we believe it may be related to differences in the eligibility rules and the service populations associated with the specific groups of community agencies that enrolled participants in each of the projects. 51 Such differences may also help explain why those in the affective/mental health category received far less service in the WI SPI project, but comparable levels to the reference group in the SSDI-EP. With one exception, we do not discuss other control variables. That exception is benefits counseling prior to study entry (PreStudyBC). That variable proves to have a strong effect on dosage over the Q0-Q8 period in the SSDI-EP study specific models. The coefficients are consistent across the three models and suggest that having prior benefits counseling increases dosage approximately four hours (that is, nearly half of the mean dosage). It is possible that consumers receiving benefits counseling services develop what economists call a taste for the service. It is also possible that prior receipt helped to motivate the better pre-enrollment employment outcomes we find associated with dosage during the Q0-Q8 period. The data available to us does not support further examination of this issue. Nonetheless, though these models are always significant at a 0.00 p-value, none of the models explain very much of the variance. In no case is the r 2 much greater than 0.1, suggesting that the models are either incompletely specified or that there is a great 50 For the SSDI-EP study specific model the Incqn1 variable was nearly significant (p-value = 0.069). The beta coefficient was again 0.001. 51 In WI SPI, each provider agency was restricted to enrolling consumers from a single disability group. A substantial proportion of participants were enrolled at agencies restricted to enrolling those defined as having a developmental disability. 43 deal of individual level variation. One important factor might be work motivation. Unfortunately, the survey data still available for both studies provides only modest resources to explore this possibility. We do not give much credence to responses to questions about whether a severely disabled person would prefer to work if possible as there is little correlation with actual labor force participation. However, there was one survey item available for both studies that might be a reasonably credible indicator of motivation, the extent that, on a five point scale, a respondent agreed that she had a career plan. We added the career plan variable to the three “common” models run for both studies. The results are mixed. For the SPI sample, the variable is always statistically significant with a positive but relatively small coefficient. Including the variable makes a slight improvement in the model’s r 2 value. For the SSDI-EP, having a career plan was never statistically significant and the variable’s inclusion resulted in a slightly smaller r 2 . Conclusion Researchers have established a reasonably strong case that better employment outcomes for persons with severe disabilities, especially earnings, are positively associated with the delivery of work incentive benefits counseling. The relevant service population is defined not only by disability in a medical or functional sense, but also by attachment to public programs that provide income support, health care, or other services as a consequence of meeting disability related program eligibility requirements. Our aim has been to better understand this association, especially whether the amount of benefits counseling services provided, that is dosage, is positively related to better employment outcomes. The descriptive data from the SSDI-EP project (where about 21% of participants did not get any benefits counseling in the Q0-Q8 period) are basically consistent with those from other studies. Consumers who receive benefits counseling have better employment outcomes. More to the point, when we look at the association between dosage and employment outcomes we see clearly positive relationships. Figure 1 in this report provides an example (from the SSDI-EP) for the positive association of benefits counseling dosage with subsequent earnings levels and growth. Table 11 displays information suggesting a strong association between dosage levels and income growth over the Q0-Q8 period in both the WI SPI and SSDI-EP projects. Indeed, we would argue that from an individual’s perspective, income growth should be expected to be more important than earnings growth. 52 52 In saying this, we are not suggesting that many individuals might not place more value on a dollar of income achieved through employment than one obtained from some other source. 44 Table 11: Repeated Measures MANOVA Estimated Mean Income by Received Amounts of Benefits Counseling Study Amount of Benefits Q0 Q8 Q8 – % Counseling Q0 Change < ½ SD BC $2672 $2866 $194 7.3% WI SPI (n =437) Within ½ SD BC $2850 $3284 $434 15.2% > ½ SD BC $3008 $3698 $690 22.9% < ½ SD BC $3356 $3378 $22 0.6% SSDI-EP (n =443) Within ½ SD BC $3512 $3879 $367 10.5% > ½ SD BC $3550 $4192 $642 18.0% However, our purpose is also to explore whether the positive relationship between the amount of benefits counseling and employment outcomes would remain after controlling for other factors, such as indicators of previous employment outcomes and personal characteristics such as age, gender, educational attainment, and broadly defined disability categories. Ideally, we would have liked to have reported dosage impacts in a standard form such as “one additional hour of benefits counseling leads, on average, to x dollars difference in earnings.” As WI SPI participants typically received many more service hours than those in the SSDI-EP, we decided to utilize dosage categories relative to the mean dosage for each of the projects. Thus, results must always be interpreted as dosage relative to that which was typical for project participants. While we hope that other research efforts will be implemented that will more directly measure the impact of a unit of benefits counseling services, as we explain below, we are not sure that this is the best way to conceptualize how the service impacts employment outcomes. Though conclusions based on data from two relatively small projects located in a single state cannot fully settle the issue, our results suggest that receiving higher dosages of benefits counseling services leads to better employment outcomes. This is unequivocally true for UI earnings. The results for the combined samples model suggest that, after controlling for other factors, quarterly earnings at the end the study period for those in the medium dosage category increased about $275 relative to those for participants in the low dosage category. On average, Q8 earnings for those in the high dosage category grew about $550 relative to those for participants in the low dosage category. Our results are less consistent for other employment outcomes, though higher dosages are usually associated with higher probabilities of employment and higher mean income. Though higher dosage levels are generally associated with better outcomes, they are not as consistently associated with outcome change rates over the study period. Much the same could be said for our subgroup analyses of the earnings trends for relatively high earners in the year prior to study enrollment and those who had reported no workforce participation in the period between going on Social Security benefits and enrolling in one of the projects. The basic patterns are discernable, but without clear differentiation of earnings trends across the dosage categories. One of the claims made for the efficacy of benefits counseling is that it encourages consumers to use available work incentives. The results from our analysis are mixed, with the strongest evidence coming in the positive relationship of benefits counseling dosage to TWP completion and benefit offset use in the SSDI-EP. As noted 45 earlier, our results for WI-SPI may have as much to do with the lack of good data about work incentive use over the full Q0-Q8 period as with the impact of benefits counseling. It is crucial to acknowledge that the explanatory power of our regression and MANOVA models is very modest and the benefits counseling dosage categories (including their interaction with time) never explains over a few percentage points of observed variance. Generally, the pre-enrollment earnings variable is, by far, the strongest predictor variable. Based on prior work, this possibility had been anticipated. Thus, we performed an analysis of what factors best explain how many hours of benefits counseling a participant received. Participants with stronger pre-enrollment employment outcomes received more service than those who had weaker ones. Additionally (for the SSDI-EP only) those who reported receiving benefits counseling services before enrollment also received more service subsequent to project entry. Nonetheless, the explanatory power of the models estimating how many hours of benefits counseling participants received is low, with r 2 values never rising appreciably above 0.1. Part of this may be the lack of potentially important attitudinal variables in our models. However, we suspect that how one thinks about causality may also be an important issue. We doubt that one should think about the impacts of benefits counseling as one might think about the impacts of a drug. It might be better to view the service as a tool that those already committed to achieving better employment outcomes take up to pursue their goals. It might also be important to think about the context in which the service is used. The results of a standardized unit of service might prove to be more strongly affected by labor market conditions or program rules (such as the SSDI cash cliff) than by many individual level characteristics. Finally, our samples are limited to volunteers who appear to contain a far greater proportion of “work oriented” individuals than the overall working age population of those getting Social Security disability benefits. We cannot say with any certainty whether we would have observed similar patterns between benefits counseling dosage and employment outcomes if the typical WI SPI or SSDI-EP participant had entered the projects with the same likelihood of employment or mean earnings as those of the overall SSDI and SSI populations in Wisconsin or nationally. Much needs to be done to determine the efficacy of benefits counseling services for the “non-work oriented,” though there may be other good reasons to provide the service. In particular, we find purchase in the argument that no person in clear need of retaining cash benefits, access to health care, and other support services should be asked to risk their eligibility unless they are fully aware of the risks they face and what might be done to mitigate them. In any case, our paper adds to the body of evidence that supports the value of benefits counseling services for facilitating “return to work.” To the extent that the public and policymakers are interested in encouraging fuller labor market inclusion for those with severe disabilities, whether the motivation be general economic development, lowering outlays for government programs, or a concern for the economic welfare and dignity of persons with disabilities, the continuation, indeed expansion, of work incentive benefits counseling services appears to us to provide an effective policy tool for helping to achieve that goal. 46 Appendix A: Graphs of Mixed Linear Model Predicted Outcomes by Amount of Benefits Counseling Received for Combined Sample Figure 16: Predicted Mean UI Earnings by Amount of Benefits Counseling Received, Q0-Q8 Predicted Mean Quarterly Earnings by Amount of Benefits Counseling Mean Predicted Earnings 2000 HalfSDBCSplit 1500 Greater than 1/2 SD of BC Mean Within 1/2 SD of BC Mean Less than 1/2 SD of BC Mean 1000 500 0 0 1 2 3 4 5 6 7 8 Quarter Compared to Enrollment 47 Figure 17: Predicted Mean UI Employment Rates by Amount of Benefits Counseling Received Q0-Q8 Mean Predicted Employment Rates study2: Study Participant 1.00 0.80 HalfSDBCSplit 0.60 Greater than 1/2 SD of BC Mean Within 1/2 SD of BC Mean Less than 1/2 SD of BC Mean 0.40 0.20 0.00 0 1 2 3 4 5 6 7 8 Quarter Compared to Enrollment 48 Figure 18: Predicted Mean Income by Amount of Benefits Counseling Received Q0-Q8 study2: Study Participant 4500 Mean Predicted Income 4000 3500 3000 HalfSDBCSplit 2500 Greater than 1/2 SD of BC Mean Within 1/2 SD of BC Mean 2000 Less than 1/2 SD of BC Mean 1500 1000 500 0 0 1 2 3 4 5 6 7 8 Quarter Compared to Enrollment 49 Appendix B: MANOVA Model Summary Statistics Table 12: Repeated Measures Earnings MANOVA – WI SPI Participants (n = 437) With-In Subject (Wilks’ Lambda) Between Subject Sig Partial Eta Sig Partial Eta Squared Squared Intercept Quarter 0.776 0.011 < 0.001 0.032 Benefits Counseling *Quarter 0.125 0.026 0.013 0.020 PreEarnings *Quarter 0.045 0.037 < 0.001 0.118 Benefits Counseling * *Quarter 0.042 0.031 0.281 0.006 PreEarnings Gender *Quarter 0.837 0.010 0.351 0.002 Age *Quarter 0.916 0.008 0.202 0.004 Education *Quarter 0.212 0.025 0.420 0.002 Disability *Quarter 0.831 0.010 0.824 < 0.001 Table 13: Repeated Measures Earnings MANOVA – SSDI-EP Participants (n = 443) With-In Subject (Wilks’ Lambda) Between Subject Sig Partial Eta Sig Partial Eta Squared Squared Intercept Quarter 0.872 0.009 0.012 0.014 Benefits Counseling *Quarter 0.059 0.029 0.021 0.018 PreEarnings *Quarter 0.043 0.037 < 0.001 0.122 Benefits Counseling * *Quarter 0.110 0.027 0.924 < 0.001 PreEarnings Gender *Quarter 0.016 0.043 0.721 < 0.001 Age *Quarter 0.111 0.030 0.025 0.012 Education *Quarter 0.822 0.010 0.107 0.006 Disability *Quarter 0.296 0.022 0.200 0.004 Table 14: Repeated Measures Employment MANOVA – WI SPI Participants (n = 437) With-In Subject (Wilks’ Lambda) Between Subject Sig Partial Eta Sig Partial Eta Squared Squared Intercept Quarter 0.321 0.022 < 0.001 0.190 Benefits Counseling *Quarter 0.839 0.012 0.376 0.005 PreEarnings *Quarter < 0.001 0.068 < 0.001 0.198 Benefits Counseling * *Quarter 0.423 0.019 0.392 0.004 PreEarnings Gender *Quarter 0.757 0.012 0.120 0.006 Age *Quarter 0.807 0.011 0.092 0.007 Education *Quarter 0.047 0.036 0.350 0.002 Disability *Quarter 0.396 0.020 0.001 0.025 50 Table 15: Repeated Measures Employment MANOVA – SSDI-EP Participants (n = 443) With-In Subject (Wilks’ Lambda) Between Subject Sig Partial Eta Sig Partial Eta Squared Squared Intercept Quarter 0.755 0.012 < 0.001 0.102 Benefits Counseling *Quarter 0.693 0.015 0.004 0.025 PreEarnings *Quarter 0.102 0.031 < 0.001 0.236 Benefits Counseling * *Quarter 0.555 0.017 0.240 0.007 PreEarnings Gender *Quarter 0.050 0.035 0.965 < 0.001 Age *Quarter 0.436 0.018 0.007 0.017 Education *Quarter 0.313 0.022 0.613 0.001 Disability *Quarter 0.143 0.028 0.205 0.004 Table 16: Repeated Measures Income MANOVA – WI SPI Participants (n = 437) With-In Subject (Wilks’ Lambda) Between Subject Sig Partial Eta Sig Partial Eta Squared Squared Intercept Quarter 0.284 0.023 < 0.001 0.061 Benefits Counseling *Quarter 0.216 0.023 0.020 0.018 PreEarnings *Quarter 0.125 0.029 < 0.001 0.082 Benefits Counseling * *Quarter 0.100 0.027 0.053 0.014 PreEarnings Gender *Quarter 0.899 0.008 0.119 0.006 Age *Quarter 0656 0.014 0.044 0.009 Education *Quarter 0.238 0.024 0.010 0.015 Disability *Quarter 0.773 0.011 0.044 0.009 Table 17: Repeated Measures Income MANOVA – SSDI-EP Participants (n = 443) With-In Subject (Wilks’ Lambda) Between Subject Sig Partial Eta Sig Partial Eta Squared Squared Intercept Quarter 0.825 0.010 0.001 0.024 Benefits Counseling *Quarter 0.110 0.027 0.118 0.010 PreEarnings *Quarter 0.014 0.044 < 0.001 0.072 Benefits Counseling * *Quarter 0.071 0.029 0.561 0.003 PreEarnings Gender *Quarter 0.019 0.042 0.028 0.011 Age *Quarter 0.127 0.029 0.919 < 0.001 Education *Quarter 0.838 0.010 0.004 0.019 Disability *Quarter 0.327 0.021 0.169 0.004 51 Table 18: Employment Since SSA Benefits Repeated Measures Earnings MANOVA – WI SPI Participants (n = 437) With-In Subject (Wilks’ Lambda) Between Subject Sig Partial Eta Sig Partial Eta Squared Squared Intercept Quarter 0.594 0.017 0.093 0.007 Benefits Counseling *Quarter 0.802 0.014 0.048 0.001 Employment Since SSA *Quarter 0.986 0.005 < 0.001 0.077 Benefits Benefits Counseling * *Quarter 0.944 0.007 0.380 0.005 Employment Since SSA Benefits Gender *Quarter 0.747 0.013 0.570 0.001 Age *Quarter 0.735 0.013 0.211 0.004 Education *Quarter 0.250 0.026 0.220 0.004 Disability *Quarter 0.784 0.012 0.458 0.001 Table 19: Employment Since SSA Benefits Repeated Measures Earnings MANOVA – SSDI-EP Participants (n = 443) With-In Subject (Wilks’ Lambda) Between Subject Sig Partial Eta Sig Partial Eta Squared Squared Intercept Quarter 0.790 0.011 0.297 0.003 Benefits Counseling *Quarter 0.606 0.016 0.200 0.007 Employment Since SSA *Quarter 0.465 0.018 < 0.001 0.041 Benefits Benefits Counseling * *Quarter 0.905 0.011 0.428 0.004 Employment Since SSA Benefits Gender *Quarter 0.022 0.041 0.815 < 0.001 Age *Quarter 0.120 0.029 0.048 0.009 Education *Quarter 0.797 0.011 0.054 0.009 Disability *Quarter 0.211 0.025 0.245 0.003 52 Appendix C: Q8-Q0 Change in Employment Outcomes Predicted by MANOVA models Table 20: Repeated Measures MANOVA Estimated Mean Earnings by Received Amounts of Benefits Counseling Study Amount of Benefits Q0 Q8 Q8 – Q0 Counseling < ½ SD BC $900 $1270 $370 WI SPI (n =437) Within ½ SD BC $1200 $1324 $124 > ½ SD BC $1346 $1809 $463 < ½ SD BC $1224 $1060 -$164 SSDI-EP (n =443) Within ½ SD BC $1490 $1751 $261 > ½ SD BC $1683 $1887 $204 Table 21: Repeated Measures MANOVA Estimated Mean Employment by Received Amounts of Benefits Counseling Study Amount of Benefits Q0 Q8 Q8 – Q0 Counseling < ½ SD BC 0.28 0.31 0.03 WI SPI (n =437) Within ½ SD BC 0.28 0.39 0.11 > ½ SD BC 0.29 0.48 0.19 < ½ SD BC 0.32 0.28 -0.04 SSDI-EP (n =443) Within ½ SD BC 0.36 0.47 0.11 > ½ SD BC 0.37 0.53 0.16 Table 22: Repeated Measures MANOVA Estimated Mean Income by Received Amounts of Benefits Counseling Study Amount of Benefits Q0 Q8 Q8 – Q0 Counseling < ½ SD BC $2672 $2866 $194 WI SPI (n =437) Within ½ SD BC $2850 $3284 $434 > ½ SD BC $3008 $3698 $690 < ½ SD BC $3356 $3378 $22 SSDI-EP (n =443) Within ½ SD BC $3512 $3879 $367 > ½ SD BC $3550 $4192 $642 Table 23: Repeated Measures MANOVA Estimated Mean Earnings for PreEnrollment High Earners by Received Amounts of Benefits Counseling Study Amount of Benefits Q0 Q8 Q8 – Q0 Counseling < ½ SD BC $1539 $1996 $457 WI SPI (n =437) Within ½ SD BC $2166 $1805 -$361 > ½ SD BC $2178 $2516 $338 < ½ SD BC $1937 $1571 -$366 SSDI-EP (n =443) Within ½ SD BC $2412 $2422 $10 > ½ SD BC $2754 $2549 -$205 53 Table 24: Repeated Measures MANOVA Estimated Mean Earnings for PreEnrollment Low Earners by Received Amounts of Benefits Counseling Study Amount of Benefits Q0 Q8 Q8 – Q0 Counseling < ½ SD BC $261 $544 $283 WI SPI (n =437) Within ½ SD BC $235 $843 $608 > ½ SD BC $513 $1101 $588 < ½ SD BC $511 $547 $36 SSDI-EP (n =443) Within ½ SD BC $567 $1079 $512 > ½ SD BC $613 $1224 $611 Table 25: Repeated Measures MANOVA Estimated Mean Earnings for Participants who at Enrollment were not Employed after Receiving Benefits by Received Amounts of Benefits Counseling Study Amount of Benefits Q0 Q8 Q8 – Q0 Counseling < ½ SD BC $0 $246 $246 WI SPI (n =437) Within ½ SD BC $0 $589 $589 > ½ SD BC $22 $891 $869 < ½ SD BC $83 $306 $223 SSDI-EP (n =443) Within ½ SD BC $67 $562 $495 > ½ SD BC $38 $1160 $1122 Table 26: Repeated Measures MANOVA Estimated Mean Earnings for Participants who at Enrollment were Employed after Receiving Benefits by Received Amounts of Benefits Counseling Study Amount of Benefits Q0 Q8 Q8 – Q0 Counseling < ½ SD BC $824 $1144 $320 WI SPI (n =437) Within ½ SD BC $968 $1263 $295 > ½ SD BC $1352 $1938 $586 < ½ SD BC $1085 $938 -$147 SSDI-EP (n =443) Within ½ SD BC $1348 $1676 $328 > ½ SD BC $1559 $1768 $209