Statement of Christopher J. O'Leary, Ph.D., Senior Economist, W.E. Upjohn Institute for Employment Research, Kalamazoo, Michigan

Testimony Before the Subcommittee on Income Security and Family Support
of the House Committee on Ways and Means

April 10, 2003

I.  Introduction

This year marks the tenth anniversary of legislation that established the Worker Profiling and Reemployment Services (WPRS) system.  Passage of PL 103-152 in November 1993 required each state to implement a WPRS system, with the purpose of promoting speedy reemployment for unemployment insurance (UI) beneficiaries at risk of long-term unemployment.  The system includes two key components:  1) identification of UI beneficiaries who are most likely to exhaust their regular UI benefits and 2) referral of those individuals to reemployment services.  WPRS is based on a large body of research conducted by states and the federal government that found targeted job search assistance to be a highly cost effective means for promoting return to work (Wandner 1994, Meyer 1995).  WPRS has been operational in all states since 1995, and recent evaluations of the program show that it has provided an effective incentive for reducing UI benefit receipt (Dickinson, Decker, and Kreutzer 2002).  There is also some evidence that WPRS has led to increased earnings during the UI benefit year (Black, Smith, Berger and Noel 2001).

The purpose of my remarks is to describe the profiling and referral process, to offer evidence of its effectiveness, and to suggest areas that need improvement.  I will also briefly describe two innovative extensions of the profiling concept to other workforce programs. 

II.  Formal Selection Process

One of the innovative aspects of WPRS is the formal approach it takes in selecting customers into employment programs.  The administrative process by which individuals are selected to participate in employment is referred to as targeting.  When program participation is not an entitlement and existing capacity of the program cannot accommodate all those who may desire to participate, a selection process must be adopted.  To achieve an efficient and effective program, one must devise a selection process that directs customers to services that best meet their needs.[1] 

Through its statistical profiling model, WPRS offers a systematic referral process using objective data which is applied equally to all eligible customers.  Traditionally, the selection process has been informal, relying upon the judgment of frontline staff or the queuing principle of first-come, first-served.[2]  Formal methods like WPRS provide for systematic selection based on objective criteria applied equally to all customers.    Evaluations of WPRS have shown that the statistical models are able to distinguish among those most likely to exhaust UI benefits from those least likely to exhaust with significant precision (Dickinson et al. 1999, 2002).

 III.  Concept and Purpose of WPRS

Through WPRS, states have taken preemptive action to help unemployment insurance (UI) beneficiaries shorten their duration of UI compensation.  A state WPRS system identifies, primarily through statistical methods, those UI recipients who are most likely to exhaust their benefit entitlement and refers them to required reemployment services.  The profiling and referral process is performed in three stages.  First, unemployment insurance recipients who are expecting to be recalled to their previous job or who are members of a union hiring hall awaiting their next assignment are dropped from the process.^[3]  Second, the remaining unemployment insurance recipients are ranked by their likelihood of exhausting regular unemployment insurance benefits as determined by a statistical model.  Third, beneficiaries are then referred to reemployment services in order of their ranking until the capacity of local agencies to serve them is filled.^[4]

To profile workers, most states have adopted a statistical methodology that assigns a probability of benefit exhaustion to each UI beneficiary who is eligible for profiling.[5]  A few states, which lacked sufficient data or expertise to estimate a probability model, started with a simple screening device based on one or two characteristics.  Some of these states have moved to statistical models once the data deficiencies were corrected.  Today, about 85 percent of the states use statistical models.  The probability of exhausting benefits is derived from estimating the effects of personal characteristics and economic factors on the likelihood that a UI recipient will exhaust benefits.  Personal characteristics typically include:  educational attainment; tenure, wages, industry and occupation of last job held; and exhaustion of benefits in prior benefit years.  Civil rights legislation prohibits using a claimant’s age, race, and gender as variables in the model.  Local labor market conditions are also included to reflect the likelihood of reemployment in the various local labor markets within a state.  In essence, the probability assigned to each eligible UI recipient is a weighted average of the effect of each of these characteristics on the likelihood that an individual exhausts UI benefits.[6]

IV.  Background

WPRS can trace its roots to research sponsored by the U.S. Department of Labor during the 1980s.  Those studies revealed several common characteristics about dislocated workers, which could be used to identify those who would have the most difficulty finding employment.  For example, workers with longer job tenure (more than three years) and who were employed in manufacturing industries were more likely to experience long durations of unemployment and significant earnings reductions than those with shorter tenure and in industries other than manufacturing, particularly nondurable industries.  In addition, demonstration projects conducted in New Jersey, Nevada, Minnesota, and Washington, offered convincing evidence that supported the profiling and referral concept (Meyer 1995).  The demonstrations in New Jersey and Minnesota established the efficacy of using statistical methods and administrative data to identify, early in their unemployment spell, those who are likely to experience long periods of joblessness.  Results from all four states showed that providing more intensive job search assistance reduces the duration of insured unemployment and UI expenditures.  The magnitude of the effects were large enough to make a difference in program costs:  Reduction in UI receipts ranged from 4 weeks in Minnesota to a half week in Washington, and the government benefit-to-cost ratio varied from 4.8 in Minnesota to 1.8 in New Jersey.  At the same time, workers’ earnings were higher because job search assistance accelerated their reemployment and thus increased the number of hours worked (Corson, Dunstan, Decker, and Gordon 1989). 

Encouraged by the prospect of UI benefit savings from the early identification and referral of long-term unemployed to reemployment services along with the persistent increase in the number of long-term unemployed, Congress passed legislation in November 1993 that mandated states to implement WPRS programs.  The legislation gained broad bipartisan support in part because of the large and convincing body of prior research findings and the estimates by the Congressional Budget Office that the WPRS would generate significant savings for the federal government over the first five years of the program.  The bill did not create new services for displaced workers, and states were required to provide only those services that were already available.  Workers who were referred to available services were required to participate in the program or risk losing their UI benefits.   

Although WPRS is federally mandated, each state was asked to implement the program themselves.  The federal government provided states with one-time funds to build capacity and expertise and offered state agencies limited technical assistance.  After that, states were expected to finance the program out of ongoing employment and training program funds.  Consequently, the ability of the states to serve claimants depends upon the capacity of the existing reemployment services.   For some states, the demands of designing and testing a statistical profiling model were beyond the technical expertise of their staff, and they elicited the assistance of universities and other research groups to help develop a model.  Therefore, successful implementation of the program required cooperation and coordination among a variety of federal and state agencies, including UI, the employment service, the Economic Dislocated Worker Adjustment Assistance (EDWAA) training programs, and research groups.  

V.  Evaluations of the Effectiveness of WPRS      

Two evaluations have been conducted to determine the success of WPRS.  A multi-state evaluation of WPRS, sponsored by the U.S. Department of Labor, was based on claimant-level data from a sample of states (Dickinson et al. 1999, 2002).  In each of the states included in the study (Connecticut, Illinois, Kentucky, Maine, New Jersey, and South Carolina), labor market outcome data were compiled from administrative records on all new initial UI claimants between July 1995 and December 1996 who were eligible for referral to mandatory WPRS job search assistance (JSA).  The combined samples included 92,401 profiled and referred claimants, and 295,920 claimants who were profiled but not referred to WPRS JSA.  The impact estimates were statistically significant in all states except South Carolina.  For those five states with statistically significant results, the largest impact was !0.98 weeks in Maine with the other impacts ranging from !0.21 to !0.41 weeks of UI benefits.

The State of Kentucky also sponsored an assessment of its WPRS system.  A feature of the Kentucky evaluation that sets it apart from the national evaluation was that the evaluation design was incorporated into the profiling modeling and implementation process.  This allowed for the randomized assignment of claimants to treatment and control groups--an improvement over the design of the multi-state evaluation.  A team of economists at the Center for Business and Economic Research at the University of Kentucky developed the profiling model and conducted the evaluation (Berger et al. 1997, 2001).    

The impact estimates for WPRS in the Kentucky evaluation were more dramatic than those found in the multi-state evaluation.  With regard to the three outcomes of interest, the estimated impacts were a reduction of 2.2 weeks of UI, a reduction of $143 in UI benefits per beneficiary, and an increase of $1,054 per beneficiary in earnings during the UI benefit year.  The differences in these estimates from those of the multi-state WPRS evaluation are most likely due to the fact that Black et al. (2001) essentially confined their comparisons within narrow intervals of exhaustion probabilities, thereby achieving a closer counterfactual.  Dickinson et al. (1999) compared those assigned to WPRS, who had the highest probability of benefit exhaustion, with all those profiled but not referred, including many with very low exhaustion probabilities.  This meant that the comparison group in the multi-state evaluation was likely to have shorter mean benefit duration than program participants, even in the absence of WPRS services.  The ideal approach is to use beneficiaries from the same percentile group to make the comparison between the outcomes of those who were referred to orientation with those who were not.

VI.  Issues Requiring Attention and Improvement

Two aspects of WPRS require particular attention and improvement.  The first issue is the ability to provide reliable estimates of a beneficiary’s likelihood of exhausting benefits.  At the heart of WPRS is a statistical model that predicts the probability that a UI beneficiary will exhaust his or her benefits.  The model is based on the relationship between the event that a UI beneficiary exhausts benefits and key personal characteristics and local labor market conditions.  Using the experience of UI beneficiaries who have recently filed claims, estimates of the relative contribution of each of the characteristics embedded in the model are obtained.  These estimates are then combined with a claimant’s personal characteristics to generate that person’s probability of exhaustion. 

In order to ensure that the predictions are as accurate as possible, states must be diligent in updating their statistical models on a regular basis. The WPRS policy workgroup established in 1998 by USDOL recommended that states update their models so that they reflect current labor market conditions and worker behavior (Messenger, Schwartz and Wandner 1999).  The USDOL also provided Significant Improvement Demonstration Grants to 11 states, half of which used the funds to update their models (Needels, Corson, and Van Noy 2002).  Unfortunately, limited funds were available to assist only a handful of states.  More resources, both at the state and federal levels, should to be provided to ensure the quality of these models and to make sure they reflect current labor market conditions.  One approach is for state workforce agencies to establish linkages between economic research units at universities and other research institutions.  Such collaboration can leverage government funds and benefit everyone involved.

The second issue is the integration of the identification process with the provision of services.  Adequate reemployment services are the critical step between profiling and getting the unemployed back to work.  Worker profiling alone is not sufficient to yield the intended results of the program. WPRS has made significant strides in placing greater emphasis within the UI system on the work test by requiring UI beneficiaries to participate in services and to actively search for jobs, and has prompted claimants to undertake these activities earlier than later in their unemployment spell.  One office manager we talked with during our evaluation of Michigan’s WPRS offered that WPRS gave his staff the opportunity to do what they were supposed to do—assist the unemployed in finding a job.  Previously, staff was frustrated because too few people were requesting assistance (Eberts and O’Leary 1997). 

Yet, reemployment services require funding.  Since the inception of WPRS, the funding of services has come from sources outside of WPRS.  The federal legislation assumes that states will provide the services from other federal funds, mainly ES grants.  ES grants are the primary source of funding of public labor exchange and job search assistance services.  Congress has provided $35 million for FY 2003 and in several prior years for “Reemployment Services Grants,” which are part of “Employment Service Grants to States.  However, these grants are not proposed in the Administration’s budget for FY2004. 

VII.  Extension of Statistical Targeting Tools to Other Programs

Although WPRS is entering its second decade, the use of statistical methods to target resources is only in its infancy.  These statistical management tools have great potential, particularly in the one-stop environment established by the Workforce Investment Act (WIA).  WIA has established a hierarchy of services, from core to intensive to training.  Given the extensive number of services available, one-stop staff is faced with the challenge of directing customers to services that best meet their reemployment needs.  Currently, the Upjohn Institute is collaborating with the U.S. Department of Labor and the Georgia Department of Labor to develop a statistical assessment and targeting methodology that assists frontline staff in evaluating available job openings and making referrals to services.  This system, termed the Frontline Decision Support System (FDSS), offers a systematic framework for staff to quickly assess the needs of customers, to target services that meet customers’ needs, and to deliver services in an effective and efficient manner.  The FDSS tools are similar to worker profiling models in that statistical relationships are estimated between a customer’s outcomes and personal characteristics and other factors.  In the case of FDSS, the outcome is employment rather than UI benefit exhaustion (Eberts, O’Leary, and DeRango 2002). 

Despite the similar methodologies, FDSS’s referral decision process is more complex than that of WPRS.  With WPRS, the decision is whether or not to refer a UI claimant to a predetermined set of services.  Under FDSS, the decision is which among a large number of services best meets the needs of a specific customer.  FDSS provides a customized list of services, ranked from most effective to least effective for each individual.  The list is customized for each individual in that it reflects the effectiveness of services for past participants with characteristics similar to the customer that a staff person is currently serving.  FDSS also provides specific information about job prospects and wage potential for each customer.  Thus, FDSS serves all customers who enter the one step, not simply UI claimants.  Yet, like WPRS, FDSS promises to reduce the length of time job seekers are out of work by helping staff and customers make more informed decisions about services and job prospects.  FDSS is currently in operation at two sites in Georgia and is scheduled to go statewide in a few months. 

Prior to developing and implementing FDSS, the Upjohn Institute with support from the U.S. Department of Labor, extended the statistical assessment methods of WPRS to welfare-to-work programs.  The success of this project provided the basis for developing FDSS.  Welfare-to-work programs typically treat all recipients the same, providing the same basic services regardless of a participant’s skills, aptitudes, and motivation.  Yet, barriers vary widely among participants.  Some customers require little assistance in finding a job, while others have multiple barriers and stand to benefit from more intensive, targeted services.  The Upjohn Institute developed and conducted a pilot that used administrative tools to target services to customers without changing the nature of the program or significantly raising costs.  Statistical techniques were developed to estimate the likelihood of employment based on participants’ demographic and work history information found in administrative records.  An employability score was computed for each customer and was then used to assign each participant to one of three service providers.  Each provider offered the same basic set of services but differed in the mix of services and in their approach to delivering services.  The pilot used these differences to determine the best provider for each customer.   

An evaluation, based on random assignment, provided evidence that the pilot was successful in using statistical tools to improve program outcomes by placing more welfare recipients into jobs.[7]  It showed that the statistical assessment tool successfully distinguished among participants with respect to barriers to employment.  It also found that referring participants to service providers according to their individualized statistical needs assessment (employability score) increased the overall effectiveness of the program by 27 percent as measured by the program goal of customers finding and retaining a job for 90 consecutive days. 

VIII.  Conclusion

WPRS has introduced an innovative management tool into the workforce development arena.  The statistical targeting methodology has provided staff with an effective means of directing reemployment services to those unemployed workers who need them most.  Evaluations have shown that such a tool has benefited both the UI system by reducing unemployment duration and the worker by increasing earnings.  Furthermore, statistical tools have also been successfully used in workforce programs that are broader in scope.  I believe that with the proper support for WPRS and continued encouragement for states to develop and implement additional tools to help staff and customers make more informed decisions, we can continue to improve the efficiency and cost-effectiveness of the UI and workforce development systems in this country.   

References

Berger, Mark C., Dan A. Black, Amitabh Chandra and Steven N. Allen.  1997.  "Profiling Workers for Unemployment Insurance in Kentucky," The Kentucky Journal of Business and Economics 16: 1-18.

Black, Dan, Jeffrey Smith, Mark Berger and Brett Noel.  2001.  "Is the Threat of Reemployment Services More Effective than the Services Themselves?  Experimental Evidence from the UI System." unpublished manuscript.  College Park, MD: University of Maryland.

Black, Dan, Jeffrey Smith, Miana Plesca and Suzanne Plourde.  2002.  "Estimating the Duration of Unemployment Insurance Benefit Recipiency."  Final Technical Report.  Contract Number UI-10908-00-60.  Washington, DC: U.S. Department of Labor (February).

Corson, Walter, Shari Dunstan, Paul Decker, and Anne Gordon, 1989.  “New Jersey Unemployment Insurance Reemployment Demonstration Project,” Unemployment Insurance Service, Occasional Paper 89-3.  Washington, DC: U.S. Department of Labor (April).

Dickinson, Katherine P., Paul T. Decker, Suzanne D. Kreutzer, and Richard W. West.  1999.  "Evaluation of Worker Profiling and Reemployment Services: Final Report."  Research and Evaluation Report Series 99-D.  Washington, DC: Office of Policy and Research, Employment and Training Administration, U.S. Department of Labor.

Dickinson, Katherine P., Paul T. Decker and Suzanne D. Kreutzer.  2002.  "Evaluation of WPRS Systems" in Targeting Employment Services, Randall W. Eberts, Christopher J. O'Leary, and Stephen A. Wandner, eds.  Kalamazoo, MI: W.E. Upjohn Institute for Employment Research. 

Eberts, Randall W. 2002.  Design, Implementation, and Evaluation of the Work First Profiling Pilot Project.   Washington D.C.: U.S. Department of Labor, Employment and Training Administration Occasional Paper 2002-07.

Eberts, Randall W., and Christopher J. O'Leary.  1997.  "Process Analysis of the Worker Profiling and Reemployment Services (WPRS) System in Michigan." Prepared for the Michigan Employment Security Agency.  Kalamazoo, MI:  W.E. Upjohn Institute for Employment Research (April).

Eberts, Randall W., and Christopher J. O'Leary.  2003.  "A New WPRS Profiling Model for Michigan,” Report to the Michigan Unemployment Agency prepared by   W.E. Upjohn Institute for Employment Research (March).

Eberts, Randall W., Christopher J. O’Leary, and Kelly J. DeRango.  2002. “A Frontline Decision Support System for One-Stop Centers,” in Targeting Employment Services, Randall W. Eberts, Christopher J. O'Leary, and Stephen A. Wandner, eds.  Kalamazoo, MI: W.E. Upjohn Institute for Employment Research. 

Eberts, Randall W., Christopher J. O'Leary, and Stephen A. Wandner, eds. 2002. Targeting Employment Services, Kalamazoo, MI: W.E. Upjohn Institute for Employment Research. 

Gueron, Judith, and Edward Pauly with Cameran M. Lougy.  1991.  From Welfare to Work.  New York: Russell Sage Foundation.

Messenger, Jon, Suzanne Schwartz, and Stephen A. Wandner.  1999.  The Worker   Profiling and Reemployment Services Policy Workgroup:  Policy Analysis and   Recommendations.  Washington, D.C.: U.S. Department of Labor, Office of Policy and Research. 

Needels, Karen, Walter Corson, and Michelle Van Noy.  2002.  "Evaluation of the Significant Improvement Demonstration Grants for the Provision of Reemployment Services for UI Claimants," Final Report to the U.S. Department of Labor, Mathematica Policy Research, Inc. July. 

OECD.  1998.  Early Identification of Jobseekers at Risk of Long-Term Unemployment: The Role of Profiling.  Paris: Organisation for Economic Co-Operation and Development.

Wandner, Stephen A., ed.  1994  “The Worker Profiling and Reemployment Services System: Legislation, Implementation Process and Research Findings.”   Unemployment Insurance Occasional Paper No. 94-4.   Washington, DC: UI Service, Employment and Training Administration, U.S. Department of Labor. 

Wandner, Stephen A. 1997.  “Early Reemployment for Dislocated Workers in the United States,” International Social Security Review 50(4): 95-112.