reducing child support debt and its consequences: can ... wicsec conference materials...reducing...

Abstract

As child support debt owed nationally persists at enormous levels, both noncustodi-al parents and the custodial families who are not receiving support suffer significanthardships, and states are forced to expend greater resources on collection andenforcement efforts. This paper presents findings from an evaluation of a demon-stration program developed to help noncustodial parents with large child supportdebts reduce their debt while simultaneously increasing child support paid to fami-lies, through gradual forgiveness of arrears conditional on payment of current childsupport obligations. The evaluation employs a randomized experimental design,nonexperimental analyses using propensity score matching and multilevel modelingtechniques, and focus groups and follow-up interviews. Results show a pattern ofeffects that suggests individuals responded to the program as intended. State- andfamily-owed child support debt balances decreased for program participants, andparticipants paid more toward their child support obligations and arrears and mademore frequent child support payments. The study findings suggest promise for theeffectiveness of this program model in reducing child support debt burdens and inincreasing families’ receipt of child support, and they also point to ways in whichthe implementation of the program might be improved. © 2011 by the Association forPublic Policy Analysis and Management.

INTRODUCTION

Despite federal and state policy efforts to address nonpayment of child support, theamount of child support debt owed nationally, over $107 billion in Fiscal Year (FY)2009 (U.S. Department of Health and Human Services, Administration for Childrenand Families, Office of Child Support Enforcement [USDHHS ACF, OCSE], 2010),remains alarming. The fact that low-income noncustodial parents (NCPs) owe adisproportionate share of this debt suggests that traditional collection and enforce-ment policies may be inadequate, and that much of the amount may ultimatelyprove “uncollectible” (Sorensen, Sousa, & Schaner, 2007). Additionally, child sup-port debt presents significant hardships for NCPs and custodial families who arenot receiving support, as well as for states that expend resources on collection andenforcement efforts and face federal performance standard consequences for non-collection.

We present findings from an experimental and nonexperimental evaluation of ademonstration program, Families Forward, developed to help NCPs with largechild support debts to reduce their debt levels while simultaneously increasingchild support paid to custodial parents (CPs). The Families Forward programdesign was unique in providing for gradual forgiveness of debt owed to the CP, the

Reducing Child Support Debtand Its Consequences: CanForgiveness Benefit All?

Carolyn J. HeinrichBrett C. Burkhardt

Hilary M. Shager

Journal of Policy Analysis and Management, Vol. 30, No. 4, 755–774 (2011)© 2011 by the Association for Public Policy Analysis and Management Published by Wiley Periodicals, Inc. View this article online at wileyonlinelibrary.com/journal/pamDOI: 10.1002/pam.20599

756 / Reducing Child Support Debt and Its Consequences

Journal of Policy Analysis and Management DOI: 10.1002/pamPublished on behalf of the Association for Public Policy Analysis and Management

state (or both) over time, conditional on NCP payments toward current child sup-port or debt. The outcomes evaluated include payments made toward current support obligations or debt, the frequency of payments made, and changes in state-owed and family-owed debt balances. We begin with a review of the literature onchild support debt accumulation, its consequences, and policy and program strate-gies developed to address this problem, and then describe the distinctive features ofFamilies Forward and the evaluation samples and data. We present the experimen-tal and nonexperimental methods used in the evaluation and findings from theanalyses. Experimental analyses comparing those intending to receive treatmentwith the control group do not show statistically significant program impacts.Alternatively, nonexperimental analyses comparing actual participants to nonpartic-ipants find statistically significant program effects in increasing NCPs’ paymentstoward child support or arrears, increasing the frequency of payment, and reducingstate-owed debt balances. We conclude with a discussion of a planned expansion ofthe program and changes that are being considered to improve program enrollmentand effectiveness.

LITERATURE REVIEW

How Child Support Debt Accumulates and Its Consequences

Child support debt owed to custodial families may be composed of unpaid currentsupport, retroactive support, and interest. Arrears owed to the government derivefrom a mix of policies that pass costs on to NCPs, including reimbursement for pub-lic assistance benefits (paid to CPs), lying-in (medical) costs, interest charges, andother fees associated with case processing or genetic testing (Bartfeld, 2003).1 NCPsmay also enter the child support system with high debt due to retroactive supportorders, in which NCPs are responsible for the lump sum accumulated between thebirth of a child and establishment of a formal support order (U.S. Department ofHealth and Human Services, Office of Inspector General [USDHHS, OIG], 2000).Research suggests a correlation between establishment debt and subsequent non-payment of child support obligations (Cancian, Heinrich, & Chung, 2009).Furthermore, federal regulations require payment of current support before debt,so that debt and interest may accumulate quickly even for NCPs who pay their cur-rent obligations.

Child support order establishment guidelines may also impose support obliga-tions or default orders that low-income parents cannot afford. Some states requiremonthly minimum support orders that may be high compared to actual income,particularly for NCPs who are incarcerated or unemployed (USDHHS, OIG, 2000).For NCPs who fail to appear in court or provide current earnings information,default orders may be established based on income imputation, often assuming thatthe NCP is working full time and using a state’s minimum wage or an average stateor industry wage (Roberts, 2001). Even if established orders are within the meansof NCPs, subsequent changes in employment or health status, or periods of incar-ceration may call for order modification, which rarely occurs automatically andmay be difficult for low-income parents to secure due to lack of information or legalrepresentation (Roberts, 2001). Other policy factors that may contribute to theaccumulation of child support debt include the system’s failure to recognize informalcustody changes (Roberts, 2001), double-digit interest rates and other operationalfees charged by states, and parents’ poor understanding of child support policies(Meyer, Cancian, & Nam, 2007).

Analyses of child support arrears suggest that much of the debt is held by poorNCPs. A national study found that of the 7 million nonresident fathers who did not

1 Many child support policies are determined at the state or county level; therefore, not all practices existin all jurisdictions.

Reducing Child Support Debt and Its Consequences / 757


pay child support, approximately 36 percent had incomes below the poverty line(Sorensen & Zibman, 2001). Another study in nine large states found that half of the obligors reported no annual income or earnings of less than $10,000 a year, and thedebt owed by these obligors accounted for 70 percent of the total arrears (Sorensen,Sousa, & Schaner, 2007).2 Other factors contributing to NCPs’ inability to pay childsupport include poor job skills and employment opportunities, low levels of educa-tion, incarceration, unstable health, and a lack of assets to negotiate large lump sumpayments for conventional debt forgiveness (Pate, 2002). Given the poor economiccircumstances of many obligors, researchers have posited that it may be unrealisticto expect additional financial support from these parents (Cancian & Meyer, 2004).Simulations for seven states predicted that only 40 percent of the arrears currentlyowed would likely be collected within the next decade and total arrears would growby at least 50 percent over the same time period (Sorensen, Sousa, & Schaner,2007).3 A California analysis produced even more dire results, with simulation mod-els suggesting that over ten years, NCPs would be able to pay only 25 percent of the$14.4 billion in arrears accumulated as of March 2000, and arrears would continueto more than double over the same time period (Sorensen, 2004).

The accumulation of child support debt poses problems for custodial families,low-income NCPs, and states (Bartfeld, 2003). In 2007, approximately 25 percent ofCPs had incomes below the federal poverty level, and for poor CPs who received fullpayments, child support represented approximately 48 percent of their averageincome. Approximately 31 percent of poverty-level families, however, received nopayments at all (Grall, 2009). Nonpayment of child support may also exacerbateconflict between CPs and NCPs, and reduce NCPs’ contact with children (Bartfeld,2003). For NCPs, debt may deter work, as automatic wage withholding may encour-age parents to go into the underground economy to avoid cooperating with the formal child support system (Waller & Plotnick, 2001).4 NCPs with delinquent pay-ments may also be subject to enforcement actions, including loss of driver’s license,liens against vehicles or property, and incarceration that may depress their earningcapacity and further exacerbate debt levels (Holzer, Offner, & Sorensen, 2005).States face increasing expenditures for collection and enforcement efforts andlower state scores on federal performance measures for payments collected and dis-tributed (Bartfeld, 2003).

Policy Responses to the Child Support Debt Problem: Debt Forgiveness Programs

With large and growing debt already on the books, debt forgiveness strategies areincreasingly employed by some counties and states (USDHHS, OIG, 2007). Federallaw allows states to accept less than full payment of state-owed arrearages and alsoallows reduction of custodial-owed arrears if both parties agree (USDHHS, ACF,OCSE, 2000). State or local courts or child support enforcement (CSE) agencies canaccept reduced debt payments or expunge debt altogether in exchange for otherdesired behaviors, such as on-time payment of current support obligations, or par-ticipation in employment or parenting programs.

2 The nine states included were Arizona, Florida, Illinois, Michigan, New Jersey, New York, Ohio,Pennsylvania, and Texas. Collectively, these states accounted for 39 percent of the country’s child supportarrears in FY 2006. An analysis of California’s child support debt yielded remarkably similar findings: 70percent of the state’s arrears were owed by NCPs reporting annual incomes of less than $10,000, and mostof these parents owed more than $20,000 (Sorensen, 2004).3 These states provided detailed administrative data on obligors and their arrears, which were matchedby OCSE to six quarters of national quarterly wage and unemployment insurance data. Collection rateswere estimated using current individual state child support policies.4 In contrast, using data from the Fragile Families study, Rich, Garfinkel, and Gao (2007) found thatstricter child support enforcement was associated with fewer hours of underground employment.



Previous studies have identified small-scale or pilot debt forgiveness programsoperating in a number of states, primarily at the county level (Bartfeld, 2003;Hennessey & Venohr, 2000; Pearson & Griswold, 2001), and a recent report by theOffice of Inspector General (OIG) (2007) describes child support debt compromiseprograms operating in at least 20 states.5 To reduce or expunge debt, 19 statesrequire NCPs to pay a lump sum toward the arrearage; 19 states require regularpayments on arrearages and current support orders; eight states require involve-ment with the child; seven states require participation in fatherhood programs; fivestates require participation in a parenting program; and four states require mainte-nance of employment, or some combination of these requirements.

Despite increasing adoption, little empirical evidence of the effectiveness of childsupport debt forgiveness programs exists (USDHHS, OIG, 2007). The OIG study offive states with fully implemented debt compromise programs6 found that the aver-age arrears per case was $22,029, and the average amount by which debt wasreduced through compromise agreements was $9,383. Only three other pilot pro-grams have been formally evaluated (using nonexperimental methods). Colorado’sArrears Forgiveness Demonstration Program provided forgiveness of all state-owedchild support debt for participants who paid their current support orders consis-tently and in full over a period of ten months (Pearson & Davis, 2002). Minnesota’sDebt Compromise Demonstration Program for low-income NCPs in HennepinCounty also provided full forgiveness of state-owed public assistance arrears (eitherin two lump sums or more gradually) if participants paid existing child supportorders on a regular basis over a period of 12 consecutive months (Pukstas et al.,2004).7 Maryland’s Arrears Leveraging Pilot Project (ALPP) (in Baltimore) enrolledobligors who were currently working with community-based organizations (CBOs)on employment issues and rewarded consistent current support payers by gradual-ly reducing or eliminating their state-owed arrears (in five stages). Each of theseprograms reported some success in collecting more child support from participat-ing NCPs; however, the Colorado and Minnesota programs experienced challengeswith low and selective enrollment, and the Maryland program participants encoun-tered substantial delays between eligibility and actual debt forgiveness. Also, theevaluation could not distinguish between the impact of the CBO programs and thearrears leveraging component (Ovwigho, Saunders, & Born, 2007). In each of thesecases, success in meeting program requirements was correlated with prior earningsor child support payment, making it difficult to reach definitive conclusions on pro-gram success. New, more rigorously designed research is clearly needed to deter-mine whether and how such programs work to reduce debt and increase formalchild support payments.

THE FAMILIES FORWARD PILOT PROGRAM, EVALUATION SAMPLES, AND DATA

The State of Wisconsin Bureau of Child Support (BCS) and the Institute forResearch on Poverty (IRP) at the University of Wisconsin–Madison began investigat-ing policy options to address growing child support debt in Wisconsin in 2003.

5 According to the OIG survey, 12 states have CSE agencies that operate fully implemented debt compro-mise programs, and eight states have agencies that are currently running pilot programs. Another 23states allow debt compromise on a case-by-case basis, with individually determined amounts, eligibilitycriteria, and conditions. Eight states (Arkansas, Idaho, Indiana, Mississippi, Missouri, New York,Tennessee, Virginia) reported that they do not allow compromise of debt owed to the state, although allstates plus the District of Columbia acknowledge the legality of compromise of unassigned debt owed toCPs, per request and agreement between the two parties (USDHHS, OIG, 2007).6 The states selected were California, Massachusetts, New Mexico, Texas, and Washington. 7 Although originally designed as a randomized controlled trial, some noncustodial parents assigned totreatment groups chose not to enroll. Enrollees had significantly higher incomes and lower monthly childsupport obligations, factors found to be predictive of program success (Pukstas et al., 2004).



Following exploratory research on possible program features, the pilot program,Families Forward, was designed to reduce child support debt while increasing childsupport paid. Racine County (in southeastern Wisconsin) was chosen as the experi-mental site because of the substantial number of cases with high levels of debt andcounty demographics (e.g., race, income and poverty, employment rates, single-parentfamilies) that were comparable to other large urban U.S. counties (such asMilwaukee), where child support debt and enforcement challenges are considerable.

Program Features, Eligibility, and Enrollment

The Families Forward program includes features that are relatively unique com-pared to other arrears forgiveness program models. Rather than forgiving child sup-port debt in a lump sum after specific requirements are satisfied, Families Forwardprovides for gradual forgiveness of debt. For an NCP with state-owed debt, the stateagrees to reduce the debt owed to the state by 50 cents for every dollar of currentsupport the NCP pays. If the NCP has family-owed debt, the NCP may receive a debtreduction of 50 cents for every dollar of current support paid, although reductionof family-owed debt requires the consent of the custodial parent (CP).8 NCPs whohave both state-owed and family-owed debts may receive reductions of both typesof debt. Additionally, NCPs with multiple child support cases may have differentdebt reduction arrangements across cases (e.g., reductions for state-owed debt forcase A, state- and family-owed debt for case B, and neither for case C).9 For all typesof debt reduction, credit is applied quarterly, first to an NCP’s interest balance andthen to the principal of the debt. Interest charging on debt is also suspended dur-ing program participation, which could continue for up to two years, unless theNCP fails to make a current support payment in two consecutive quarters.

To ensure that the size of the pilot program would be manageable, the eligiblepopulation was restricted to NCPs with total debt burdens of at least $2,000 owedto the state or the family on at least one child support case (excluding foster care,kinship, and interstate cases) and who had a recent history of nonpayment, definedas the following: (1) no payment on current support in the past three months; (2) current support paid in less than six of the last 12 months; or (3) paid less thanone-half of annual amount owed in current support over the past 12 months. Inadvance of the start of the pilot, the sample frame was identified by applying thesecriteria to NCPs attached to Racine County IV-D cases open at the end of December2004. All NCPs (and their IV-D cases) in the eligible sample frame (approximately5,000 cases) were randomly assigned to the control or experimental group using the last two digits of NCPs’ social security numbers; seven of every 10 cases wereassigned experimental status prior to the program roll-out. New IV-D cases weresimilarly randomly assigned to the control or experimental groups each quarter ifthey met eligibility requirements.

Enrollment of eligible NCPs in the pilot program began in May 2005 and ended inNovember of 2007, with participation continuing for up to two years. Eligible NCPshad to go through a number of steps before enrolling and receiving debt reduction.Because Families Forward was a new program, NCPs first had to learn of its exis-tence. IRP mailed letters of invitation to eligible NCPs (both in the experimental andcontrol group) and their CPs, and posters and brochures were distributed at retail,government, and service provider locations in Racine County. NCPs then had to con-tact the child support agency to confirm their eligibility and express their willingnessto participate; control cases were informed that they did not qualify for program

8 An NCP may pay more current support than he or she owes in a quarter. When such lump sum pay-ments occur, the NCP receives debt reduction credit for the entire amount paid. For both state- and family-owed debt, NCPs do not receive credit for payments made via federal or state tax intercept.9 If an NCP makes payments that are divided among multiple child support cases, arrears are reducedonly for the portion of the payment that is applied to the enrolled case(s).



participation at that time. An experimental NCP looking to receive reduction onstate-owed debt then had to wait for the child support agency to generate and maila stipulation agreement outlining the terms of the debt forgiveness. An NCP look-ing to receive reduction on family-owed debt also had to contact the CP anddescribe the program.10 If the CP agreed to participate in the program, he or shehad to communicate this to the child support agency, and both parties then had towait for the child support agency to generate and mail the stipulation agreements.Upon receiving the stipulation agreement(s), the NCP (and CP) had to sign andreturn the forms to the agency, and only then did the agency formally enroll theNCP and enable him or her to begin to receive debt reduction.

In practice, these steps to enrollment were sometimes difficult to navigate, andmany eligible, experimental NCPs failed to complete the process. For example, weknow that some NCPs (and CPs) did not receive or read the invitation letter andwould therefore not be aware of the opportunity to participate. We also know thatothers who read the letter and made some effort to enroll were unsuccessful in theirefforts to reach the appropriate contact at the child support agency.11 In addition,CP cooperation was an obstacle to enrollment for some NCPs with family-oweddebt, and some NCPs who had family-owed debt and state-owed debt were notenrolled to receive forgiveness on state-owed arrears after their unsuccessful effortsto secure CP participation. Furthermore, a follow-up survey (with a sample of eligi-ble NCPs, described in the Appendix12) found that some NCPs did not followthrough with the entire enrollment process, as they mistakenly thought that theyhad already been enrolled. The end result was a low take-up rate, an issue we fur-ther discuss.

Analytic Sample and Data

The sample used in the analyses of pilot program impacts consists of 531 NCPsattached to 1,976 eligible IV-D cases (in the sample frame) who contacted theRacine County Child Support Agency and expressed a willingness to participate inthe Families Forward program. Within this analytic sample, 376 (71 percent) wereexperimental cases, and the other 152 were assigned to the control group. Of thesewilling NCPs, two originally assigned to the control group participated in FamiliesForward, along with 118 others, for a total of 120 participants. Of these 120 partic-ipants, nearly three-quarters (88) received forgiveness on state-owed arrears only,25 received forgiveness on family-owed arrears only, and 7 received forgiveness onboth state- and family-owed arrears.

The Kids Information Data System (KIDS), used by BCS for tracking child sup-port orders and payments, provides the primary data for the impact analysis. TheKIDS data contain information on child support orders, payments, receipts, andarrearages; the method of payment and destination of the payment (family, state);and demographic information about the parents and children in the cases (includ-ing birth dates, residential location of both parents, and dates of marriage, divorce,

10 The Racine County child support agency did not provide NCPs with assistance in locating or contact-ing CPs. In order to gain CP consent, an NCP either had to personally persuade the CP to participate orthe CP may have proactively contacted the child support agency to give consent (perhaps in response toan invitation letter from IRP).11 At the start of Families Forward, there was a designated phone line at the Racine County child supportagency that eligible NCPs could call; however, this line was discontinued in early 2006, and all calls werererouted to a Milwaukee call center where only 5 percent of phone calls were being answered. One NCP,who participated in a focus group, described his experience in trying to get enrolled as follows: “Well, like[name], same thing: call, call, call, call, nothing happened. And I finally went down to the child supportoffice, and they said, you have to call this number. I said, ‘But nobody answers this number.’ ”12 All appendices are available at the end of this article as it appears in JPAM online. See the completearticle at wileyonlinelibrary.com.



and paternity establishment). These data were merged with UnemploymentInsurance (UI) Wage Record Files, which track the wages of workers eligible forstate unemployment insurance.13 Measures constructed with these data (e.g.,employment and earnings histories and other characteristics of parents) are used ascontrols in the nonexperimental analysis of program impacts.

The outcome measures, constructed from the KIDS data, assess how well theFamilies Forward program was meeting its goals of reducing state- and family-owed debts while simultaneously increasing NCP payments toward current supportand arrears. These primary outcomes of interest include the following: (1) thechange in the family child support debt balance (from the month of program enroll-ment to the final month of participation); (2) the change in the state child supportdebt balance (from the month of program enrollment to the final month of partic-ipation); (3) the change in the average monthly (amount of) payments made by theNCP toward current support or debt accounts; (4) the change in the percent ofmonths and change in probability that any payment was made by the NCP towardcurrent support or debt accounts; (5) the change in the percent of months andchange in probability that any payment toward family arrears was made; and (6) thechange in the percent of months and change in probability that any paymenttoward state arrears was made. For each of these outcomes, program impacts aremeasured as “differences in differences”; that is, the difference in the pre- to post-intervention changes in outcomes for program participants compared to changes inthese outcomes for the control or comparison groups of nonparticipating NCPs. Forthe control and comparison group members, the “pseudo-participation” period overwhich changes in their outcomes is measured begins with their first contact, inwhich they expressed their interest in participating in the program and continuesfor the ensuing 24 months (given the two-year time limit on participation).14 ForFamilies Forward participants, the participation period starts with their enrollmentdate and ends with their program termination date, which is two years later formost participants. In the multilevel models, the structure of the data and models(with time nested within individual NCPs) allows us to use all months prior to themonth of program enrollment (or first contact for comparison group members)going back to January 2004, as well as months following their participation, tomeasure changes in average monthly payments made by NCPs toward current sup-port or arrears.

STUDY METHODS

Experimental Methods

The primary advantage of an experimental approach that randomly assigns participa-tion in the intervention is that it assures that participation in the intervention is theonly factor that differs between those in the treatment group and those excluded

13 In the IRP match of NCP KIDS records with UI wage records, UI earnings information was missing forapproximately 19 percent of NCPs. Generally, exceptions from UI wage reporting total about 9 percentand include the self-employed, federal employees, church employees, employees of small nonprofitorganizations, and wages that are earned outside of Wisconsin. It is possible that the higher rate of missinginformation in this sample reflects more NCPs out of the labor force or working in the informal sector,or more NCPs working outside of Wisconsin, given Racine’s proximity to the Wisconsin–Illinois border.In light of this, we performed sensitivity tests in our impact estimation using measures with (and with-out) the missing values in earnings recoded to zeros. The recoding had negligible implications for theestimation results.14 The “pseudo-participation” period is defined in the following ways. For the first two measures ofchanges in family and state debt balances, we simply use the balance in the month of enrollment (or firstcontact for NCPs in the control and comparison group) as the pre-program measure. For the other out-come measures (focusing on NCP payments), we compare changes in the outcomes during the participa-tion period to changes in payments during all months prior to the NCP’s program enrollment month (orfirst contact month if in the control and comparison group), going back to January 2004.



from participating (the control group). Using standard terminology, we denote Y1 asthe outcome for an individual assigned to the experimental group and willing toparticipate in the program, Y0 as the outcome for that individual over the same peri-od in the absence of participation, and D � 1 for those in the treatment group andD � 0 for those in the control group. With random assignment assuring that treat-ment status is independent of Y0 and Y1 and factors influencing them (X), we calcu-late the program impact as the average difference in outcomes between treatmentand control group members (Y1 � Y0):

E(Y1 � Y0 � D � 1) � E(Y1 � D � 1) � E(Y0 � D � 0) (1)

The first two columns of Table 1 show that with regard to individual characteris-tics and key measures of interest in this study—total family debt balances, statedebt balances, and recent payment histories at program start—the 376 willingexperimental NCPs who contacted Racine County to sign up for the FamiliesForward program were statistically equivalent to the 150 control group memberswho also expressed an interest in participating but did not enroll due to their con-trol status. Thus, we calculate “intent to treat” impacts as the average difference inchanges in outcomes between the 376 experimental NCPs and the 150 controlgroup members.

However, all willing NCPs with experimental status did not complete all steps toenroll in the Families Forward program, and those who did enroll were not statis-tically equivalent to willing NCPs assigned to the control group. The third columnin Table 1 presents descriptive statistics on actual participants in the pilot program.Comparing actual participants with NCPs assigned to the control group, statistical-ly significant differences are evident for NCP age and for two key characteristicsconcerning the program intervention (family and state debt balances prior to pro-gram start). As random assignments to the treatment and control groups were madeprior to the mailing of invitations to participate in Families Forward, we are not

Table 1. Characteristics of noncustodial parents in Families Forward evaluation.

Experimental Control(Willing to (Willing to ExperimentalParticipate) Participate) Group and

Group Group ParticipatingNoncustodial Parent Characteristics (N � 376) (N � 150) (N � 120)

Age in years 37.9 38.2 40.5*Percent female 5.3 5.0 2.5Percent white 36.0 35.0 27.5Paternity case 70.0 68.6 73.1UI earnings in 2004 $4,970 $5,059 $4,630

(9,060) (11,822) (8,644)Number of children (across all cases) 3.2 3.1 3.5State child support debt balance, $10,823 $9,847 $16,907*

March 2005 (17,875) (18,227) (17,924)Family child support debt balance, $24,399 $21,087 $27,987*

March 2005 (24,013) (21,572) (25,502)Percent with no child support paid in 2004 11.4 11.3 10.8Number of months made payment in 2004 6.2 6.1 6.6

Notes: *A t-test or chi-square test confirmed that the difference between this value for participatingexperimentals and that for those in the control group was statistically significant at p � 0.05. Standarddeviations are in parentheses below the measures of earnings and state and family debt balances.



able to make the same assumption that participation15 in the program (which wewill call D�) is independent of factors influencing Y0. That is, calculating average dif-ferences in outcomes between actual participants and control group members willnot identify the impact of the “treatment on the treated.”

Nonexperimental Methods

The use of random assignment in the Families Forward program evaluation alsofacilitates the application of rigorous nonexperimental evaluation methods, whererandom assignment status contributes an exogenous source of variation in participa-tion that may be useful in estimating program impacts. Assuming that randomizationdoes not influence program outcomes, we use propensity score (differences-in-differences) matching (PSM) and include random assignment to treatment as anexogenous predictor variable in the first-stage estimation of the probability of pro-gram participation. As there is still debate in the literature as to whether or notmatching on covariates that satisfy instrumental variables assumptions improves orbiases impact estimation (Heckman & Navarro-Lozano, 2004; Wooldridge, 2009),we also estimated PSM models without using random assignment status as a con-trol variable.16

In applying matching methods, we invoke the conditional independence assump-tion, which implies that after controlling for observable characteristics (X), a per-son’s treatment status is unrelated to what his outcome would have been in theabsence of treatment (Rosenbaum & Rubin, 1983):

Y0 › D� � X (2)

The validity of this assumption depends largely on the set of variables (X) availablefor the estimation and how the comparison group is chosen. We expect that theremay be some unmeasured factors that influence program participation; what isimportant is that participation not be predictive of the outcome that would haveoccurred without the program. In addition, because our outcome variables aredefined as the difference between a pre-program and post-intervention measure, weuse a panel form of the matching estimator (difference-in-differences matching)that allows for time-invariant, unobserved differences between participants andnonparticipants without biasing estimates of program impacts. In estimating thismodel, we make the assumption that conditional independence holds for the peri-ods both before (t) and after (t�) treatment:

E(Y0t� � Y0t � D� � 1, X) � E(Y0t� � Y0t � D� � 0, X) (3)

Propensity score matching (PSM) is a two-step process in which we first estimatethe probability of participation based on the conditioning variables and then comparetreated and untreated NCPs with similar propensity scores (rather than requiringmatches on all of the X variables). Let P(X) be the probability that an individual withcharacteristics X is a participant, which we estimate with a probit function. We include

15 Participation in the Families Forward program is defined by the signing of a stipulation agreement thatformally enrolls the NCP, stops interest charging on debt, and begins providing credit toward debt basedon current support payments made.16 Another possible sensitivity test involves estimating a standard instrumental variables (IV) model andapplying the Bloom (1984) correction for the proportion of experimental group members who enrolledin the program. However, program take-up in Families Forward was low (32 percent of willing experi-mental NCPs participated), and although first-stage results confirm that random assignment is a statis-tically significant predictor of participation, the F-statistic suggests that it is a relatively weak instrument.Staiger and Stock (1997) show that weak instruments may produce severely biased estimates of programimpacts.



all eligible NCPs who expressed an interest in participating in Families Forward,whether assigned to treatment or control status. This allows us to use randomassignment treatment to introduce exogenous variation into this model.17 Asexpected, random assignment to the experimental group was the most influentialpredictor of participation in Families Forward. Other statistically significant pre-dictors of NCPs’ participation included age, race, earnings in the year prior to pro-gram start, and child support payment frequency prior to program start.

The particular PSM matching technique we apply in the second stage model isradius matching, which specifies a “caliper” or maximum propensity score distance(0.01 in our analysis) by which a match can be made. It uses not only the nearestneighbor within each caliper, but all comparison cases within the caliper (based onthe specified distance), and the common support condition is imposed to excludepoor matches from the analysis.18 We also employ a nearest-neighbor matchingprocedure with bias adjustment, which was recently developed in response to con-cerns first articulated by Abadie and Imbens (2002) that even after removing theconditional bias, matching estimators with a fixed number of matches may notreach the semiparametric efficiency bound for average treatment effects. Theirapproach implements a bias-correction that removes the conditional bias asymptot-ically (Abadie et al., 2004).

We also take advantage of the longitudinal, hierarchical structure of our data (withobservations over 72 months nested within individuals) to estimate multilevelgrowth curve models of changes over time in the outcome variables and their rela-tionship to the timing of events (such as expression of interest in the program andthe start and length of program participation), while controlling for other time-varyingand stable characteristics. This modeling approach allows both the number of meas-urement occasions and their timing to vary across individuals (i.e., they do not needto be balanced in the sample). For example, NCPs responded to the opportunity toparticipate in Families Forward over a period of 33 months, so that the timing of keyevents (e.g., their first contact and dates of program enrollment for participants) variedconsiderably. As the timing of these events determines the pre- and post-interventionperiods for NCPs, it is important to accurately capture this in modeling individualoutcomes over time. We also assume that NCPs’ responses to these events are on thesame temporal cycle as everything else that might affect patterns in the program out-comes (and for which we do not have controls in the model).

Our choice of model specification is informed largely by our contextual knowl-edge of child support and related programs, gained through the literature and ourobservation of the implementation of Families Forward. For example, we know thatthe debt balances of most NCPs in our sample are growing over time due to stateinterest charges, but as participation in Families Forward stops interest charging,we expect a change in this growth rate for participants. We also expect the newincentives offered by the program to encourage NCPs to pay more support oncethey are enrolled, although we also anticipate possible lags in NCP responses tothese new incentives, particularly if prospective participants hold off on makingmore or bigger payments (after their first contact to express interest in the

17 One concern in doing this is that experimental NCPs who were eligible and willing to participate butdid not enroll might be significantly different from those who were eligible and participated. We there-fore checked for statistically significant differences between the characteristics of these two groups ofNCPs. Similar to the comparison of participants and controls, these NCPs differ in their age (participantsare older) and in state debt balances (participants have higher state debt balances than experimentalNCPs who did not enroll). As a sensitivity test, we estimated the PSM models on the subsample includ-ing only participants (n � 120) and controls (n � 150), excluding the experimental NCPs who did notparticipate. The substantive conclusions from these analyses were the same; indeed, including all exper-imental NCPs (who expressed an interest in participating) tended to produce slightly more conservativeestimates of program impacts.18 Standard errors are calculated using bootstrapping procedures.



program) until they are enrolled and getting extra credit toward their debt for eachdollar paid. Finally, we anticipate possible drop-offs in payments over time, as par-ticipants exhaust financial reserves or reach the end of program participation.

In the multilevel models we estimate, Yti is the response at measurement occasiont(t � 1, . . ., Ti) for individual i(i � 1, . . ., n). At level one, we model the average slopein pre- and post-program (or pre- and post-contact) periods (p1i and p2i) with timetrend variables, and we use indicator variables to capture shifts (increments ordecrements) in outcomes (p3i and p4i) at specific times after the first contact and thethree-month periods (March to May) each year when tax returns are typicallyreceived (p5i to p9i):19

Yti � p0i � p1iTime_preti � p2iTime_postti � p3i1st3mos_postti

� p4i2nd3mos_postti � p5itaxtime05ti � p6itaxtime06ti (6)

� p7itaxtime07ti � p8itaxtime08ti � p9itaxtime09ti � rti

At level two, the intercept (p0i, where time � 0, i.e., initial status) from the level-one model is specified as random and a function of individual characteristics(demographic, number of children, monthly earnings, whether a current supportorder is in effect, and their cohort as defined by timing of first contact) and treat-ment status (di � 1 for participants):

p0i � b00 � b01di � b02Xi(1) � . . . � b0nXi(n) � u0i (7)

The time trend (growth rate) variables and indicators for capturing post-contactshifts (p1i to p4i) are also specified as random and a function of treatment status;for example:

p1i � b10 � b11di � u1i (8)

and correspondingly for p2i to p4 i. The random effects equations (p1i to p4i) modelthe effects of the program on the outcome trajectories of participants (relative tothose who were eligible and expressed an interest in the program but did not participate); p5i to p9i are modeled as fixed effects. These level-one and level-twomodels are estimated simultaneously, with an autoregressive error structure (with-in-person error covariance) of lag one, AR(1).20

ANALYSIS FINDINGS

Table 2 presents the findings of experimental, intent to treat calculations of pro-gram impacts (comparing interested NCPs assigned experimental status with thoseassigned to the control group); the simple difference in outcome calculations com-paring actual participants with the control group; and the propensity score andnearest-neighbor matching estimation results for the six outcome measuresdescribed earlier. Looking at the first column in Table 2, which presents the resultsof experimental calculations of Families Forward program impacts, one sees that

19 Research by Kathryn Edin and colleagues at Harvard University suggests higher levels of consumptionand debt reduction in low-income households during the three months in which tax returns and creditsare typically received. The tax time indicators are the same calendar time periods for all individuals inthe sample, while calendar time of the other indicators differs across individuals according to theirmonth of enrollment (or for nonparticipants, their first contact to express an interest) in FamiliesForward.20 We tested a standard (variance components) specification, as well as six other error covariance struc-tures that other researchers (Singer & Willett, 2003) have found most useful (unstructured, compoundsymmetric, heterogeneous compound symmetric, autoregressive, heterogeneous autoregressive, andToeplitz), and concluded that the AR(1) structure was the best fit.

Tab

le 2

.Im

pac

t es

tim

ates

fro

m e

xper

imen

tal

and

mat

chin

g m

eth

od a

nal

yses

.

Dif

fere

nce

:D

iffe

ren

ce:

PS

M M

eth

ods

Exp

erim

enta

ls

Par

tici

pan

ts

Rad

ius

Cal

iper

NN

M

(n�

376)

– C

ontr

ols

(n�

120)

– C

ontr

ols

(120

par

tici

pan

ts v

s.(n

�15

0)(n

�15

0)41

0 n

ot p

arti

cip

atin

g)

�in

deb

t b

alan

ce:

enro

llm

ent

mon

th t

o p

rogr

am e

nd

mon

thS

tate

-ow

ed�

$353

.36

�$1

,732

.54

�$

2,7

42

.70

�$

2,8

22

.92

(512

.44)

(678

.14)

(643

.00)

(681

.96)

Fam

ily-

owed

$840

.43

�$1

,054

.98

�$

2,5

64

.20

�$

2,3

29

.84

(948

.27)

(117

7.33

)(1

047.

32)

(108

9.90

)

�in

ave

rage

mon

tly

pay

men

t�

$14.

93$2

3.85

$7

0.4

4$

70

.72

(24.

26)

(32.

73)

(29.

49)

(33.

41)

�in

% m

onth

s m

ade

any

pay

men

t0.

012

0.06

40

.06

50.

065

(0.0

30)

(0.0

35)

(0.0

32)

(0.0

53)

�in

% m

onth

s m

ade

stat

e ar

rear

s 0.

003

0.1

71

0.2

29

0.2

34

pay

men

t(0

.023

)(0

.035

)(0

.034

)(0

.042

)

�in

% m

onth

s m

ade

fam

ily

arre

ars

0.01

50.

069

0.0

83

0.1

24

pay

men

t(0

.031

)(0

.041

)(0

.040

)(0

.058

)

Pro

pen

sity

sco

re m

atch

ing

(PS

M)

met

hod

s ar

e ra

diu

s ca

lip

er a

nd

nea

rest

nei

ghb

or m

atch

ing

(NN

M).

Exp

erim

enta

l ca

lcu

lati

ons

(N�

526)

exc

lud

e tw

o N

CP

s or

igin

ally

ass

ign

ed t

o th

e co

ntr

ol g

rou

p w

ho

sub

sequ

entl

y b

ecam

e p

arti

cip

ants

. M

atch

ing

and

OL

S e

stim

atio

ns

(N�

530)

in

clu

de

the

two

NC

Ps

orig

inal

ly a

ssig

ned

to

the

con

trol

gro

up

wh

o su

bse

quen

tly

bec

ame

par

tici

pan

ts,

as w

ell

as t

wo

oth

er N

CP

s w

ho

wer

e en

roll

ed b

ut

wer

en

ot o

rigi

nal

ly a

ssig

ned

to

eith

er t

he

exp

erim

enta

l or

con

trol

gro

up

.

Not

es o

n s

ensi

tivi

ty t

ests

:1.

Mat

chin

g sp

ecif

icat

ion

s th

at d

id n

ot i

ncl

ud

e ex

per

imen

tal

stat

us

as a

pre

dic

tor

vari

able

in

th

e fi

rst

stag

e m

odel

(of

par

tici

pat

ion

in

Fam

ilie

s F

orw

ard

) p

rod

uce

d e

sti-

mat

ed i

mp

acts

th

at w

ere

very

clo

se o

r sl

igh

tly

smal

ler.

For

exa

mp

le,

the

esti

mat

e fo

r ch

ange

s in

sta

te d

ebt

bal

ance

s w

ith

out

exp

erim

enta

l st

atu

s as

a f

irst

-sta

ge m

odel

pre

dic

tor

is �

$2,7

41.5

4 (s

tan

dar

d e

rror

�67

7.61

), a

dif

fere

nce

of

only

$1.

16.

2. M

atch

ing

esti

mat

ion

s in

clu

din

g on

ly p

arti

cip

ants

an

d c

ontr

ols

pro

du

ced

est

imat

es t

hat

wer

e ty

pic

ally

clo

se t

o th

ose

rep

orte

d a

bov

e, a

nd

dep

end

ing

on t

he

outc

ome,

wer

e sl

igh

tly

larg

er o

r sm

alle

r in

mag

nit

ud

e. F

or e

xam

ple

, th

e es

tim

ate

for

chan

ges

in s

tate

deb

t b

alan

ces

com

par

ing

par

tici

pan

ts w

ith

con

trol

s on

ly w

as �

$3,0

30.4

7(s

tan

dar

d e

rror

�79

8.00

), a

dif

fere

nce

of

$287

.77.

Alt

ern

ativ

ely,

th

e es

tim

ated

ch

ange

in

th

e p

erce

nt

of m

onth

s p

arti

cip

atin

g N

CP

s m

ade

pay

men

ts t

owar

d f

amil

yar

rear

s (c

omp

ared

to

con

trol

NC

Ps)

was

0.0

72,

a d

iffe

ren

ce o

f 0.

011,

wh

ich

su

gges

ts a

sm

alle

r im

pac

t.



there are no statistically significant differences in the changes in outcomes betweenthe experimental and control group NCPs. For example, the first estimated differ-ence shows that the state debt balances of experimental NCPs declined more thanthose of control group NCPs (by $353) during the period of participation (or the 24months after first contact for the control group NCPs), but the difference in thechange in outcomes was not statistically significant. Indeed, the standard errors ofall of the experimental differences are large, suggesting that we cannot draw con-clusions about the direction of these impact estimates. This was not unexpected,given that only 32 percent of the experimental NCPs were actually getting debtreduction credits through program participation. Comparing only participatingNCPs with the control group, without adjusting for known differences in their char-acteristics, the pattern of program effects is noticeably different (see column 2 inTable 2). The magnitude of estimated effects is larger, and all are in the directionexpected if the program incentives were working (i.e., NCP debt balances decreas-ing and payments increasing), although only the change in the frequency of pay-ments toward state arrears is statistically significant.

The next two columns of results present the nonexperimental estimates of treat-ment on the treated program effects that adjust for pre-intervention differencesbetween actual participants and comparison group members. The third columnshows the estimated effects from PSM radius matching, and the fourth column pres-ents results from nearest-neighbor matching analyses (NNM). After matching, theestimated impacts are considerably larger; all but one of the differences between par-ticipants and comparison group members is statistically significant, and the radiusand NNM results are very close in magnitude. Specifically, these results show thatstate-owed debt balances of participating NCPs declined by more than $2,700 morethan the state-owed debt balances of nonparticipating NCPs from the month ofenrollment (or first contact) to their last month of program participation (or approx-imately 24 months later). Similarly, family-owed debt balances of participating NCPsdeclined by more than $2,500 more than those of nonparticipating NCPs (again,between the month of enrollment and their last month of program participation).The PSM models also show that participating NCPs paid on average $70 per monthmore during the months of their participation than during the months prior to theirenrollment (compared to nonparticipating NCPs). In addition, the frequency (or per-cent of months that they were paying) increased by 6.5 percent more (than nonpar-ticipants) for any type of payment, by 8.3 percent more for payments toward familyarrears, and by almost 23 percent more than nonparticipants for payments towardstate-owed debt, compared to their rate of payments in 2004. This latter finding isnot unexpected, given that a majority of enrolled NCPs received forgiveness on state-owed arrears only. In addition, reductions (or slower growth) in debt balances werealso expected, in part because interest charging was stopped during program partic-ipation. Results of after-matching balancing tests (available from the authors)showed that there were no statistically significant differences between participantand nonparticipant characteristics after matching, with reductions in bias rangingfrom 31 to 99 percent, and there was also no loss of observations due to the imposi-tion of a common support. The notes on sensitivity tests at the bottom of Table 2indicate that differences in estimated effects associated with excluding experimentalstatus as an exogenous predictor variable in the first stage of PSM and NNM analy-ses were small, as were differences when including only NCPs randomized out ofparticipation (in the control group) as comparison cases. In all sensitivity tests, thesubstantive conclusions from the estimation were unchanged.

Multilevel Longitudinal Model Results

The multilevel longitudinal models were estimated for four payment outcomes(average monthly payment made, the probability of making a payment in a given



month, and the probability of making a payment toward state-owed debt or familydebt); results from these models are presented in Table 3, with the key parameter ofinterest highlighted in bold.21 Looking at the statistically significant findings in col-umn 1, the first two variables suggest that in the months before NCPs made contactto enroll in the program, as well as in the months after that first contact, NCP pay-ments were increasing by only $2 and $1 per month, respectively. The FamiliesForward participation coefficient estimate suggests that across the months NCPswere participating in the program, on average, they paid $105 more per monthtoward their current support or debt than NCPs who were not participating inFamilies Forward, a fairly dramatic increase.22 However, to fully interpret this keyestimate of interest in the impact evaluation, the coefficient for participation inFamilies Forward, we also need to assess the interactions of this variable with thepost-contact/enrollment growth rates and cohort indicators. The participant growthrate just below the Families Forward participation variable (i.e., the interactionbetween the post-contact growth rate and the participation indicator) was intendedto measure differences in the rate of change in payments by participants (vs. non-participants) following their first contact, although this coefficient estimate is notstatistically significant. Likewise, the interactions between participation and thefirst three months post-contact and second three months post-contact, included tocapture possible delays in participant responses to the program incentives as theywaited for stipulations to become effective, are not statistically significant.

Looking to the interactions with cohort indicators, only one interaction, withcohort 3, is statistically significant (and negative), suggesting that the impact wassignificantly smaller for participants who began the program in the first or secondquarters of 2006. Specifically, NCPs in the third cohort paid approximately $53 moreper month toward their current support or debt than NCPs in other cohorts, but theNCPs who were participating in Families Forward paid just $10 per month morethan nonparticipants in the third cohort, (i.e., $104.56 � $52.86 � $94.73 � $62.29),so the differential impact is smaller.23 The results in column 1 also show that NCPspaid more per month during the months in which tax returns arrive each year, andthose with a current support order, with more children and higher monthly earnings(and higher annual earnings in 2004) also paid significantly more each month.

The model results in the other three columns in Table 3 show the probability thatan NCP made a payment in a given month (any payment or specifically towardstate- or family-owed debt). It is important to note that before any payments areapplied to arrears, the money paid by an NCP first goes toward current support;that is, current support has to be paid in full before any part of the payment isapplied to arrears (if the NCP has a current support order). The results in column 2suggest that the probability that any NCP made a payment after first contact toenroll (the post-contact growth rate) was increasing by just 0.01 percent, while thecoefficient for the probability of making any payment during program participationsuggests an increase of 3.5 percent for participants, although this is not statisticallysignificant. Alternatively, the coefficients showing the probability of making pay-ments toward state-owed arrears and family-owed arrears for Families Forward

21 The random effects components in all four outcome models confirm that there is statistically signifi-cant variation in both intercepts and slopes that is usefully explained by covariates at the NCP level.22 Any extra credit toward debt received by participating NCPs (the program incentive) was not includedin this payment measure.23 The first two quarters of 2006 encompassed the period of program implementation with the mostextensive administrative problems. As discussed earlier, the direct line at the county child support agencyfor enrolling in the program was unexpectedly discontinued, and one of the agency staff who was new tothe program was turning away eligible NCPs with a poor payment history, contrary to the rules and intentof the program. Given that these problems were subsequently corrected, it is not surprising that the thirdcohort of participating NCPs may have been different or have had a different experience in the programthan the others.

Tab

le 3

.M

ult

ilev

el l

ongi

tud

inal

est

imat

es o

f F

amil

ies

For

war

d p

rogr

am i

mp

acts

.

3. M

ade

Pay

men

t4.

Mad

e P

aym

ent

1. M

onth

ly2.

Mad

eon

Sta

te-O

wed

on F

amil

y-O

wed

Ou

tcom

e M

easu

reP

aym

ent

Am

oun

tA

ny

Pay

men

tA

rrea

rsA

rrea

rs

Std

.S

td.

Std

.S

td.

Pre

dic

tor

Coe

f.E

rror

Coe

f.E

rror

Coe

f.E

rror

Coe

f.E

rror

Inte

rcep

t�

220.

3747

.743

2*�

0.15

70.

0676

*�

0.15

740.

0343

5*�

0.09

00.

053

Gro

wth

rat

e b

efor

e fi

rst

con

tact

to i

nit

iate

2.

180.

85*

0.00

370.

0007

*0.

003

0.00

1*0.

007

0.00

1*p

rogr

am e

nro

llm

ent

Pos

t-co

nta

ct g

row

th r

ate

1.26

0.41

*0.

001

0.00

0*0.

000

0.00

0*0.

002

0.00

0*F

irst

3 m

onth

s p

ost-

con

tact

in

dic

ator

8.72

25.6

90.

015

0.00

9�

0.00

50.

007

0.00

60.

012

Sec

ond

3 m

onth

s p

ost-

con

tact

in

dic

ator

�18

.82

25.8

4�

0.00

10.

010

�0.

012

0.00

70.

012

0.01

2M

arch

–May

05

47.7

122

.11*

0.01

20.

008

0.01

40.

006*

0.01

40.

010

Mar

ch–M

ay 0

637

.12

22.5

90.

011

0.00

80.

005

0.00

60.

014

0.01

0M

arch

–May

07

61.9

321

.86*

0.01

40.

008

0.00

00.

006

0.00

50.

010

Mar

ch–M

ay 0

846

.64

21.8

0*0.

005

0.00

80.

002

0.00

60.

004

0.01

0M

arch

–May

09

72.6

222

.55*

0.01

80.

008*

0.01

40.

006*

�0.

010

0.01

0P

artc

ipat

ing

in F

amil

ies

Fo

rwar

d1

04

.56

40

.58

*0

.03

50

.02

30

.13

00

.01

7*

0.1

09

0.0

28

*P

arti

cipa

nt

post

-con

tact

gro

wth

rat

e�

2.31

1.30

0.00

00.

001

�0.

002

0.00

1*�

0.00

20.

001*

Par

tici

pan

t fi

rst

3 m

onth

s po

st-c

onta

ct�

96.2

059

.28

�0.

020

0.02

4�

0.05

20.

018*

�0.

034

0.02

9P

arti

cipa

nt

seco

nd

3 m

onth

s po

st-c

onta

ct4.

7157

.60

0.02

30.

021

�0.

002

0.01

60.

000

0.02

7W

hit

e70

.70

15.6

4*�

0.01

70.

023

�0.

020

0.01

10.

065

0.01

7*N

onw

hit

e45

.33

20.7

9*�

0.02

20.

030

�0.

021

0.01

50.

008

0.02

3M

ale

36.3

130

.08

0.03

00.

043

0.04

80.

022*

0.01

50.

033

Age

in

yea

rs4.

130.

85*

0.01

00.

001*

0.00

40.

001*

0.00

50.

001*

Nev

er m

arri

ed�

23.5

620

.75

�0.

041

0.03

0�

0.01

30.

015

�0.

041

0.02

3L

og o

f 20

04 e

arn

ings

2.53

1.12

*�

0.00

50.

002*

0.00

10.

001

0.00

00.

001

Nu

mb

er o

f ch

ild

ren

17.7

13.

27*

0.01

10.

005*

0.00

70.

002*

�0.

006

0.00

4M

onth

ly e

arn

ings

9.64

1.00

*0.

007

0.00

0*0.

003

0.00

0*0.

010

0.00

1*C

urr

ent

sup

por

t or

der

266.

1610

.75*

0.84

30.

004*

0.04

90.

003*

0.25

20.

005*

Coh

ort

2: f

irst

con

tact

Ju

ly–D

ec.

0521

.34

19.6

20.

022

0.02

60.

015

0.01

40.

022

0.02

1C

ohor

t 3:

fir

st c

onta

ct J

an.–

July

06

52.8

620

.30*

0.01

30.

027

0.01

30.

014

0.03

00.

022

Coh

ort

4: f

irst

con

tact

Ju

ly–D

ec.

0668

.77

22.9

5*0.

061

0.03

20.

033

0.01

6*0.

075

0.02

5*C

ohor

t 5:

fir

st c

onta

ct i

n 0

767

.14

34.5

8�

0.00

60.

062

0.10

80.

027*

0.13

60.

044*

Coh

ort

2 pa

rtic

ipan

ts�

42.6

244

.69

�0.

036

0.02

6�

0.04

00.

020*

�0.

013

0.03

1C

ohor

t 3

part

icip

ants

�94

.73

45.7

1*0.

043

0.03

00.

041

0.02

20.

037

0.03

5C

ohor

t 4

part

icip

ants

�79

.67

59.5

7�

0.02

60.

041

0.02

60.

031

�0.

134

0.04

8*C

ohor

t 5

part

icip

ants

�75

.57

107.

23�

0.08

50.

073

0.10

80.

053*

0.11

40.

084

Not

es:

Key

par

amet

er o

f in

tere

st i

s h

igh

ligh

ted

in

bol

d.

* in

dic

ates

coe

ffic

ien

t is

sta

tist

ical

ly s

ign

ific

ant

at a

�0.

05.

Ital

iciz

ed v

aria

ble

s ar

e in

tera

ctio

ns

bet

wee

n a

par

tici

pan

t in

dic

ator

an

d o

ther

cov

aria

te i

n t

he

mod

el.

N�

530.



participants suggest large, statistically significant responses to program incentives.For all NCPs, there was no change (0 percent growth rate) in the probability of mak-ing a payment toward state arrears after their first contact to enroll, but the proba-bility that Families Forward participants made payments toward their state-owedarrears in a given month was 13 percent higher during participation. Likewise, forall NCPs, the probability of making payments toward family-owed debt wasincreasing by just 0.02 percent in the months following their first contact to enroll,whereas the probability that Families Forward participants made payments towardtheir family-owed debt was 11 percent higher on average (than for nonpartici-pants). The larger estimated effects for payments toward arrears suggest that theprogram impact is probably driven less by differences in rates of payments towardmonthly current support owed, and more so by the increase in participating NCPswho were paying above and beyond their current support order (relative to nonpar-ticipants). These findings affirm those of the PSM models that showed significant-ly higher frequencies of paying debt (especially state-owed debt) among FamiliesForward participants.

In summary, the comparisons of NCPs participating in Families Forward to non-participants (using propensity score and nearest-neighbor matching and multilevellongitudinal methods) showed important effects of the program: Program partici-pants made larger payments toward current child support and debt balances, weremore likely to make payments and to pay more frequently, and reduced their state-and family-owed debt balances. Overall, we conclude that the consistency of theestimates (from rigorous differences-in-differences estimators and subject to multi-ple sensitivity tests) combined with a relatively long period of observation (i.e., 72months) suggests promise for this pilot program model in future state efforts toreduce child support arrears.

CONCLUSION AND POLICY IMPLICATIONS

The promising results for NCPs who participated in the Families Forward pilot pro-gram raise the question of whether or not they would extend to other NCPs withlarge debt burdens. It is possible that these results, which statistically adjust for theselective nature of program take-up, might not reflect the true potential of the pro-gram model to increase child support payments and reduce arrears across the pop-ulation of NCPs with poor payment histories. The experimental estimates, whichcombined both treatment effects for participants and null effects for the 68 percentof experimental cases who did not participate, showed no statistically significanteffects. If participation remained low in future implementations of the program, ashas been the experience in other debt forgiveness programs, then a majority ofNCPs would be unaffected by the program.

The nature of selective take-up in the Families Forward program was that NCPswith larger debts and who were older (likely because it takes time to accumulatevery large debts) were more likely to participate in the program. Although other dif-ferences were not statistically significant, participants generally appeared more dis-advantaged (e.g., lower earnings, more likely to be nonwhite). We also suspect thatthere was some degree of chance (randomness)—outside of whether NCPs wereassigned experimental or control status—in their likelihood of enrolling, given thatonly 5 percent of the call center calls (essential for enrolling) were being answered(after the switch to the call center at the end of 2005). For these reasons, we are lessconcerned about the typical selection problem where more advantaged personsprogress to participation, implying that the same effects might not be attainable forthose who are less advantaged.

In addition, low take-up is a problem common to many important social programinterventions. An MDRC survey (Wallace, 2002) reported that relatively small per-centages of eligible low-income families avail themselves to work supports such as



education and training (23 percent) and food stamps (43 percent), and studies ofhousing programs have shown that among the 77 classes of interventions, take-uprates never rise above 25 percent and are lowest among the poorest eligible house-holds (Currie, 2006). Still, the Families Forward pilot program experience shouldencourage policymakers to address two questions before moving forward with sim-ilar programs (and the planned expansion in Wisconsin). First, what steps might betaken to increase awareness of the opportunity to participate and eligible parents’understanding of the program? Second, what modifications might be made to pro-gram processes to increase enrollment among eligible, interested parents?

In the Families Forward pilot program, responses indicating willingness to partici-pate were received from approximately 40 percent of NCPs who were sent an invita-tion. Many of the letters were returned from the post office as undeliverable or withalternate addresses (so many that we stopped tracking them over time). Other NCPshad prison addresses, and letters without prisoner identification numbers werereturned. In focus groups, a number of NCPs admitted their fear that the letter of invi-tation to participate in the pilot program was actually a set-up for a child supportsting operation; that is, they had taken a risk (of possible imprisonment) in respond-ing to the invitation. In summary, our experience in the pilot program implementationsuggests that communication with and outreach to NCPs is challenging, especially forNCPs who have not recently been cooperating with the child support system.

Future programs should strive for increased outreach and an expanded scale toinvolve greater numbers of highly indebted NCPs. The low response rate of eligibleNCPs who were mailed a letter belied high levels of interest among those who becameaware of the program—over 82 percent of surveyed NCPs said they were immediate-ly willing to sign up for Families Forward upon hearing about it. Although proactivecommunication on the part of child support agency staff is most important, beyondthe child support agency, we found that promotional materials distributed through-out the community (e.g., in libraries, unemployment offices, churches), including inlanguages other than English as locally appropriate, were effective not only inincreasing awareness, but also in substantiating the veracity of the program. Focusgroup findings also suggested the importance of using advertisements in media suchas radio, newspapers, and the Internet, along with testimonials from parents whowere in the program, to allay fears that it might be a sting operation. Partnershipswith other governmental and community organizations could also increase knowl-edge about the program. During the pilot program, prison social workers and men’sand women’s shelter staff, who often had direct contact with potential enrollees,inquired about Families Forward and were enthusiastic about providing informationabout the program and encouraging clients to enroll.24 Child support workers andfocus group participants also recommended providing information at relevant courtproceedings (e.g., paternity, child support order, or custody hearings).

A second way to increase participation among eligible NCPs is to reach out to thecustodial parents. The Families Forward pilot program was relatively unique in allow-ing reductions of both state- and family-owed debt. Yet comparatively fewparticipants received debt reduction on family-owed debt (only 34 CPs signed stipu-lation agreements to participate). In focus groups, some CPs indicated that they expe-rienced the same challenges as NCPs in attempting to reach Racine County ChildSupport to find out more about the program and to enroll. Others did not have currentcontact information for the NCP. This presents a real obstacle to reducing family-owedarrears balances. Focus groups with CPs, conducted in the early stages of the pilotprogram design, suggested two means of increasing CP participation. First, outreachmust clearly state what the CP might reasonably hope to gain from participating 24 Coordination with prisons and jails could be especially valuable, as many NCPs in the target popula-tion are likely to have been incarcerated at some point, possibly for child support nonpayment. In oursurvey sample, over 80 percent of NCP respondents said they had been incarcerated at least once sincethey were first ordered to pay child support.



(i.e., increased current support payments). One way of doing this would be to use tes-timonials from CPs who have benefited from the program. Second, contact with CPsshould be made through a third party such as a child support agency. Both NCPs andCPs in focus groups described several reasons why this might be more effective thanrelying on the parents to contact each other: (1) they often did not have current con-tact information for the other parent, (2) parents who had antagonistic relationshipswere concerned about heightening conflict,25 and (3) parents indicated that they hada hard time explaining the details of the program to the other parent. If CP participa-tion can be increased, the incentives for NCPs to make payments will likewiseincrease (as they receive additional credits for payments).

Another approach to increasing NCP participation is to improve communicationswith NCPs, both during and after the enrollment process. Survey results showedthat participants had difficulty ascertaining whether they were enrolled in FamiliesForward and whether it was working to reduce their arrears. For a conditional debtreduction program to maximize its impact, NCPs must be able to easily see howtheir payments link to reductions in their arrears. To this end, we recommend thatfuture implementations provide unambiguous notifications to NCPs that indicatewhen participation begins, and once enrolled, that statements include how muchcredit they received toward their arrears. This would help to avoid situations inwhich NCPs prematurely assume that they are enrolled and fail to complete the fullenrollment process, while making clear the monetary benefits that NCPs gainthrough their payments made while enrolled in the program.

Even if our expectation that aggregate participation levels could be increased isnot realized, the low take-up rates of many other social program interventions sug-gest that this alone is not a reason to discard the program, particularly in this case,when it appears to be reaching those most in need of its benefits. The pilot studyfindings offer considerable guidance for achieving greater take-up among eligiblepopulations in future implementations by increasing outreach to NCPs and CPs,improving the enrollment process for state-owed debt, and improving communica-tions during and after the enrollment process. Additionally, Racine County,Wisconsin, represented a particularly challenging test case for this conditional debtreduction program, as staff persons were reluctant to implement the program asdesigned, and the county faced exceptionally challenging economic conditions. Incounties or states that are more receptive to the principle of forgiving arrears,implementation might proceed more smoothly, making it easier for NCPs to enrollin the program (Brodkin, 2007). The State of Wisconsin is currently moving for-ward with a plan to implement a statewide version of this program—newly namedthe “Payment Incentive Program”—and the pilot program findings have been dis-seminated in national conference settings and are drawing the interest of otherstates as well. In this regard, the pilot program evaluation is already informingfuture policy decisions that could contribute importantly to reducing the persistent,high levels of child support debt in the United States.

CAROLYN J. HEINRICH is Sid Richardson Professor of Public Affairs and Affiliated Professorof Economics, Lyndon B. Johnson School of Public Affairs, the University of Texas at Austin.

BRETT C. BURKHARDT is Assistant Professor of Sociology, School of Public Policy,Oregon State University.

HILARY M. SHAGER is a PhD candidate in public policy, Wisconsin Department ofChildren and Families, University of Wisconsin–Madison.

25 We also recognize that there may be instances (e.g., parents with a history of domestic violence) whereit would be inappropriate to facilitate any contact between CPs and NCPs, and child support agenciestypically have the information necessary to make these types of judgments.



ACKNOWLEDGMENTS

This paper has been prepared under a contractual agreement between the WisconsinDepartment of Workforce Development and the Institute for Research on Poverty (IRP). HilaryShager’s work on this project was supported by the Institute of Education Sciences, U.S.Department of Education, through Award #R305C050055 to the University ofWisconsin–Madison. Any views expressed in this paper are those of the authors and not neces-sarily those of the sponsoring institutions. We thank the leadership and staff of the WisconsinBureau of Child Support, particularly Susan Pfeiffer and Carol Chellew, for their vision and sup-port of this project; we thank Racine County Child Support staff, whose extra efforts made thepilot program possible, including Pat Birchell-Sielaff, Karen Day, and Julie Busch; and we great-ly appreciate the support of IRP staff who contributed enormously to this project, especiallyMaria Cancian, Tom Kaplan, Ingrid Rothe, Lynn Wimer, Bill Wambach, and Dawn Duren, andthe research assistance of Samuel Hall, Katie Maguire, and Alan Paberz. In addition, we areespecially grateful to Dan Meyer at IRP and to anonymous referees for their review and com-ments on an earlier version of this manuscript.

REFERENCES

Abadie, A., & Imbens, G. W. (2002). Simple and bias-corrected matching estimators for aver-age treatment effects. NBER Working Paper No. T0283. Cambridge, MA: National Bureauof Economic Research.

Abadie, A., Drukker, D., Herr, J. L., & Imbens, G. W. (2004). Implementing matching estima-tors for average treatment effects in Stata. Stata Journal, 4, 290–311.

Argys, L., Peters, E., Brooks-Gunn, J., & Smith, J. (1988). The impact of child support on cog-nitive outcomes of young children. Demography, 35, 159–173.

Bartfeld, J. (2003). Forgiveness of state-owed child support arrears. Special Report No. 84.Madison, WI: Institute for Research on Poverty.

Bloom, H. S. (1984). Accounting for no-shows in experimental evaluation designs.Evaluation Review, 8, 225–246.

Brodkin, E. Z. (2007). Bureaucracy redux: Management reformism and the welfare state.Journal of Public Administration Research and Theory, 17, 1–17.

Cancian, M., & Meyer, D. R. (2004). Fathers of children receiving welfare: Can they providemore child support? Social Service Review, 78, 179–206.

Cancian, M., Heinrich, C. J., & Chung, Y. (2009). Does debt discourage employment and pay-ment of child support? Evidence from a natural experiment. Discussion Paper 1366-09.Madison, WI: Institute for Research on Poverty.

Currie, J. (2006). The take-up of social benefits. In A. Auerbach, D. Card, & J. Quigley (Eds.),Poverty, the distribution of income, and public policy (pp. 80–148). New York: RussellSage.

Grall, T. S. (2009). Custodial mothers and fathers and their child support: 2007. CurrentPopulation Report P60-237. Washington, DC: U.S. Census Bureau.

Heckman, J. J., & Navarro-Lozano, S. (2004). Using matching, instrumental variables, andcontrol functions to estimate economic choice models. Review of Economics andStatistics, 86, 30–57.

Hennessey, J. A., & Venohr, J. (2000). Exploring options: Child support arrears forgivenessand passthrough of payments to custodial families. Report to the Minnesota Departmentof Human Services Child Support Enforcement Division. Denver, CO: Policy Studies, Inc.

Holzer, H. J., Offner, P., & Sorensen, E. (2005). Declining employment among young blackless-educated men: The role of incarceration and child support. Journal of Policy Analysisand Management, 24, 329–350.

Meyer, D. R., Cancian, M., & Nam, K. (2007). Welfare and child support program knowledgegaps reduce program effectiveness. Journal of Policy Analysis and Management, 26,575–597.

Ovwigho, P. C., Saunders, C., & Born, C. E. (2007). Arrears forgiveness: A strategy for childsupport’s $100 billion problem? Journal of Policy Practice, 6, 23–44.



Pate, D. (2002). An ethnographic inquiry into the life experiences of African American fatherswith children on W-2. In D. R. Meyer & M. Cancian (Eds.), W-2 child support demonstra-tion evaluation: Report on nonexperimental analyses (pp. 29–118). Madison, WI: Institutefor Research on Poverty.

Pearson, J., & Davis, L. (2002). An evaluation of the Colorado Arrears ForgivenessDemonstration Project: Final report. Denver, CO: Center for Policy Research.

Pearson, J., & Griswold, E. A. (2001). New approaches to child support arrears: A survey ofstate policies and practices. Denver, CO: Center for Policy Research.

Pukstas, K., Albrecht, D., Auten, L., Drew, V., & Dabruzzi, S. (2004). Arrears management forlow-income noncustodial parents evaluation report (Grant No. 90FD0071). St. Paul, MN:Minnesota Department of Human Services.

Rich, L. M., Garfinkel, I., & Gao, Q. (2007). Child support enforcement policy and unmarriedfathers’ employment in the underground and regular economies. Journal of Policy Analysisand Management, 26, 791–810.

Roberts, P. (2001). An ounce of prevention and a pound of cure: Developing state policy onthe payment of child support arrears by low income parents. Washington, DC: Center forLaw and Social Policy.

Rosenbaum, P. R., & Rubin, D. B. (1983). The central role of the propensity score in obser-vational studies for causal effects. Biometrika, 70, 41–55.

Singer, J. D., & Willett, J. B. (2003). Applied Longitudinal Data Analysis: Modeling Changeand Event Occurrence. Oxford: Oxford University Press.

Sorensen, E. (2004). Understanding how child-support arrears reached $18 billion inCalifornia. American Economic Review, 94, 312–316.

Sorensen, E., & Zibman, C. (2001). Poor dads who don’t pay child support: Deadbeats or dis-advantaged? Assessing the New Federalism, Policy Brief B-30. Washington, DC: The UrbanInstitute.

Sorensen, E., Sousa, L., & Schaner, S. (2007). Assessing child support arrears in nine largestates and the nation. Retrieved January 27, 2009, from http://aspe.hhs.gov/hsp/07/assessing-CS-debt/.

Staiger, D., & Stock, J. (1997). Instrumental variables regression with weak instruments.Econometrica, 65, 557–587.

U. S. Department of Health and Human Services, Administration for Children and Families,Office of Child Support Enforcement. (2009). Child support enforcement, FY 2008,Preliminary report. Retrieved July 8, 2009, from http://www.acf.hhs.gov/programs/cse/pubs/2009/reports/preliminary_report_fy2008/#tables.

U. S. Department of Health and Human Services, Administration for Children and Families,Office of Child Support Enforcement. (2000). PIQ-00-03. Washington, DC: Author.

U. S. Department of Health and Human Services, Office of Inspector General. (2000). Theestablishment of child support orders for low income non-custodial parents. OEI-05-99-00390. Retrieved July 8, 2009, from http://oig.hhs.gov/oei/reports/oei-05-99-00390.pdf.

U. S. Department of Health and Human Services, Office of Inspector General. (2007). Stateuse of debt compromise to reduce child support arrearages. OEI-06-06-00070. RetrievedSeptember 16, 2008, from http://oig.hhs.gov/oei/reports/oei-06-06-00070.pdf.

Wallace, J. (2002). Work support centers: A framework. New York: MDRC.

Waller, M. R., & Plotnick, R. (2001). Effective child support policy for low-income families:Evidence from street level research. Journal of Policy Analysis and Management, 20,89–110.

Wooldridge, J. M. (2009). Should instrumental variables be used as matching variables?Working Paper. East Lansing: Michigan State University.

Reducing Child Support Debt and Its Consequences


APPENDIX

Survey Data Collection

Between March 2007 and April 2008, we conducted a follow-up telephone survey ofNCPs to gather in-depth information about those eligible for Families Forward,enrollees’ experiences with and attitudes about Families Forward, and potentialeffects of the program not identifiable in the administrative data. The sample framefor the survey included NCPs in the experimental group who had enrolled byFebruary 2007 (n � 79), NCPs assigned to the experimental group who did notenroll in Families Forward (n � 192), and all willing NCPs in the control group (n � 123). This resulted in a total raw sample frame of 394 unique NCPs who wereattached to 604 IV-D child support cases.26 From this original sample frame weexcluded NCPs who had addresses listed as prisons or jails or no address listed inthe BCS administrative database, resulting in an eligible sample frame of 327 NCPs:68 participants, 151 willing but unenrolled experimentals, and 108 willing controls.Of these, 108 eligible NCPs responded to letters of invitation by calling to schedulea telephone survey, and 97 of these completed a survey.27 Overall, 30 percent of the327 NCPs in the eligible sample frame completed a survey.

We used administrative data from KIDS and UI wage data to compare the sur-veyed NCPs with the remaining NCPs in the analytic sample. T-tests of differencesin means showed no systematic differences across the groups in terms of currentsupport orders, family-owed debt balances, state-owed debt balances, amount ofstate- or family-owed debt reduction, total child support payments, or wages. Therewere also no differences across groups in terms of sex, race, or group assignment(treatment vs. control). Survey respondents are therefore representative of the larg-er analytic sample on these important variables.

26 The survey was designed to reference one specific IV-D child support case per NCP. Yet 116 of the 394NCPs had multiple cases. The rules used to select one focal case for each NCP with multiple cases prior-itized cases in which a CP had enrolled with the NCP and cases with higher debt balances.27 We offered $25 for completing a survey during the first six months of the invitation process andincreased the compensation to $35 after the first six months in an attempt increase the incentive to par-ticipate in the survey.

reducing child support debt and its consequences: can ... wicsec conference materials...reducing...

Documents