Child support compliance during economic downturns

Children and Youth Services Review 65 (2016) 127–139

Contents lists available at ScienceDirect

Children and Youth Services Review journal homepage: www.elsevier.com/locate/childyouth

Child support compliance during economic downturns Ronald B. Mincy a, Daniel P. Miller b,⁎, Elia De la Cruz Toledo a a b

Columbia University School of Social Work, 1255 Amsterdam Avenue, New York, NY 10027, United States Boston University School of Social Work, 264 Bay State Road, Boston, MA 02215, United States

a r t i c l e

i n f o

Article history: Received 19 October 2015 Received in revised form 25 March 2016 Accepted 25 March 2016 Available online 29 March 2016 Keywords: Child support Child support compliance Unemployment Economic downturns Non-custodial fathers

a b s t r a c t Prior studies have found little evidence of an association between unemployment and child support compliance. However, few such studies used sample periods including a recession as severe as the one that occurred in 2007– 2009 or a period following the Congressional mandate requiring states to adopt immediate wage withholding for all child support orders established after January 1992. While virtually assuring compliance by steadily employed nonresident fathers, this requirement imposes hardships on unemployed nonresident fathers, especially during recessions, because modifying child support orders is costly, difficult, and uncertain. Using the CPS-CSS, this study provides reduced form estimates of the association between unemployment and child support compliance over a period (1993–2011) with severe business cycle fluctuations and immediate wage withholding in full effect. Despite controls for fixed effects (state and year) and a state-specific linear time trend, we found that local unemployment rates were associated with decreases in some measures of compliance in our full sample. In models using non-pass through child support payments, which minimized measurement error due to misreporting, there was a much more consistent relationship between unemployment and compliance. Further, after restricting the sample to cash assistance recipients to avoid bias due to selection into the child support enforcement system, we found that local unemployment rates were consistently, strongly, and negatively associated with compliance. Given the volatility in unemployment rates during recent recessions and in the Great Recession in particular, these findings suggest the potential for a large-scale impact of macro-economic factors on the consistent provisions of child support. © 2016 Elsevier Ltd. All rights reserved.

1. Introduction Economic trends and policy changes since the early 1980s have created a particularly harsh climate for noncustodial parents (NCPs), most of whom are fathers. During (and shortly after) the twin recessions of the early 1980s, men began to experience larger increases in unemployment rates than women. The gender gap in unemployment rates has become larger with each successive recession (Sahin, Song, & Hobijn, 2010). During the years bracketing the recent recession for example, annual unemployment rates for men over the age of 16 increased from 4.6% in 2006 to 10.5% in 2010 (a 128% increase), while the comparable increase in female unemployment rates was from 4.6% to 8.6% (an 87% increase) (U.S. Bureau of Labor Statistics, 2014a). What is more, improvements in the child support enforcement system required by several amendments to the Social Security Act since the early 1980s have exacerbated the consequences of economic downturns for noncustodial parents. These improvements culminated in the 1988 amendments that made immediate wage withholding mandatory for all child support orders initiated after January 1994. Since then, immediate wage ⁎ Corresponding author. E-mail addresses: [email protected] (R.B. Mincy), [email protected] (D.P. Miller), [email protected] (E. De la Cruz Toledo).

http://dx.doi.org/10.1016/j.childyouth.2016.03.018 0190-7409/© 2016 Elsevier Ltd. All rights reserved.

withholding has become the primary and most effective method for collecting child support (Sorensen & Hill, 2004). Unfortunately, immediate wage withholding can reduce disposable incomes of NCPs who experience declines in earnings (rather than unemployment spells) during economic downturns, because these NCPs continue having the full amount of their original child support orders withheld from their paychecks. Bartfeld and Meyer (2003) showed that these changes had different effects on compliance among noncustodial parents with strong and weak attachments to the labor force. Among noncustodial parents with strong attachments to the labor force, changes in compliance became more strongly associated with the child support enforcement system and more weakly associated with individual preferences toward compliance. By contrast, individual preferences were more strongly predictive of compliance among noncustodial parents with weak attachments to the labor force who were affected little by variations in the child support enforcement system. Though enforcement and preferences for compliance were their main independent variables of interest, Bartfeld and Meyer (2003) also controlled for the local unemployment rate and demographic characteristics, which were proxies for father's earnings. They found that variations in unemployment rates played no significant role in explaining variation in compliance among either group. Other studies reached similar conclusions about the association

128

R.B. Mincy et al. / Children and Youth Services Review 65 (2016) 127–139

between unemployment rates and compliance. However, few of these studies included recent time periods during which the national unemployment rate was very high, local unemployment rates varied widely, and the child support orders of most noncustodial parents were subject to immediate wage withholding. During such periods, the association between unemployment rates and child support compliance is likely to be stronger than the results estimated in previous studies. Using data from the Current Population Survey-Child Support Supplement (CPS-CSS), this paper estimated the association between unemployment rates and child support compliance of noncustodial fathers between 1993 and 2011. Our estimates controlled for enforcement of child support orders by the IV-D program, marital vs. nonmarital births, and demographic characteristics of custodial parents, which under the assumption of assortative mating, are good proxies for the earnings of noncustodial parents. Our initial models found little evidence of a relationship between unemployment and the proportion of child support received, the probability that a mother received any payments, or the probability that a mother received all that was due in the full sample of custodial mothers. However, when we focused on a group of compliance outcomes related to non-pass through child support payments, which are less likely to exhibit measurement error, there was a much more consistent relationship between unemployment and compliance. In particular, we found that a one percentage point increase in the lagged unemployment rate was associated with a marginally significant reduction (p b 0.10) in the probability that mothers reported receiving all non-pass-through child support payments and a significant reduction (−0.006, p b 0.05) in the probability that all non-pass-through payments were for the full amount. These results became even stronger and more consistent after we restricted the sample to cash assistance recipients in order to avoid selection bias, which could arise because some mothers are induced to seek child-support orders because of their own job separations and lost earnings during economic downturns. For example, we found that a one point increase in lagged unemployment rates was associated with a decrease of 0.014 in the probability that non pass-through payments to cash-assistance recipients were for the full amount. Given the volatility in unemployment rates during recent recessions and in the Great Recession in particular, these findings suggest the potential for a large-scale impact of macro-economic factors on the consistent provision of child support. We organize the paper as follows. The next section reviews the literature on the determinants of child support compliance to see how noncustodial parents adjust their child support payments in response to economic downturns and what variables we must control for to obtain consistent estimates of associations between economic downturns and child support compliance. The third section describes the data and the methods we use to generate estimates of these associations. The fourth section presents our findings and the final section discusses their implications for future research and policy. 2. Theory and empirical evidence Before reviewing the empirical evidence on the effects of unemployment on child support compliance, we first examine theoretical explanations of the determinants of child support compliance. Theoretical explanations emphasize three sets of factors that determine child support compliance: economic factors like unemployment, the main determinants of child support compliance; the aggressiveness or effectiveness of child support compliance enforcement measures, and non-economic factors (Beller & Graham, 1986). 2.1. Economic factors Unemployment affects child support compliance through its effects on earnings, child support orders, and child support payments. These effects are illustrated in Fig. 1 by the arrows (pathways), which we

identify in the figure and the following discussion by their corresponding numbers in brackets []. In the figure, the positive and negative signs indicate the expected direction of the pathways between the variables in this system. For instance, when the unemployment rate rises, many NCPs experience declines in earnings [1]. The declines could result from a reduction in wages or hours of work at their current jobs or layoffs followed by new jobs at lower wages. Because of relationships among earnings, child support payments and child support orders, the reduction in earnings [1] can lead to two additional effects: a decrease in child support payments [2], which probably involves immediate non-compliance as NCPs are unable to meet the full amount of their child support obligations; and, an eventual downward modification of child support orders [3]. Because downward modification of child support orders takes time (Hatcher & Lieberman, 2003; Henry, 1999), a second round of reductions in child support payments [4] occurs once downward modifications of child support orders eventually take effect. However, this round occurs without further reductions in compliance. Instead, lower payments reflect the lower amount due on their modified child support orders, allowing NCPs to pay the reduced amount in full compliance with the new order. Finally, unemployment might affect child support payments through the effect of unemployment on child support orders. Two possibilities arise here. Unemployment insurance claims typically rise when the unemployment rate is high. Once such claims are granted, fathers can petition for a downward modification of their existing child support orders [5]. If their petitions are granted, their child support orders will be lower and the payments will be automatically deducted from their unemployment insurance benefits. Because the amount paid is directly related to the size of the order, pathway [4] is positive. So, fathers come back into compliance, but their payments are lower than they were while employed. This effect will also likely occur with a lag. The second possibility involves fathers who do not have orders. Unemployment may also result in job separations and lower earnings of mothers, driving some mothers to seek formal child support payments from fathers who are paying support informally, if at all. In these cases, the father will owe child support to the mother. Other mothers will be forced to seek public assistance (Council of Economic Advisers, 1997; Ziliak, Figlio, Davis, & Connolly, 2000). As a condition of receiving such assistance, mothers are required to sign over their rights to child support to the state, which in turn, seeks formal child support orders from fathers. In these cases, the father owes child support to the state, not to the mother. Mothers belonging to population groups with employment and earnings that are more sensitive to business cycles are attached to fathers who are equally vulnerable (Hoynes, Miller, & Schaller, 2012). Nevertheless, we can expect variations in the formal child support orders these fathers will face once they are brought into the formal child support system by the mother or the state. Fathers brought into the child support enforcement system who owe child support to the mother will have lower ability to pay and lower child support orders than fathers who owed child support to mothers before the unemployment rate rose. This suggests that rising local unemployment rates are negatively associated with child support orders [5]. However, fathers brought into the child support enforcement system who owe child support to the state will have greater ability to pay (and higher child support orders) than fathers who owed child support to the state before the unemployment rate rose. This suggests that local rising unemployment rates are positively associated with child support orders [5]. Below we discuss how we examine the degree to which our results are sensitive to these opposing selection effects. Since the association between unemployment and child support compliance is of primary interest to our study, a few preliminary comments are in order before we review the empirical literature. As Fig. 1 illustrates, the association between unemployment and child support compliance is embedded in a system of at least three pathways (structural equations) in which the outcomes are the three variables encircled

R.B. Mincy et al. / Children and Youth Services Review 65 (2016) 127–139

129

Fig. 1. How does unemployment affect child support payments? Note: [Pathway numbers] and expected direction of influence (+ or −).

in Fig. 1: (1) earnings, (2) child support orders, and (3) child support payments. There is considerable interdependence among the outcomes in this system, which is to say that these three outcomes are endogenous. Fortunately, under reasonable simplifying assumptions, there is only one arrow between each of the outcome variables (i.e., neither earnings nor child support orders depend upon payments, nor do earnings depend upon child support orders). Therefore, when endogenous variables are among the regressors in the structural equations determining other endogenous (outcome) variables in this system, researchers can safely assume that these endogenous variables are uncorrelated with (independent of) the respective error terms in the structural equations, and obtain consistent estimates of the coefficients in the system using maximum-likelihood estimators (such as ordinary least-squares or probit). Even if these independence assumptions are not met, researchers can still obtain estimates of the effects of (exogenous) variables determined outside of the system, such as the unemployment rate, on outcome variables for purposes of forecasting and policy analysis (Edge, Kiley, & Laforte, 2010). These estimates come from manipulating the structural equations to eliminate endogenous variables from the regressors of each of the equations. The result is a reduced form equation for each of the (endogenous) outcome variables. The coefficients of the exogenous variables in these reduced form equations do not represent the original coefficients (represented by the individual arrows like [1] in Fig. 1) in each of the structural equations. They cannot tell us, for example, about the size of the effect of unemployment on earnings or child support orders. Instead, the reduced form coefficients represent the overall effects of exogenous variables on endogenous outcome variables, expressed as functions of the various structural coefficients in one or more of the structural equations. Because of this, the signs of these reduced form coefficients can be ambiguous, even when the underlying structural model has a strong theoretical basis. As we show in Appendix 1, however, in the reduced form equation for child support payments—the equation of interest here—the sign of the unemployment rate is unambiguously negative. With these preliminary comments, we can interpret the findings of prior empirical studies about the association between unemployment and child support compliance.

2.1.1. Previous research Empirical studies in this area suffer from several limitations. First, few nationally representative surveys include data on child support orders, child support payments, and individual earnings of NCPs. Several studies relying on such data include a control for the unemployment rate (or other measure of the economic climate). Without data on child support orders, the unemployment rate coefficients in these studies are estimates of the reduced form parameters, not the structural parameters in Fig. 1. However, studies are rarely clear on this point, nor are they always clear about the expected sign of the unemployment rate coefficient. For example, Hanson, Garfinkel, McLanahan, and Miller (1996) included the employment rate in the mother's state of residence in an equation used to predict mother's income, which was then used as a regressor in a model of child support receipt. Though the coefficient of this predictor was positive and statistically significant, the authors did not provide a separate estimate of the association between employment rates and compliance. To represent the probability that the mother worked, Sorensen and Hill (2004) included the current and lagged values of the unemployment rate in a multinomial logit model in which combinations of welfare and child support receipt were the outcome variable of interest. Neither the current nor the lagged value of the unemployment rate was statistically significant in these analyses. Studies in which child support enforcement was the main independent variable of interest also found that the unemployment rate was not significantly associated with child support compliance (Huang, 2009; Nepomnyaschy & Garfinkel, 2010). However, neither of these studies included periods with unemployment rates as high as those reached during the early 1980's or during the recent recession. While studies using nationally representative data typically do not find a significant association between compliance and unemployment rates, they do find associations between compliance and proxies for mothers' earnings (e.g., her age, race, and education). Assuming assortative mating, these proxies are highly correlated with the corresponding characteristics of fathers, and therefore are good proxies for the fathers' earnings as well (Beller & Graham, 1986; Case, Lin, & McLanahan, 2003; Freeman & Waldfogel, 2001; Hanson et al., 1996; Sorensen & Hill, 2004).

130

R.B. Mincy et al. / Children and Youth Services Review 65 (2016) 127–139

The second limitation of empirical studies on compliance is that studies with the requisite data on child support orders, child support payments, and individual earnings of NCPs tend to focus on fathers with low earnings in one state. Administrative data from the State of Wisconsin allows researchers to merge information about child support orders (or payments) with data on individual earnings of NCPs. So estimates of the structural parameters in Fig. 1 could be obtained from these data. Unfortunately, much of this administrative data focuses on the fathers of children receiving some form of public assistance (Ha, Cancian, & Meyer, 2010; Wu, 2011). The fathers of these children have lower earnings and skills than the general NCP population. Therefore estimates of the association between unemployment (or other predictors) and child support compliance that rely on these data may also overestimate the corresponding effects in the population more generally. In addition, most of the Wisconsin studies focus on periods of economic growth or mild recessions, when relatively small numbers of workers are displaced (Bartfeld & Meyer, 2003; Ha, Cancian, Meyer, & Han, 2008; Ha et al., 2010; Ha, Cancian, & Meyer, 2011; Meyer, Ha, & Hu, 2008). During such periods, variation in unemployment or earnings may be so low that compliance is little affected. This may help to explain why there is little evidence from the Wisconsin studies in support of negative structural effects of unemployment on child support compliance, after controlling for earnings. Recall that Bartfeld and Meyer (2003) found that county unemployment rates were not significantly associated with compliance among NCPs with firm or weak attachments to the regular labor market. Meyer et al. (2008) found that this unemployment rate was not significantly associated with payments or the compliance rate in fixed effects models. However, in their Tobit model, which accounted for the clustering of payments at zero in the last year of their study, they also found that the unemployment rate had a positive and statistically significant association with payments. Meyer and colleagues offered no explanation for this unexpected result. Studies using Wisconsin data also focus on changes in earnings over the business cycle and child support compliance. Bartfeld and Meyer (2003) show that fathers with higher earnings have higher compliance rates, though the association between earnings and compliance is nonlinear. For example, when orders are a high percentage of the NCPs income, increases in earnings are not associated with increases in child support compliance. However, Meyer et al. (2008) find that although payments and compliance rates were positively associated with earnings, compliance rates were negatively associated with the proportion of earnings owed in child support. Several studies use Wisconsin data to undertake bivariate analysis of changes in earnings over the business cycle and changes in compliance. While of interest, these bivariate analyses do not provide estimates of the structural link between earnings and child support compliance [2]. Nevertheless, using Wisconsin data from 2000 to 2005, Ha and colleagues (Ha et al., 2011) found a small positive association between changes in earnings and changes in child support payments. Wu (2011), who used more recent Wisconsin data, including the 2008– 2009 recession, observed relatively stable child support payments, though the largest declines (and increases) in child support payments occurred among NCPs with large declines (or large increases) in earnings. Indeed, the proportion of NCPs who paid in full actually grew slightly between 2006 and 2009, while the proportion making partial payments declined. Nevertheless, the size of earnings decreases was associated with the likelihood that fathers did not pay in full. Put differently, fathers who paid the full amount due in 2006 tended to continue doing so through 2009, unless they experienced large reductions in earnings. These findings very likely reflected selection bias, because NCPs earning less than $10,000 represented about 40% of the sample. These NCPs lived in counties with the largest increases in unemployment rates and experienced the greatest declines in earnings.

In summary, because of the severe data limitations that arise when studying the effects of unemployment on child support compliance, there is much we still do not understand about how economic downturns affect child support compliance. We know that payments and compliance rates fall in response to reductions in earnings primarily among NCPs most severely affected by economic downturns (i.e., those who experience earnings declines of 50% or more). We also know that on average, compliance falls in response to changes in proxies for fathers' earnings such as mothers' age, race, income or education. However, there is little evidence from studies including such proxies that higher unemployment rates are associated with lower payments or compliance rates, and much of this evidence relies on data on NCPs with very low earnings who live in one state (Wisconsin). 2.2. The child support enforcement environment The aggressiveness or effectiveness of governmental efforts to enforce child support orders also affects child support compliance. For example, several studies showed that per capita expenditures on child support enforcement affect child support compliance (Freeman & Waldfogel, 2001; Huang, Garfinkel, & Waldfogel, 2004). Other studies showed that the number of child support enforcement provisions designed to increase the effectiveness of child support enforcement also affected child support compliance (Case et al., 2003; Hanson et al., 1996; Nepomnyaschy & Garfinkel, 2010; Sorensen & Hill, 2004). These provisions included new-hire registries, immediate wage withholding, presumptive guidelines for child support orders, in-hospital paternity establishment and state income-tax intercepts. Huang (2009) and Huang and Edwards (2009) argued that three dimensions of child support enforcement (adoption of strict laws, adequate expenditures, and strong implementation) were necessary to properly assess its effects on paternity establishment, child support order establishment, and child support collections. They found that a comprehensive index of the three dimensions of child support enforcement was significantly and positively associated with collection on child support orders. However, analyses of individual components of this index revealed that it was only caseload per full time child support enforcement staff (a measure of state-level expenditures) that was significantly associated with order collection. Moreover, analyses substituting a more traditional measure of expenditures – state level per capita child support expenditures – arrived at the same conclusion (Huang & Edwards, 2009). The impact of economic downturns on compliance likely differs according to fathers' participation in the formal child support system. First, as stated above, economic downturns are likely to draw some fathers without orders into the child support enforcement system, because custodial mothers or states opening new cash assistance cases on behalf of these mothers will seek formal child-support payments. Second, immediate wage withholding has applied to child support collections for all NCPs with child support orders established or modified by the IV-D program (hereafter, IV-D orders) during the period of our study. Therefore, it is possible that many NCPs with IV-D orders and unemployment spells will cease paying child support in full, because they no longer have employers to withhold the full amount due. They may continue paying some child support from loans, savings, or other sources of income. They may also skip payments (paying fewer than the full number of payments due). Finally, they may reduce the amount they pay if they are unable or unwilling to comply with their orders once immediate wage withholding ceases. Once NCPs begin receiving unemployment benefits, they will resume full payment of their child support, because the IV-D program automatically deducts child support payments from such benefits, or they might pay less than the full amount owed if their unemployment benefits are insufficient to cover the amount of their child support orders. Discretionary obligors with low attachment to the formal labor force may choose not to resume working, particularly if the accumulation of arrears during their lay off

R.B. Mincy et al. / Children and Youth Services Review 65 (2016) 127–139

period has made the marginal tax on their earnings sufficiently high enough to discourage their participation in the labor force (Miller & Mincy, 2012). Nonetheless, even if fathers' earnings fall (either because they are forced to cut back on hours at their current jobs or become underemployed at new jobs), most NCPs with IV-D orders who remain employed will continue having the full amount of child support deducted by their employers as withholdings do not automatically adjust to changes in income. When compared with custodial parents with orders enforced by the IV-D system, mothers with orders established and enforced outside that system (non IV-D orders) have much higher incomes, are less likely to receive public assistance, and are much less likely to have been nevermarried (Lippold & Sorensen, 2013). As a result, economic downturns are less likely to disrupt compliance among NCPs who have children with the latter custodial parents. However, because they are not subject to immediate wage withholding, NCPs with non IV-D orders have much more discretion about compliance than their counterparts with IV-D orders. The former can elect to reduce payment amounts or frequency even when employed. For most of the period of our study (1993–2011), states had adopted many or all of the most important provisions designed to increase the child support collections (Case et al., 2003; Huang, 2002; Sorensen & Hill, 2004). Therefore, expenditures on child support enforcement and the proportion of custodial mothers covered by the IV-D program were the two primary factors responsible for the variation in the child support enforcement environment for which consistent measures were available for the period of our study (Huang & Edwards, 2009). Thus, and in light of previous findings, we expect compliance to be higher in states with higher child support expenditures (Case et al., 2003; Huang & Edwards, 2009). The size of the metropolitan area in which custodial mothers reside might also influence child support compliance, but the direction of the effect is ambiguous. To the extent that population density and urbanicity are correlated with divorce and non-marital birth rates, the potential child support caseload will be larger in the jurisdictions with larger metropolitan areas. Public expenditures for children in motheronly families will also be larger in such areas. Both factors could motivate states and counties with larger metropolitan areas to adopt more aggressive child support enforcement policies. For this reason, compliance might be higher for custodial mothers in larger metropolitan areas. On the other hand, the resources available to enforce those policies would have to compete with many other demands. For this reason, compliance might also be lower for cases in larger metropolitan areas.

131

Custodial parents and NCPs may wish to rely upon informal, rather than formal, child support payments. For example, never-married parents are more likely to rely upon informal child-support arrangements than divorced or separated parents (Nepomnyaschy & Garfinkel, 2010). Minority custodial parents and NCPs may be discouraged from using the formal child support system because of discrimination or language barriers (Beller & Graham, 1986). As a result, we expect compliance among minority NCPs will be lower than among white NCPs. 3. Contributions of this study Aware of the limitations in previous work identified above, this paper extends the literature in four important ways. First, using nationally representative data from the CPS-CSS, we update estimates of the effects of economic downturns on child support compliance. We estimate the association of the unemployment rate on child support compliance over a period including the second half of the 1990s economic boom and the national adoption of the provisions of the 1996 Amendments to the Social Security Act, which strengthened child support enforcement. The study period also includes the brief recession of 2001 and the 2007–2009 recession. Second, we focus on reduced form estimates of the effect of unemployment rates and compliance, which incorporate the effects that unemployment rates have on earnings and child support orders. These estimates incorporate direct and indirect effects operating through earnings and child support orders. Unlike prior studies we are explicit in our expectation that the effect of unemployment rate on compliance is unambiguously negative, assuming that neither earnings nor child support orders depend upon child support payments, and that earnings do not depend upon child support orders. Third, we examine multiple measures of child support compliance for both families that receive cash assistance and those that do not. Doing so allows us to test our expectations regarding the short-term, exogenous impact of changes in unemployment on child support compliance as detailed in Fig. 1 and to address concerns about selection and measurement error. Last, we generate estimates of the association between local unemployment rates and compliance for different subgroups, which represent populations of interest to policy makers. Given previous evidence regarding the role of non-economic factors, this allows us to explicitly test whether the effect of the business cycle on compliance varies for subgroups of the population that are more or less vulnerable to the substantial economic fluctuations that occurred during the period of our study. 4. Data and methods

2.3. Non-economic factors 4.1. Data Besides economic factors and the child support enforcement environment, the willingness of parents to pay or receive formal child support affects child support compliance. NCPs may withhold child support payments because they are not committed to the welfare of their non-resident children (Furstenberg & Hughes, 1998), they are unable to control how custodial mothers spend formal child support payments (Weiss & Willis, 1985), or because of conflicts with the mothers of their children surrounding visitation or other matters. As noted above, Bartfeld and Meyer (2003) argue that immediate wage withholding leaves little room for such factors to influence child support compliance among NCPs with firm attachments to work. However, immediate wage withholding is less effective among NCPs with weak attachments to the formal labor market. Two of their findings support their hypotheses. First, predictors of ability to pay (father's age and mother's education) have larger effects on compliance among NCPs with weak attachments to formal employment. Second, proxies for the desire to comply have statistically significant associations with compliance among NCPs with weak attachments to formal employment (Bartfeld & Meyer, 2003).

This study used data from the CPS-CSS. Each March, the Current Population Survey includes the Annual Social and Economic Supplement (ASEC), which in addition to collecting typical labor force data, also collects information on income, work experience, government program participation and benefits, and migration (U.S. Census Bureau, 2007). The CPS-CSS is a separate bi-annual survey administered in April, which collects additional data from biological and adoptive custodial parents ages 15 years and older who were also surveyed in the March CPS. Data from the CPS-CSS are commonly used to describe the situation of children and parents in divorced, separated, or otherwise unmarried families (Grall, 2011). Because of changes to the CPS-CSS beginning in the early 1990s, data collected in or after 1994 were not compatible with those from earlier survey years (Freeman & Waldfogel, 2001). Thus, we used data from the available waves of the CPS-CSS collected in 1994 or later, that cover events in the previous calendar years: 1993, 1995, 1997, 2001, 2003, 2005, 2007, 2009, and 2011. These years included almost the entire 1990s economic boom, when the unemployment rate fell from 6.9%

132

R.B. Mincy et al. / Children and Youth Services Review 65 (2016) 127–139

in 1993 to 4.0% in 2000, which was followed by a brief recession in 2001, when unemployment rose from 4.7% to 6% in 2003 (U.S. Bureau of Labor Statistics, 2014a). In 2005, there was a jobless recovery, when the unemployment rate declined to 5.1% in 2005 and then 4.6% in 2007, shortly before the Great Recession, which began at the end of 2007 and concluded in the middle of 2009. Though it was 9.6% at the official end point of the Great Recession, the unemployment rate actually climbed to a high of 10% in October 2009 and remained above or close to 9.5% until the end of 2011 (U.S. Bureau of Labor Statistics, 2014a). We pooled these nationally representative samples, and because our population of interest is mothers who have a child with a noncustodial father, we dropped custodial fathers from our analytic sample (nearly 7800 male respondents). In addition, so as not to arrive at results driven by families where there are not child support obligations, we dropped an additional approximately 17,960 cases where mothers reported that they were not owed child support. These two restrictions left us with an initial sample of 20,960 mothers who reported being owed some child support in the previous year. We describe the process for arriving at our final analytic sample below. 4.2. Variables 4.2.1. Child support payments For each year of available data, custodial mothers reported on a variety of topics related to child support. As noted in the technical papers accompanying the release of the CPS-CSS, approximately 30% of cases in each year were missing data on child support, largely due to survey nonresponse (U.S. Census Bureau, 2007). When missing, data for child support variables were imputed by the Census Bureau. Despite the relatively large proportion of missing data, our confidence in the CPSCSS data is underscored by its continued use in research and by the fact that the data form the basis for the federal government's reports on child support participation (Grall, 2011). First, for all years, the CPS-CSS created variables recording the amount of child support mothers were due and the amount that they received. Using these variables, we created three initial measures of compliance: the proportion of child support that was paid (calculated as the amount received divided by the amount due), which we log transformed, a dichotomous indicator for whether a mother received any support, and an indicator for whether mothers received the full amount due to them. Also, in each year, all mothers who were due child support responded to the following questions in the supplement, which referred to support they were supposed to receive in addition to any support passed through a welfare agency: “Other than the child support passed through the welfare agency [in the past calendar year], did you ACTU ALLY receive ANY child support payments – even one – for [covered children]?”; “In [the past calendar year] did you receive EVERY SINGLE ONE of the child support payments you were supposed to receive for the children?”; “Of the child support payments you received in [the past calendar year], how many were received on TIME? Would you say (1) all of them were on time, (2) most of them, (3) some of them or (4) none of them?” and, “And for child support payments you received, how many of them were for the FULL amount you were supposed to receive? Would you say (1) all of them, (2) most of them, (3) some of them, or (4) none of them?” Based on these questions we developed three additional indicators of child support compliance. The first of these was a dichotomous variable coded as equal to one if mothers reported receiving all non-passthrough child support payments in the previous calendar year and zero otherwise. Second, we created a variable coded as equal to one if mothers reported that all non-pass-through payments received arrived on time, and last, we created an indicator coded as equal to one if mothers reported that all payments received were for the full amount. Because all mothers were due some amount of child support in the previous year, each of these variables was coded as equal to zero if mothers

reported receiving no payments in addition to a pass-through. Collectively, these questions on child support compliance allow us to address potential concerns about sample selectivity and measurement error. We discuss our specific use of these variables in the analysis section below. 4.2.2. Unemployment rates We lagged our unemployment rate by one year for all analyses. In the CPS, custodial mothers reported on child support compliance in the previous calendar year. So, for example, to implement our lag we used data on unemployment from 1992 to predict child support compliance in 1993, which mothers reported in the 1994 CPS-CSS. Our study thus relies on unemployment data collected from the BLS from 1992 to 2010. We used the unemployment rate lagged one year for three reasons. First, while we expect compliance to decline during economic downturns as NCPs experience unemployment spells, the full response to economic downturns could take some time. Second, applications for both child-support modifications and unemployment insurance take time before they are granted, so, even NCPs who qualify for unemployment insurance may pay less than the full amount owed at least initially. By using the lagged unemployment rate our models capture more than the short-run response. Finally, younger NCPs and those who work parttime or part year may not qualify for unemployment insurance and therefore may not come back into full compliance until their unemployment spells end. However, recall to a previous job has occurred much less frequently during recent recessions, so that unemployment spells typically end when workers find new jobs at lower wages (Vroman, 2005). This may prolong periods during which workers without unemployment insurance pay no child support. Unless and until the child support orders are modified downward, these NCPs may pay less than the full amount due for some time. To adequately assess the impact of unemployment rates on child support payments, we set out to collect data on local area unemployment rates (LAUs) for all NCPs in the CPS-CSS. However, this effort was complicated by data limitations. Because of the structure of the CPS-CSS, data were not collected directly on NCPs. Rather, mothers reported on fathers' state of residence only. In addition, the CPS-CSS did not record information on the local area of residence for some mothers. Most often (97.5% of the time), this occurred when mothers' residences were known but outside metropolitan areas. Only a small minority of mothers lacking a local area identifier had a county of residence identifier, rendering the use of county unemployment rates ineffective. As previously mentioned, male unemployment rates have changed more drastically than those of females, but we were not able to use male LAUs because local rates for men are not available prior to 1997 and only for selected metropolitan areas thereafter. Thus, to assign unemployment rates for NCPs, we were forced to make a number of assumptions. Using non-seasonally adjusted unemployment rate data retrieved from the BLS (U.S. Bureau of Labor Statistics, 2014b), we assigned unemployment rates to NCPs using the following rules to arrive at what we term a pooled unemployment rate. If a mother lived in a metropolitan area and reported that a father lived in the same state, we assumed the father lived in the same area and assigned the LAU for the mother's metropolitan area. If a mother did not live in a metropolitan area and reported that a father lived in the same state, we assigned the state-level unemployment rate for the mother's (and thus the father's) state of residence. Finally, if a mother reported that a father lived in a different state, we assigned the father's state-level unemployment rates. Two thousand and sixty five fathers either lived abroad or their residence was unknown, and these cases were dropped from the analysis. We discuss the implications of excluding these fathers below. Five additional cases were missing information on mother's race and ethnicity and were also dropped. After our restrictions, we were left with a final analytic sample size of 18,887. With this sample, our pooled unemployment rate consisted of local area unemployment information for 11,110 fathers (approximately 58.8% of the sample), shared state

R.B. Mincy et al. / Children and Youth Services Review 65 (2016) 127–139

unemployment rates for 4703 fathers (24.9%), and fathers' separate state unemployment rates for the remaining 3074 fathers (16.3%). Information on the amount of child support due was not available for roughly 200 fathers, and so in some analyses (described below) we focus on a smaller sample of 18,686. Our pooled unemployment rate is likely subject to measurement error from two sources: the use of mothers' LAUs and state unemployment rates to proxy fathers' LAUs, and the use of state unemployment rates when mothers live outside metropolitan areas. In both cases, we believe that measurement error will bias our unemployment rate coefficient toward zero, meaning that the estimates we present in the results should be interpreted as lower bounds. In the first case, utilizing mothers' LAUs and state unemployment rates will likely introduce random error into our measure: in some instances, the rate we use will be lower than the true rate for fathers and will be higher in others. This random error results in attenuation bias, driving the coefficient for unemployment toward zero. In the second case, if the majority of NCPs linked to mothers in non-metropolitan areas likewise live outside urban areas, the use of state unemployment rates will likely underestimate true unemployment rates, which are traditionally higher than those in metropolitan areas and will also decrease the variance (Kusmin & Parker, 2006). If NCPs who had children with mothers who live in non-metropolitan areas are spread across both urban and nonurban areas, the measurement error in our measure will likely be random, also biasing the coefficient for the unemployment rate toward zero.

133

Support Performance and Incentive Act of 1998 (CSPIA; (Huang & Edwards, 2009). Based on previous research summarized above, we indexed the state-level strength of the child support enforcement system with a measure of per capita spending on child support, lagged by two years. Supplemental analyses using a measure based on state expenditures per single mother arrived at nearly the same results as those reported below. Data on state child support expenditures were collected from the OCSE Annual Reports to Congress and were divided by state estimates of population size for each year (Office of Child Support Enforcement, 1997, 2002, 2004, 2006, 2008a, 2009). We classified states as being part of one of the three following categories: lowest third (“low expenditures”), middle third (“medium expenditures”), or highest third (“high expenditures”) of per capita expenditures. The third variable is a set of indicators measuring the size of the metropolitan statistical area (MSA) a mother lives in (not a MSA – omitted; 100 thousand to 249 thousand people, 250 thousand to 499 thousand people, 500 thousand to 999 thousand people, 1 to 2.49 million people, 2.5 to 5 million people, and N5 million people,). 4.3. Methods 4.3.1. General estimation strategy For our analyses, we specified state- and year-fixed effects reduced form models, with measures of child support compliance as our dependent variables. Our models took the general form of Y ist ¼ β0 þ β1 Unempist−1 þ β2 X it þ β3 CSEst−2 þ δs þ γt þ nst þ εist ð1Þ

4.2.3. Controls To control for possible confounding, the analysis included a number of family- and person-level variables from the CPS-CSS as informed by the literature review above. These included a group of demographic controls, which are proxies for mothers' (and fathers') earnings: indicator variables for mother's marital status (not married, never married, married); mother's age in years and age in years-squared, and indicator variables for mother's educational attainment (less than high school degree, high school degree, some college but no degree, associate's degree, bachelor's degree, graduate school,). We also included indicator variables for mothers' race/ethnicity (black not-Hispanic, Asian/Pacific Islander not-Hispanic, white not-Hispanic, and other not Hispanic). Besides their role in predicting mothers' (and fathers' earnings), race/ ethnicity also proxy non-economic factors that affect compliance. Next we include a variable measuring the number of children under the age of 18 in the household who belonged to the mother, which predicts how aggressively mothers may seek child support income. Because custodial family income is correlated with NCP income, the former should increase child support compliance. However, consistent with our reduced form estimates, the results we report did not control for custodial family poverty in our models, as any measure of income would be endogenous to the relationship between unemployment rates (which affect the earnings of custodial mothers as well) and compliance. However, analyses (not reported here) show that inclusion of custodial family poverty status does not alter our results. To control for factors that affect the IV-D environment for child support collections, we included six other variables. The first measures whether a respondent's child support order is being enforced through the IV-D system. This variable was created by replicating the coding scheme created by the Office of the Assistance Secretary for Planning and Evaluation within the Department of Health and Human Services as a technique for estimating participation in the IV-D program by respondents in the CPS-CSS data file (Hong, Mellgren, Desale, & SinclairJames, forthcoming). The second variable indexes child support enforcement efforts in mothers' states of residence. The use of a comprehensive index of state child support enforcement efficiency (as in (Huang & Edwards, 2009) is precluded for our study by the changes in state data reporting requirements and procedures following the passage of the Child

where Yist is a child support compliance outcome (natural log of child support received, any payment received, full support received, all nonpass-through payments received, all non-pass-through payments on time, all non-pass-through payments for full amount) from individual i in state s in year t, Unempist-1 is the unemployment rate assigned to individual i in state s in the year before compliance is measured (year t − 1), Xist is a vector of exogenous, individual-level control variables described above, CSEst-2 is a state- and year-specific measure of per capita spending on child support enforcement two years before compliance is measured, δs is the state fixed effect, γt is the year fixed effect, nst is a state-specific linear time trend, and εist is the error term. Our coefficient of interest is β1. Our fixed effects modeling approach has been commonly adopted in previous research on child support enforcement and addresses bias by controlling for both observed and unobserved factors that are specific to states and to years of observation. Our state-specific linear time trend accounts for the unobserved factors that proceed linearly over time within states, an approach that has likewise been used in numerous previous analyses of child support or other policies (Argys & Peters, 2001; Garfinkel, Huang, McLanahan, & Gaylin, 2003; Huang et al., 2004). By including this term in our analyses (along with our fixed effects terms), our models only examine the impact of unemployment rates that deviate from linear time trends Huang et al., 2004). An added benefit of the linear trend is that it captures the increase in child support enforcement activities that occurred at the state level over the period of our study, which we could not directly include in our models because of inconsistencies in the data series on state enforcement activities. An alternative and more restrictive approach that has also been employed in some previous work (Freeman & Waldfogel, 2001) would be to expand Eq. (1) to include state-by-year fixed effects (often operationalized as indicator variables for each state-year observation). However, our information on unemployment precludes this method; as noted above, state unemployment rates are assigned to over 40% of our cases, which would be effectively eliminated from a state-by-year fixed effects analysis. For all of our compliance outcomes, we specified ordinary least squares regressions, generating linear probability models for our compliance model. All models were specified using Stata 14 MP with robust

134

R.B. Mincy et al. / Children and Youth Services Review 65 (2016) 127–139

clustered standard errors at the state level to adjust for the nonindependence of observations within a father's state of residence (Rogers, 1994; Williams, 2000). For our main analyses, we specified an increasingly restrictive set of analyses, adding in state dummies,

Table 1 Descriptive statistics (n = 18,887). Variables

Range

Child support compliance Amount of child support received (2011 dollars) Proportion of child support order received (n = 18,686) Mother received any support Mother received full amount due (n = 18,686) Mother received all non-pass through payments All non-pass through payments received were on time All non-pass through payments received were for full amt. Unemployment Lagged unemployment rate Control variables

Range

IV-D program participant Per capita child support expenditures Low Medium High Custodial mother's race/ethnicity White, not Hispanic Black, not Hispanic Hispanic any race Asian/Pacific Islander, not Hispanic Other, not Hispanic Custodial mother's marital status Married Not married Never married Custodial mother's education Less than high school High school degree Some college Associate's degree Bachelor's degree or more Graduate degree Custodial mother's age in years Number of custodial mother's children in household Metropolitan area size Not a metropolitan area 100,000–249,000 250,000–499,000 500,000–999,000 1 million–2.5 million 2.5 million–4.99 million 5 million or more Survey year 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011

Mean or SD for continuous proportion variables

0–98,241.08 4011.54 0–320

0.76

0–1 0–1

0.78 0.46

0–1

0.50

0–1

0.35

0–1

0.49

1.7–30.9

5.59

5623.12 3.66

2.12

Mean or SD for continuous proportion variables

0–1

0.72

0–1 0–1 0–1

0.34 0.33 0.33

0–1 0–1 0–1 0–1

0.68 0.18 0.10 0.02

0–1

0.02

0–1 0–1 0–1

0.25 0.52 0.23

0–1 0–1 0–1 0–1 0–1 0–1 15–88 0–9

0.12 0.35 0.25 0.11 0.12 0.05 36.39 1.65

0–1 0–1 0–1 0–1 0–1 0–1 0–1

0.31 0.08 0.09 0.11 0.16 0.09 0.16

0–1 0–1 0–1 0–1 0–1 0–1 0–1 0–1 0–1 0–1

0.12 0.10 0.10 0.10 0.11 0.11 0.10 0.09 0.09 0.09

8.29 1.14

state linear time trends, and year dummies in turn. Though the final model (which includes all controls) is our preferred one, we also note here that it is highly restrictive given the strong link between year of measurement and the unemployment rate. All told, the state dummies, state linear time trends, and year dummies account for nearly 70% of the variance in the lagged unemployment rate, and under these circumstances, it is difficult for our data to reject the null hypothesis that β1 =0. 4.3.2. Analyses steps As mentioned above, our first three measures of child support compliance (proportion of the order paid, whether the full amount of the order was paid, whether any of the order was paid) are based on questions from the CPS-CSS regarding the amount of child support received by custodial parents, rather than the amount of child support the nonresident parent actually paid. Significantly, these amounts may differ for custodial parents receiving cash assistance, who were required to sign over their rights to child-support to the state as a condition of receiving benefits. Amendments to the Social Security Act in 1976 required all states to disregard up to $50 each month of child support paid on behalf of custodial parents receiving Aid to Families with Dependent Children (AFDC). The 1996 Amendments further affected the discrepancy between child support received and child support paid in two important ways. On the one hand, because the amendments replaced AFDC with Temporary Assistance to Needy Families, the cash assistance caseload became less responsive to the business cycle than other benefit programs. So, the number of low-income fathers drawn into the child support enforcement system during economic downturns was presumably smaller after the 1996 amendments than before. On the other hand, the 1996 Amendments also gave states latitude to set their own disregard policies. They could retain all child support paid on behalf of families receiving cash assistance. They could implement disregards by passing up to $50 of the child support paid on behalf of AFDC/TANF families to the custodial family. Finally, they could send these families a separate check in the amount of the disregard (Cassetty, Cancian, & Meyer, 2002). Thus, although the 1992 amendments to the Social Security Act required states to notify custodial parents receiving cash assistance of the amount of child support paid by nonresident parents on their behalf each month, using the amount of child support received by custodial mothers is likely to understate the amount of child support paid by nonresident fathers on behalf of custodial mothers receiving cash assistance (Meyer & Hu, 1999). Further, this source of measurement error likely varies across states and over time. Given the likely presence of measurement error, our analyses proceeds in a number of specific steps. First, we estimate associations between unemployment and our first measures of child support, defined above, using our full sample of custodial parents. Second, we conduct the same analyses among the group of custodial parents who had not received cash assistance. Third, we use our full sample to explore associations between unemployment rates and our second set of child support compliance outcomes, which focused on the receipt of non-pass through support. By estimating models over samples that exclude cash assistance recipients and that focus on non-pass through support in steps 2 and 3, we avoid the measurement error that occurs when custodial mothers who are subject to pass through rules report the amount of child support they receive, but this amount differs from the amount of child support fathers actually pay. However, neither of these steps purges our results of the bias arising because of selection into the formal child support system. In fact, the subsample of custodial mothers not receiving cash assistance includes precisely those who are likely to apply for welfare or seek a formal child support order when unemployment rises as they become more likely to lose their jobs in the face of a downturn. Because these mothers are better off than those receiving cash assistance, they are likely to be paired with non-custodial fathers who are more likely to comply with

R.B. Mincy et al. / Children and Youth Services Review 65 (2016) 127–139

135

Table 2 Unemployment rates and child support compliance. Coefficient for the lagged unemployment rate (robust SE) Proportion of child support received (1) Full sample (n = 18,686) (2) No cash assistance (n = 16,748)

−0.006 (0.009) −0.003 (0.010)

0.000 (0.009) 0.002 (0.010)

0.016+ (0.009) 0.018+ (0.010)

0.019 (0.016) 0.022 (0.019)

Mother received all non-pass through payments (3) Full sample (n = 18,887) (4) Yes cash assistance (n = 1974) Controls State dummies State linear time trends Year dummies

Mother received any payments −0.005⁎⁎⁎ −0.004⁎ (0.001) (0.002) −0.004⁎ −0.003 (0.002) (0.002)

−0.002 (0.002) −0.001 (0.002)

−0.005 (0.003) −0.003 (0.004)

All non-pass-through payments were received on time

−0.001 −0.002 −0.002 −0.007⁎⁎⁎ −0.005⁎ −0.004+ −0.006+ −0.003⁎ (0.002) (0.002) (0.002) (0.003) (0.001) (0.001) (0.002) (0.002) ⁎⁎⁎ ⁎⁎ ⁎⁎ ⁎⁎ ⁎⁎⁎ ⁎⁎ ⁎⁎ −0.015 −0.013 −0.013 −0.013 −0.014 −0.013 −0.014 −0.017⁎⁎⁎ (0.004) (0.004) (0.004) (0.004) (0.003) (0.005) (0.005) (0.005) x

x x

x x x

x x x x

x

x x

x x x

x x x x

Mother received all that was due −0.010⁎⁎⁎ (0.002) −0.008⁎⁎⁎ (0.002)

−0.007⁎⁎⁎ (0.002) −0.007⁎⁎ (0.002)

−0.007⁎⁎ (0.002) −0.006⁎ (0.003)

−0.003 (0.003) −0.002 (0.004)

All non-pass-through payments were for the full amount −0.009⁎⁎⁎ (0.002) −0.013⁎⁎⁎ (0.004) x

−0.007⁎⁎⁎ (0.002) −0.014⁎⁎ (0.005) x x

−0.006⁎⁎ (0.002) −0.014⁎⁎ (0.005) x x x

−0.006⁎ (0.003) −0.014⁎ (0.006) x x x x

Robust clustered standard errors in parentheses. ⁎⁎⁎ p b 0.001. ⁎⁎ p b 0.01. ⁎ p b 0.05. + p b 0.1.

a child support order than fathers with children already receiving public assistance. This is particularly true given our use of a lagged unemployment rate, which allows for fathers to recover from any reductions in earnings or job separations that might have affected them directly during a downturn. Thus, we would expect that selection into the child support enforcement system by this group of mothers would weaken whatever (predicted negative) association there is between unemployment and child support compliance among existing cash assistance recipients. For this reason, as a fourth step we estimate models over the subsample that includes only cash assistance recipients reporting on non-pass through payments. Because this group is already in the formal child support system and is reporting on support received above and beyond whatever is passed-through, estimates from these models should be free from selection bias and measurement error from the sources discussed above. As an additional set of analyses we also generate models reflecting sample restrictions designed to identify subgroups of NCPs of special interest. For these analyses, we focus exclusively on our second set of child support outcomes, which reference non-pass through payments and are thus less likely to include measurement error. As above, we also present separate results for the subsample of cash assistance recipients to address concerns about selection bias. We first re-estimated the model on subsamples of custodial mothers with child support orders covered and not covered by the IV-D program. Immediate wage withholding more frequently applies to child support orders covered by the IV-D program, so, we expect full compliance with child support to be more responsive to unemployment rates in the first (IV-D program) subsample. Though child support orders covered by the IV-D program are more likely subject to automatic wage-withholding, NCPs with IV-D orders may be more likely to be laid off during economic downturns and in turn may not be covered by unemployment insurance. Next, we re-estimated the models after restricting the samples to married, not married, and never-married mothers. As compared with evermarried mothers, never-married mothers are more likely to have children by disadvantaged fathers, (e.g. young, black and Hispanic, less-educated males), who experience large increases in unemployment during economic downturns (Hoynes et al., 2012). Therefore, we expect compliance to be more responsive to unemployment rates in the subsample of never married mothers.

We also hypothesize that the level of education will differentiate responses among NCPs. Education is a good predictor of wealth. During economic downturns more educated -wealthier- fathers typically have more certainty of their jobs, and even if they do not, these individuals have a wider variety of assets, access to credit and savings that let them maintain economic stability. On average, the employment statuses of educated NCPs are less responsive to economic fluctuations, and as a consequence their compliance should not change abruptly. This group could overpower the response of less educated fathers in partially or totally adjusting their compliance in economic downturns. Again assuming assortative mating, we include mother's education to infer education levels for NCPs. Finally, we examined the association between the unemployment rate and our measures of compliance in states classified as having low, medium, or high child support expenditures. States with high levels of expenditures may be more successful than other states in deterring noncompliance, because they can detect and respond to delinquencies more rapidly and effectively. We used our preferred model that included state dummies, state linear time trends, and year dummies for all sub-group analyses. 5. Results Table 1 presents summary statistics for the sample. The average mother received about $4000 dollars per year in child support. Though not shown in the table, the median amount of support received was roughly $2590, the average among those receiving any support was around $5204; and fewer than 5% of mothers reported receiving more than about $12,800. Just over three quarters of the mothers received any child support payment in the previous year though only 46% received the full amount due to them. Half of mothers reported receiving all non-pass through payments; 35% reported that all the non-passthrough payments they received were on time, and 49% reported that all the non-pass through payments were for the full amount. Though on average mothers reported receiving just about three quarters of all support due to them, about 5% of mothers received more than was due. In many of these cases, mothers were paid a fractional amount more than what was due. For example, nearly a third of this group were paid between 100.1% and 110% of what they were owed. In a small number of cases, mothers reported receiving proportionally far

136

R.B. Mincy et al. / Children and Youth Services Review 65 (2016) 127–139

Table 3 Unemployment rates and child support compliance - subgroup analyses. Coefficient for the lagged unemployment rate (robust SE) IV-D program participation

Mother received all Non-pass-through payments Cash assistance YES

Marital status

Not IV-Da

Married

Not married

Never married

HS degree or less

Some college or associates

BA degree or higherb

Low

Medium

High

−0.008+ (0.004) −0.013⁎⁎

−0.001 (0.004)

−0.009+ (0.005) 0.008 (0.047)

−0.006 (0.005) 0.004 (0.009)

−0.004 (0.004) −0.020⁎⁎

−0.005+ (0.003) −0.014⁎⁎

−0.006 (0.006)

(0.007)

(0.005)

−0.009 (0.007) −0.012 (0.018)

−0.006+ (0.003) −0.013+ (0.007)

−0.009 (0.006) 0.006 (0.013)

−0.003 (0.007) −0.020+ (0.011)

0.001 (0.004) 0.014 (0.048)

−0.005 (0.003) −0.020⁎

−0.001 (0.002) −0.013⁎

−0.007+ (0.004) −0.025⁎

0.002 (0.007)

0.002 (0.004) −0.014⁎

(0.009)

0.001 (0.004) −0.011+ (0.007)

(0.006)

(0.011)

(0.006)

−0.008 (0.005) −0.009 (0.013)

−0.004 (0.005) −0.017 (0.010)

−0.006 (0.005) −0.017 (0.012)

−0.005 (0.004) −0.010 (0.009)

−0.005 (0.004) −0.016⁎⁎

−0.007⁎ (0.003) −0.017 (0.018)

−0.012⁎⁎ (0.004) −0.015+ (0.008)

−0.002 (0.006) 0.016 (0.014)

−0.002 (0.005) −0.023 (0.014)

−0.007⁎⁎ (0.002) −0.017⁎⁎⁎

0.010⁎⁎ (0.004)

(0.005) All non-pass-through payments Were for full amount Cash assistance YES

Child support enforcement expenditures

IV-D

(0.004) All non-pass-through payments Were received on time Cash assistance YES

Education

−0.009⁎⁎ (0.003) −0.014⁎ (0.006)

0.001 (0.004)

+

−0.007 (0.004) −0.054 (0.079)

(0.005)

−0.008 (0.007)

Robust clustered standard errors in parentheses. All analyses include all controls, state and year dummies, and linear time trends. ⁎⁎⁎ p b 0.001. ⁎⁎ p b 0.01. ⁎ p b 0.05. + p b 0.1. a There are no cash assistance recipients for those not in the IV-D program, and so analyses by cash assistance are omitted for this column. b Sample sizes were too small for cash assistance recipients with a BA degree or higher and results are omitted.

more than what was due, accounting for the high maximum value for this variable. There was great variability in our pooled unemployment rate; the average lagged unemployment rate was 5.59%, but varied from a low of 1.7% (Danbury, Connecticut in 2000 and Fargo, North Dakota in 1998) to over 30% (Yuma, Arizona in 1996). Over 70% of custodial mothers were participants in the Federal IV-D program. Nearly 70% were White, not Hispanic. Only a quarter were married at the time of the survey. Over half had received at least some college education. Mothers were about 36 years-old on average and they had about 1.7 of their own children living in their homes. Finally, about 70% of the sample lived in a MSA of some size. Table 2 presents the results of our main analyses. For the ease of presentation, we report only results related to the lagged unemployment rate here and in the subsequent Table. Full results are available upon request. Row (1) presents the results of analyses with the full sample and examining our first set of child support compliance outcomes. The lagged unemployment rate was not significantly associated with the proportion of child support received in any model. Similarly, unemployment was associated with the probability of receiving any payments only in the models with the most basic set of controls. Unemployment was significantly and negatively associated with the probability that a mother received all that was due even after the inclusion of state dummies and state linear time trends. However, the coefficient was no longer significant in the most restrictive model, which included year dummies. Row (2) presents results from analyses with the subsample of custodial mothers who did not receive cash assistance and would be less likely to misreport the amount of child support that would be paid. The pattern of results in this row are nearly identical to those for the full sample: the lagged unemployment rate was not significantly associated with any measure of compliance in the most restrictive models, which included controls, state and year indicators, and state linear time trends. Results from our second set of compliance measures, which focused on the receipt of non-pass through support are shown in Row (3). Like Row (2), this set of analyses is meant to address the potential for measurement error in mother's reports of child support compliance. Contrary to earlier results, in even the most restrictive models, higher unemployment rates were associated with marginally significant

decreases in the probability that a mother received all non-pass through payments (− 0.006, p b 0.10) and statistically significant decreases in the probability that all such payments were for the full amount (−0.006, p b 0.05). Finally, row (4) shows the results from models that examined nonpass through support among the sub-sample of cash assistance recipients. As noted, these analyses have the benefits of removing the potential for selection into the child support system in a down economy as well as the likely source of measurement error associated with maternal reports. In these analyses, the unemployment rate was negatively and significantly associated with each measure of compliance in all models. Coefficients were much larger than among the full sample of custodial mothers. In the most restrictive models, a one percentage point increase in the lagged unemployment rate was associated with a 0.013 decrease in the probability that mothers received all non-pass through payments (p b 0.01), a 0.017 point decrease in the probability that mothers received all payments on time (p b 0.001) and a 0.014 point decrease in the probability payments were all for the full amount (p b 0.05). Table 3 presents the results of subgroup analyses, which focused on non-pass through child support payments among the full sample and among the subsample of cash assistance recipients. As indicated in the Table footnotes, separate analyses by cash assistance receipt were not possible for those not in the IV-D program (cash assistance recipients are necessarily in the IV-D program), and sample sizes were too small to generate estimates for the group of cash assistance recipients who had a college degree or greater. Generally, results in the table suggest that the relationship between the unemployment rate and receipt of non-pass through support was strongest (most negative) and significant among the most disadvantaged groups of mothers: those who were in the IV-D program and those with a high school degree or less. Further, unemployment was most consistently negatively associated with compliance among mothers living in states that spent the least on child support enforcement. There was also evidence of a negative association between unemployment and some measures of compliance among mothers with some college or an associate's degree. In most cases, when an association was present among the full sample, the relationship was more pronounced among the sub sample of cash assistance recipients.

R.B. Mincy et al. / Children and Youth Services Review 65 (2016) 127–139

6. Summary and implications Surprisingly few studies find significant associations between local unemployment rates and child support compliance after controlling for characteristics that proxy earnings. Most studies use earnings rather than unemployment to measure the effect of economic downturns on compliance. Since data on earnings and child support compliance is generally unavailable, these studies do not provide nationally representative estimates of the association between unemployment rates and child support compliance. Other studies that incorporate local unemployment rates have covered periods of economic growth or mild recessions. This study takes a macroeconomic perspective that uses nationally representative data from the CPS-CSS to provide reduced form estimates of the association between unemployment rates and child support compliance. These estimates incorporate effects of unemployment operating through earnings and child support orders. We use an empirical strategy that examines various measures of compliance among cash assistance recipients and non-cash assistance recipients to develop estimates of the relationship between unemployment and the receipt of child support that are free from likely sources of measurement error and selection bias. We also provide estimates for mothers residing in states with varying levels of expenditures on child support enforcement and for subgroups of the population that are more or less vulnerable to the substantial economic fluctuations that occurred between 1993 and 2011. This is an important period in the literature on compliance for two reasons. First, except for the first year of our data, immediate wage withholding (the most effective provision for enforcing child support) was in effect for all new child support cases by Congressional mandate. Second, our data includes most of the 1990s economic boom, the brief recession of 2001, the jobless recovery of 2003–2006, and all of the 2007–2009 recession. Our preferred models control for state fixed effects, year fixed effects and a state-specific linear time trend, to account for unobserved factors, which could affect child support compliance. Despite the restrictiveness of these models, we find that the unemployment rate is associated with decreases in some measures of compliance in our full sample. Our initial models found little evidence of a relationship between unemployment and the proportion of child support received, the probability that a mother received any payments, or the probability that a mother received all that was due in the full sample of custodial mothers. However, as we note above, these estimates were likely subject to both measurement error (because of misreporting from mothers who received cash assistance and were thus subject to state disregard and past through policies) and selection bias because of the presence of more affluent mothers who are most likely to be drawn into the child support system in the face of economic downturns. In models that attempted to address concerns about measurement error by focusing on the subset of respondents who did not receive cash assistance, the unemployment rate was not significantly associated with compliance in any of the most restrictive models. This is consistent with the intuition we develop earlier in the paper; because mothers not receiving cash assistance are better off than their counterparts, they are paired with non-custodial fathers who are better able to respond to a child support order. Thus, selection into the child support enforcement system by these mothers results in a weaker predicted association between unemployment rates and compliance compared with the association that exists among existing cash assistance recipients. However, when we focused on a group of compliance outcomes related to non-pass through child support payments, which should also be free of error due to misreporting, there was a much more consistent relationship between unemployment and compliance. In particular, we find that a one percentage point increase in the lagged unemployment rate is associated with a 0.006 reduction in the probability that all non-pass-through payments were for the full amount and a marginally significant association between the unemployment rate and the

137

probability that mothers reported receiving all non-pass-through child support payments (Table 2). Further, in models that sought to address concerns about selection into the child support system by more affluent mothers by restricting analyses to the sample of cash assistance recipients, we found that the unemployment rates were consistently, strongly, and negatively associated with compliance. For example, we found that a one point increase in the lagged unemployment rates was associated with a decrease of 0.014 in the probability that passthrough payments were for the full amount. A potential concern related to these results is that recipients of public assistance have become an increasingly select group over time (particularly because of the large purges in public assistance caseloads after the welfare reform), and that our results are driven by this selectivity. However, we note that our preferred models include both state-specific linear time trends and year fixed effects, which should account for any trend toward selectivity among this group. Given the volatility in unemployment rates during recent recessions and in the Great Recession in particular, these findings suggest the potential for a large-scale impact of macroeconomic factors on the consistent provisions of child support. Though our focus on cash recipients was intended to address selection, our results also point to the strongest associations between unemployment and compliance among custodial mothers (and non-custodial fathers) who are most disadvantaged. Indeed, our subgroup analyses (Table 3) seem to confirm this general observation and the intuition we develop earlier in the paper. Our findings here suggest that the relationship between unemployment and child support receipt is most robust among those mothers who are attached to fathers whose employment is most sensitive to economic downturns. This includes mothers whose child support orders are enforced by the IV-D program and particularly those with the lowest level of education. In addition, we find evidence that child support compliance is negatively associated with increases in the unemployment rates among mothers living in states that were in the lowest third for per capita child support expenditures. Though a less-consistent set of findings related to never married mothers is counterintuitive at first glance, we note that these results may reflect the fact that never married mothers have become more common over time, and thus less selected as a group. Indeed, analyses of our dataset suggest that the proportion of never married mothers in our sample grew from roughly 19% in 1993 to nearly 29% in 2011. In our efforts to address concerns about selection into the childsupport enforcement system, we focus on a sample of cash-assistance recipients. The fathers of children born to these mothers are likely to be more disadvantaged than the fathers of children born to non-cash assistance recipients. These former fathers are at greater risk of layoffs and reduced hours during an economic downturn than the latter. Earlier, we were critical of prior studies that focused on samples drawn from administrative records including families receiving some form of public assistance. Doing so, we argued, would overestimate the association between unemployment and compliance. Thus, a limitation of our study is that some of our estimates, which reduce the risk of selection bias, are for the same reason likely to overestimate the association between unemployment and compliance. It is also worth noting that a possible alternative explanation for the large association between unemployment and compliance is a general decline in the cash assistance caseload over the period of our study and the increase in the proportion of highly disadvantaged mothers who remain and who are likely paired with fathers with less ability to pay. That this population is so distinct from the general population of mothers receiving child support might be construed as diminishing the policy significance of our results. However, an alternative framing is that policy might more easily be targeted to protect a well-defined and small vulnerable population from reduced child support payments during economic downturns. Indeed, there is a historical precedent for this sort of targeted protection. After the recession of 2001, a number of states implemented procedures (including meetings with parents, employment assistance, reminders about upcoming payments, etc.) to

138

R.B. Mincy et al. / Children and Youth Services Review 65 (2016) 127–139

encourage payments among cases deemed at high risk for non-compliance (Office of Child Support Enforcement, 2008b). These efforts were especially effective in increasing the amount of child support orders established, payments received and the proportion of payments paid in cases receiving public assistance and were feasible because of the relatively small number of high risk cases. Similar approaches could be successful among recipients of public assistance during economic downturns. Another limitation is that our study, like most studies of child support compliance relies upon mother's reports of the child-support they receive, rather than the amounts fathers actually pay. However, in the absence of administrative data, which has its own limitations including its availability, there are few alternatives to this source of measurement error. Likewise, our limited information about the specific areas of residence for non-resident fathers forced us to make a number of assumptions in order to assign unemployment rates. However (and as we note above), we believe that the nature of these assumptions would tend to bias our estimates toward zero, meaning that the association between unemployment rates and child support compliance may be stronger than we find in this paper. Our findings suggest that reductions in child support compliance is likely a factor linking cyclical increases in child poverty to cyclical increases in male unemployment, which have been larger than cyclical increases in female unemployment since the early 1980s. Using these estimates, policymakers can get a more accurate estimate of the full cost of recessions. However, policymakers should also reflect on this link when they consider conditioning subsidies on full compliance with child support, especially during recessions, when such benefits are needed most. For example, in recent years policymakers have proposed expanding the Earned Income Tax Credits to low-income single adults, including nonresident fathers. To date, all such proposals require that recipients of such expanded credits must have paid the full amount of child support due. Our results suggest that recession-induced reductions in compliance would make some nonresident fathers ineligible for these expanded credits. An important objective of the Affordable Care Act was to decrease the number of uninsured using a refundable tax credit to subsidize health care insurance purchased by adults with incomes above 133% of the poverty line. Nonresident fathers were an important target population for this subsidy. Unfortunately, under federal law, the IRS would intercept this refundable tax credit (along with any other tax refunds) owed to a nonresident father who failed to pay the full amount of child support due after being laid off during a recession. Thus, our findings suggest that the link between unemployment and compliance with child support could undermine efforts to increase work and access to healthcare for nonresident fathers who are most at risk during economic downturns. In addition, there might be other links between unemployment and fathers' non-economic involvement and parents relationship that can have long term consequences on child wellbeing. Appendix 1

P ¼ α 1 þ β11 O þ β12 E þ E ¼ α2 þ

Xn i¼1

Xn i¼1

γ 1i X i þ δ1 U þ ε1

X i þ δ2 U þ ε2

O ¼ α 3 þ β31 E þ

Xn i¼1

γ 3i X i þ δ3 U þ ε3 ;

ð2Þ ð3Þ ð4Þ

where P, O, and E are endogenous variables denoting: payments, orders, and earnings, respectively, with coefficients βkj, (k = 1, 3 and j = 1, 2). U is the (exogenous) unemployment rate with coefficients δk b 0, (k = 1, 2, 3). The Xi are exogenous variables, especially demographic characteristics, with coefficients γki (k = 1…3 and i = 1…n). Finally, εk (k = 1, 2,

3) are disturbances. Substituting Eqs. (3) and (4) into Eq. (2) yields: P ¼ α 1hþ β 11 ðα 3 þ β 31 α 2 Þ  i Xn Xn Xn Xn þ β11 γ þ β31 i¼1 γ 2i þ β12 i¼1 γ 2i þ 1i X i i¼1 3i i¼1 þ ½β11 ðδ3 þ β31 δ2 Þ þ ðβ12 δ2 Þ þ δ1 U þ ½β11 ðβ31 ε2 þ ε3 Þ þ β12 ε2 þ ε1 : ð5Þ While the signs of these complex coefficients are generally ambiguous, the coefficient of U is not.

References Argys, L. M., & Peters, H. E. (2001). Interactions between unmarried fathers and their children: The role of paternity establishment and child-support policies. American Economic Review, 91, 125–129. Bartfeld, J., & Meyer, D. R. (2003). Child support compliance among discretionary and nondiscretionary obligors. Social Service Review, 77, 347–372. Beller, A. H., & Graham, J. W. (1986). Child support awards: Differentials and trends by race and marital status. Demography, 23, 231–245. Case, A. C., Lin, I. F., & McLanahan, S. (2003). Explaining trends in child support: Economic, demographic, and policy changes. Demography, 40, 171–189. Cassetty, J., Cancian, M., & Meyer, D. (2002). Child support disregard policies and program outcomes: An analysis of data from the OCSE. W-2 child support demonstration evaluation report on nonexperimental analyses, vol. 3, . Council of Economic Advisers. 1997. Explaining the decline in welfare receipt, 1993–1996. Washington: White House (URL: http://www.whitehouse.gov/wh/eop/cea/welfare/ technicaLreport.html). Edge, R. M., Kiley, M. T., & Laforte, J. (2010). A comparison of forecast performance between federal reserve staff forecasts, simple reduced-form models, and a DSGE model. Journal of Applied Econometrics, 25, 720–754. Freeman, R. B., & Waldfogel, J. (2001). Dunning delinquent dads: The effects of child support enforcement policy on child support receipt by never married women. The Journal of Human Resources, 36, 207–225. Furstenberg, F. F., & Hughes, M. E. (1998). Social capital and the role of fathers in the family. In A. Booth, & A. C. Crouter (Eds.), Men in families. When do they get involved (pp. 295–301). Mahway, NJ: Lawrence Erlbaum. Garfinkel, I., Huang, C., McLanahan, S. S., & Gaylin, D. S. (2003). The roles of child support enforcement and welfare in non-marital childbearing. Journal of Population Economics, 16, 55–70. Grall, T. S. (2011). Custodial mothers and fathers and their child support: 2009. Current population reports (pp. 60–240). U.S. Census Bureau. Ha, Y., Cancian, M., Meyer, D. R., & Han, E. (2008). Factors associated with nonpayment of child support. University of Wisconsin - Madison: Institute for Research on Poverty (retrieved 2009, retrieved from: http://www.irp.wisc.edu/research/childsup/ cspolicy/pdfs/2007-09/T7-FactorsNonPayCS-Report.pdf). Ha, Y., Cancian, M., & Meyer, D. R. (2010). Unchanging child support orders in the face of unstable earnings. Journal of Policy Analysis and Management, 29, 799–820. Ha, Y., Cancian, M., & Meyer, D. R. (2011). The regularity of child support and its contribution to the regularity of income. Social Service Review, 85, 401–419. Hanson, T. L., Garfinkel, I., McLanahan, S. S., & Miller, C. K. (1996). Trends in child support outcomes. Demography, 33, 483–496. Hatcher, D. L., & Lieberman, H. (2003). Breaking the cycle of defeat for deadbroke noncustodial parents through advocacy on child support issues. Clearinghouse Review, 37, 5–22. Henry, R. K. (1999). Child support at a crossroads: When the real world intrudes upon academics and advocates. Family Law Quarterly, 33, 235–264. Hong, C., Mellgren, L., Desale, S., & Sinclair-James, B. (2016). Characteristics of families using title IV-D services in 2005 and over time. Washington, D.C.: U.S. Department of Health and Human Services, Office of the Assistant Secretary for Planning and Evaluation (forthcoming). Hoynes, H., Miller, D. L., & Schaller, J. (2012). Who suffers during recessions? The Journal of Economic Perspectives, 26(3), 27–47. Huang, C. (2002). The impact of child support enforcement on nonmarital and marital births: Does it differ by racial and age groups? Social Service Review, 76, 275–301. Huang, C. (2009). Trends in child support from 1994 to 2004: Does child support enforcement work? Journal of Policy Practice, 9, 36–53. Huang, C., & Edwards, R. L. (2009). The relationship between state efforts and child support performance. Children and Youth Services Review, 31, 243–248. Huang, C., Garfinkel, I., & Waldfogel, J. (2004). Child support enforcement and welfare caseloads. Journal of Human Resources, 39, 108–134. Kusmin, L. D., & Parker, T. S. (2006). Rural employment at a glance. United States Department of Agriculture, Economic Research Service (retrieved 2009, retrieved from http://www.ers.usda.gov/media/866662/eib21_002.pdf).

R.B. Mincy et al. / Children and Youth Services Review 65 (2016) 127–139 Lippold, K., & Sorensen, E. (2013). Characteristics of families served by the child support (IVD) program: 2010 census survey results. The Urban Institute (retrieved 2014, retrieved from: http://www.urban.org/uploadedpdf/412926-characteristics-of-familiesserved-by-the-child-support-iv-d-program.pdf). Meyer, D. R., & Hu, M. C. (1999). A note on the antipoverty effectiveness of child support among mother-only families. Journal of Human Resources, 225–234. Meyer, D. R., Ha, Y., & Hu, M. (2008). Do high child support orders discourage child support payments? Social Service Review, 82, 93–118. Miller, D. P., & Mincy, R. B. (2012). Falling further behind? Child support arrears and fathers' labor force participation. Social Service Review, 86, 604–635. Nepomnyaschy, L., & Garfinkel, I. (2010). Child support enforcement and fathers' contributions to their nonmarital children. The Social Service Review, 84, 341–380. Office of Child Support Enforcement (1997). FY 1997 annual report to Congress. Administration for Children and Families, Office of Child Support Enforcement. Office of Child Support Enforcement (2002). FY 2002 annual report to Congress. Administration for Children and Families, Office of Child Support Enforcement. Office of Child Support Enforcement (2004). FY 2004 annual report to Congress. Administration for Children and Families, Office of Child Support Enforcement. Office of Child Support Enforcement (2006). FY 2006 annual report to Congress. Administration for Children and Families, Office of Child Support Enforcement. Office of Child Support Enforcement (2008a). FY 2008 annual report to Congress. Administration for Children and Families, Office of Child Support Enforcement. Project to avoid increasing delinquencies recap. (2008).Office of Child Support Enforcement (retrieved from: http://www.acf.hhs.gov/sites/default/files/ocse/dcl_08_05a.pdf). Office of Child Support Enforcement (2009). FY 2009 annual report to Congress. Administration for Children and Families, Office of Child Support Enforcement. Rogers, W. (1994). Regression standard errors in clustered samples. Stata Technical Bulletin, 3, 19–23.

139

Sahin, A., Song, J., & Hobijn, B. (2010). The unemployment gender gap during the 2007 recession. Current Issues in Economics and Finance, 16(2). Sorensen, E., & Hill, A. (2004). Single mothers and their child-support receipt how well is child-support enforcement doing? Journal of Human Resources, 39, 135–154. U.S. Bureau of Labor Statistics (2014a). Employment status of the civilian noninstitutional population 16 years and over by sex, 1970s to date. (retrieved, 2012, retrieved from http://www.bls.gov/cps/cpsaat02.pdf). U.S. Bureau of Labor Statistics (2014b). Local area unemployment statistics. (retrieved, 2012, retrieved from http://www.bls.gov/lau/#data). U.S. Census Bureau (2007). Current population survey, March/April 2006 match file: Child support. (retrieved, 2009, retrieved from http://www.census.gov/prod/techdoc/cps/ cpsmarapr06.pdf). Vroman, W. (2005). The recession of 2001 and unemployment insurance financing. Economic Policy Review, 11, 61–79. Weiss, Y., & Willis, R. J. (1985). Children as collective goods and divorce settlements. Journal of Labor Economics, 3, 268–292. Williams, R. L. (2000). A note on robust variance estimation for cluster-correlated data. Biometrics, 56, 645–646. Wu, C. (2011). Child support in an economic downturn: Changes in earnings, child support orders, and payments. (retrieved, 2011, retrieved from http://www.irp.wisc.edu/ research/childsup/cspolicy/pdfs/2009-11/T8b2011ChiFangWu.pdf). Ziliak, J. P., Figlio, D. N., Davis, E. E., & Connolly, L. S. (2000). Accounting for the decline in AFDC caseloads: Welfare reform or the economy? Journal of Human Resources, 570–586.