Single mothers working at night: standard work and child care subsidies.

Tekin, Erdal

A Single parent is constantly scrambling for someone to care for her two preschool daughters while she processes paperwork and inmates at a county jail in Winston-Salem, N.C. Hers is an unpredictable, ever-shifting schedule. Some days, she picks up the phone and calls in her sister, who's often busy with her own teen-ager. Some nights, she persuades a friend with two little ones of her own to let the girls sleep over. Her predicament is shared by millions of Americans who find themselves working long, nonstandard, or erratic hours and having to hunt for child care to match.

--Hays (1995)

I. INTRODUCTION

Working outside the "standard" weekday hours of 8 A.M. to 6 P.M. between Monday and Friday is an increasingly common practice in the United States. For example, 34.3% of all female workers in the United States were nonstandard workers in 1995 (Kalleberg et al. 1997). The investigation of nonstandard work is important for a number of reasons. First, there is evidence suggesting that workers engaged in nonstandard work are more likely to be assigned to routine jobs and to receive less training and fewer promotions than others (Barker 1993; Tilly 1996). Consequently, these workers tend to earn less, and they are less likely to have health insurance and pension benefits than standard workers (Hipple and Stewart 1996; Loprest 1999 and 2002). There also exists a positive link between the quality of an initial job and the likelihood of maintaining employment over time (Cancian and Meyer 2000; Rangarajan et al. 1998; Strawn and Martinson 2000). Second, nonstandard work is linked to a number of adverse outcomes for parents and children, such as work and family conflicts, marital instability, health problems for both parents and children, and poor educational outcomes for children (Heymann 2000; Presser 2000; Staines and Pleck 1983). Finally, the majority of nonstandard workers view employment during nonstandard hours as an accommodation to labor market needs, not as a personal preference. According to the Current Population Survey, more than half of the workers with nonstandard schedules report the nature of their jobs as the reason for their choice. Only about 6% of nonstandard workers report working such schedules for better pay, and only 4% give better child care as their reason for working nonstandard schedules (Beers 2000).

With the passage of welfare reform in 1996, child care assistance has become a more significant tool for helping welfare recipients move into the workforce and for helping other low-income families stay off welfare. (1) According to the General Accounting Office (GAO), the majority of states make welfare recipients and families transitioning from welfare to work eligible for child care assistance or give them priority over other low-income families when resources are insufficient to cover all who apply (GAO 2003). Almost a decade after the passage of welfare reform, Congress now debates legislation to reauthorize welfare reform, and child care funding remains a key issue. However, very little is known about whether child care subsidies have in fact played a role in increasing employment among welfare recipients, or in general, low-income individuals in the post-welfare reform period (Blank 2002). Even less is known about the effect of these subsidies on standard-nonstandard employment decisions of these individuals.

Since the passage of welfare reform, the employment rate of single mothers has risen dramatically (Jones-DeWeer et al. 2003). However, leaving welfare does not necessarily mean gaining adequate work and increasing economic serf-sufficiency. In fact, only 8% of welfare leavers have been able to sustain employment over a period of four years (Martinson 2000). Over three-quarters (78%) of employed low income single mothers are concentrated in typically low-wage and low benefit occupations (Jones-DeWeer et al. 2003). These occupations typically demand a greater number of hours outside the standard weekday times of 8 A.M. to 6 P.M. (2) About one quarter of all welfare leavers worked night shifts or had irregular schedules on a regular basis in 1999 (Loprest 2002). Despite gains in employment, about 52% of those who left welfare in 1999 had incomes below the poverty level (Nightingale 2002). Welfare reform might have been successful so far in helping welfare participants secure entry-level jobs. However, there is a great deal of concern over the possibility that many former welfare recipients who have gone to work are having difficulty finding stable employment and are working at jobs with low wages and few benefits. This article examines the usefulness of child care subsidies in helping mothers find jobs with conventional or standard schedules, the kind of jobs that usually pay higher wages, provide better benefits, and lead to long-term economic self-sufficiency of parents.

Using data from the 1999 National Survey of America's Families (NSAF), a binary model of standard/nonstandard employment is estimated jointly with the binary models of subsidy receipt and labor force participation to control for potential endogeneity and selectivity problems. To address whether there exists a differential effect of subsidy receipt on standard work between welfare recipients and nonrecipients, the equations are reestimated with an interaction of welfare and subsidy indicators in the standard work model, accounting for the potential endogeneity of welfare receipt. This investigation is particularly important because many states give priority to families leaving welfare for child care assistance (Schumacher and Greenberg 1999). (3)

The rest of the article is organized as follows. Section II reviews the previous literature. The econometric approach is described in section III. Section IV introduces the data. Section V presents the results, and section VI concludes.

II. PREVIOUS LITERATURE

Although this is the first publication to examine the effect of child care subsidies on standard work decision, it is not the first to consider the relationship between child care subsidies and employment in general. However, most of the literature on the impact of child care subsidies on employment focuses on the pre-welfare reform period. Because welfare reform changed the systems of welfare and child care assistance dramatically, results from the pre-welfare reform period may be a less relevant guide to the impact of current subsidies.

One body of evidence on the association between child care subsidies and employment comes from several demonstration projects designed to help economically disadvantaged families. These projects include child care subsidies along with other benefits and services. Most of these projects are conducted as randomized experiments prior to the 1996 welfare reform legislation and they typically find that employment increased as a result of the treatment. However, the child care subsidy is only one of a large number of services and benefits provided to the treatment group. Therefore, it is not possible to isolate the actual subsidy impact from the overall program impact. (4)

The largest source of evidence on the effect of child care subsidies on employment comes from the studies estimating the effect of the cost of child care on employment. These studies typically exploit the variation in child care costs across individuals and the geographic variation in the cost of child care. (5) They implicitly rely on the assumption that there are no costs to taking up a subsidy in the form of either the time costs required to deal with the bureaucratic system or the stigma of participating in a means-tested program. If this assumption is not true however, then the price effect would not be a reliable guide for the subsidy effect.

Research on the impact of actual child care subsidies has been limited, primarily due to a lack of data. Berger and Black (1992) and Gelbach

(2002) examine the effect of child care subsidies by comparing the employment of two groups of mothers who differ in their access to child care subsidies by a natural experiment. Both of these studies find positive impacts of child care subsidies on maternal employment. Meyers et al. (2002) use data from a sample of low-income single mothers (current and recent welfare recipients in California between 1992 and 1995) to estimate the probability of their child care subsidy receipt and the effect of this probability on labor market activity. The authors find that the probability of subsidy receipt is associated with an increase in the probability of employment. Blau and Tekin (forthcoming) analyze the determinants of receipt of child care subsidy and the effect of subsidy receipt on employment, unemployment, school attendance, and welfare participation using data from the 1999 NSAF. The authors control for the endogeneity of child care subsidy receipt using instrumental variables. They find positive effects of child care subsidy receipt on employment.

The information on the link between the standard/nonstandard work and child care decisions of mothers is very limited and mostly descriptive in nature. A key difference between the present analysis and the previous research is that the focus in this article is on the standard/nonstandard work decision while the previous studies concentrated on the effect of standard or nonstandard work on some other outcome measure like modes of child care or the decision to receive a child care subsidy (Brayfield 1995; Burstein et al. 2001; Casper and O'Connell 1998; Chaplin et al. 1999; Georges et al. 2001; Kimmel and Powell 2001; Presser 1986 and 1988). With the exception of Kimmel and Powell (2001), none of them addressed the endogeneity of standard/ nonstandard work status. Kimmel and Powell (2001) examine the impact of standard work on child care choices of single mothers and find that work patterns play an important role in mothers' decisions regarding the mode of child care.

Ill. ECONOMETRIC FRAMEWORK

The objective of the article is to examine the effect of child care subsidy receipt on standard employment. Based on the theoretical model described in the appendix, the econometric model can be expressed by the following equations:

(1) [S.sub.i] = [X.sub.i][beta] + [Z.sub.i][delta] + [T.sub.si][mu] + [[epsilon].sub.i],

(2) [ST.sub.i] = [alpha][S.sub.i] + [X.sub.i][gamma] + [Z.sub.i][zeta] + [v.sub.i] if [E.sub.i] = 1,

where [S.sub.i] is a binary indicator of subsidy receipt for mother i, [ST.sub.i] is the binary outcome of standard employment, [X.sub.i] is a vector of family characteristics, [Z.sub.i] is vectors of policy variables and other characteristics of the location of the residence of the family, and [T.sub.s] is a binary variable indicating whether a single mother who is eligible for a subsidy is actually offered one by subsidy administrators. The [[epsilon].sub.i] and [v.sub.i] are disturbances, and [beta]'s, [delta]'s, [alpha], [gamma]'s, and [zeta]'s are the parameters. As the theoretical model in the appendix implies, the demand for child care subsidies is determined by the price of child care, the mother's wage rate, nonwage income, preferences for consumption relative to leisure, the parameters of the subsidy program, the stigma of participating in a means-tested program, and so on. These factors are determined in turn by family characteristics (X), the observed features of the state child care subsidy system (Z and [T.sub.s]), and unobserved family and state characteristics ([epsilon]). Because equation (1) is a reduced form, it is not possible to identify the supply and demand effects of X, Z, and [T.sub.s]. Therefore, [beta]'s, [delta]'s and [mu]'s are the net effects of demand and supply forces on the subsidy receipt. (6)

Identifying the causal impact of child care subsidy receipt on work schedule in equation (2) is complicated by the possibility that [[epsilon].sub.i] and [v.sub.i] are correlated. For example, a mother who is strongly motivated to work during standard hours may also be motivated to seek a child care subsidy to better accommodate her child care needs, generating a positive correlation between [[epsilon].sub.i] and [v.sub.i]. Alternatively, administrators of the subsidy system may give priority to the least employable mothers (Blau and Tekin forthcoming), imparting a negative correlation. It is also possible that the administrators of the child care subsidy program may think that a child care subsidy would most effectively help those who are standard workers because child care is available mostly during standard hours and mothers are more likely to find a relative for child care during nonstandard hours.

The behavioral model implies that the vector [T.sub.s] is a valid identifying instrument because it can be appropriately excluded from the standard work equation. It is assumed that the average amount of Child Care Development Fund (CCDF) spent per child in a state and the state's percentage of eligible children served by child care subsidies are positively related to [T.sub.s]. These variables would both determine the degree of the state's generosity in providing child care assistance and capture the factors that determine how subsidies are rationed by states. Furthermore, an eligible mother is more likely to receive a child care subsidy in states where mass media are used as a consumer education strategy in child care because she is more likely to be informed about the bureaucratic process, application procedures, and various opportunities for child care assistance. (7) Based on the theoretical model, these factors should not influence equation (2) directly. One may argue that the parameters of the state child care subsidy system, such as the income eligibility limit and the reimbursement rate for subsidies, could also serve as identifying instruments by affecting whether a mother receives a subsidy, but conditional on receiving a subsidy, not affecting the standard employment decision. However, as indicated by the behavioral model, these variables affect the standard employment decision. This is because the parameters that determine eligibility for a child care subsidy would also affect how much a mother can earn and thus the value of being employed and receiving a subsidy (Blau and Tekin forthcoming). (8)

Another complication arises from the fact that standard work status is observed only for workers (i.e., for those with E = 1). To control for selection into employment, equations (1) and (2) are estimated jointly with a binary employment equation. The employment equation can be obtained from the theoretical model similar to the way the standard work model is derived. Then, it can be expressed as a function of all the exogenous variables in the model (9)

(3) [E.sub.i] = [X.sub.i][zeta] + [Z.sub.i][pi] + [T.sub.si][phi] + [eta].sub.i].

Estimation of equations (1), (2), and (3) using full information maximum likelihood requires evaluating a trivariate normal integral. To avoid the difficulty of estimating a multivariate integral, the article employs a quasi-maximum likelihood estimator with discrete approximations. This method is computationally less expensive than the regular maximum likelihood estimator and has been increasingly used recently (e.g., Blau and Hagy 1998; Hu 1999; Lokshin 2004; Mocan and Tekin 2003; Mroz 1999; Picone et al. 2003). Mroz (1999) shows in a Monte Carlo study that this estimator is more robust to deviations from normality and quality of instruments than two-stage methods.

To implement the quasi-maximum likelihood estimator with discrete approximations, the following structure is imposed on the disturbances in equations (1)-(3):

[[epsilon].sub.i] = [rho].sub.1] u + [[lambda].sub.1i],

[v.sub.i] = [rho].sub.2] u + [[lambda].sub.2i],

[[eta].sub.i] = [rho].sub.3] u + [[lambda].sub.3i],

where [[lambda].sub.1], [[lambda].sub.2], [[lambda].sub.3] are independently normally distributed errors and u is a random variable with equation-specific factor loading parameters [[rho].sub.1], [[rho].sub.2], and [[rho].sub.3]. This structure places the restriction that all heterogeneity or the correlation across the error terms enters the model through the common factor u that is assumed to follow a discrete distribution (Heckman and Singer 1984). Specifically, Pr(u = [[omega].sub.k]) = [p.sub.k] [less than or equal to] 0 for k = 1, ..., K and [summation.sup.K][p.sub.k] = 1. The number of points of support K, the location of the support point [[omega].sub.k], and their probabilities [p.sub.k] are called incidental parameters and are estimated jointly with the other parameters of the system of equations. (10) Then the likelihood function for the system of equations can be written as:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII],

where Pr stands for probability.

IV. DATA

The data used in this article are drawn from the second round of the NSAF. The second round of NSAF is conducted by the Urban Institute between February and October 1999. (11) The NSAF sample is representative of the United States civilian, noninstitutionalized population under age 65. Residents of 13 states and households with income below 200% of the federal poverty line were oversampled. The oversampled states contain more than half of the United States population. Interviews were conducted with over 42,000 households. (12)

The NSAF is an ideal data source for the purpose of this study for several reasons. First, it is specifically designed to analyze the consequences of devolution of responsibility for social programs from the federal government to the states. Second, it provides one of the few nationally representative household data sets on child care subsidies. Third, the second round of the NSAF is conducted three years after the enactment of the welfare reform legislation. In this respect, it represents a more comprehensive picture of the post-welfare reform environment. Finally, the NSAF provides a large sample of single mothers. The sample is limited to single mothers because the standard/nonstandard work behavior of married mothers may be quite different from that of single mothers as fathers are likely to be an important source of child care when their spouses are at work. Also, single mothers are the primary target for assistance under welfare law. They accounted for over 90% of Temporary Assistance to Needy Families (TANF) program cases in 1998 (Committee on Ways and Means 2000).

The sample used in the analysis contains 4,405 single mothers with at least one child under age 13. The NSAF contains information on child care subsidy receipt for children under age 13, which is the federal cut-off age for eligibility under CCDF. The mother is asked whether she receives any assistance paying for child care, including assistance from a welfare or a social services agency, her employer, and a noncustodial parent. A mother is coded as receiving a child care subsidy if she reports that a welfare or a social services agency pays for all or part of the cost of child care for any of the children in the family. She is coded as working at a standard job if she reports performing her work during traditional hours of 8 A.M. to 6 P.M. during business days (Monday to Friday). Those who perform their work outside these traditional hours are coded as working at nonstandard jobs. This group may include mothers who work weekends, evenings, split shifts, or irregular daily or weekly schedules because the NSAF does not distinguish between various types of nonstandard hours.

Definitions and the descriptive statistics of variables used in the analysis are presented in Table 1. Column 1 shows the means for the full sample, and column 2 reports them for workers only. Columns 3 and 4 display the means for standard and nonstandard workers, respectively. Column 5 displays the means for standard workers who are subsidy recipients and column 6 displays the means for standard workers who are non-recipients. As shown in column 1, 11.6% of the sample receives a child care subsidy. The Administration for Children and Families (2000) predicts that between 12% and 15% of all eligible families received a CCDF subsidy in 1998-99. The sample in this study includes all single mothers regardless of their income, and some of these mothers are certainly ineligible for subsidies as their incomes exceed the threshold level. Thus, 11.6% subsidy coverage rate is not unreasonable. The employment rate in the sample is 71.1%. Among those who are employed, 20.7% work nonstandard hours. Among workers, the subsidy receipt is higher for those who work standard hours than those who work nonstandard workers (13.1% versus 10.8%).

Welfare recipients constitute about 16% of the sample. This figure matches closely to the Current Population Survey, which indicates a 15% welfare utilization rate for 1999 (Grogger 2003). A higher percentage of nonstandard workers receive welfare than standard workers (14.2% versus 9.3%). This is reasonable given that standard workers usually have higher nonwage income and education than nonstandard workers, which would make it harder for them to be eligible for welfare. Furthermore, among standard workers, child care subsidy recipients are much more likely to be on welfare than nonrecipients, 30.5% versus 6.1%. Given the emphasis of the CCDF on giving priority to welfare recipients, the size of this difference is not surprising.

As Table 1 illustrates, there are major differences in occupations between standard and nonstandard workers. A description of the occupational indicators is provided in Appendix Table A4. Nonstandard workers are concentrated mostly in sectors with high demand for off-hour services. For instance, standard workers are more likely to be employed in managerial, professional specialty, and administrative support occupations than nonstandard workers. On the other hand, they are less likely to work in sales, protective services, service occupations, and occupations such as machine operators, assemblers, inspectors, handlers, helpers, or cleaners. The percentage of single mothers with less than a high school degree is approximately 8% for standard workers and 15% for nonstandard workers. This pattern is entirely reversed for college graduates with 8% of nonstandard and 17% of standard workers holding a bachelor degree or more. These figures are consistent with those documented in previous studies using different data sources (e.g., Kalleberg et al. 1997; Presser and Cox 1997).

Equations (1), (2), and (3) condition on a number of characteristics of the mother that reflects both demand and supply factors. These include age, ethnicity, health status, education, presence of children, family structure, nonwage income, and region of residence. In addition to these variables, the occupation fixed effects are included in the nonstandard employment equation to control for any unobserved differences in demand for standard workers across various occupations. The models also include state's median income, unemployment rate for females, state's percentage of female-headed households with children living under poverty, maximum reimbursement rate for licensed child care, maximum annual income for subsidy eligibility, and monthly copayment for child care for a family of three.

V. RESULTS

The results of the employment equation that is estimated to control for selection into the labor force are displayed in Appendix Table A2. They are mostly consistent with those usually found in the relevant literature. Because this equation is not the central focus of the article, the results are not discussed in the text.

Table 2 presents the estimates of the model for child care subsidy receipt. (13) The first column presents the coefficient estimates, and the second column displays the standard errors. Linear probability models are estimated for ease of interpretation. (14) Blacks are more likely to receive a child care subsidy than both whites and other races. The likelihood of subsidy receipt also increases with the number of children between ages 0-5 and 6-13 and the effect is stronger for the younger age group (5 percentage points versus 2.4 percentage points). High school graduates and those with some college education are 3.6 and 6 percentage points more likely to receive a child care subsidy than high school dropouts, respectively. High school graduates and those with some college education may be more likely to be aware about the availability of child care subsidies than high school dropouts. A mother with a higher nonwage income is less likely to receive a child care subsidy than others. A $10,000 increase in nonwage income results in a three percentage point decrease in the probability of receiving a child care subsidy. The presence of an additional relative in the household decreases the probability of subsidy receipt by about two percentage points.

It is important to note that the coefficients of the variables used as identifying instruments have the expected signs. As displayed in Table 2, living in a state where mass media are used as a consumer education strategy for child care subsidies is associated with a 3.5 percentage point increase in child care subsidy receipt. A 1 percentage point increase in the number of eligible children served by child care subsidies in a state increases the likelihood of subsidy receipt by a single mother by 0.78 percentage point. An increase in the CCDF funds per child by $1,000 is associated with only a 0.49 percentage point increase in the probability of subsidy receipt, and the coefficient estimate is statistically insignificant. It is not surprising that this coefficient is not significant because the model also conditions on the percentage of eligible children. The coefficients on the parameters of the state's subsidy program (copayment, reimbursements rate, and income eligibility limit) also have the expected signs, however, none of the coefficients is statistically significant.

Table 3 displays the results of the model for standard/nonstandard employment equation. The variable of the primary interest, the receipt of a child care subsidy, has a positive and significant coefficient. Single mothers who receive a child care subsidy are 6.9 percentage points more likely to work standard hours than nonstandard hours, all else equal. This finding underscores the importance of child care subsidies on facilitating the transition from nonstandard work to standard work for single mothers.

Mothers with at least a bachelor's degree are more likely to work at standard jobs than others. This is consistent with the fact that standard jobs are more demanding of human capital than nonstandard jobs. Whites are about four percentage points more likely to work at standard jobs than are other races, but the coefficient is not significant. The number of children in the household is associated with a decrease in the likelihood of standard work, although the effect is significant only for younger children. One possible explanation would be that the younger children require more time with their mothers, and standard jobs are the type of jobs that usually require higher number of hours of work than the nonstandard jobs. Looking at the number of hours of work per week in Table 1, one can see that standard workers work longer hours than the nonstandard workers do.

As displayed in Table 3, occupational status is a significant determinant of a single mother's work schedule. Mothers working in technical, sales, and support occupations, as well as protective services, precision production, craft, repairs, farming and fishing, or as machine operators, assemblers, handlers, equipment cleaners and helpers are less likely to work standard schedules, compared to the omitted category (executive, administrative, managerial occupations). This result is not surprising because these are the types of occupations in which the demand for nonstandard hours is usually high (Presser and Cox 1997).

As the descriptive statistics indicate, welfare recipients are more likely to work at nonstandard jobs than are nonrecipients. This may have unintended consequences in the long run as welfare recipients try to advance in their careers over time, especially if it is usually the standard jobs that lead to permanent employment. Therefore, it is important to consider whether the impact of subsidy receipt differs between welfare recipients and nonrecipients. Normally an indicator for mother's welfare receipt and its interaction with the subsidy receipt variable included in the standard/ nonstandard work equation would provide the answer to this question. However, welfare receipt is likely to be endogenous to both subsidy receipt and the standard work decision. (15) Therefore, including welfare receipt as an explanatory variable in the standard work equation might introduce bias to the estimates. To address this problem, the predicted probability of welfare receipt is constructed from a first-stage regression. Then this predicted probability and its interaction with the child care subsidy receipt are included in the standard work equation, which is estimated jointly with the labor force participation and child care subsidy receipt equations using the random effects estimator explained previously. (16) The state's earnings eligibility limit for TANF for a single-parent family of three and the maximum monthly TANF benefit level for a family of three are used as identifying instruments in the first stage.

The results of the first-stage welfare equation are reported in Appendix Table A3. Less educated parents and parents with young children are more likely to use welfare than others. Whites, Hispanics, parents with better health and higher nonwage income are less likely to receive welfare than others. The identifying instruments, the state's TANF earnings eligibility and the maximum benefit level, are statistically significant determinants of welfare receipt. A $100 increase in the earnings eligibility limit for a single parent applicant increases her probability of welfare receipt by 1.18 percentage points. A $100 increase in the maximum TANF benefit level is associated with about 0.7 percentage point increase in welfare receipt.

The results of the standard work equation with the welfare variable and its interaction with subsidy receipt are presented in Table 4. The coefficient estimates on welfare and its interaction with subsidy receipt indicate that child care subsidies serve as a major incentive for welfare recipients to work at standard jobs but have a much smaller impact on nonrecipients. A subsidy-receiving mother is only 1.8 percentage points more likely to work at a standard job than a nonrecipient mother if she is not on welfare. However, if the mother is on welfare, she is 14.0 (1.8 + 12.2) percentage points more likely to work at a standard job when she receives a child care subsidy. Similarly, welfare recipients are about 15 percentage points less likely to work at standard jobs than nonrecipients if they do not receive a subsidy. However, the effect becomes 2.7 percentage points (-0.149 + 0.122) if they receive a subsidy. These results suggest that child care subsidies induce welfare receiving mothers to work at standard jobs, but have a much smaller effect for those who do not receive welfare.

One potential explanation for this result would be that priority for CCDF funds is given to families who are on welfare, and most of these funds have work requirements. In addition to a work requirement for subsidy eligibility, there is also an hours of work requirement to participate on welfare, which was set at 25 hours per week in 1999 and at 30 hours per week in 2000. The welfare reform reauthorization bill that is currently debated introduces even tougher work requirements. For example, 90% of all two-parent families receiving TANF would have to participate in work activities for at least 35 hours per week (Parrott et al. 2006). This underscores that welfare recipients face a stronger pressure to work at longer hours than others. The number of hours of work is higher for standard workers than nonstandard workers in the analyses sample. To the extent that most standard jobs require longer hours of work, this may explain why welfare recipients are more responsive to the child care subsidies than others.

Specification Checks

As discussed earlier, the set of occupation dummies are strong determinants of standard work decision. These dummies are included in the analysis to control for the variation in the demand for standard hours and variability in the labor market conditions among occupations. To ensure that the coefficient estimate of the child care subsidy receipt is not influenced by the possibility of the endogeneity of occupation indicators, the system of equations is estimated without these occupation indicators. Once these indicators are dropped, the effect of child care subsidy receipt is still statistically significant and is equal to 0.059, which is close to the current estimate.

According to the theoretical model described in the appendix, the parameters of the state's subsidy program (reimbursement rate, copayment, income eligibility limit, etc.) must enter all the equations. However, it can be argued that these parameters are endogenous. To address the possibility that the child care subsidy coefficient may be biased due to the endogeneity of the parameters of the state's subsidy program, the system of equations is estimated without these parameters in all three models. In a fully reduced-form model, these parameters are determined by observed parent characteristics, observed features of the state economy, and unobserved parent and state characteristics. Dropping these variables had no substantial effect on the estimates. This result is not surprising because none of the coefficients on these variables was statistically significant in the original model.

The choice of identifying instruments for the coefficient of child care subsidy receipt is theoretically justified by the model presented earlier. The validity of these variables as instruments is further supported by testing whether their coefficients are jointly significant. As mentioned earlier, the p-value from this test is less than 0.001, indicating that they are jointly significant. This is not surprising given the fact that two of the three coefficients are highly significant individually in the subsidy equation as displayed in Table 2.

VI. CONCLUSION

The evidence linking the quality of the initial job to the probabilities of maintaining employment and promoting career advancement suggests that finding a job itself may not necessarily result in moving single mothers toward economic self-sufficiency in the long run. It is therefore important to encourage low-income parents to seek jobs with a potential to move them up the income ladder. This article examines the effectiveness of child care subsidies for accomplishing this goal. Child care subsidies are an integral part of the new welfare system. Though subsidies are not usually limited to parents who are on welfare, they are especially vital for the success of welfare reform because of their role in helping parents make the transition from welfare to work and staying off welfare.

This article provides evidence on the relationship between child care subsidies and standard work using data from the 1999 NSAF. The findings suggest that child care subsidies induce mothers to work at standard jobs. Specifically, single mothers with a child care subsidy are about seven percentage points more likely to work standard hours than others, all else being equal. When the impact of subsidy receipt is allowed to differ between welfare recipients and nonrecipients, results indicate that subsidies generate a relatively substantial incentive for single mothers to work at standard jobs, whereas they have a much smaller impact on nonrecipients. These results underscore the importance of child care subsidies in helping low-income parents, especially welfare recipients, find jobs with a potential for long-term economic self-sufficiency. These findings are particularly meaningful given the states' efforts to prioritize TANF recipients for child care assistance. For example, during 1999, 27 states guaranteed child care assistance to families transitioning from TANF to work, and 15 gave priority to those families (State Policy Demonstration Project 1999). However, according to the GAO, 23 states made changes to their child care assistance programs and decreased the availability of assistance since January 2001, mainly because of the financial crisis they were facing and the exhaustion of TANF surplus from prior years (GAO 2003).

APPENDIX

The simple behavioral model developed here serves as a guide for the econometric model used for estimating the effect of child care subsidies on standard work. Suppose that a single mother allocates her time between leisure and work. She either works during standard hours or nonstandard hours, but not both. If she does not work, she is assumed to provide child care during her leisure hours. During her work hours, she can use market care or receive free care from a relative. Although the choice of paid versus unpaid care and the employment decision of the relative are not part of the empirical model, they are considered in the theory to account for the use of unpaid child care. The mother can receive a child care subsidy if she is eligible for one and is offered a subsidy by the subsidy program administrators. In addition to satisfying the income condition, a mother must either be employed or in a work-related activity to be eligible for a subsidy as required by the law. Finally. it is assumed that the mother may derive disutility from receiving a child care subsidy due to psychic costs or stigma of participating in a means-tested program.

The mother's utility is determined by consumption. hours of work. leisure of her relative, child quality, and child care subsidy receipt. The utility is an increasing function of consumption, child quality, and the leisure of relative, and it is a decreasing function of subsidy receipt. If a mother works during standard hours, she will rely on market care or receive free care from a relative. Similarly. a mother who works during nonstandard hours can use either market care or free care from a relative. It is further assumed that market care during standard hours has better quality than the market care during nonstandard hours and that better quality care is more expensive than the lower quality care. To simplify matters, one can think of market care during standard hours coming from a regulated provider and market care during nonstandard hours coming from an unregulated provider (e.g.. a babysitter). For example, center care. which constitutes the majority of regulated child care market, is more difficult to find during nonstandard hours than standard hours (Cochi-Ficano and Peters 2001; Giannarelli et al. 2001: Layzer 2001). This would cause many single mothers to choose unregulated child care during nonstandard hours. which may be cheaper but has lower quality. Finally. the reimbursement rates are higher for regulated child care than unregulated child care (Cabrera et al. 2002).

Under these assumptions, a mother maximizes her utility subject to her budget and time constraints. Suppressing the subscript that refers to mother i. the utility and the budget constrains can be expressed as follows:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].

where:

U = utility.

C = consumption.

[d.sub.st] = binary indicator of standard work.

[H.sub.st] = work hours during standard hours.

[H.sub.nst] = work hours during nonstandard hours.

Q = child quality,

[L.sub.r] = relative's leisure hours.

[q.sub.s] = the disutility of receiving a subsidy.

S = binary indicator of subsidy receipt,

[L.sub.m] = mother's leisure hours,

J = hours of free care received from a relative,

R = the relative' hours of work,

[M.sub.st] = hours of market during standard hours.

[M.sub.nst] = hours of market during nonstandard hours,

[p.sub.t] = price of child care per hour during standard hours,

[p.sub.nst] = price of child care per hour during nonstandard hours.

Y = nonwage income.

T = total time available to a mother or a relative.

w = hourly wage rate.

[E.sub.s] = the income eligibility limit for child care subsidy.

[r.sub.st] = the subsidy rate per hour of child care at standard hours.

[r.sub.nst] = the subsidy rate per hour of child care at nonstandard hours,

[t.sub.s] = the rate at which child care assistance is reduced as earnings rise,

[T.sub.s] = a binary variable indicating whether an eligible mother is actually offered a subsidy.

Note that [p.sub.nst] > [p.sub.st] and [r.sub.st] > [r.sub.ns] by assumption, and suppose that Q = [k.sub.1][d.sub.st] + [k.sub.2](1 - [d.sub.st]) where [k.sub.1] and [k.sub.2] are exogenous levels of child quality at standard and nonstandard hours with [k.sub.1] > [k.sub.2].

Based on this model, a mother who works during standard hours and uses more expensive but higher quality regulated market care will continue to do so with a subsidy. If she uses free care from a relative, she will again move to market care if the utility gain from releasing her relative from child care outweighs any utility loss due to child care costs net of the subsidy amount. A mother who works during nonstandard hours and uses market care may move to standard hours in response to a subsidy if the utility gain from moving to a higher quality care outweighs any utility loss due to a potential increase in the cost of care. Similarly, a mother working during nonstandard hours and using relative care may move to unregulated care and continue to work at nonstandard hours if the utility gain from freeing a relative of child care duty outweighs any utility loss due to increased cost of child care. Alternatively, she can move to regulated care and work at standard hours if the utility gain from higher quality is high enough to outweigh any utility loss from increased cost of care.

The mother chooses C, [d.sub.st], [H.sub.st], [H.sub.nst], [L.sub.r], S, [L.sub.m], [M.sub.st], [M.sub.nst], and J to maximize her utility subject to her constraints. The set of alternatives available to a single mother are displayed in Table A1. The mother compares her utility in each of these alternatives and then chooses the one that gives her the highest utility.

Let [V.sub.i] be the indirect utility associated with alternative i, which can be derived by solving the optimization problem. Given Table A1, the value of receiving a subsidy can be written as:

V(S = 1) = max{[V.sub.4](Y, w,[p.sub.st], [r.sub.st], [E.sub.s], [q.sub.s], [t.sub.s]), [V.sub.7](Y, w, [p.sub.nst], [r.sub.nst], [E, sub.s], [q.sub.s], [t.sub.s])}.

Similarly, the value of not receiving a subsidy is

V(S = O)= max{[V.sub.1](Y). [V.sub.2](Y, w), [V.sub.2](Y, w, [p.sub.st]). [V.sub.5](Y, w), [V.sub.6](Y, w, [p.sub.nst])}.

A single mother will receive a subsidy if V(S = 1) > V(S = 0). Y + hw [less than or equal to] [E.sub.s], and [T.sub.s] = l. Thus, a reduced-form model of subsidy receipt is a function of nonwage income, prices, all the exogenous variables in the model, and the [T.sub.s].

S = S(Y, [P.sub.st], [P.sub.nst], [r.sub.st], [r.sub.nst], w, [q.sub.s], [E.sub.s], [T.sub.s]).

Let Pr([d.sub.st] = 1|S - E = l) be the probability of standard employment conditional on receiving a subsidy and being employed, where E is a binary indicator of employment. Then

Pr([d.sub.st] = 1|S = 1, E = 1) Pr([V.sub.4](Y, w, [P.sub.st], [r.sub.st], [E.sub.s], [q.sub.s], [t.sub.s]) > [V.sub.7] (Y, w, [p.sub.nst], [r.sub.nst], [E.sub.s] [q.sub.s] [t.sub.])).

Similarly. the probability of standard employment conditional on not receiving a subsidy but being employed is

Pr(d.sub.st] = 1|S = 0, E = 1) = Pr(max{[V.sub.2](Y, w), [V.sub.3](Y, w, [p.sub.st])} > max{(V.sub.5](Y, w). [V.sub.6](Y, w, [p.sub.nst)}).

Therefore, the probability of standard employment conditional on subsidy status can be expressed as

[d.sub.st] = [d.sub.st](S, E, Y, [p.sub.st], [p.sub.nst], [r.sub.nst], [r.sub.st], w, [E.sub.s], [q.sub.s], [t.sub.s].

TABLE A1

Discrete Alternatives in the Theoretical Model

Alternative Work Child Care Subsidy

1 None None None
2 Standard Relative None
3 Standard Market None
4 Standard Market Yes
5 Nonstandard Relative None
6 Nonstandard Market None
7 Nonstandard Market Yes

Alternative Choice Variables Parameters

1 -- Y
2 [H.sub.st], C, J Y, w
3 [H.sub.st], C, [M.sub.st] Y, w, [p.sub.st]
4 [H.sub.st], S, C, Y, w, [p.sub.st],
 [M.sub.st] [E.sub.s], [r.sub.st],
 [q.sub.s], [t.sub.s]
5 [H.sub.nst], C, J Y, w
6 [H.sub.nst], C, [M.sub.nst] Y, w, [p.sub.nst]
7 [H.sub.nst], S, C, Y, w, p.nst], [E.sub.s],
 [r.sub.nst], [q.sub.s],
 [t.sub.s]

TABLE A2
Results from the Employment Equation

Variable Coefficient Robust SE

Mother's age 0.036 *** 0.009
Age (2) (/1000) -0.514 *** 0.128
Black 0.041 0.035
White 0.033 0.031
Hispanic ethnicity -0.028 * 0.016
Mother is in good health 0.185 *** 0.018
Number of relatives living in 0.017 ** 0.008
the household
High school 0.214 *** 0.023
Some college 0.285 *** 0.027
Bachelor+ 0.312 *** 0.025
Number of children between -0.101 *** 0.015
ages 0-5
Number of children between -0.045 *** 0.012
ages 6-13
South 0.019 0.032
West 0.021 0.023
Midwest 0.074 *** 0.025
Nonwage income (/1000) -0.004 *** 0.001
State's unemployment rate for -0.192 * 0.112
females
State's percentage of female- -0.131 0.122
headed households with children
living under poverty (/100)
Maximum state reimbursement -0.035 0.051
rate for licensed child care (/1000)
Maximum annual income for -0.383 ** 0.166
subsidy eligibility (/100,000)
Monthly copayment for child care -0.005 0.029
for a family of three (/100)
State's median income for a -0.039 0.281
family of three (/100,000)
Percentage of eligible children -0.174 0.216
served in the state (/100)
State uses mass media as a -0.002 0.018
consumer education strategy
Amount of CCDF funds spent 0.026 0.054
per child (/10,000)
Constant 0.087 0.249
Log likelihood -4,567.2
Sample size 4,405

Note: *, **, and *** indicate that the estimated coefficients
are statistically significant at the 10%, 5%, and
1% levels, respectively.

TABLE A3
Results from the Welfare Receipt Equation

Variable Coefficient Robust SE

Mother's age -0.025 *** 0.007
Age (2) (/1000) 0.286 *** 0.093
Black 0.009 0.036
White -0.048 * 0.028
Hispanic ethnicity -0.051 *** 0.015
Mother is in good health -0.070 *** 0.018
Number of relatives living 0.003 0.007
in the household
High school -0.058 *** 0.019
Some college -0.079 *** 0.021
Bachelor+ -0.099 *** 0.020
Number of children 0.057 *** 0.012
between ages 0-5
Number of children 0.039 *** 0.010
between ages 6-13
South 0.015 0.034
West 0.028 0.025
Midwest -0.062 *** 0.023
Nonwage income (/1000) -0.014 *** 0.001
State's unemployment -0.210 ** 0.099
rate for female workers
State's percentage of -0.036 0.125
female-headed households
with children living under
poverty (/100)
Maximum state -0.065 0.071
reimbursement rate for
licensed child care (/1000)
Maximum annual income 0.055 0.157
for subsidy eligibility
(/100,000)
Monthly copayment for 0.006 0.026
child care for a family of
three (/100)
State's TANF earnings 0.118 ** 0.053
eligibility for a single
parent family of three
(for applicants) (/1000)
State's maximum TANF 0.065 * 0.036
benefits for a family of
three (/1000)
State's median income 0.003 0.255
for a family of three
(/100,000)
Percentage of eligible 0.551 ** 0.246
children served in the
state (/100)
State uses mass media -0.005 0.021
as a consumer education
strategy
Amount of CCDF funds 0.206 *** 0.055
spent per child (/10,000)
Constant 0.527 ** 0.216
Log likelihood -1,525.1
Sample size 4,405

Note: *, **, and *** indicate that the estimated coefficients
are statistically significant at the 10%, 5%, and
1%, levels, respectively.

TABLE A4
Definitions of Occupation Indicators

Occupation 1 Binary indicator for executive,
 administrative, and managerial occupations
Occupation 2 Binary indicator for professional specialty
 occupations
Occupation 3 Binary indicator for technicians and related
 support occupations
Occupation 4 Binary indicator for sales occupations
Occupation 5 Binary indicator for administrative support
 occupations
Occupation 6 Binary indicator for protective service
 occupations
Occupation 7 Binary indicator for service occupations
Occupation 8 Binary indicator for precision production,
 craft, and repair occupations
Occupation 9 Binary indicator for machine operators,
 assemblers, and inspectors
Occupation 10 Binary indicator for transportation, and
 material moving equipment occupations
Occupation 11 Binary indicator for handlers, equipment
 cleaners, helpers
Occupation 12 Binary indicator for farming, forestry,
 and fishing occupations

REFERENCES

Administration for Children and Families. "New Statistics Show Only Small Percentage of Eligible Families Receive Child Care Help." Washington, DC, 2000.

Anderson, P. M., and P. B. Levine. "Child Care and Mothers' Employment Decisions," in Finding Jobs: Work and Welfare Reform, edited by R. M. Blank and D. Card. New York: Russell Sage Foundation, 2000, 420-62.

Barker, K. "Changing Assumptions and Contingent Solutions: The Costs and Benefits of Women Working Full and Part Time." Sex Roles. January 1993, 47-71.

Beers, T. M. "Flexible Schedules and Shift Work." Monthly Labor Review. 123(6), 2000. 33-40.

Berger, M. C., and D. A. Black. "Child Care Subsidies, Quality of Care. and the Labor Supply of Low-Income Single Mothers." Review of [Economics and Statistics, 74(4), 1992. 635-42.

Blank, R. "Evaluating Welfare Reform in the United States." Journal of Economic Literature, 40(4). 2002, 1104-66.

Blau, David M. "Child Care Subsidy Programs," in Means-Tested Transfer Programs in the United States, edited by R. A. Moffitt. Chicago: University of Chicago Press, 2003. 443-516.

Blau, D. M., and A. Hagy. "The Demand for Quality in Child Care." Journal of Political Economy, 106(1). 1998. 107-46.

Blau D. M., and E. Tekin. "The Determinants and Consequences of Child Care Subsidies for Single Mothers." Journal of Population Economics, forthcoming.

Bloom, D.. M. Farrell. J. J. Kemple, and N. Verma. The Family Transition Program: Implementation and Three-Year Impacts of Florida's Initial Time-Limited Welfare Program. New York: Manpower Demonstration Research, 1999.

Bos, J. M., A. C. Huston, R. C. Granger. G. J. Duncan, T. W. Brock, and V. C. McCloyd. New Hope for People with Low Incomes: Two-Year Results of a program to Reduce Poverty and Reform Welfare. New York: Manpower Demonstration Research, 1999.

Brayfield. A. "Juggling Jobs and Kids: The Impact of Employment Schedules on Fathers' Caring for Children." Journal of Marriage and the Family. 57(2), 1995, 321 32.

Burstein, N., J. I. Layzer, K. Cahill, A. Werner. and N. McGarry. National Study of Child Care for Low-Income Families: Patterns of Child Care Use among Low-Income Families. Cambridge, MA: ABT Associates, 2001.

Cabrera, Na.. R. Hutchens, and H. E. Peters. "From Welfare to Work." Poverty Research News. 6(2), 2002.

Cancian, M., and D. R. Meyer. "Work after Welfare: Work Effort, Occupation. and Economic Well-Being." Social Work Research. 24(2), 2000. 69-86.

Casper, L. M., and M. O'Connell. "Work, Income, the Economy, and Married Fathers as Child-Care Providers." Demography, 35(3), 1998, 243-250. I

Chaplin, D. D., P. K. Robins, S. Hofferth, D. A. Wissoker, and P. Frostin. "The Price Sensitivity of Child Care Demand: A Sensitivity Analysis." Working Paper, Urban Institute, 1999.

Cochi-Ficano, C., and H. E. Peters. "Work and Child Care Choice among Low-Income Women: Does Policy Matter?" Paper presented at the conference "From Welfare to Child Care: What Happens to Infants and Toddlers When Single Mothers Exchange Welfare for Work." Washington, DC, May 2001.

Committee on Ways and Means, U.S. House of Representatives. 2000 Green Book. Washington DC, 2000.

Gelbach, J. "Public Schooling for Young Children and Maternal Labor Supply." American Economic Review, 92(1), 2002, 307-22.

General Accounting Office. "Recent State Policy Changes Affecting the Availability of Assistance for Low-Income Families." 6AO-03-588. May 2003.

Georges A., R. L. Wasmiller Jr., and H. H. Lu. "Child Care Subsidy Receipt and State of Residence Comparisons by Age and Parent Work Schedule." National Center for Children in Poverty, 2001.

Giannarelli, L., F. Sonenstein, and M. Stagner. "Child Care Arrangements and Help for Low-Income Families with Young Children: Evidence from the National Survey of America's Families." Paper presented at the conference "From Welfare to Child Care: What Happens to Infants and Toddlers When Single Mothers Exchange Welfare for Work." Washington, DC, 2001.

Grogger, J. "The Effects of Time Limits, the EITC, and Other Policy Changes on Welfare Use, Work, and Income among Female-Headed Families." Review of Economics and Statistics, (85)2, 2003, 394-408.

Hays, Constance L. "Increasing Shift Work Challenges Child Care." New York Times, June 8, 1995.

Heckman, J. J., and B. Singer. "A Method for Minimizing the Impact of Distributional Assumptions in Econometric Models for Duration Data." Econometrica, 52(2), 1984, 271-320.

Heymann, J. The Widening Gap: Why America's Working Families Are in Jeopardy and What Can Be Done about It. New York: Basic Books, 2000.

Hipple, S., and J. Stewart. "Earnings and Benefits of Contingent and Noncontingent Workers." Monthly Labor Review, October 1996, 22-30.

Hu, W. Y. "Child Support, Welfare Delinquency, and Woman's Labor Supply." Journal of Human Resources, 34(I), 1999, 71-103.

Jones-DeWeer, A., J. Peterson, and X. Song. "Before and after Welfare Reform: The Work and Well-Being of Low-Income Single Parent Families." Institute for Women's Policy Research Report, 2003.

Kalleberg, L. A., E. Rasell, K. Hudson, D. Webster, B. F. Reskin, N. Cassirer, and E. Appelbaum. "Nonstandard Work, Substandard Jobs: Flexible Work Arrangements in the U.S." Economic Policy Institute, Washington, DC, 1997.

Kimmel, J., and L. M. Powell. "Nonstandard Work and Child Care Choices of Married Mothers." W. E. Upjohn Institute for Employment Research, Working Paper 01-74, December 2001.

Kisker, E. E., A. Rangarajan, and K. Boller. Moving into Adulthood: Were the Impacts of Mandatory Programs for Welfare-Dependent Teenage Parents Sustained after the Programs Ended? Princeton, N J: Mathematica Policy Research, 1998.

Layzer, J. "The National Study of Child Care for Low-Income Families." Paper presented at the conference "From Welfare to Child Care: What Happens to Infants and Toddlers When Single Mothers Exchange Welfare for Work." Washington, DC, 2001.

Lokshin, M. M. "Household Childcare Choices and Women's Work Behavior in Russia." Journal of Human Resources, 39(4), 2004, 1094 115.

Loprest, P. "Families Who Left Welfare: Who Are They and How Are They Doing?" Urban Institute. 1999.

--. "Making the Transition from Welfare to Work: Successes but Continuing Concerns," in Welfare Reform: the Next Act, edited by A. Weil and K. Finegold. Washington DC: Urban Institute Press, 2002.

Martinson, K. "'The Experiences of Welfare Recipients Who Find Jobs." Manpower Demonstration Research Corporation, New York, 2000.

Meyers, M. K., T. Heintze, and D. A. Wolf. "Child Care Subsidies and the Employment of Welfare Recipients." Demography, 39(1), 2002, 165-80.

Miller, C., V. Knox, P. Auspos, J. A. Hunter-Means, and A. Orenstein. "Making Welfare Work and Work Pay: Implementation and 18-Month Impacts of the Minnesota Family Investment Program.'" Manpower Demonstration Research Corporation, New York, 1997.

Mocan, N. H., and E. Tekin. "Nonprofit Sector and Part-Time Work: An Analysis of Employer-Employee Matched Data of Child Care Workers." Review of Economies and Statistics, 85(1), 2002, 38-50.

Mroz, T. A. "Discrete Factor Approximations in Simultaneous Equation Models: Estimating the Impact of a Dummy Endogenous Variable on a Continuous Outcome." Journal of Econometrics, 92(2), 1999, 233-74.

Nightingale, D. S. "Work Opportunities for People Leaving Welfare," in Welfare Reform: The Next Act, edited by A. Weil and K. Finegold. Washington. DC: Urban Institute Press, 2002.

Parrott, S., E. Park, and R. Greenstein. "Assessing the Effects of the Budget Conference Agreement on Low-Income Families and Individuals." Washington, DC: Center on Budget and Policy Priorities, January 9, 2006.

Picone, A. G., F. Sloan, S.-Y. Chou, and D. H. Taylor Jr. "Does Higher Hospital Cost Imply Higher Quality of Care?" Review of Economics and Statistics, 85(1), 2003, 51-62.

Presser, H. B. "Shift Work among American Women and Child Care." Journal of Marriage and the Family, 48(3), 1986, 551-63.

--. "Shift Work and Child Care among Young Dual-Earner American Parents." Journal of Marriage and the Family, 50(I), 1988, 133-48.

--. "Nonstandard Work Schedules and Marital Instability." Journal of Marriage and the Family, 62(1), 2000, 93-110.

Presser, H., and A. G. Cox. "The Work Schedules of Low-Educated American Women and Welfare Reform." Monthly Labor Review, April 1997, 25-34.

Quint, J. C., J.M. Bos, and D. F. Polit. "New Chance: Final Report on a Comprehensive Program for Young Mothers in Poverty and Their Children." Manpower Demonstration Research, New York, 1997,

Rangarajan, A., P. Schochet, and D. Chu. Employment Experience of Welfare Recipients Who Find Jobs: Is Targeting Possible? Princeton, NJ: Mathematica, 1998.

Schulman. K., H. Blank, and D. Ewen. A Fragile Foundation: State Child Care Assistance Policies. Washington, DC: Children's Defense Fund, 2001.

Schumacher. R., and M. Greenberg. "Child Care after Leaving Welfare: Early Evidence from State Studies." Center for Law and Social Policy Working Paper, 1999.

Staines. G. L., and J. H. Pleck. The Impact of Work Schedules on the Family. Ann Arbor: University of Michigan Press, Survey Research Center, Institute for Social Research, 1983.

Strawn, J., and K. Martinson. "Steady Work and Better Jobs: How to Help Low Income Parents Sustain Employment and Advance in the Workforce." Manpower Demonstration Research Corporation, New York, 2000.

Tilly, C. Half a Job. Bad and Good Part-Time Jobs in a Changing Labor Market. Philadelphia: Temple University Press, 1996.

ABBREVIATIONS

CCDF: Child Care Development Fund

GAO: General Accounting Office

NSAF: National Survey of America's Families

TANF: Temporary Assistance to Needy Families

doi:10.1093/ei/cb1004

(1.) The welfare reform legislation combined the previously existing four child care funding programs into the CCDF and increased federal funding for child care substantially. It also gave states greater flexibility in setting up and administering their programs. In fiscal year 1999, states spent all of their CCDF allocation of around $5 billion and spent directly on child care or transferred another $4 billion from the Temporary Assistance for Needy Families (TANF) funds, See Blau (2003) for a summary of the system of child care assistance under welfare reform.

(2.) It is important to note that the term nonstandard is not used to describe workers who are employed on a temporary basis. Rather, it refers to the individual's reported work schedule.

(3.) Prior to 1996, welfare recipients had the priority in receiving child care subsidies, This is no longer a requirement under federal law, although it is still often the case in practice.

(4.) Several examples of these demonstration projects are New Hope (Bos et al. 1999), the Teenage Parent Demonstration (Kisker et al. 1998), New Chance (Quint et al. 1997), the Minnesota Family Investment Program (Miller et al. 1997), and the Florida Family Transition Program (Bloom et al. 1999).

(5.) See Anderson and Levine (2000) for an excellent summary of these studies.

(6.) A structural model would contain the price of child care and the mother's wage rate. However, the determinants of these variables are substituted in the empirical analyses. Thus, equation (2) is really a quasi-reduced form model. This is done to avoid the problems of estimating wages and prices of nonworkers and nonpayers, respectively. Also the locus of the article is the effect of subsidy receipt, not the effects of the price of child care and the wage rate.

(7.) CCDF funds are provided to the states in three streams: discretionary, mandatory, and matching. The fist two streams are distributed according to the number of children and income. To receive funds from the matching stream, a state must maintain its child care expenditure at specified previous levels (maintenance-of-effort spending). Only 12-15% of eligible families are served by a CCDF subsidy in 1998-99 (Administration for Children and Families 2000). One-third of states report that they have a waiting list for a child care subsidies. Two-fifths of all states report that most eligible families are not aware of their eligibility. Only four states report that they could serve all eligible families (Schulman et al. 2001). Also, the absence of a waiting list may be due to the fact that states simply turn away clients for whom funds are not available without putting them on a waiting list (Blau and Tekin forthcoming).

(8.) It must be cautioned that any state-level unobserved heterogeneity that is correlated with the instruments would still cause bias in a cross-sectional framework.

(9.) Note that subsidy receipt is not in the work/no-work equation. This is because the determinants of the subsidy receipt are substituted, which results in a fully reduced-form equation. This is preferred because the focus of the article is the subsidy receipt standard work relationship, but not the effect on employment/no employment. Therefore, the [T.sub.s] appears in equation (3) not as an identifying instrument but rather because it is a determinant of subsidy receipt, which is substituted into equation (3).

(10.) The location and the scale of the distribution of u are not identified. Because each model contains an intercept and the factor loading parameters [[rho].sub.1], [[rho].sub.2], and [[rho].sub.3] are estimated in the parameterization, [[omega].sub.k] is restricted to be between 0 and 1 (Picone et al. 2003). Further parameterization is specified as follows: [[omega].sub.k] = exp([a.sub.k])/[1 + exp([a.sub.k])], k = 2, ..., k - 1, and [[omega].sub.0] = 0 and [[omega].sub.k] = 1. [p.sub.k] = exp([b.sub.k])/ [summation.sup.K-1] exp([b.sub.k]). The a's the b's are free parameters to be estimated. The likelihood function is maximized with respect to all the parameters including those representing heterogeneity.

(11.) The first round of the NSAF was conducted in 1997 with a different sample. There is a third wave conducted with another sample in 2001. The third wave was not available to the public at the time this article was written.

(12.) These 13 states are Alabama, California, Colorado, Florida, Massachusetts, Michigan, Minnesota, Mississippi, New Jersey, New York, Texas, Washington, and Wisconsin.

(13.) The results presented herein are taken from a model estimated with tour points of support. A model with five points of support did not provide a significant improvement in the likelihood over a model with four points of support. Although there is no standard theory about how to select the number of points of support in a finite sample, the consensus is to add points of support until the likelihood fails to improve significantly (Blau and Hagy 1998; Mocan and Tekin 2003; Picone et al. 2003). Mroz (1999) shows that the likelihood ratio test performs well when determining the number of points of support. The estimates of the heterogeneity parameters are available from the author.

(14.) The coefficients in linear probability models are consistent estimates of average probability derivatives, but standard errors are biased as a result of heteroscedasticity. In the article, standard error estimates that are robust to any form of heteroscedasticity are reported. It must be noted that the linear probability model is most reasonable away from the tails.

(15.) However, the problem of endogeneity may be less severe once the model is conditioned on employment. One can argue that once someone decides to work, whether she works standard or nonstandard hours is less likely to be correlated with the unobserved factors that are also correlated with welfare.

(16.) To convert the predicted welfare to a binary variable, a value of one is assigned to predicted welfare receipt if" the predicted welfare receipt for a mother is in the top [mu]th percentile where [mu] is the sample mean for welfare receipt, and a zero is assigned otherwise. This ensures that the binary predicted welfare receipt has a mean equal to the sample mean.

ERDAL TEKIN *

* I thank Naci Mocan, David Ribar, and seminar participants at the 2004 European Society of Population Economics Meetings in Norway and 2004 Association for Public Policy Analysis and Management Meetings (APPAM) for helpful comments, and the APPAM and W. E. Upjohn Institute for Employment Research for financial support. Roy Wada provided excellent research assistance. All errors are mine.

Tekin: Assistant Professor, Andrew Young School of Policy Studies, Georgia State University, P.O. Box 3992, Atlanta, GA 30302, NBER, and IZA. Phone 1-404-651-3968, Fax 1-404-651-4985, E-mail [email protected]

TABLE 1
Descriptive Statistics

 1. 2.
Variable Name Full Sample Work

Mother works 0.711 1.000
 (0.147) (0.000)
Mother works at a standard job -- 0.793
 (0.405)
Mother works at a nonstandard job -- 0.207
 (0.405)
Mother receives a child care subsidy 0.116 0.126
 (0.321) (0.332)
Hours of work per week -- 38.401
 (11.151)
Mother receives welfare 0.158 0.103
 (0.365) (0.305)
Mother's age 31.780 32.174
 (7.182) (7.079)
Mother's race
Black 0.314 0.307
 (0.464) (0.461)
White 0.652 0.661
 (0.476) (0.473)
Other race (a) 0.035 0.033
 (0.183) (0.178)
Hispanic ethnicity 0.172 0.149
 (0.378) (0.356)
Mother is in good health 0.835 0.879
 (0.371) (0.327)
Number of relatives living in the 2.398 2.286
household (1.379) (1.315)
Mother's education
Less than high school (a) 0.149 0.093
 (0.356) (0.290)
High school 0.366 0.363
 (0.482) (0.481)
Some college 0.360 0.394
 (0.480) (0.489)
Bachelor+ 0.126 0.150
 (0.332) (0.358)
Number of children
Between ages 0-5 0.774 0.692
 (0.787) (0.736)
Between ages 6-13 1.207 1.189
 (1.070) (1.029)
Mother's region of residence
South 0.290 0.284
 (0.454) (0.451)
West 0.197 0.191
 (0.398) (0.393)
Midwest 0.271 0.300
 (0.445) (0.458)
Northeast (a) 0.242 0.225
 (0.428) (0.418)
Nonwage income (/1000) 3.399 3.387
 (7.158) (6.808)
Mother's occupation
Occupation1(a) 0.076 0.106
 (0.264) (0.308)
Occupation2 0.087 0.123
 (0.282) (0.328)
Occupation3 0.032 0.045
 (0.175) (0.207)
Occupation4 0.079 0.112
 (0.270) (0.315)
Occupation5 0.175 0.246
 (0.380) (0.431)
Occupation6 0.009 0.012
 (0.094) (0.111)
Occupation7 0.146 0.205
 (0.353) (0.404)
Occupation8 0.023 0.032
 (0.150) (0.177)
Occupation9 0.050 0.072
 (0.218) (0.256)
Occupation10 0.009 0.013
 (0.096) (0.114)
Occupation11 0.020 0.027
 (0.138) (0.163)
Occupation12 0.005 0.007
 (0.072) (0.085)
State's unemployment rate for females (b) 4.170 4.108
 (0.949) (0.957)
Maximum annual income for subsidy 0.284 0.283
eligibility (/100,000) (c) (0.053) (0.054)
Monthly copayment for child care for 0.514 0.518
a family of three (/100) (c) (0.384) (0.378)
Maximum state reimbursement rate for 0.622 0.624
licensed child care (/1000) (c) (0.178) (0.173)
State's TANF earnings eligibility for a 0.641 0.643
single parent family of three (for (0.220) (0.218)
applicants) (/1000) (d)
State's maximum TANF benefits for a 0.446 0.451
family of three (/1000) (d) (0.188) (0.188)
State's percentage of female-headed 0.370 0.367
households with children living under (0.083) (0.083)
poverty (/100) (c)
Percentage of eligible children served in 0.116 0.114
the state (/100) (f) (0.041) (0.041)
State uses mass media as a consumer 0.714 0.718
education strategy (f) (0.452) (0.450)
Annual amount of CCDF funds spent 0.529 0.535
per child (/10,000) (f) (0.183) (0.182)
State's Median Income for a family of 0.452 0.454
three (/100,000) (e) (0.055) (0.059)
Sample size 4,405 3,132

 3. 4.
 Standard Nonstandard
Variable Name Work Work

Mother works 1.000 1.000
 (0.000) (0.000)
Mother works at a standard job 1.000 0.000
 (0.000) (0.000)
Mother works at a nonstandard job 0.000 1.000
 (0.000) (0.000)
Mother receives a child care subsidy 0.131 0.108 *
 (0.337 (0.310)
Hours of work per week 40.070 32.014 ***
 (13.154) (8.921)
Mother receives welfare 0.093 0.142 ***
 (0.291) (0.349)
Mother's age 32.530 30.812 ***
 (6.985) (7.274)
Mother's race
Black 0.297 0.342 **
 (0.457) (0.475)
White 0.672 0.619 **
 (0.470) (0.486)
Other race (a) 0.031 0.039
 (0.173) (0.193)
Hispanic ethnicity 0.148 0.153
 (0.355) (0.360)
Mother is in good health 0.884 0.858 *
 (0.320) (0.349)
Number of relatives living in the 2.231 2.496 ***
household (1.268) (1.464)
Mother's education
Less than high school (a) 0.079 0.146 ***
 (0.269) (0.354)
High school 0.359 0.379
 (0.480) (0.486)
Some college 0.393 0.396
 (0.489) (0.489)
Bachelor+ 0.169 0.079 ***
 (0.375) (0.269)
Number of children
Between ages 0-5 0.657 0.824 ***
 (0.723) (0.772)
Between ages 6-13 1.193 1.188
 (1.146) (0.996)
Mother's region of residence
South 0.281 0.294
 (0.450) (0.456)
West 0.195 0.179
 (0.396) (0.383)
Midwest 0.298 0.305
 (0.458) (0.461)
Northeast (a) 0.226 0.222
 (0.418) (0.416)
Nonwage income (/1000) 3.492 2.983 *
 (6.992) (6.040)
Mother's occupation
Occupation1(a) 0.121 0.051 ***
 (0.326) (0.220)
Occupation2 0.140 0.069 ***
 (0.344) (0.254)
Occupation3 0.045 0.043
 (0.208) (0.203)
Occupation4 0.100 0.157 ***
 (0.300) (0.364)
Occupation5 0.271 0.149 ***
 (0.445) (0.357)
Occupation6 0.010 0.020 *
 (0.102) (0.140)
Occupation7 0.177 0.314 ***
 (0.382) (0.465)
Occupation8 0.033 0.028
 (0.180) (0.164)
Occupation9 0.062 0.102 ***
 (0.242) (0.302)
Occupation10 0.014 0.008
 (0.120) (0.088)
Occupation11 0.022 0.049 ***
 (0.146) (0.217)
Occupation12 0.007 0.009
 (0.082) (0.096)
State's unemployment rate for females (b) 4.105 4.116
 (0.958) (0.953)
Maximum annual income for subsidy 0.283 0.285
eligibility (/100,000) (c) (0.054) (0.052)
Monthly copayment for child care for 0.518 0.515
a family of three (/100) (c) (0.379) (0.374)
Maximum state reimbursement rate for 0.625 0.615
licensed child care (/1000) (c) (0.173) (0.174)
State's TANF earnings eligibility for a 0.640 0.656 *
single parent family of three (for (0.216) (0.223)
applicants) (/1000) (d)
State's maximum TANF benefits for a 0.451 0.449
family of three (/1000) (d) (0.187) (0.193)
State's percentage of female-headed 0.366 0.369
households with children living under (0.082) (0.085)
poverty (/100) (c)
Percentage of eligible children served in 0.114 0.116
the state (/100) (f) (0.041) (0.042)
State uses mass media as a consumer 0.718 0.715
education strategy (f) (0.450) (0.452)
Annual amount of CCDF funds spent 0.534 0.536
per child (/10,000) (f) (0.186) (0.186)
State's Median Income for a family of 0.454 0.453
three (/100,000) (e) (0.055) (0.056)
Sample size 2,483 649

 5. 6.
 Standard Standard
 Work Work
 and Receive and Do Not
 a Subsidy Receive a
Variable Name Subsidy

Mother works 1.000 1.000
 (0.000) (0.000)
Mother works at a standard job 1.000 1.000
 (0.000) (0.000)
Mother works at a nonstandard job 0.000 0.000
 (0.000) (0.000)
Mother receives a child care subsidy 1.000 0.000
 (0.000) (0.000)
Hours of work per week 38.911 38.019
 (11.510) (11.679)
Mother receives welfare 0.305 0.061 ***
 (0.256) (0.005)
Mother's age 28.898 33.077 ***
 (6.153) (6.940)
Mother's race
Black 0.412 0.280 ***
 (0.493) (0.449)
White 0.557 0.689 ***
 (0.498) (0.463)
Other race (a) 0.031 0.031
 (0.173) (0.173)
Hispanic ethnicity 0.157 0.146
 (0.364) (0.354)
Mother is in good health 0.858 0.888
 (0.349) (0.316)
Number of relatives living in the 2.397 2.206 **
household (1.264) (1.267)
Mother's education
Less than high school (a) 0.071 0.080
 (0.257) (0.271)
High school 0.437 0.348 ***
 (0.497) (0.473)
Some college 0.428 0.388
 (0.496) (0.487)
Bachelor+ 0.065 0.184 ***
 (0.246) (0.388)
Number of children
Between ages 0-5 1.089 0.592 ***
 (0.774) (0.692)
Between ages 6-13 0.969 1.221 ***
 (1.036) (0.086)
Mother's region of residence
South 0.203 0.293 ***
 (0.403) (0.455)
West 0.212 0.192
 (0.410) (0.394)
Midwest 0.351 0.291 **
 (0.478) (0.454)
Northeast (a) 0.234 0.225
 (0.424) (0.418)
Nonwage income (/1000) 1.226 3.833
 (3.994) (7.278) ***
Mother's occupation
Occupation1(a) 0.074 0.128 ***
 (0.262) (0.334)
Occupation2 0.086 0.145 ***
 (0.281) (0.352)
Occupation3 0.031 0.047
 (0.173) (0.212)
Occupation4 0.126 0.096 *
 (0.333) (0.295)
Occupation5 0.357 0.258 ***
 (0.480) (0.438)
Occupation6 0.006 0.011
 (0.078) (0.105)
Occupation7 0.243 0.167 ***
 (0.430) (0.373)
Occupation8 0.025 0.035
 (0.155) (0.183)
Occupation9 0.022 0.069 ***
 (0.145) (0.253)
Occupation10 0.009 0.015
 (0.096) (0.123)
Occupation11 0.018 0.022
 (0.135) (0.148)
Occupation12 0.003 0.007
 (0.055) (0.086)
State's unemployment rate for females (b) 3.946 4.129 ***
 (9.403) (9.585)
Maximum annual income for subsidy 0.290 0.282 ***
eligibility (/100,000) (c) (0.057) (0.053)
Monthly copayment for child care for 0.454 0.528 ***
a family of three (/100) (c) (0.371) (0.379)
Maximum state reimbursement rate for 0.665 0.619 ***
licensed child care (/1000) (c) (0.152) (0.176)
State's TANF earnings eligibility for a 0.683 0.633 ***
single parent family of three (for (0.219) (0.215)
applicants) (/1000) (d)
State's maximum TANF benefits for a 0.498 0.444
family of three (/1000) (d) (0.173) (0.188)
State's percentage of female-headed 0.357 0.368 **
households with children living under (0.088) (0.081)
poverty (/100) (c)
Percentage of eligible children served in 0.118 0.113 *
the state (/100) (f) (0.044) (0.041)
State uses mass media as a consumer 0.738 0.715
education strategy (f) (0.440) (0.451)
Annual amount of CCDF funds spent 0.570 0.529 ***
per child (/10,000) (f) (0.164) (0.183)
State's Median Income for a family of 0.463 0.452 ***
three (/100,000) (e) (0.047) (0.056)
Sample size 325 2,158

Notes: SDs are in parentheses. *, **, and *** indicate statistically
significant difference in means between "standard work" and
"nonstandard work" or "standard work and receive a subsidy" and
"standard work and do not receive a subsidy" at 10%, 5%, and 1%
levels, respectively. Nonwage income includes all income during 1996
except the mother's earnings and income from means-tested programs.

(a) Omitted category.

(b) Source: Urban Institute's State Database.

(c) Source: Children's Defense Fund and Child Care Bureau.

(d) Source: State Policy Documentation Project.

(e) Source: Bureau of Labor Statistics.

(f) Source: Child Care Bureau.

TABLE 2

Results from Child Care Subsidy
Receipt Equation

Variable Coefficient Robust SE

Mother's age -0.015 *** 0.006
[Age.sup.2] (/100) 0.018 ** 0.009
Black 0.085 *** 0.027
White 0.012 0.026
Hispanic ethnicity 0.004 0.014
Mother is in good health -0.007 0.124
Number of relatives living -0.018 *** 0.006
in the household
High school 0.036 *** 0.014
Some college 0.060 *** 0.015
Bachelor+ 0.002 0.00
Number of children 0.050 *** 0.011
between ages 0-5
Number of children 0.024 *** 0.008
between ages 6-13
South 0.027 0.022
West 0.051 ** 0.023
Midwest 0.002 0.016
Nonwage income (/1000) -0.003 *** 0.001
State's unemployment -0.212 *** 0.076
rate for females
State's percentage of -0.087 0.072
female-headed households
with children living
under poverty (/100)
Maximum state 0.042 0.034
reimbursement rate for
licensed child care (/1000)
Maximum annual income 0.106 0.707
for subsidy eligibility
(/100,000)
Monthly copayment for 0.111 0.106
child care for a family of
three (/100)
State's Median Income 0.056 0.208
for a family of three
(/100,000)
Percentage of eligible 0.782 *** 0.167
children served in the
state (/100)
State uses mass media 0.035 *** 0.013
as a consumer education
strategy
Amount of CCDF funds 0.049 0.038
spent per child (/10,000)
Constant 0.292 * 0.163
Log likelihood -4,567.2
Sample size 4,405

Note: *, **, and *** indicate that the estimated coefficients
are statistically significant at the 10%, 5%, and
1% levels, respectively.

TABLE 3

Results from the Standard/Nonstandard
Employment Equation

Variable Coefficient Robust SE

Child care subsidy receipt 0.069 ** 0.035
Mother's age 0.031 *** 0.009
[Age.sup.2] (/1,000) -0.475 *** 0.132
Black 0.023 0.039
White 0.042 0.043
Hispanic ethnicity 0.023 0.019
Mother is in good health 0.015 0.021
Number of relatives living -0.004 0.008
in the household
High school 0.051 * 0.028
Some college 0.032 0.030
Bachelor+ 0.095 *** 0.036
Number of children -0.032 ** 0.015
between ages 0-5
Number of children -0.014 0.011
between ages 6-13
South -0.016 0.033
West 0.001 0.025
Midwest -0.003 0.021
Nonwage income (/1000) -0.001 0.001
Occupation2 -0.035 0.026
Occupation3 -0.102 ** 0.048
Occupation4 -0.175 *** 0.032
Occupation5 -0.014 0.021
Occupation6 -0.229 *** 0.089
Occupation7 -0.201 *** 0.029
Occupation8 -0.073 * 0.040
Occupation9 -0.188 *** 0.044
Occupation10 0.008 0.053
Occupation11 0.246 *** 0.061
Occupation12 -0.142 * 0.075
State's unemployment -0.008 0.131
rate for females
State's percentage of -0.076 0.118
female-headed households
with children living
under poverty (/100)
Maximum state 0.027 0.053
reimbursement rate for
licensed child care (/1000)
Maximum annual income for -0.281 0.183
subsidy eligibility (/100,000)
Monthly copayment for child 0.011 0.024
care for a family of three (/100)
State's median income for -0.367 0.268
a family of three (/100,000)
Constant 0.427 ** 0.229
Log-likelihood -4,567.2
Sample size 3,132

Note: *, **, and *** indicate that the estimated coefficients
are statistically significant at the 10%, 5%, and
1% levels, respectively.

TABLE 4

Results from the Standard/Nonstandard
Employment Equation with Predicted
Welfare Receipt

Variable Coefficient Robust SE

Child care subsidy receipt 0.018 * 0.011
Welfare receipt -0.149 ** 0.074
Welfare receipt x Child care 0.122 * 0.067
subsidy receipt
Mother's age 0.029 *** 0.009
[Age.sup.2] (/1000) -0.488 *** 0.136
Black 0.020 0.043
White 0.031 0.040
Hispanic ethnicity 0.017 0.016
Mother is in good health 0.011 0.021
Number of relatives living -0.002 0.009
in the household
High school 0.044 0.030
Some college 0.024 0.031
Bachelor+ 0.085 *** 0.027
Number of children -0.029 ** 0.138
between ages 0-5
Number of children -0.011 0.014
between ages 6-13
South -0.016 0.028
West 0.001 0.025
Midwest -0.003 0.023
Nonwage income(/1000) -0.001 0.001
Occupation2 -0.032 0.022
Occupation3 -0.096 *** 0.045
Occupation4 -0.156 *** 0.029
Occupations -0.012 0.020
Occupation6 -0.209 *** 0.078
Occupation7 -0.183 *** 0.027
Occupation8 -0.072 * 0.038
Occupation9 -0.176 *** 0.036
Occupation10 0.006 0.050
Occupation11 -0.238 *** 0.056
Occupation12 -0.120 0.092
State's unemployment -0.007 0.111
rate for females
State's percentage of -0.083 0.129
female-headed households with
children living under
poverty (/100)
Maximum state 0.024 0.054
reimbursement rate for
licensed child care (/1000)
Maximum annual income for -0.264 0.168
subsidy eligibility (/100,000)
Monthly copayment for child 0.009 0.025
care for a family of three (/100)
State's median income for -0.315 0.254
a family of three (/100,000)
Constant -0.465 *** 0.236
Log-likelihood -4,558.5
Sample size 3,132

Note: *, **, and *** indicate that the estimated coefficients
are statistically significant at the 10%, 5%, and
1% levels, respectively.