- all W^M Ii3§§ JBBBHhI ■"■".'-■.•-'■ I "~ 1 ' ^ FACULTY WORKING PAPER NO, 988 Variations in the Economic We!!-6eina of Divorced Women and Their Children Andrea H. Beller John W. Graham THE LIBRARY OF IH6 OEC 5 6€ AT URfiANA-CHAMPAlGf* College of Commerce and Business Administration Bureau of Economic and Business Research University of Illinois, Urbana-Champaign BEBR FACULTY WORKING PAPER NO. 988 College of Commerce and Business Administration University of Illinois at Urbana-Champaign November 1983 DRAFT - Not for Quotation - Comments Welcome Variations in the Economic Well-Being of Divorced Women and Their Children Andrea H. Beller, Asst. Professor Department of Family and Consumer Economics John W. Graham, Asst. Professor Department of Economics Prepared for presentation at the National Bureau of Economic Research Conference on Research in Income and Wealth: "Horizontal Equity, Uncertainty, and Economic Well-Being," December 8-9, 1983, Baltimore, MD. This research was supported by Hatch funds from the University of Illinois Agricultural Experiment Station and by funds from the University of Illinois at Urbana-Champaign Research Board. We gratefully acknowledge the excellent research assistance of Kee-ok Kim Han and Lorraine Maddox, and the excellent computer programming assistance of John Boyd. Thanks are also due to Elizabeth Peters for sharing with us her computer program for preparation of the data extract and to Morey MacDonald for some early discussions. We alone take responsibility for errors and omissions. Digitized by the Internet Archive in 2011 with funding from University of Illinois Urbana-Champaign http://www.archive.org/details/variationsinecon988bell Abstract In this paper we investigate variations in the economic well-being of ever-divorced or currently separated mothers with children from an absent father — a quickly growing and disportionately poor segment of the population. In large measure, the economic well-being of a divorced mother depends upon the amount of family support (child support and alimony) she receives from her ex-husband, her labor force participation, and how quickly she remarries, if at all. One unique aspect of our study is that we pay particular attention to the effects of child support and alimony upon the probabilities of remarriage and labor force participation. The data set used in our analysis is the March-April 1979 Match File of the Current Population Survey (CPS). Within the limitations imposed by our data, we find the likelihood of being awarded child support depends upon the needs of the mother and her children, and upon the absent father's long-term ability to pay. In contrast, the likelihood of receiving child support due depends less upon the circumstances of the woman, and more upon the current financial well-being of her ex-husband. His income is found to affect whether or not any child support is actually paid, but not how much is paid. Like other forms of nonlabor income, child support payments are found to lower the probability of remarriage. But unlike other forms of nonlabor income, child support payments are found to raise labor force participation. I. Introduction Female-headed families are a large and growing proportion of all families. Special concern for this population stems from the limited amount of resources available to the family unit. Divorce almost always results in a decline in the level of living for all parties involved, but the decline is larger for women than for men (Hoffman, 1979). Female-headed families make up a disproportionate share of the poverty population and many have little choice but to rely upon the welfare system (Bradbury, et al . , 1979). Between 1970 and 1981 the number of married-couple families increased only 10.1 percent while the number of female-headed families increased 62.4 percent, so that by 1981 over 15 percent of all families were headed by a woman only. This rising rate of female headship is of special concern in part because female-headed families have lower incomes and are more likely to be classified as poor than married-couple families. In 1980, the median income of married-couple families was $23,180 compared to $10,830 for female-headed families. Of the 6.2 million families with incomes below the poverty level in that year, almost 3 million were female-headed, a phenomenon recently dubbed the "feminization of poverty." Female headship also may be of special concern because of the large number of children involved. Between 1970 and 1980 the number of children living with two parents (not necessarily their natural ones) declined 20.6 percent, while the number living with one parent (overwhelmingly their mother) increased 67.1 percent. By 1982, 15.3 percent of white and 47.2 percent of black children lived with only their mothers. For most this is a temporary arrangement, since a large majority of divorced women remarry. However, it has been estimated that "children born in the mid-1970' s have about 45 chances in 100 of living in a one-parent family for a period of at least several months before they reach the age of 18 years." Most mothers with children from an absent father head their own families at -2- least for a time. Figure 1 shows the distribution by household status in April 1979 of mothers agea 18 and over with children (under 21 years of age) from an absent father. Of these 6.4 million mothers, 5.5 million had previously been married to the child's father, while 0.9 million had not. Among the ever- married, 3.5 million are divorced or separated: 3.2 million of these head their own families and 0.3 million live as subfamilies with parents or other relatives. Another 2.0 million are remarried and 54 thousand are widowed after remarriage. Among the never-married, most head their own families. In this paper, we focus on the two largest of these groups, divorced or separated mothers heading their own families ana remarried mothers, excluding the widowed. We investigate variations in several key components of the economic well-being of ever-divorced or currently separated mothers with children from an absent father. In large measure, the economic well-being of a divorced mother depends upon the amount of family support (child support and alimony) she receives from her ex-husband, her labor force participation, and how ouickly she remarries, if at all. Whether family support is received or not and how much is received is subject to considerable uncertainty, both in amount and regularity of payments. Because many divorced mothers are awarded child support while few are awarded alimony, we shall devote much more attention to the former. A divorced mother is more likely to participate in the labor force than a married mother, since her earnings become crucial to the family's economic status. She also can seek to improve the position of her family by searching for a new mate to regain the economic well-being lost by divorce. One unioue aspect of our study is that we pay particular attention to the effects of child support and alimony upon the probabilities of remarriage and labor force participation. Figure 1 also presents four measures of well-being for each household -3- status group: average total family income; the percentage of families in poverty; the percentage of families receiving child support; and the lahor force participation rate of mothers. In terms of economic status, ever-married are oetter off than never-married mothers, and remarriage clearly improves economic status. While the total family income of divorced (or separated) mothers is only $9678 and 36.2 percent of them live in poverty, family income of remarried mothers is $23,044 and only 5.6 percent of them live in poverty. The low economic position of currently divorced compared to remarried mothers occurs despite a higher proportion of them receiving child support (43.9 compared to 39.0 percent) and participating in the labor force (73.3 compared to 60.6 percent). These figures portray strong economic incentives for divorced mothers to remarry. The data set upon which these population estimates are based and from which our sample is drawn is the March/April 1979 Match File of the Current Population Survey (CPS). Data on marital status, divorce history and child support and alimony were collected on all women 18 years of age and older in a special supplement to the April 1979 CPS. A file was created containing these data along with the income information from the March 1979 CPS. This file contains 1579 currently divorced or separated female family heads and 1005 remarried mothers. Population estimates for all mothers eligible for child support and alimony as well as descriptive analyses of them are presented in U.S. Bureau of the Census (1981). 9 A divorced mother will be better off if she is awarded and receives child support, if she participates in the labor force and/or if she remarries. Women who are awaraed child support are better off than those who are not, and their income is higher by more than the amount of child support received. In our sample, the total personal income in 1978 of women awarded child support was -4- $7970 of which child support comprised $1115. For those not awarded child support, total personal income was $5568. The difference is even greater between those who received child support and those who did not. Total personal income of those who received child support in 1978 was $9425 while for those who did not it was $5742. The average amount of child support received for those who received any was $1901, much less than the difference in income. Women who participate in the labor market have higher personal income and receive more child support than those who do not. In 1978, the average personal income of currently divorced or separated mothers who worked outside the home was $9,632 while for those who dia not it was $4197; if they worked, they received $998 in child support but if they did not, only $570. As shown in Figure 1 above, although remarriage improves the economic well-being of aivorcea mothers the most, it appears to be associated with less child support received, $1683 compared to $2021 for currently divorced or separated mothers. An important question raised by these figures is whether child support payments have a direct causal effect upon the decision to work and/or remarry, or whether the observed association is simply simultaneously determined by other factors? To examine the interrelationships between child support and these other components of economic well-being, we postulate the theoretical model presented in Section II. Based upon that model, we examine what factors influence the awara and receipt of child support in Section III. Then, we examine the impact of child support payments (holding constant other factors) upon remarriage in Section IV and upon labor force participation in Section V. II. Theoretical Framework Before turning to our empirical work, we offer a short sketch of the theory upon which our analysis is based. We posit the following recursive system: -5- CSDUE = f^H 1 , W 1 , L 1 ) (1) CSREC = f 2 (H 2 , W 2 , L 2 , CSDUE) (2) RM = g[Y d (CSREC d ,...), Y r (CSREC r , . . . ) , C] (3) LF Q = k[W h (CSREC d ,...), W m , V d (CSREC d> . . . )] (4) Tne first equation says that child support due (CSDUE) is a function of a vector of characteristics of the ex-husband's ability (and desire) to pay (H-"-), the financial needs of the woman and children (W 1 ) and the legal environment at divorce (!_■'■) . The second eouation says that support actually received (CSREC) depends upon the same three vectors of variables (H 2 , W, and L 2 ) as well as upon the amount of support due. We label these vectors with different superscripts to suggest that they may contain somewhat different elements from those in the first eouation. For example, W* may measure the judge's perception of the woman's needs, while W measures the ex-husband's perception of her needs. A priori W could depend in part upon the woman's current marital status (RM), but in our empirical work we offer evidence that this effect is limited. Past work (Gordon, Jones and Sawhill, 1978; Cassetty, 1978) finds that the effect of H 2 upon CSREC is particularly strong, where H 2 refers to the absent father's current financial status, measured perhaps by his current income or current employment status. L 2 includes child support enforcement laws which vary from state to state. Equation 3 states that the probability a divorced woman remarries (RM) depends negatively upon her real income (monetary plus nonmonetary) while divorced (Y Q ) ) positively upon her expected real income if remarried (Y ), ano negatively upon the costs of marital search (C). Since child support payments are supposed to continue even after a woman remarries, CSREC enters both Y d an( j y although a priori we allow expected receipts to vary with -6- marital status (CSREC d and CSREC r ). Ceteris paribus , an increase in child support payments will reduce the expected increase in income from remarriage (because, although portable, child support payments are a larger fraction of Y d than of Y r ), but will raise the woman's attractiveness to potential marriage partners. Therefore, the net impact of child support on remarriage is ambiguous. It should be noted that public assistance benefits may be another important component of Y . The final equation states that the probability a divorced woman is in the labor force (LF^) depends negatively upon the value of her time at home ( w n ), positively upon the value of her time in the market (W m ), and negatively upon nonlabor income (V d ), which itself depends upon child support payments (CSREC d ). if a woman uses child support payments to purchase for her children market goods which can be substituted for her home time, then w n may depend negatively upon CSREC d . In this case, the net effect of an increase in CSREC^ n LF d is ambiguous: it raises V d which reduces Lfr d , but it lowers W h which raises LF d . To summarize, economic theory cannot determine the net impact of child support payments on either the probability of remarriage or the probability of labor force participation. Instead, this remains to be determined by empirical analysis. III. Empirical Analysis of Child Support Child support payments are in many cases an uncertain and inadequate source of income to divorced mothers. In this section, we examine what factors determine whether a woman is awarded child support, and then given its award whether she actually receives it, as well as how much she receives. Our study significantly improves upon previous studies of child support (Cassetty, 1973; Gordon, Jones and Sawhill, 1978; and Sorensen and MacDonald, 1981) -7- by separating the question of award from receipt and by use of a large national data set sampling the entire eligible population. Excluding observations with missing values, our sample consists of up to 2416 ever-divorced or currently separated women (18 years of age or over) with children (under 21 years of age) from an absent father. There are 389 black women and 2027 nonblacks, of which 7 percent are of Spanish origin. At divorce or separation, 72.6 percent of these women were awaraed child support — 77.5 percent of nonblacks and 47.6 percent of blacks. For those who were due payment in 1978 the average amount due annually was $2028. Of these, 71.4 percent received partial or full payments averaging $1899. While the large majority of those due support received at least partial payment, 45.4 percent reported receiving no payments or irregular payments. Means and standard deviations for various characteristics of this sample are summarized in the Appendix, Table A-l. In this section, we estimate determinants of the award and receipt of child support for currently divorced or separated and remarried women together. z Because black women are much less likely than nonblack women to be awarded child support, we estimate the determinants of award probability separately by race. However, in analyzing the receipt of child support, we combine blacks and nonblacks because too few blacks are awarded child support to justify separate analysis. * The variables used in the regression equations are defined in the Appendix, Table A-2. Jeterminants of the Probability of Being Awarded Child Support Chila support awards may be either court-ordered or informally agreed to, although even most voluntary agreements are formalized through legal contracts. Our theoretical analysis postulates that child support is more likely to be awarded the greater is tne perceived need of the woman and children, and the -8- greater is the ability (or desire) of the absent father to pay. The probability of award also depends upon the legal environment at divorce which varies from state to state and over time, and may vary within a state across legal jurisdictions, as well as across individual judges. There is some evidence that in states with no-fault divorce laws, the amount of child support awarded is lower than in states without no-fault divorce (Peters, 1982) . For our sample we have only some of the information that would be desirable on the needs of the mother and on the ability of the father to pay, and no information on the legal environment. Let P be the probability that child support is agreed to or awarded conditional upon a woman being ever-divorced or currently separated. Then the logistic function P = 1/(1 + e~ 6X ~ u ) (5) was estimated by maximum likelihood methods, where X is a vector of independ- ent variables and 8 a vector of coefficients to be estimated. Table 1 presents the estimated partial derivatives of P with respect to each variable, found by multiplying the estimated B's by P(l-P), where P is the mean of the dependent variable. Variables which may reflect the new family's financial need include the mother's age at divorce (AGEDIV), education (EDUC and COLLGRAD), number of children by the absent father (PATERNR) and age of the children (KID6T017). While no direct information is available on the father's financial status at the time of divorce, his ex-wife's age (at divorce) and education will be positively correlated with his own and, conseouently, with his ability to pay. Another proxy for his ability to pay is the value of the property settlement reached in the divorce proceedings (SETVAL). Finally, geographic location variables (NEAST, NCENTR, SOUTH, SMSA, and CC) may control for some of the -9- variation in the legal environment that we are unable to capture. It is important to distinguish divorced from separated (SEP) women in our sample. Fourteen percent of the nonolacks are currently separated, while 48 percent of the blacks are separated. Separated women may, but are less likely to, be awardea child support. According to our estimates, ceteris paribus , separated nonblack (black) women are 21 (29) percent less likely than comparable divorced women to have been awarded child support. For nonblacks, the probability of being awarded child support increases as the number of aependent children increases (PATERNR), by 4.1 to 4.7 percent per chila, ana as one or more of tnese children tends to be older (KID6T017). For the black sample, however, neither variable is significant. EDUC is positive for both groups but significant only for nonblacks. In addition, the dummy variable COLLGRAD is negative and significant for non- blacks, suggesting that college educated women are between 2.5 and 5.0 percent less likely than otherwise similar high school graduates to be awarded child support. As noted above, EDUC may serve both as a proxy for ex-husband's lifetime earnings potential and as a measure of the new family's financial needs. Although more education may enable a woman to support the children by herself, thereby reducing needs, it also raises the standard of living to which the family is accustomed, thereby increasing needs. The positive sign on EDUC may reflect both this second effect as well as her ex-husband's ability to pay, while the negative coefficient on COLLGRAD may reflect the competing effect of a lower need for support. The coefficient on the woman's age at divorce (AGEDIV) is positive for both races and significant for blacks. In part, her age is again a proxy for her ex-husbano's ability to provide support. (It appears to be a better proxy -10- for income of black ex-husbands than is her education.) But in addition, the variable is strongly positively correlated with the duration of the marriage. ■*■-> Longer marriages may be more likely to result in child support awards for several reasons. Chief among them is that the father is likely to be more attached to his children and therefore more disposed toward providing them support. A longer duration of marriage might also suggest that the woman has invested more in this marriage, developing greater marriage-specific capital (Becker, Landes and Michael, 1977) and specialized more in home activities, developing less human capital valued in the market. Thus, she woula be less able to support the children. In neither racial sample of ever-divorced women is the dummy variable on current marital status (REMAR) significant. This result is important, since it suggests that women who have remarried do not possess some unobserved traits that make them systematically more or less likely to be awarded child support at the time of their divorce. For a slightly smaller sample of women, we were able to include an index of the value of property settlements (SETVAL) as a proxy variable for the financial well-being of the couple before divorce. Since property settlements and child support awards are determined together, the coefficient on SETVAL may be biased if any unobservable factors which affect the value of property settlements also affect the probability of being awarded child support. To the extent that SETVAL is a good proxy, its significant positive coefficient for both racial groups reinforces the hypothesis that father's ability to pay (and the original family's standard of living) is a strong determinant of child support awaras. Note that the inclusion of SETVAL reduces the coefficients on EDUC and AGEDIV, also proxies for father's ability to pay. -11- For the sample of all women both awaroed child support and expecting to receive it in 1978, we estimated eouations on factors determining the amount due (results not shown), with a few notable exceptions, factors that determine whether or not child support is awarded also affect the amount due. Few child support awards are automatically indexed to the rising price level (Krause, 1981, p. 24) and most are infreouently renegotiated; thus, we controlled for the number of years since the divorce (YEARSDIV) and found its coefficient to be negative and significant. We also introduced as an explanatory variable the woman's estimate of her ex-husband's current income (HUSINC); this considerably reduces the sample size because many women do not know HUSINC.-'- 6 While the results from this smaller sample may be biased because of this self-selection, the means of the independent variables (reported in Table A-l) changed very little. We found the coefficient on HUSINC positive and significant, suggesting that each adaitional $5000 in the absent father's 1978 income raises child support due by $219. The coefficient on EDUC declines in size, but remains significant, suggesting that EDUC stands for more than just his current ability to pay. We nave argued that it serves as a proxy for his lifetime earnings potential upon which the support award is likely to be based. It may also indicate a higher standard of living to which the family is accustomed. Determinants of the Probability of Receiving Child Support and of the Amount Received Not all women due child support actually receive it. Of those expecting payment in 1978, only 71. A percent received any. Moreover, those who receive it frequently receive less than what is due. Among women receiving some support, the mean receipt was $1899 out of the $2204 due them, around 86 percent. In this section we offer some evidence that the probability of actually receiving child support, unlike its award, depends less upon the needs -12- of the woman, and more upon the current financial status of her ex-husband. This problem consists of two related issues: first, whether a divorced mother receives any child support and second, how much she receives. This second issue introduces the possibility of sample selection bias: estimates of the determinants of support received may be biased if some omitted factors that determine whether the woman gets into the sample (i.e., receives any child support) also determine how much she receives. To correct for this sample selection bias, we use a technique developed by Heckman (1979) which eliminates the bias by introducing a new independent variable A (negatively related to the probability of receiving child support) into the regressions on amount of child support received. Using maximum likelihood probit, first we estimate determinants of the probability of receiving any child support in 1978, conditional upon some payment being due. Estimated partial derivatives of these factors are presented in cols. 1-3 of Table 2. Next we use the probit estimates to compute a x-value for each mother who receives any child support, and enter it in OLS regressions on child support received, holding constant the amount due. These results appear in cols. 4-6 of Table 2. Because previous researchers did not explicitly recognize the possibility of sample selection bias, their regression results on factors affecting child support received may be biased (see, for example, Cassetty, 1978). Ex poste , we find only limited evidence of sample selection bias: the coefficient on x is insignificantly different from zero in both cols. A and 5 and only significant at a 10 percent level in col. 6. The negative sign on X in col. 6 indicates that omitted factors which lower the probability of receiv- ing chilo support (and therefore increase x) also lower the amount received. Black women are less likely to receive child support and, when they do, receive less than nonblack women, even after controlling for the lower amount -13- they are due. However, these differences become insignificant for blacks who know their ex-husband's income. Although women of Spanish origin are eaually likely to receive child support as non-Spanish women, they do receive significantly less child support. The coefficient on SPANISH becomes even more negative for those few who know their ex-husband's income. EDUC, AGE and AGESQ may serve as proxies for the absent father's ability to pay. Education has a positive impact on the likelihood of payment but not on the amount. Advancing age increases both the probability of receiving payment and the amount paid at a decreasing rate (reminiscent of the manner in which earnings change with age). When HUSINC is included, these proxies for ability to pay become insignificant or less significant in the probability of receipt eguation (col. 3), but age remains significant in the eauation on the amount received (col. 6). Although HUSINC significantly increases the likelihood of payment, it has no significant effect upon the amount paid once we control for the amount due. Thus, we conclude that while the absent father's income is an important determinant of the amount of child support awarded (each $5000 increment raises the amount by $219) and of whether or not ne pays anything, it is not an important determinant of the portion of the award that he pays. This suggests that fathers with negative transitory income are likely to evade the support obligation altogether rather than pay a smaller portion of the amount due. ° Three measures of the impact of children are included. Holding constant the length of time since the divorce (YEARSDIV), having older children (KID6T017) increases the probability of receiving support, consistent with the hypothesis that absent fathers are more likely to pay support for children whom they lived with longer. As the number of children due support increases (PATERNR), the probability that support will be paid appears to decline although -14- this effect is significant only in the equation in col. 2. This result is difficult to explain. If the woman has other children not fathered by her most recent ex-husband (OTHERKID), this significantly reduces by 7 to 9 percent the probability that he pays child support. This result is more understandable. Given the amount due, neither the number nor ages of the children affects the amount of support received, except in the sample that knows HUSINC. This group (col. 6) receives $83 more for eacn additional child but $253 less for older chilaren, another result that is difficult to explain. Women were asked whether their child support awards were court-ordered or agreed to voluntarily. Those reporting voluntary agreements (CSVOL) are between 15 and 22 percent more likely to receive payments; however, given the amount agreed to, the effect of CSVOL on the amount received is positive but insignif- icant. Apparently voluntary initial agreement, which probably indicates some- thing about the character of the father or of the relationship, has lasting effects. If the marriage that ended in divorce was not the woman's first, her likelihood of receiving payment is reduced for each higher-order marriage (NUMMAR) oy around 8.5 percent, at least when controlling for her ex-husband's income. Moreover, she receives up to $204 less support per higher-order marriage, controlling for the amount due. This is especially interesting in light of the finding that women in higher-order marriages seem to negotiate somewhat higher support awards (results not shown). Absent fathers reported by their ex-wives to have other children to support (HUSCHD) are no less likely to pay child support. While new children may reduce his ability to support absent ones, the father's decision to remarry and have more cnilaren suggests he values family life and is apt to be more conscientious about providing support for his absent children. Because of these -15- competing effects, the insignificance of HUSCHD does not disturb our hypothesis that the probability of receiving child support is largely a function of the absent father's ability to pay. 20 The variable YEARSDIV has a significant negative effect upon the probability of receiving child support, but not on the proportion received. Each additional year since the divorce reduces the probability of receiving support by approximately 2 percent. This relationship appears to be linear. The absent father may be less likely to pay support with each passing year if he loses physical or emotional contact with his children. (We have no information on whether he has visitation rights.) This explanation is reinforced by noting that the coefficient on YEARSDIV is much smaller in magnitude for the sample of women who know HUSINC, women who may be presumed not to have lost complete contact with their ex-husbands. Whether the woman is remarried (REMAR) does not affect the likelihood she receives any child support, but does appear to lower the dollar amount she receives by up to $159, except for those (remarried) women who know HUSINC. 2 - 1 - This suggests that when his ex-wife remarries, the absent father may reduce somewhat the amount of child support he pays, but not attempt to renegotiate the amount due (REMAR is insignificant in the eouation on amount due — results not shown), nor stop paying altogether (REMAR is insignificant in cols. 1-3 of Table 2). However, another possibility is that our regression suffers from reverse causality: women who receive less of the support they are due may be more likely to remarry. We examine this possibility in the next section. IV. Empirical Analysis of Remarriage Tne majority of women in the United States who divorce eventually remarry. For example, in the 1979 CPS, 70 percent of divorced mothers remarried within -16- 15 years of their divorce. This identical remarriage rate for divorced women was found in the 1967 Survey of Economic Opportunity (SEO) data by Becker, Landes, and Michael (1977). In the CPS, one-half of all mothers remarried within 5 years of their divorce, and 29 percent within 2 years. 22 The modern theory of job search has recently been applied to marital search by, among others, Becker, Landes, and Michael (1977) and Hutchens (1979). Briefly, the probability of remarriage depends upon a woman's own searcn behavior and upon her attractiveness to potential marriage partners. A aivorced woman can decide whetner or not to engage in search, how intensively to searcn, and how long to search. These decisions depend upon her expected gain frdm remarriage which is a function of her expected future flow of real income if single compared to her distribution of offers of real income if remarried, net of search costs. A divorced woman's distribution of offers will depend upon her attractiveness to potential mates, which will be a function of some characteristics we can observe such as age, some we cannot observe such as charm or beauty, and the amount of "marital -specific" capital (investments that are significantly less valuable when aivorced, such as children) from her previous marriage. As Becker, Landes, and Michael (1977, p. 1155) point out, "positive" specific capital in one marriage may be "negative" specific capital in a subseauent marriage. Some factors affect her search behavior and her attractiveness to others in opposite directions making it difficult to determine their net effect upon the probability of remarriage a priori . One example is a woman's portable (carries into marriage) nonwage income. A higher income if single would reduce the gain from remarriage which might cause a woman to choose not to search or to search less intensely and longer. However, it also enhances her attractiveness to potential spouses, causing an increase in the mean of the -17- offer distribution. This would tend to increase her probability of participa- tion in search, and have an indeterminant impact on intensity and duration of search (Hutchens, 1979, p. 371). Hutchens cites Becker's theoretical analysis (1973, p. 891) as evidence that the latter effect tends to dominate the former. Thus, greater portable nonwage income is expected to increase both the probability of participation in and the duration of search. The presence of children is another factor with competing effects. The greater the number of children, ceteris paribus , the greater the economic gain from remarriage; thus, the woman is likely to set lower standards for a new husband and to search more intensely. However, this greater specific capital from the previous marriage also reduces her attractiveness to potential marriage partners which decreases the number and mean value of marriage offers. We hypothesize that women will remarry ouickly if they are either unusually attractive to potential mates or have established a low "reservation" set of characteristics for a new husband. Similarly, women will take longer to remarry (or never remarry) if they are either less attractive partners or have set a higher standard for a new husband. We related the probability of remarriage by the nth year after termination of the previous marriage (n = 2, 5, 10 and 15) to the expected gain from remarriage. We selected our sample so that all women were divorced (excludes separated) at least n_ years. (Thus, each successive sample is smaller than the previous one.) We estimated a logistic function for each sample by maximum likelihood methods. Results for the n = 15 group are not shown because that sample contains less than 150 observations and because most of these women had remarried by n = 10. Partial derivatives of the logistic function with respect to each independent variable, evaluated at the sample mean, for divorce durations of 2, 5 and 10 years are reported in Table 3. -18- A higher age at divorce (AGEDIV) significantly reduces the probability of remarriage of divorced mothers at all three durations. The increase in the size of the coefficient with each successive duration suggests that age at divorce becomes an increasingly important determinant of remarriage probabil- ities at higher durations, possibly because it reduces a woman's attractiveness as a marriage partner. The greater her number of previous marriages (NUMMAR), the less likely a woman is to remarry, holding AGEDIV constant. Ceteris paribus , a greater number of previous marriages might reduce her attractive- ness to potential marriage partners or her inclination to try again. Again, this effect becomes larger at higher durations. We might expect to find a secular rise in the probability of remarriage because the divorce rate has been increasing, which makes available a greater pool of potential mates and reduces the stigma associated with being a divorced woman. Indeed, we find that for durations of 5 and 10 years, given AGEDIV, the more recent the divorce (the lower YEARSDIV), the more likely is a woman to remarry. This effect is significant in the regressions in columns 3 and 5. As discussed above, a greater number of children (PATERNR) will have the competing effects of raising search intensity but lowering potential marriage offers. (The former effect may be mitigated by the fact that a greater number of children probably increases her costs of search and reduces her need for additional companionship.) We find that the probability of remarriage within 2 years of divorce (col. 1) increases as the number of children increases up to 4 and decreases thereafter. The woman may have set her reservation standard low in order to remarry Quickly for her children's sake, but too many children make it difficult to attract offers. At higher durations, the effect is linear: each additional child reduces the probability of remarriage within 5 years by 3.9 percent (col. 2) ana within 10 years by 9.6 percent (col. 4). -19- When we control for chilo support due per child (PCCSDUE), the coefficient on PATERNR is still negative and significant at both durations. This suggests that what is unattractive about a greater number of children is more than just the cost of caring for them. The increase in the size of the coefficient on PATERNR with increasing divorce duration suggests that the woman's attractiveness as a marriage partner becomes an increasingly important determinant of remarriage the longer she has been divorced. Alimony is nonportable income; that is, it stops with remarriage. As such, it should have a negative effect upon the probability of remarriage. AWARDAL is negative at all divorce durations, becomes larger at higher durations, but is significant only at 10 years in the regression controlling for child support due (col. 5). Women awarded alimony are 8.2 percent less likely to have remarried 5 years after divorce and 16.2 percent less likely 10 years after. This finding is consistent with findings by Hutchens (1979, p. 377) that nonportable transfers tend to reduce remarriage probabilities. Unfortunately, this is the only information about alimony we have that is relevant to the remarriage decision. Child support income is legally portable between the single and married status. As such, it has the competing effects discussed above of decreasing the gain from remarriage but of increasing the woman's attractiveness to potential mates; therefore, theoretically its effect upon remarriage is indeterminate. Where Hutchens (1979, p. 377) was unable to draw any conclusions on the relationship between portable nonwage income and remarriage, we are able to do so. Two additional considerations make it likely that child support income will aecrease the gain from remarriage more than other income. First, if a divorced mother spends child support income on the children, in part on goods and services substitutable for child care time, then it reduces the benefits -20- to be derived from the sexual division of labor in the care of children. Second, although legally portable, in practice child support income may be only partly portable. We showed in Table 2 that for otherwise identical women due the same amount of child support, remarried women receive up to $159 less support than divorced women. If fully anticipated, this reduction in payments would serve to further reduce the expected gain from remarriage. Women awarded child support (AWARDCS) are 5.3 percent less likely to remarry within the first 2 years after divorce, but this effect is significant only at the 10 percent level (col. 1). If not awarded child support, a woman stands to gain a great deal more from a auick remarriage than a woman awarded support. Consequently, she may search more intensely and set lower standards for a new husband. In the equations in columns 3 and 5, we add measures of whether child support was due in 1978 (CHSUP78) and of the amount of child support due per child (PCCSDUE). Women both awarded child support and to whom it was due in 1978 were 3.9 percent less likely to have remarried 5 years after the termination of their previous marriage and 11.1 percent less likely 10 years after than women not awarded child support. For those awarded and due child support in 1978, each additional $500 due per child reduces the probability of remarriage within 5 years by 4.2 percent. Within 10 years, the effect is smaller and statistically insignificant. On net, child support income reduces the gain from remarriage more than it increases the woman's attractiveness to potential mates, probably because several factors work in this direction. Thus, we conclude from these findings on child support that this type of portable nonwage income reduces the probability of remarriage. Holding constant whether and how much child support is due in 1978, AWARDCS becomes positive and significant. Women awarded child support but due none in -21- 1978 (around 20 percent of the sample) are 18-19 percent more likely to remarry than women not awarded support, even though women both awarded and due support are less likely to remarry. If a woman is due child support, or due more support, she searches longer before remarrying, and may not remarry at all. But, if her child support income stops she is much more likely to remarry than if it were never awarded at all. Two otner variables that have the competing effects of reducing the woman's gain from remarriage but of increasing her attractiveness to others are her wealth at divorce (SETVAL) and her education (EDUC). SETVAL is insignificant in all regressions; EDUC is consistently negative, but significant only at the 10 percent level for durations of 2 and 5 years. AFDC and other public assistance benefits available to single mothers are other important forms of nonportable income which would reduce the gain from remarriage and prolong the duration of divorce. Because remarriage occurs in many different years in our sample, we have no simple measure of the relevant public assistance benefits when each woman was deciding upon remarriage. However, some of our variables may be interpreted as proxies for the likelihood a divorced woman received AFDC. For example, by 5 years after termination of the marriage, black women (BLACK) are 32 percent less likely to have remarried than similar nonblack women. If black women, women in the northeast (NEAST) or in central cities (CO, and women with many children are somewhat more likely AFDC candidates than other women, then our results support previous findings (Hutchens, 1979) that welfare reduces the probability of remarriage. V. Empirical Analysis of Labor Force Participation Although differences in female labor force participation rates have narrowed recently, unmarried women are still more likely to be in the labor force than married women. Among our sample of 1503 currently divorced or -22- separated mothers, 73.9 percent were either working or looking for work in March 1979, compared with 60.3 percent of our sample of 936 remarried mothers. The greater labor force attachment of divorced mothers is an obvious response to their loss of other family income. In this section we examine factors affecting the variation in labor force participation rates among divorced mothers, paying special attention to the influence of family support — child support and alimony. The probability of labor force participation should depend negatively upon the value of a woman's time at home, positively upon the value of her time in the market, and negatively upon nonlabor income. The presence of younger children or more children raises her time value at home, while education and job experience raise her value more in the market than in the home. In addition, if market work imposes fixed entry costs, then a woman may elect not to participate unless her desired hours of employment exceed some critical minimum. And since hours of work depend negatively upon nonlabor income, a c eteris paribus increase in such income should reduce the likelihood a woman participates in the labor force. For divorced mothers, most nonlabor income consists of income from property, earnings of other family members, public welfare, and alimony and child support payments from her ex-husband. In contrast to the other sources of nonlabor income, child support payments might exert an additional and opposite influence upon labor force participation. Since this income is awarded to support the children, a woman may feel explicitly or implicitly constrained to spend it entirely on goods and services for them. If these commodities (such as day-care for small cniloren) can substitute for her own home time, then she may be able to reduce her hours of work in the home, and raise her desired hours of employment beyono the critical minimum required for labor force entry. If this occurs, -23- then the net impact of child support payments on labor force participation is ambiguous. On the one hand, child support has an income effect of raising leisure, reducing desired hours of work, and therefore reducing the probability of labor force participation. On the other hand, child support has a substi- tution effect of increasing the quantity of goods devoted to children, decreasing the amount of time spent on them, and therefore raising the likelihood of labor force participation. Table 4 reports estimated partial derivatives from logistic models on factors affecting the likelihood of labor force participation in March 1979 for our sample of divorced mothers. All three variations contain the same economic and demographic variables but different child support and alimony variables. The coefficients on the common set of independent variables change very little across functional forms. The pattern of coefficients on the number of children under 18 years by age groups is consistent with findings from studies for married, spouse-present women (Smith, 1980). Compared with having at home one child age 18 to 20, having one child age 6 to 17 (KIDLT18) reduces the probability of participation by 3 percent; having one child age 3 to 5 (KIDLT6) reduces participation another 6 percent; and having one child under 3 (KIDLT3) reduces participation another 13 percent. Small children raise the value of the woman's home time and reduce her likelihood of labor force entry. Each additional adult in the household (NADULT) who might share in child care duties increases her labor force participation by up to A. 5 percent. As the woman becomes more valuable in the market, her labor force participation increases. Each additional year of education (EDUC) raises her entry probability by over 5 percent. Advancing age (and presumably experience) raises entry at a declining rate (AGE is positive and AGESQ is -24- negative). Finally, women who have been divorced a longer time (YEARSDIV) are less likely to enter the job market. This result appears counterintuitive if labor market entry is viewed as one possible way to restore economic well-being lost at divorce (where remarriage is another possible way). One explanation consistent with this result is that mothers who are likely candidates for public assistance and therefore less likely to be working remain divorced (or separated) for a longer time than do other mothers to maintain their eligibility. Alternatively, this result may reflect the secular trend toward rising labor force participation among younger (more recently divorced) women. Black women (BLACK) are around 8 percent and women of Spanish origin (SPANISH) around 10 percent less likely to participate in the labor force than other divorced mothers. This may reflect cultural differences and/or discrimi- nation against minority women in the labor market. Women in the northeast (NEAST) are 12 to 13 percent less likely than other women to participate, while women in central cities (CC) are 6 percent less likely to participate. Since women with these characteristics make up a disproportionate share of the AFDC population, these results also suggest that an increased likelihood of being on public assistance reduces labor force participation, findings consistent with those of other studies such as Keeley et al . (1978). AFDCMAX eouals the maximum amount of monthly AFDC payments that a woman with a given number of children in a particular state could receive if she does not work (and has no other income) as of July 1978. (Data on AFDCMAX were obtained from Department of Health, Education and Welfare, 1979.) In other words, it represents potential, not actual, welfare benefits. Generous public assistance benefits, ceteris paribus , should be expected to reduce labor force participation. Our finding indicates that each additional $100 per month in benefits reduces the likelihood of labor force participation by around 5 -25- percent. When these potential benefits are replaced by actual benefits received, the regression fits even better but it then suffers from an obvious simultaneity bias. 25 As expected, the coefficient on other family income excluding child support, alimony, and public assistance (OTHFINC) is negative and significant; each additional $1000 in annual income deters labor market entry by slightly more than 1 percent. The coefficient on amount of alimony due (ALIMDUE) is also negative and significant, but each additional $1000 of alimony deters entry by 2.7 percent (col. 1). Controlling for whether or not alimony is due in 1978 (ALIM78), the coefficient on amount received (ALIMREC) is virtually identical to the one on ALIMDUE (col. 2). One reason that alimony appears to reduce participation more than other income is that its coefficient may be biased downward. Women awarded alimony are probably less likely to have worked in the labor market during their marriage and thus, with less previous job experience, less likely to be working now. Therefore, the coefficient on alimony may reflect both an income effect and an "experience" effect. Unlike alimony, child support income appears to be positively associated with labor force participation. This is consistent with past work demonstrating a positive simple correlation between labor force participation and the receipt of child support (Grossman and Hayghe, 1982), but extends it by showing that even when other characteristics (such as race, age, education, and family size) are held constant, the positive association persists. The coefficient of .0000305 on CHSUPDUE (col. 1) indicates that among women due child support in 1978, one who is due the average amount (slightly over $2100) is 6.4 percent more likely to participate in the labor force than an otherwise identical woman due no child support. Furthermore, each additional $1000 of support due raises participation by 3.0 percent. The positive sign is -26- consistent with our hypothesis that child support payments are used to purchase goods and services that substitute for mother's home time and thus facilitate her labor force participation. In cols. 2 and 3, we enter alternative measures of child support due and/or received. Women due child support in 1978 (CHSUP78) are 11 to 14 percent more likely to participate in the labor force than women not due support. One reason why this effect appears so large is that the coefficient on CHSUP78 may be biased upward. At divorce, women who anticipate future labor force participation may seek to obtain child support, which is nontaxable income, instead of alimony, which is taxable. Once we control for whether it is due, the amount of child support received (CHSUPREC) has no significant additional effect on participation (col. 2). However, for the 58 percent of divorced mothers due child support in 1978 who received it regularly (CSREGULR), the difference in labor force participation over women due no support is 6.1 percent less, but this effect is insignificant (col. 3). This suggests that the uncertainty created by receiving child support payments on an irregular basis may cause a greater increase in the labor force participation of divorced mothers than if payment were regular. VI. Summary and Conclusions In this section we summarize our most important findings and discuss their implications for the formation of public policy. Also, we suggest promising directions for future research. Like other forms of nonlabor income, child support payments were found to lower the probability of remarriage. But unlike other forms of nonlabor income, child support payments were found to raise labor force participation. Each of these findings is consistent with our hypothesis that child support income serves as a substitute for the absent father. If a divorced mother -27- receives child support, her economic incentive to remarry is reduced. This allows her to increase the duration of marital search which should improve the duality of the match and thus the likelihood that the new marriage will last. In addition to these long-run benefits, child support payments also increase a divorced woman's labor force participation, which raises her family's income and well-being in the short-run. Thus, child support payments are found to have both immediate and future benefits. It is interesting to contrast the effects of child support, a private transfer, with AFDC payments, a public transfer. Previous work (Hutchens, 1979) has found that AFDC reduces remarriage rates, just as we found that child support does. Other previous work (Keeley et al . , 1978) has found that AFDC reduces labor force participation unlike our finding that child support raises participation. While this may be evidence that child support income does not create the same work disincentive as AFDC, it is also possible that women not awarded child support, who are more likely candidates for AFDC, exhibit a fundamentally different response to nonwage income. Within the limitations imposed by our data, we found that somewhat different factors affect the award of child support than its receipt. The likelihood of being awarded child support depends upon the needs of the mother and her children, and upon the absent father's long-term ability to pay. There appears also to be substantial racial differences: black women are less likely to be awarded support. In contrast, the likelihood of receiving child support due depends less upon race or the circumstances of the woman, and more upon the current financial well-being of the ex-husband. His income was found to affect whether or not any child support is actually paid, but not how much is paid. Apparently, absent fathers with low current incomes evade payment altogether, instead of partially cutting back on payments. This suggests that -28- one effective strategy for child support enforcement may be to get nonpaying fathers to pay at least some portion of the child support they owe. Given the beneficial effects of child support upon the well-being of female-headed families, it is unfortunate that such support is not awarded to or received by more divorced mothers and their children. Recent state and federal legislation have attempted to improve the enforcement of existing child support contracts. In future work we plan to study the effects of these new laws on child support receipts. However, no degree of enforcement can improve the well-being of a family never awarded child support in the first place. More effort needs to be placed on improving the system of awards. Initial work in this direction has been accomplished by Garfinkel (1982). Our analysis of determinants may help to provide a foundation for improving both the enforcement and award of child support. -29- Notes 1. Statistical Abstract 1982-83 , Table 60. 2. Statistical Abstract 1982-83 , Table 717. 3. Statistical Abstract 1982-83 , Table 735. 4. National Organization of Women President, Judy Goldsmith, New York Times , September 1, 1983. 5. Current Population Reports , Series P-20, No. 380, p. 5. 6. Glick and Norton (1977). 7. The idea for this figure comes from Bradbury et al . (1979). 8. Subfamilies pose special problems of income sharing that we cannot readily handle. Never-married mothers differ in significant ways from ever-married mothers and thus must be considerea separately. Women who are widowed after remarriage form a small group whose economic position may depend more upon their current circumstances than upon having been divorced. 9. The population estimates published in that report differ slightly from the ones we report in Figure 1. 10. Black mothers are significantly less likely to be awarded support even after controlling for other factors. 11. Around 17.2 percent of all mothers — 16.8 percent of nonblacks and 21.6 percent of blacks — awarded child support were not due any in 1978 for a variety of reasons such as death of the previous spouse and children past the age of eligibility for support. 12. In separate regression equations we found no systematic differences between them in the estimated coefficients. 13. Equations estimated for the nonblack sample alone are nearly identical to those for all races combined, with the notable exception that the black intercept dummy is usually significant in the combined sample. -30- 14. It would be especially desirable to have some direct information on the financial circumstances of the father at the time of divorce. For our sample, it is not possible to find a simple measure of the legal environment at divorce Decause tne divorce year varies from 1945 to 1979. Ten percent of the sample were divorced in 1965 or earlier and 29 percent in 1970 or earlier. We could not in any case account for the individual variation from judge to judge in a study at the national level. For a study that takes account of such variation in a single state, see Chambers, 1979. 15. We also estimated this equation for the sample of currently divorced (or separated) women only for which we know marriage duration. When we enter it as an independent variable in place of AGEDIV, its coefficient is positive and significant for nonblacks, but insignificant for blacks. 16. This variable is believed to be subject to considerable measurement error, which would increase its standard error. It is thus surprising that it turns out as significant as it does. 17. The sample means (and standard deviations) for these a variables are as follows: x(l) .4l4(.246) A(2) .378C.281) \(3) .289C.212). 18. In a separate regression, we included child support due (CHSUPDUE) as an explanatory variable. Holding HUSINC constant, its coefficient is small but significantly positive: each additional $100 of support due increases the probability that some support will be received by two-tenths of one percent. If the absent father's permanent or lifetime income contributes positively to the amount of child support awarded, then the significance of both CHSUPDUE and HUSINC in the same regression could indicate that his permanent income (CHSUPDUE) and current income (HUSINC) have separate effects upon his -31- probability of paying support. Holding current income constant, absent fathers with higher permanent income are somewhat more likely to pay support. Holding permanent income constant, fathers with higher current income are much more likely to pay support. Again this is consistent with the hypothesis that current ability to pay has a strong impact upon child support payments. 19. Cassetty (1978) found that absent fathers who remarried were no more likely to pay child support, but paid significantly more. Since she does not control for amount due, her findings on amount received confound determinants of award amount with receipt amount, and thus are not directly comparable to ours. 20. Another reason why this coefficient is insignificant may be that this variable is measured with error. 21. Our results would appear to disagree with those of Cassetty (1978) who finds that remarried women are less likely to receive child support and also receive less child support. However, her dependent variable combines child support and alimony income together. Since alimony stops with remarriage while child support is supposed to continue, her results would inevitably be negatively biased. We are fortunate to have data that enable us to separate child support from alimony income. 22. In the 1967 SEO, only 43 percent of women remarried within 5 years and 22 percent within 2 years. Either divorced mothers remarry more rapidly than divorced women in general, or there has been a secular trend since 1967 toward shorter intervals between divorce and remarriage. We offer some evidence of the latter effect in our empirical estimates. 23. Had we restricted our sample to remarriages that took place within say the last 2 years instead of n_ years after divorce, we could have included measures of AFDC (as did Hutchens, 1979). But for the problem we are interested in our specification (similar to that of Becker, Landes, and Michael, 1977) is more appropriate. -32- 24. We also estimated equations with labor force participation for any part of 1978 as the dependent variable. The results — particularly on the child support and alimony variables — were virtually identical. 25. In addition, the size and significance of the coefficients on BLACK, SPANISH, NEAST, CC, KIDLT3, KIDLT6, KIDLT18, and YEARSDIV are reduced. This is consistent with our notion that each of these variables serves in part as a proxy for the probability of going on public assistance. 26. However, for a sample of recently divorced (or separated) mothers for whom we are able to control for whether or not they were in the labor force prior to the termination of their marriage (in March 1975), the coefficient on alimony due is virtually identical to the one reported here, suggesting that the "experience" bias may be negligible for this particular sample. -33- References Becksr, Gary S., "A Theory of Marriage: Part I." Journal of Political Economy 81 (July/August 1973): 813-846. Becker, Gary S., Elisabeth M. Landes, and Robert T. Michael, "An Economic Analysis of Marital Instability." Journal of Political Economy 85 (December 1977): 1141-87. Bradbury, Katherine, Sheldon Danziger, Eugene Smolensky, and Paul Smolensky, "Public Assistance, Female Headship, and Economic Weil-Being." Journal of Marriage and the Family 41 (August 1979): 519-35. Cassetty, Judith, Child Support and Public Policy . Lexington, MA: D.C. Heath, 1978. Chambers, David L., Making Fathers Pay . Chicago, IL: University of Chicago Press, 1979. Garfinkel, Irwin, Child Support: Weaknesses of the Old and Features of a Proposed New System , Vol. I, Institute for Research on Poverty, February 1982. Glick, Paul C. and Arthur J. Norton, "Marrying, Divorcing, and Living Together in the U.S. Today." Population Bulletin 32 (October 1977): 3-39. Gordon, Nancy M., Carol A. Jones, and Isabel V. Sawhill, "The Determinants of Child Support Payments." Working Paper #992-05. Washington D.C: The Urban Institute, June 1978. Grossman, Allyson Sherman and Howard Hayghe, "Labor Force Activity of Women Receiving Child Support or Alimony." Monthly Labor Review 105 (November 1982): 39-41. Heckman, James J., "Sample Selection Bias as a Specification Error." Econometrica 47 (January 1979): 153-61. Hoffman, Saul, "Marital Instability and the Economic Status of Women." Demography 14 (February 1977): 67-76. -34- Hutchens, Robert M. "Welfare, Remarriage, and Marital Search." The American Economic Review 69 (June 1979): 369-79. Keeley, Michael C, Philip K. Robins, Robert G. Spiegelman, and Richard W. West, "The Estimation of Labor Supply Models Using Experimental Data." The American Economic Review 68 (December 1978): 873-87. Krause, Harry D., Child Support in America, The Legal Perspective . Charlottesville, VA: The Michie Co., 1981. Peters, Elizabeth, "The Impact of Regulation of Marriage, Divorce, and Property Settlements in a Private Contracting Framework." Paper presented at the Workshop in Applications of Economics, University of Chicago, November 1982. Smith, James P., ed. Female Labor Supply: Theory and Estimation . Princeton: Princeton University Press, 1980. Sorensen, Annemette and Maurice MacDonald, "Child Support: Who Pays What to Whom?" In Child Support: Weaknesses of the Old and Features of A Proposed New System , Vol. Ill, Appendices, Institute for Research on Poverty, May 1981. U.S. Bureau of the Census, Current Population Reports, Series P-23, Special Studies No. 112, Child Support and Alimony: 1978 , Washington, D.C., September 1981. . Statistical Abstract of the United States: 1982-83 (103rd edition), Washington, D.C., 1982. . Current Population Reports, Series P-20, No. 380, Marital Status and Living Arrangements: March 1982 , Washington, D.C., May 1983. U.S. Department of Health, Education and Welfare, Social Security Administration, Office of Policy, AFDC Standards for Basic Needs, July 1978 , ORS Report D-2, March 1979. 9 Mothers with Children from an Absent Father 6,372 13,321 30. A 37.0 66.5 Ever-married Mothers 5,510 14,492 25.2 41.8 68.4 Never-married Mothers 862 5,331 63.3 6.2 54.0 Widowed after Remarriage 54 11,213 34.1 13.5 37.1 Female - Sub-Family Head Head 853 9 5,875 1,764 63.1 86.1 6.2 0.0 53.9 59.7 S q^ : Tabulation s by authors from the computer tapes of the March-April Match File of the i*79 Current Population Survey. IGJRE 1. MOTHERS 18 AND OVER WITH CHILDREN FROM AN ABSENT FATHER IN 1979: NUMBER OF WOMEN UN THOUSANDS) ; AVERAGE TOTAL FAMILY INCOME; PERCENTAGE IN POVERTY; PERCENTAGE RECEIVING -HILD SUPPORT; LABOR FORCE PARTICIPATION RATE. Table 1 Effect of Selected Factors on the Probability that Child Support is Awarded for All Ever-Divorced or Currently Separated Women, Age 18 and Over, with Own Children Under 21 Years of Age from an Absent Father as of Spring 1979, by Racea Independent Variable Nonblack (1) (2) Black (3) (4) EDUC COLLGRAD SPANISHb NEAST NCENTR SOUTH SMSA CC PATERNR KID6T017 NUMMAR REMAR AGED IV .025 (4.91) .019 (3.60) .020 (1.67) .016 (1.27) - .125 (2.79) - .126 (2.60) .022 (0.14) .027 (0.17) - .088 (2.46) - .089 (2.44) • • • • • • - .086 (3.00) - .074 (2.44) - .055 (0.54) - .092 (0.88) .018 (0.64) .004 (0.10) - .040 (0.41) - .063 (0.63) - .055 (2.06) - .059 (2.10) - .058 (0.65) - .069 (0.76) .014 (0.60) .008 (0.32) .160 (1.61) .155 (1.53) - .024 (0.88) - .016 (0.55) - .179 (2.16) - .170 (2.00) .047 (4.01) .041 (3.38) .022 (1.02) .019 (0.84) .053 (2.28) .042 (1.69) .025 (0.30) .017 (0.20) - .033 (1.38) - .024 (0.96) .042 (0.50) .053 (0.64) - .040 (1.69) - .023 (0.91) - .073 (0.86) - .094 (1.06) .001 (0.84) - .001 (0.39) .008 (2.33) .006 (1.68) Table 1 — continued Nonblack Black Independent Variable (1) (2) (3) (4) SEP - .207 (7.50) - .129 (4.39) - .291 (4.62) - .207 (3.13) SETVAL • • • .094 (6.90) • • • .151 (2.51) Constant - .097 (1.13) - .051 (0.57) - .439 (1.70) - .389 (1.49) Likelihood Ratio Test (Chi-sq) 180.9 241.7 48.6 49.8 N 2027 1907 389 373 Mean of Depenaent Variable .775 .769 .476 .453 a The logistic function P = 1/(1 + e~ bX " u ) was estimated by maximum likelihood methods and the coefficients reported above are BP(l-P) where P is the mean of the dependent variable reported in the last row of the table. Asymptotic t-values are shown in parentheses. ^SPANISH was omitted from the equation for blacks because less than 1 percent of the sample was Spanish. Table 2 Effect of Selected Factors on the Probability that Child Support is Received and on the Amount of Child Support Received by Ever-Divorced Women Probability of Receiving Child Support Amount of Child Support Received (1)' (2)' (3) : (4)' (5)' (6)' BLACK -.101 (2.69) -.123 (3.03) -.043 (0.73) SPANISH -.030 (0.61) -.020 (0.38) -.104 (1.39) NEAST .082 (2.36) .081 (2.23) .094 (1.99) NCENTR .055 (1.85) .065 (2.12) .110 (2.83) SOUTH .042 (1.43) .047 (1.50) .060 (1.55) SMSA -.018 (0.70) -.015 (0.54) -.045 (1.35) CC .010 (0.32) .012 (0.36) .079 (1.84) EDUC .015 (2.87) .016 (2.89) .013 (1.90) AGE .027 (2.66) .022 (2.05) .005 (0.46) AGESQ -.0003 (2.12) -.0002 (1.52) -.0001 (0.45) PMTERNR -.012 (1.05) -.026 (2.13) -.022 (1.54) OTHERKID -.079 (2.78) -.069 (2.35) -.087 (2.35) KID6T017 .066 (2.02) .087 (2.55) .072 (1.76) CSVOL .222 (8.33) .221 (8.03) .155 (4.55) NUMMAR -.045 (1.59) -.039 (1.40) -.085 (2.73) 299 (2.49) -259 (1.70) - 56 (0.38) 346 (2.83) -406 (3.02) -926 (3.60) 177 (1.30) 170 (1.71) 107 (0.84) 50 (0.67) - 5 (0.05) -139 (1.08) ■114 (1.58) - 71 (0.90) - 90 (0.88) 39 (0.66) 18 (0.29) - 67 (0.77) 90 (1.30) 95 (1.29) 30 (0.26) 0.2 (0.02) 6 (0.36) - 22 (1.12) 59 (1.95) 67 (2.27) 59 (2.03) -0.6 (1.64) -0.7 (1.90) -0.6 (1.61) 18 (0.63) - 13 (0.38) 83 (2.03) 89 (1.01) 49 (0.51) 140 (1.06) ■ 64 (0.74) - 20 (0.19) -253 (2.18) 173 (1.23) 271 (1.63) - 31 (0.22) ■198 (2.69) -204 (2.74) -180 (1.47) Table 2 — continued Probability of Receiving Child Support Amount of Child Support Received (l) a (2) a (3) a (4) c (5) C (6) C YEARSDIV -.020 (6.94) -.020 (5.88) -.011 (2.63) - 7 (0.42) - 15 (0.84) 21 (1.39) REMAR -.008 (0.30) -.005 (0.17) -.028 (0.80) -159 (2.63) -123 (1.92) - 53 (0.62) HUSCHD • • • -.012 (0.48) • • « • • • - 41 (0.66) • • • HUSINC • • • • • • .027 (3.33) • • • • • • -2.96 (0.10) CHSUPDUE • • • • • • • • • 0.891 (80.1) 0.897 (79.2) 0.930 (61.6) X(l) b • • • • • • • • • 198 (0.45) • • • • • • A(2) b • • • • • • • • • • • • 433 (0.82) • • • X(3) D • • • • • • • • • • • • • • • -1000 (1.70) Constant -.521 (2.89) -.430 (2.30) -.068 (0.32) -1199 (1.51) -1547 (1.88) -574 (0.86) R 2 • • • • • • • • • .877 .886 .901 Likelihood Ratio Test 219.0 624.2 270.2 • • • • • • • • • N 1461 1259 648 1043 931 521 Mean of Dependent Variable .714 .739 .804 $1899 $1918 $2146 a The coefficients reported in these columns are the probit maximum likelihood estimates (b) multiplied by the sample mean of normal density functions evaluated at X^s, where Xj_ is the vector of independent variables for the i^ n observation. These sample means are .297 for col. 1, .280 for col. 2, and .232 for col. 3. b > = f(XiS)/F(Xj_6), where f and F are, respectively, the density and distribution function for a standard normal variable, and X^ and B are defined as in footnote a. See Heckman (1979), p. 156. c Since we cannot reject the null hypothesis of no selection bias (i.e., that the coefficient on X is zero), the usual OLS standard errors and t-statistics are appropriate. See Heckman (1979), p. 158. Table 3 Factors Affecting the Probability of Remarriage, by Duration of Time Since the End of Previous Marriage 3 Duration of Time Since End of Previous Marriage (in years) Independent 2 5 10 Variables (1) (2) (3) (4) (5) EDUC -.010 (1.78) -.014 (1.89) -.011 (1.47) -.010 (0.97) -.008 (0.73) BLACK -.215 (4.20) -.312 (5.21) -.320 (5.46) -.178 (2.25) -.149 (1.93) SPANISH -.042 (0.73) -.107 (1.44) -.120 (1.58) .097 (0.87) .105 (0.91) NEAST -.081 (2.04) -.122 (2.33) -.116 (2.27) -.115 (1.53) -.081 (1.08) NCENTR .001 (0.00) -.024 (0.53) -.052 (1.15) .043 (0.61) .036 (0.53) SOUTH .003 (0.10) -.025 (0.56) -.035 (0.79) .089 (1.25) .105 (1.49) SMSA -.036 (1.26) -.067 (1.69) -.042 (1.08) -.038 (0.61) -.018 (0.30) CC -.049 (1.39) -.107 (2.31) -.090 (1.98) -.230 (3.39) -.221 (3.29) PATERNR .116 (2.16) -.039 (2.19) -.042 (2.35) -.096 (3.49) -.082 (2.99) PATSQ -.024 (2.16) • • • • • • • • • AGEDIV -.012 (6.24) -.017 (6.59) -.019 (7.63) -.021 (5.24) -.023 (5.78) YEARSDIV -.0002 (0.10) -.004 (1.09) -.012 (2.93) -.009 (1.31) -.017 (2.44) NUMMAR -.052 (1.45) -.168 (3.22) -.142 (2.71) -.253 (3.08) -.237 (2.85) AWARD AL -.007 (0.20) -.041 (0.81) -.082 (1.59) -.105 (1.28) -.162 (2.02) Table 3 — continued Duration of Time Since End of Previous Marriage (in years) Independent 2 5 10 Variables (1) (2) (3) (4) (5) SETVAL -.007 (0.69) .002 (0.17) • • • -.006 (0.26) • • • AWARDCS -.053 (1.72) -.042 (1.00) .188 (3.42) -.047 (0.80) .180 (2.39) CHSUP78 ■ • • • • • -.227 (3.81) • • • -.291 (3.40) PCC5DUE • • • • • • -.834* (2.53) io- -4 • • • .327*10- A (0.66) Constant .350 (3.00) 1.129 (7.28) 1.202 (7.85) 1.542 (6.01) 1.608 (6.22) Likelihooa Ratio Test (Cni-sq) 123.2 170.0 237.1 112.2 132.0 N 1570 1069 1150 474 500 Mean of Dependent Variable .298 .512 .504 .658 .656 aSee note a in Table 1. Table 4 Factors Affecting the Likelihood of Laoor ^orce Participation of Currently Divorced (or Separated) Women in March 1979 a Independent Variables (1) (2) (3) BLACK SPANISH NEAST NCENTR SOUTH SMSA CC KIDLT3 KIDLT6 KIDLT18 NUMMAR EDUC NADULT AGE AGESQ -.085 (2.31) -.076 (2.05) -.076 (2.06) -.101 (1.99) -.103 (2.02) -.105 (2.06) -.130 (3.36) -.127 (3.28) -.126 (3.25) -.027 (0.70) -.030 (0.78) -.033 (0.84) .015 (0.30) .028 (0.53) .025 (0.47) .021 (0.60) .019 (0.54) .019 (0.55) -.062 (1.73) -.057 (1.58) -.059 (1.62) -.127 (3.21) -.128 (3.22) -.128 (3.23) -.063 (2.19) -.062 (2.13) -.062 (2.16) -.026 (1.81) -.025 (1.75) -.025 (1.72) -.037 (1.27) -.033 (1.17) -.040 (1.38) .053 (8.72) .053 (8.79) .052 (8.76) .044 (1.75) .045 (1.76) .044 (1.73) .023 (2.33) .023 (2.25) .024 (2.35) -.0003 (2.57) -.0003 (2.43) -.0003 (2.52) Table 4 — continued Independent Variables (1) (2) (3) YEARSDIV OTHFINC AFDCMAX CHSUPDUE ALIMDUE CHSUP78 ALIM78 CHSUPREC ALIMREC CSREGULR Constant Likelihood Ratio Test (Chi-sa) N -.006 (2.07) -.006 (2.08) -.007 (2.19) -.115*10-4 (2.98) -.117*10-* (2.98) -.120*10- A (3.04) -5.35*10" 4 (2.77) -4.84*10- A (2.51) -4.87*10-4 (2.52) .305*10-* (2.99) • • • • • • -.272*10~ 4 (2.37) • • • • • • • • • .113 (3.70) .144 (4.12) • • • .006 (0.10) -.040 (0.88) • • • -.016*10" 4 (0.17) • • • • • • -.254*10-4 (1.75) • • • • • • • • • -.061 (1.61) -.448 (2.22) -.506 (2.49) -.508 (2.51) 361.8 1503 368.6 1503 368.0 1503 Mean of Dependent Variable .739 .739 .739 a See note a in Table 1. I >.e C — re c <- £ & II 51 c (U — 0) 1/5 X) re 0) > x re re a; o tj >■ .c a> i u Z u C O u > : E in 5 r> o I! II vc — S— I iA O ia lA <» VC CM ' —4 "W lA **~ CO "A vC VC *■* ^-1 vc CO CO CM — vc a — -* rA — i VA -3- lA ~». 0\ iA "A IA ■•« 00 X C 1 ' or ia -co cc tr c C > CO a re T C re ■c c re re 0) > B3 U c o U CCO it J2 c — u r. § ° 5 I "D — i CO U — i w — co -j c y t. CJ OJ o Sh Q, i c , x o r — I M ! s c co — " > -j » H C C CO TJ 1 ui I Aj IA ^~ OJ ° re o u 1 — ■> \ u ^ IT co I re ! ^H o 2 a tr a z a — c a CM ec LA "^ VN lA O C CM -j r^ CM fl" CM •? IA -^ A -T (^ A' rA 1^ « i-» co co O CM —> rA o c> (V ls» CO O VC f» IA CA CM Cv CM « AI •?• iA -3- A' C CM- co IT O CM *v -- rA Al — A' » h 0\ iA Cm 52 So O IA r^ vc O 0> ia r^ CO CO rA CM MO Cv o av -* rA CM <• ■c « eo -a vc r^ VC -3- O IA. VC — 1 _ __, CO — 1 O CO VC vc — ' o r^ lA CM CJv CO CO B A' r« vc CO CA LA r^ tf> wv WN fN o O "A O CM — 1 -A CM « CM 3- lA CJ cr — — 3 (T CJ cr CJ CJ _.' E ^ -^ re s LjJ 3 ^3 LA •c ry L. 1 u CJ ■3 cc - > c 0) *J a m .0 co 6 a> —. > 10 re — * (4 .c re re co o "a >. £ i * o z t- c o 0) > o -o z re c u 6 a CD iA CM VC CM ~i cm <■ o ~ r- so s r*- >^ -3- -5 o « O cc — ( so SO CM -3; ~H ia pa CO ^H ia f^ CM — II NO Os o rA SO "^ 0\ CD *-< , 1—1 «^ > _, SO so 2 CM w ^^ / - s ^ ^-s ~H Os 8 PA — i F\ VA CO O •-I ^ fA PA CT. "■> .-s *-» •coo <• o —tv* m -c CM pa pa -a 1 ec pa fh pa r^ pa ^ OS — i 1) -J c cj <-. 0)1)0 rs X O 3 — 10 ■n 0) c U D S >. £ O -i o > -J X 73 a; I CD > CO IS - — 1 re re o cu ■a- p^ oo o p-. pa CD — 00 o O so pa r^ 00 pa r»% r«> so -co CM w OS -H CM PA eo eo CD O O IA •C CM CM P^ so lA CM OS ■? -a- so —i o cr 0, >— « «r u Ul _l LJ in to CD >- > o S- LJ p o c u 0) o s > •— s -c i- -a >~ o a> ~— — ; 4-> (0 c: (4 0) co u a. 1* 01 -J U) o i J3 co ■o & ex c 0) 4-> 0) V) x: co 0) (1) 1—1 > (/) CO J-> (-1 £ co CO » n co X CJ :* c c 0J > 5 -3 X II 2 lA II ■a — — □ i— I C -H cu jz 5 o o cao 3 ON ■D T3 -H CO cu -I r~ w ir o O OH &« = D > -H I/) — CD —J C O U D 11 C §^& X C 1 — (A dJ c j* u 5 CJ u ^<5 co a .-h > — i ■»j * CD ■o i 0) TJ U S3 u 1- ~ 0) 2 CO J* > o u U 0) CO CO t- a »H mi O CD s — i to c a o CM -*T fA CM pa w ia a o fA -3- rv CM fA On CO CM CD in *-» lA lA lA CP> -h iv ia rv O —I PA -a- co o cn -a vc o CM O IA NO o n 82 o r«. fA fA I— I fA c lA o i*\ rg PA no IV PA o iv PA -a NO PA —I CM iv ON CO pa NO -a cm -a SH vo v lAC o no r^ co no on \T>£) no cjv-h rv co k*o -a on -co co- -HO CM NO h r -C-a w «— PA w csiec pa no O c fvfA ia -a cdia iaon s co -ho O O i-hj Ov cm pa -- -a rv i— i PA CO PA NO CM -a «a •* CM •« ON PA P*. <• . . -H -w •• •• CMCM --CM •• IA fV w w *_. w w fA w ON NO no r» PA (V ia rv fA >_• «H NO -H OS IA IA co co O On -h PA ir> OS on co PA -a IA -C o PA O • li ID 3 o en C .c co > to -U .C CO IT O "O >> ■fe » — » o z u c o CD > §3 II in ll 2 -o e > ~- — > 0.' l-i o o 0; a. CO i-i ar- O I/1H n * "O C I— I -»H C — < a) £ 73 T3 iH CD 0) n-l P^ C ON O O -h 5 "H I/) — 0J — I c o »- 0) CO o § U R —1 (ft E c O CO <- ">• s -^ o •a 6 tj 1 fc- 0) t-i u, _) co 3 co > O M CO ■— I m CJ re — ' -£ o 2 CM — i — i a — vr m r> o f> CK o o O -h a- <» 00 pH r^ r^. m o — 1 CO CM CM -H C CM — ( r-~ <• — 1 V£ —H Qg 2 r^ h— h- ^ z C_) U. _l _l 1 a _l 3 Q a _ 1 £ Z CI ^ S£ x 2 u. o •a 03 n re C 03 c c IT 03 03 sz. c OJ 1- CO a c o > "D 03 S3 r>. • f\ i*n in os O r*> iA ir\ » -a- CN' w\ O Pk. »-h k> <— i — ■» «— * ^H ^ t/l (fl CD CU 0)