The Wage Penalty for Motherhood

by Michelle J. Budig, Paula England
The Wage Penalty for Motherhood
Michelle J. Budig, Paula England
American Sociological Review
Start Page: 
End Page: 
Select license: 
Select License



University of Arizona University of Pennsylvania University of Pennsylvania

Motherhood is associated with lower hourly pay, but the causes of this are not well understood. Mothers may earn less than other women because having children causes them to (I) lose job experience, (2) be less productive at work, (3) trade off higher wages for mother-friendly jobs, or (4)be discriminated against by employers. Or the relationship may be spurious rather than causal-women with lower earning potential may have children at relatively higher rates. The authors use data from the 1982-1993 National Longitudinal Survey of Youth with fixed-effects models to exam- ine the wage penalty for motherhood. Results show a wage penalty of 7percent per child. Penalties are larger for married women than for unmarried women. Women with (more) children have fewer years of job experience, and after controlling for experience a penalty of 5 percent per child remains. "Mother-friendly " characteristics of the jobs held by mothers explain little of the penalty beyond the tendency of more mothers than non-mothers to work part-time. The portion of the motherhood penalty unexplained probably results from the effect of motherhood on productivity and/or from discrimination by employers against mothers. While the benefits of mothering diffuse widely-to the employers, neighbors, friends, spouses, and chil- dren of the adults who received the mothering-the costs of child rearing are borne disproportionately by mothers.

their children leave them exhausted or dis- tracted at work, making them less produc- tive. Fourth, employers may discriminate against mothers. Finally, perhaps the asso- ciation is not really a penalty resulting from motherhood and its consequences at all. What appears in cross-sectional research to be a causal effect of having children may be a spurious correlation; some of the same un- measured factors (such as career ambition) that discourage child-bearing may also in- crease earnings.

We build on Waldfogel's 1997 study. She uses a fixed-effects model to avoid spurious- ness. Analyzing panel data spanning 1968 to

Direct all correspondence to Michelle J. Budig 1988, and after controlling for marital status, or Paula England, Population Studies Center,

experience, and education, she finds a wage

University of Pennsylvania, 3718 Locust Walk,

penalty of 6 percent for mothers with one

Philadelphia, PA 19104-6298 (budig@ssc.

child and 13 percent for mothers with two or , or We

more children. (She does not provide infor-

thank Jane Waldfogel for helpful comments. Re-

mation on the size of the motherhood effect

search was supported in part by a grant from the MacArthur Foundation to the Research Network before partialling out that portion caused by

on the Family and Economy. motherhood reducing job experience.) We


use a similar statistical model, but analyze more recent data (1982 through 1993) and include more detailed measures to assess whether the loss of full-time experience and seniority caused by motherhood explains most of the penalty. Ours is the first analysis to distinguish among the four categories of years of full-time experience and seniority and years of part-time experience and senior- ity. (Seniority refers to experience with one's present employer.) We include a measure of number of employment breaks. By adding controls for a large number of job character- istics, we attempt to assess how much, if any, of the motherhood penalty results from moth- ers choosing or being confined to lower pay- ing but more "mother-friendly" jobs. We also examine whether the motherhood penalty varies by marital status since a growing num- ber of mothers are single.

A wage penalty for motherhood is relevant to larger issues of gender inequality. Most women are mothers, and women do most of the work of child rearing. Thus, any "price" of being a mother that is not experienced by fathers will affect many women and contrib- ute to gender inequality. Of course, lower pay for employed mothers at one point in time is only the tip of the iceberg. Lifetime earnings are also lowered for those women who have a period with no earnings because they stay home caring for children full time (Davies and Joshi 1995; Joshi 1990). Gen- der inequality in earnings affects other gen- der inequalities. Lifetime earnings affect pri- vate pension income. For married women, lower earnings may affect their bargaining power with their husbands (Blumstein and Schwartz 1983; England and Kilbourne 1990). For single mothers, the motherhood penalty contributes to the gap in poverty rates between households headed by a single woman and those containing an adult male (McLanahan and Kelly 1999).

Penalties for motherhood are also relevant for theory and policy because child rearing creates broad social benefits (Coleman 1993; England and Folbre1999; Folbre1994a, 1994b; Risman and Ferree 1995). All work confers benefits on those who consume what is produced. Work that produces a physical product or a business service often has few beneficiaries beyond those who buy the product and thus indirectly pay the worker. In contrast, "caring" labor also benefits those who make no payments to the worker. Good parenting, for example, increases the likelihood that a child will grow up to be a caring, well-behaved, and productive adult. This lowers crime rates, increases the level of care for the next generation, and contrib- utes to economic productivity. Most of those who benefit-the future employers, neigh- bors, spouses, friends, and children of the person who has been well reared-pay nothing to the parent. Thus, mothers pay a price in lowered wages for doing child rearing, while most of the rest of us are "free riders" on their labor.


Several recent studies find a wage penalty for motherhood in the United States (Lund- berg and Rose 2000; Neumark and Koren- man 1994; Waldfogel 1997, 1998a; 1998b). A motherhood penalty has also been found in the United Kingdom (Harkness and Waldfogel 1999; Joshi and Newel1 1989) and Germany (Harkness and Waldfogel 1999). Men suffer no such penalty-their wages are either unaffected (Loh 1996580) or even in- crease after having a child (Lundberg and Rose 2000).


Mothers have high rates of employment to- day-for example, over 40 percent of women with children under one year of age are in the labor force (Klerman and Leibowitz 1999). Nonetheless, many women lose at least some employment time to child-rearing (Klerman and Leibowitz 1999). One explanation of the wage penalty for motherhood is based on this fact; some mothers take time out of employ- ment, and loss of work experience affects later wages. Human capital theory predicts that experience and seniority have positive returns because they involve on-the-job training that makes workers more productive.

In this view, workers pay for a part of this training with an initially lower wage, but employers raise wages with seniority to re- tain their more productive, experienced workers. A return to experience is also com- patible with institutional theories that see the reward as a result of organizational policies and inertia that reward experience for rea- sons other than its link to productivity. Ei- ther view implies that mothers will earn less if they lose any job time in child rearing.

Past studies are unclear about what part of the child penalty is explained by work expe- rience because some authors report only pen- alties with or without controls for experience. Also, studies vary in how they measure ex- perience. An early study by Hi11 (1979:589) reports that controlling for experience and tenure explained all the negative effect of children on women's pay. Lundberg and Rose (2000) find a 5-percent penalty for women's first birth, but they did not include controls for job experience. Waldfogel (1997:214) finds a penalty net of experience: 6 percent for one child, declining to 4 per- cent if controls are added for whether the current job is part-time and how much of past experience was part-time. But she doesn't report the gross penalty that includes the ef- fects of experience that would be obtained by leaving experience out of the regression. Korenman and Neumark (1992: 246-47) find no net or gross penalty for motherhood; they find no difference in wage change across a two-year period (1980-1982) between women who experienced a birth during the period and those who had not, regardless of whether women's work experience during that interval was controlled. Perhaps the two- year interval they examined was too short to reveal the effects of motherhood on wages. None of the prior studies distinguished full- time from part-time work experience and whether the experience is general (i.e., with any employer) or is entirely with the current employer. These types of work experiences may differ in their returns. Corcoran, Duncan, and Ponza (1984) find smaller re- turns to part-time work experience compared with full-time experience. Waldfogel (1997) distinguishes between full-time and part- time experience, and finds almost identical returns, but she doesn't distinguish between general employment experience and senior-

ity with one's current employer. Korenman and Neumark (1992) make this distinction but not between full-time versus part-time within either category. Waldfogel(1997) also distinguishes between whether the woman's current job was part-time, but Korenman and Neumark (1992) do not. Hill's (1979) analy- sis makes most of these distinctions (but not

I whether seniority was part-time). However, she uses much older data (1976) and applies an OLS statistical model that does not con- trol for unobserved differences between

1 mothers and non-mothers. We distinguish among years of full-time experience, part-time experience, full-time seniority, and part-time seniority. We also include a measure of the number of employ- ment breaks the woman has taken because continuity may influence wages-that is, among women with equal years of experi- ence, those with more continuous experience may have higher earnings. For example, Felmlee's (1995) analysis of 1968-1973 data shows that women who changed employers but maintained continuous employment (de- fined as a break of no more than a month) were less likely to have a reduction and more likely to see an increase in wages compared with women who were out of the labor force between jobs. We examine the "gross" moth- erhood penalty, and then we estimate a "hu- man capital model" that controls all these measures of experience.


According to human capital theory, losing job experience adversely affects mothers' wages because the mothers are less produc- tive; that is, more experienced workers are more productive, therefore they are paid more. However, is there a link between motherhood and productivity that exists even among women with equal human capi- tal? Becker's (1991) "new home economics" argues that mothers may be less productive on the job than non-mothers because they are tired from home duties or because they are "storing" energy for anticipated work at home. The assumption is that non-mothers spend more of their nonemployment hours in leisure instead of in child care or other household work and that leisure takes less

energy-thus leaving more energy for paid work. In this same vein, mothers may spend time while at work worrying about their chil- dren, calling them at home, or scheduling appointments for them. They may take sick leave to deal with children's illnesses. Moth- ers also may choose or be relegated to less demanding occupations because of this ex- tra burden of the "second shift" (Hochschild 1989). This second mechanism, operating via occupational choice or placement, is ex- plored in a later section on "motherfriendly" jobs.

No study has directly measured the effort or productivity of mothers versus non-moth- ers, or men versus women; prior research has approached these questions only indi- rectly. Bielby and Bielby (1988) analyze data from a national survey that asked re- spondents how "hard" their jobs require them to work, how much "effort, either physical or mental" their jobs require, and how much "effort they put into their jobs be- yond what is required." Women reported slightly more effort than men. This is strik- ing since other research finds that men gen- erally overestimate and women underestimate their merit or performance (Colwill 1982). As far as we know, no research has compared mothers to non-mothers on effort measures. Since it is women's responsibility for the care of children that is presumed to create differences in effort between women and men, the absence of sex differences in effort in past research suggests that mothers and non-mothers may not differ in effort.

Our data, too, lack measures of productiv- ity or effort. Thus, we must see this "effort" explanation of the motherhood penalty as consistent with a residual effect of mother- hood not explained by other variables for which we do have measures.



Mothers may seek "mother-friendly" jobs. The features of these jobs that make them easier to combine with motherhood may compensate for their lower earnings, as pre- dicted by economists' theory of compensat- ing differentials. For example, following Becker (1991), mothers may choose jobs that require less energy or that have parent-

friendly characteristics, such as flexible hours, few demands for travel or weekend or evening work, on-site day care, or availabil- ity of a phone to check on children. The 1 theory of compensating differentials states 1 that competition eventually requires all jobs to be equally attractive to the worker at the margin when both pecuniary (wage) and nonpecuniary benefits are taken into account. In this view, employers can fill jobs for lower wages if they offer nonpecuniary amenities that some workers will trade off

1 against wages. How much the amenity re- duces the market wage is determined by the preferences of the worker at the margin (En- gland 1992:69-72). "Mother-friendliness" is just one of many nonpecuniary amenities that the theory predicts can compensate for lower wages. If mothers are more willing than other workers to trade off wages f& "mother-friendly" jobs, then mothers will earn less. The most obvious mother-friendly job characteristic is being able to work part- time. Waldfogel (1997) finds that, net of ex- perience and kducation, the wage penalty of 6 percent for having one child was reduced to 4 percent when she added a control for whetier the job was part-time and whether past experience was part-time; the penalty for having two or more children was reduced from 15 percent to 12 percent. No study has tested whether or how much other job characteristics explain the mother- hood penalty. However, two studies use data from the Quality of Employment Study (QES), which contain workers' self-reports of characteristics of their jobs, to explore whether women or mothers are especially likely to be employed in parent-friendly jobs. Glass (1990) found that predominantly male jobs had more flexible schedules, un- supervised break time, and paid sick leave and vacation, all features seen as parent- friendly. Glass and Camarigg (1992) con- structed indices of schedule flexibility and ease of job performance. Among workers employed roughly full-time, and net of edu- cation, experience, tenure, marital status, and firm size, mothers were no more likely to be in jobs with these characteristics than non-mothers, nor were these characteristics more common for those in predominantly female occupations.

Direct measures of features of jobs that make them more compatible with parenting would be ideal for studying mother-friendly jobs-for example, whether employers al- low flexible hours or choice about overtime work, provide on-site day care, or allow par- ents to make personal phone calls during work. But few direct measures exist, espe- cially for national probability samples. Given data limitations, our strategy is indi- rect-we try a large number of the available job measures, enter them into our models and determine if they explain any observed motherhood penalty. Our hope is that this broad array of job characteristics includes job features that determine or correlate with mother-friendliness, even if they do not di- rectly measure this construct.

One approach we take is to examine whether mothers are employed in more heavily "female" jobs, to see if this explains the lower earnings of mothers compared with non-mothers. Prior research shows that "fe- male" jobs pay less than "male" jobs, even after controlling for skill levels (England 1992; Kilbourne et al. 1994). Is some portion of the reduced wages of female jobs a differ- ential compensating for "mother-friendly" features not controlled for in previous analy- ses? The studies by Glass (1990) and Glass and Camarigg (1992), discussed above, do not support this idea that "female" jobs are more conducive to parenting. Desai and Waite (1991) indirectly explore the possibil- ity that "female" jobs are mother-friendly by examining whether such jobs helped women stay employed continuously around a birth. They did not find that women who worked in occupations employing a higher percentage of females stay employed longer during their pregnancies or return to work sooner after a birth than do those in other jobs. However, Okamoto and England (1999) find that moth- ers are more likely to be employed in occu- pations with a high percentage of females than are other women. We assess here whether the sex composition of jobs explains any of the motherhood wage penalty.

Some employer policies are "motherfriendly" (Glass and Estes 1997; Glass and Fujimoto 1995; Glass and Riley 1998). Based on their work, we suspect that large firms and public-sector organizations offer more family-friendly policies, and that the presence of unions has equivocal effects (more provision for leave, but less child-care aid). Mothers may turn to self-employment to accommodate child-care responsibilities (Connelly 1992; Glass and Fujimoto 1995; Presser 1994). Mothering may also be ac- commodated by working in child-care, either because it is done at home or because the mother can enroll her child where she works, sometimes at a discount. We include many job characteristics in our models to see if they explain some of the wage penalty for motherhood.


Another possible explanation of the mother- hood penalty is employer discrimination- treating women differently because of their motherhood status (e.g., placing mothers in less rewarding jobs, promoting them less, or paying them less within jobs). Such dis- crimination is distinct from sex discrimina- tion that is based on the probabilistic assumption that most women are or will be- come mothers. Sex discrimination creates a sex gap in pay, but not a gap between moth- ers and other women.

Economists distinguish between discrimi- nation based on "taste" and on statistical dis- crimination. In the taste model, an employer makes no assumption about mothers' lesser productivity but simply finds it distasteful to employ them. Sometimes it is co-workers or customers who have this taste, and employ- ers find it expensive to offend them. If such differential treatment of mothers exists, it should show up in our models as a residual effect of motherhood after human capital and the mother-friendliness of jobs have been controlled. (Of course prior discrimination could affect the accumulation of ex- perience, encouraging labor force withdraw- als.) Or if some of the motherhood penalty is reduced by controlling for job character- istics that determine reward level, discrimi- nation could explain why mothers were rel- egated to lower paying jobs; in this case dis- crimination could explain more than just the residual penalty after controlling for job characteristics.

A second discrimination model is statisti- cal discrimination. Suppose that, net of types

of human capital that employers can screen cheaply, such as education and experience, mothers are, on average, less productive. The statistical discrimination model is part of economists' consideration of costs of in- formation. The idea is that it is expensive to measure individual productivity before hir- ing, so employers use averages based on in- formal or formal data gathering to predict how individuals will perform. On this basis, they might treat women with (more) children less favorably. In economists' thinking, em- ployers would create the degree of pay gap between mothers and non-mothers (or any other two groups to whom statistical dis- crimination applies) that is commensurate with their estimated productivity gap. In most statistical discrimination models of- fered by economists, the group that is dis- criminated against is paid, on average, ap- proximately commensurate with the groups' average productivity; in taste discrimination the group's average pay is less than that based on their average productivity.' Of course, in such a scheme, individual moth- ers who are more productive than the aver- age mother are being paid less than commensurate with their productivity. How would this show up in our regression mod- els? If we had accurate measures of indi- vidual productivity (measures assumed to be expensive for employers to acquire before hiring), productivity was controlled, and sta- tistical discrimination was the only source of the motherhood penalty, we would find a co- efficient of 0 for the presence of children. However, if productivity is unmeasured, then the regression coefficient for the pres- ence of children would pick up any statisti- cal or taste discrimination. Social psycho- logical research on stereotyping suggests that a more realistic model, similar to the statistical discrimination model, features employers observing real differences, exag- gerating them, and thus producing an aver-

'Theoretical reasoning suggests a strong ten- dency for the group pay gap to equal the produc- tivity gap under statistical discrimination. But the two may not be equal when mismatches between worker and employer are costly (Aigner and Cain 1977), employers are risk averse (Aigner and Cain 1977), or human capital accumulation is en- dogenous to the discrimination (Lundberg and Startz 1983).

age pay gap between groups more than com- mensurate with group differences in produc- tivity. It is also possible to perceive group differences where none exist. These types of discrimination would show up in the residual effect of motherhood on wages.

U.S. federal law prohibits sex and race discrimination in two forms. Differential treatment involves treating women differently than men because of their sex rather than any individual qualification. This stan- dard prohibits both taste and statistical dis- crimination. U.S. law, however, does not ex- plicitly prohibit discrimination based on par- enthood status, but if differential treatment on the basis of parenthood were applied only to women, the courts might well see such treatment as sex discrimination, provided that qualifications and productivity were equivalent between the groups of women.

A second kind of legal claim of sex or race discrimination involves disparate impact. This doctrine states that policies are consid- ered discriminatory and illegal if they use some screening criterion for hiring or pro- motion that screens out more women than men and the screening criterion is not a "business necessity." "Business necessity" is defined loosely to include anything that re- sults in more productive workers or reduces costs. Consider the analogous concept of business policies that have a disparate im- pact on mothers: Policies that require long or inflexible work hours, do not allow sick days to care for children, do not permit per- sonal phone calls from the job, and do not provide for maternity leave will adversely affect mothers. A disparate impact claim of discrimination against mothers parallel to the present legal standard regarding sex and race would prohibit any such policies, unless having such policies saves employers money or increases output.

Acker's (1990) and Williams's (1995) no- tion of gendered organizations can be seen as a kind of disparate impact model. Both researchers argue that many workplace poli- cies are gendered-that they were formed around an idealized image of a male worker who has a wife at home and no family re- sponsibilities other than to contribute money. Few people, male or female, have a full-time homemaker backing them up today, but some careers have requirements that seem most consistent with this gendered im- age. Such policies probably do have a dis- parate impact on mothers, and as such have a disparate impact by sex. But most such policies probably would not be deemed dis- criminatory by the courts because, despite their disparate impact, employers probably get more output from the employee, and thus employers could probably meet their burden of proof under the loosely defined "business necessity" standard.

But we are interested in a broader notion of disparate-impact discrimination than courts would allow. Thus, we ask where the effects of policies that have a disparate im- pact on mothers would show up in our statis- tical models. Such policies should affect the motherhood penalty net of work experience (although this could be an underestimate to the extent that experience is endogenous to such policies, i.e., they force women out of employment upon a birth). Effects of poli- cies that limit mothers to lower paying (more mother-friendly) jobs would be netted out when relevant job characteristics were con- trolled.

The foregoing discussion should serve as a caveat that the interpretation of motherhood effects net of human capital depends on a number of assumptions. Researchers' inabil- ity to directly measure productivity or em- ployer discrimination means that either may show up in our analysis as an unmeasured residual effect of motherhood on wages.

It is possible that there is no causal effect of motherhood on wages, but rather that some of the same individual characteristics that cause lower earnings for mothers also lead to childbearing at higher rates. For example, women with lower academic skills may be more likely to have children early because they know their career prospects are not good and thus think children will yield more satisfaction. Or perhaps women who care less about affluence are more likely to have (more) children and are more apt to trade earnings for other job values. Or perhaps a "present" orientation (e.g., an inability to delay gratification) makes it more likely that women will become pregnant unintention- ally and that they exhibit low self-discipline at work, which leads to lower wages. Each of these hypotheses involves some charac- teristic that is exogenous to both fertility and earnings that affects both, thereby creating a correlation between earnings and the pres- ence of children that is not causal.

Past studies have dealt with this possible heterogeneity through the explicit inclusion of control variables or by using fixed-effects models. All studies include some control variables, but data sets lack measures of many relevant characteristics, such self-dis- cipline or the taste for affluence. We believe that the best way to deal with heterogeneity on unmeasured characteristics is to combine the inclusion of available control variables with person-specific fixed-effects modeling. Three studies have used person fixed-effects models: Korenman and Neumark (1992), Lundberg and Rose (2000), and Waldfogel (1997).2 We use the same approach.

Person fixed-effects models require panel data that measure variables at least two points in time. Although computing algo- rithms vary, the coefficients obtained are those one would get if dummy variables for persons and years were entered into an OLS model run on the pooled sample of person- years. The inclusion of dummy variables for persons controls for unchanging characteris- tics of the person that are unmeasured but have additive effects on earnings. Person fixed-effects models have the limitation that if an unmeasured characteristic affects num- ber of children and interacts with another variable in affecting wages, the models will not eliminate bias. For example, if career ambition lowers fertility and interacts with job experience to create steeper wage trajec- tories rather than having a simple additive wage increment of a certain percentage at each year, the coefficient purporting to rep-

Another approach is sibling fixed-effects models. Such models assess how differences in sisters' wages are related to fertility differences, assuming that the relevant sources of heterogene- ity that bias models seeking to estimate child pen- alties are held constant within pairs of siblings. This is a questionable assumptiin, but, using this method. Neumark and Korenman (19941 found a child penalty of about 7 percent, which fell to 4 to 5 percent when job experience was controlled.

resent the effect of the presence of children on the log of wages would be biased. Coef- ficients for motherhood could also be biased if women decide to become pregnant when they see a period of low wages coming (e.g., if their industry or town is in recession). In this case, the anticipated low wage would be causing the birth rather than the child caus- ing the low wage. Yet fixed-effects models, by removing certain classes of bias arising from omitted variables, are a vast improve- ment on OLS models.


We examine whether the child penalty dif- fers between married and unmarried moth- ers (unmarried mothers are divided into never-married and divorced or separated). Most prior research has simply entered child status and marital status additively. Married mothers might be more able to spend time at home or to choose a more mother-friendly job, given another adult's income for sup- port, leading to a higher child penalty for married women. But, absent a sex-based di- vision of market labor versus household la- bor, we might hypothesize the opposite, that married women have someone to share child rearing duties, thus enabling them to better optimize earnings. Moreover, unmarried women with children may find that they can't earn enough after paying for child-care expenses to do better than welfare, which could lead to a greater experience deficit for mothers relative to non-mothers among the unmarried. And we might find no difference in the wage penalty by motherhood status if employer discrimination were the mecha- nism, unless employers single out married or unmarried mothers for discrimination. We also examine interactions of number of chil- dren with race, and whether any race differ- ences in child penalties result from racial differences in marital status. Two studies have found smaller child penalties for black women compared with white women (Hill

1979; Waldfogel 1997).

We pooled the 1982-1993 waves of the Na- tional Longitudinal Survey of Youth

I (NLSY), a national probability sample of in- dividuals aged 14 to 21 in 1979; blacks and 1 Latinos are oversampled.' NLSY respon- dents are interviewed annually. We limit our sample to women employed part-time or full-time during at least two of the years from 1982 to 1993, since fixed-effects mod- els require at least two observations on each person. Out of the total of 6,283 women in the 1979 NLSY, we had at least two years of employment for 5,287 women. After dele- tions of person-years with missing values on one or more variables, our analyses are based on 41,842 person-years as units of analysis, which is an average of 7.9 years (waves) of data for each of the 5,287 women. Only 6 percent of person-years were lost because of missing values and women with less than two years of employment.

From the 1990 census, we calculated the percent female in each detailed occupation1 industry cell (U.S. Bureau of the Census 1993). NLSY data are coded into 1980 oc- cupation and industry codes starting in 1982, but these codes were easily mapped onto 1990 occupation and industry codes. Be- cause pre-1982 occupations and industries were coded using 1970 census codes, which do not easily map onto 1990 census codes, we limited our sample to the 1982 through 1993 years.

The Dictionary of Occupational Titles

(U.S. Department of Labor 1977) contains data on approximately 12,000 occupations. Department of Labor observers coded occu- pations according to their demands. DOT variables were transformed into averages for each 1980 detailed census occupation (En- gland 1992, chap. 3).

Measures of effort (occupational average) were provided from the 1977 Quality of Em- ployment Survey (Quinn and Staines 1979).4 These averages were merged with our data according to 1980 census occupation codes.

Waldfogel (1997) used 1968-1988 waves of the NLS-Young Women. Neumark and Koren- man (1994) also used NLS-YW, 1973-1982; the 1980 and 1982 waves were used by Korenman and Neumark (1992). Lundberg and Rose (2000) used the Panel Study of Income Dynamics (PSID) 1980-1992. Hill (1979) used the 1976 wave of the PSID.

The authors thank Randall Filer for these data.

The dependent variable is the natural log of hourly wage in the respondent's current job. We omitted person-years whose hourly wages appeared to be outliers (i.e., below $1 or above $200 per hour). The principal inde- pendent variable is the total number of chil- dren that a respondent reported by the inter- view date of each year (1982 through 1993). In alternate specifications, we measure chil- dren with dummy variables for one child, two children, and three or more children (with "no children" as the reference category). Dummy variables for marital status include married and "divorced." (The divorced cat- egory actually includes divorced, separated, and widowed respondents, although in this young sample there were few widows.) "Never-married" is the reference category.

Measures of human capital include educa- tion, years of full-time and part-time work experience, and years of full-time and part- time seniority (i.e., experience in the organi- zation for which one currently works). These measures cover the entire life cycle back to 1978. Experience includes seniority in one's present workplace. Finally, the total number of breaks in employment is included. A break is defined as time out of employment lasting longer than 6 weeks since one's first full-time job of at least 6 weeks duration. Models controlling for human capital vari- ables also include a measure of whether the respondent is currently enrolled in school, since this is likely to affect employment and type of job.

We include a large number of job charac- teristics. A dummy variable is included for whether the respondent's current job is part- time, defined as less than 35 hours per week. (In results not shown we substituted hours worked per week and its square for the dummy variable for part-time work, and the coefficients for presence of children were virtually unchanged.) Union status is a dummy variable coded 1 if the respondent reported that wages in her job were set by collective bargaining. A dummy variable is coded 1 for work in the public sector (local, state, or federal government). Another dummy variable is coded 1 if the respond- ent's job is one of the two census occupa- tional titles for child care (child-care worker, private household; other child-care work- ers). Authority is a dummy variable coded 1 for census detailed occupational categories with titles containing the words "manage- ment," "supervisor," or "foreman" (England 1992: 137-39).

We measure the cognitive skill demanded by an occupation with a scale created by En- gland (1992:134-35). The scale was created from a factor analysis of numerous items, most taken from the Dictionary of Occupa- tional Titles. The scale score was merged with NLSY respondents' records according to their detailed (1990) census occupational category. Measures of specific vocational preparation, the physical strength demanded by the job, and the physical hazards associ- ated with one's occupation are averages of variables taken from-the Dictionary of Oc- cupational Titles and are merged with these data according to NLSY respondents' detailed occupation.

Finally, several measures, created from the Quality of Employment Survey (QES) are included as continuous variables. Two were measures used by Bielby and Bielby (1988) measuring how much "effort they put into their jobs beyond what is required" and "how much effort, either physical or mental" their jobs require. A third measure is the ratio of the amount of effort their job requires to the amount of effort respondents said it takes to watch television. We also in- clude two more indirect measures: percent- age of time spent not actually working while at work (e.g., waiting), and the per- centage of time spent goofing off while at work.

The percentage of females in respondent's job in 1990 is calculated from 1990 census data. It is the percent female among persons employed in a cell of a matrix cross-classi- fying detailed 1990 three-digit occupational category with detailed three-digit industry category.

A measure of the number of employees in the respondent's current work location is in- cluded-to model the effects of firm size on the motherhood penalty. The NLSY began collecting data for this measure in 1986. Thus, this measure is not included in our main models but is used in a supplementary analysis on those years for which we had the measure-1986 through 1993.

A measure of time (in minutes) spent com- muting to one's current place of employ- ment, a variable available only in the NLSY 1988 and 1993 surveys, is not included in our main models. But we do include it in a supplementary analysis using only 1988 and 1993 data.

We use fixed-effects regression models to analyze NLSY data arranged in a pooled time-series cross-section with person-years as units of analysis. Effects are fixed for years and persons. Person fixed-effects mod- els eliminate bias created by the failure to include controls for unmeasured personal characteristics that have additive effects. Thus, fixed-effects models control for ef- fects of unchanging aspects of cognitive ap- titude, preferences resulting from early so- cialization, life cycle plans, tastes for afflu- ence, future orientation, and other unmea- sured human capital. The model is:


Regression coefficients are denoted by b, k indexes measured independent variables (X's), i indexes individuals, t indexes time periods, r is the error term, u is the cross- sectional (individual) component of error, v is the timewise component of error, w is the purely random component of error, and bo is the intercept. The dependent variable, Y, is the natural logarithm of hourly earnings.

For all models, the Hausman test was con- ducted to assess whether random-effects models were adequate. In each case, the test indicated a need for fixed-effects models. We also present results from ordinary-least- squares (OLS) regression models for com- parison. Because OLS models presumably contain greater omitted-variable bias, the comparison provides some insight into whether those who have (more) children have lower earning-potential based on their unobserved characteristics. Because the mul- tiple observations on each individual are not independent, we use the Huber-White method to correct the standard errors in the OLS models (although this correction never substantially changed standard errors). We place more confidence in the fixed-effects models for causal inference.

We do not add Heckman-type selectivity corrections to our models. However, if women for whom the motherhood penalty would be the worst are the most likely to re- main out of the labor force, our models will underestimate the motherhood penalty.


Table 1 presents means for variables used in the analysis by marital status and mother- hood status. Models intended to capture causal effects begin in Table 2. We refer to models with fixed-effects, except where oth- erwise noted. Table 2 presents only those coefficients indicating the effect of the total number of children a woman has on the natu- ral log of hourly wage. (The complete regres- sion results for the model including all vari- ables are presented in Appendix Table A.)

The models capturing the "gross" effect of motherhood include no controls other than person-specific and year-specific fixed-ef- fects. They indicate that the wage penalty for each child is 7 percent. The OLS models show only a slightly higher gross child pen- alty, 8 percent. This suggests only slight negative selectivity into having (more) chil- dren on unmeasured pay-relevant character- istic~.~

Adding marital status to the model in- creases the estimated motherhood penalty slightly (by -.005 in fixed-effects models). Inspection of the full regression results (Ap- pendix Table A) shows that marriage actu- ally increases women's earnings, so mar- riage is a suppressor. These are average ef- fects of marriage across child statuses; we will see below that presence of children and marriage interact to affect wages.

If we compute gross OLS models on a cross- section of the most recent year, 1993, the gross child penalty is 11 percent, which is larger than the penalty in OLS pooled data. This suggests that selectivity into motherhood creates a worse bias for cross-sectional than for pooled OLS models.

Table 1. Means and Standard Deviations (in Parentheses) for Variables Used in the Analysis, by Marital Status and Motherhood Status: NLSY, 1982 to 1993

Variable     Childless     Mother
Hourly wage Ln hourly wage Human Capital Variables Education (in years)     5.99 (1.7 1) 1.73 (54)     5.29 (1.63) 1.67 (.49)
Enrolled in school         
Number of breaks in employment Full-time seniority (in years) Part-time seniority (in years) Full-time experience (in years) Part-time experience (in years)         
Family Characteristics Number of children         
One child         
Two children         
Three or more children         

Black (non-Hispanic)

Table 2 shows that reduced experience is clearly part of the explanation of the moth- erhood penalty. Controlling for the human capital variables shown in Table 1, reduces the child penalty by 36 percent, from about 7 percent to 5 per~ent.~

OLS models show an even larger reduction in the child penalty when the human capital vari- ables are added (from 8 percent to 2 percent com- pared with from 7 percent to 5 percent in fixed- effects models). The larger drop in OLS suggests that some of the unobserved human capital dif-
Married     Divorceda
Childless     Mother     Childless     Mother
6.48 (1.74) 1.87 (.55)     6.35 (1.77) 1.85 (57)     6.42 (1.77) 1.86 (57)     6.08 (1.66) 1.81 (31)

(Table 1 continued on next page)

We next explore whether something about the jobs held by mothers explains their lower wages. Mothers may trade wages for "mother-friendly7' jobs, or lowered produc- tivity could cause women to choose less de-

ference between mothers and non-mothers is ex- ogenous to both motherhood and measured hu- man capital, and affects each. Because this com- ponent is netted out of both the gross and human capital models under fixed-effects, the mother- hood coefficients differ less. It is unclear why in the models that control human capital variables,

(Table I continued from previous page)

Never-Married Married Divorced3 Variable Childless Mother Childless Mother Childless Mother

Job Chumcteristics

Part-time job

Work effort ratio (QES)

Work effort required (QES Extra work effort (QEs) Percent of time waiting on job (QES)

Percent of time goofing off on job (QES)

Hazardous conditions (DOT) Strength requirement (DOT) Specific vocational training (DOT)

Cognitive skill (DOT)

Authority (DOT)

Percent female in occupation/industry

Government job

Union member

Child-care job


Number of person-years

a Category includes separated, divorced, and widowed. Variables are coded such that low scores indicate more average effort reported.

manding jobs. Or job characteristics could less well. Table 2 shows that support for explain the motherhood penalty if employ- these ideas is weak. Including all the job ers discriminated against mothers, exclud- characteristics lowers the (marital status- ing them from high-paying jobs with de- and human capital-adjusted) penalty for mands they believed mothers would fulfill each child from -.047 to -.037. Although this is a 21-percent reduction, a decline in the a child penalty to wages from about 5

child coefficients are larger in the fixed-effects model than in the OLS model, whereas in the percent to about 4 percent seems small. The gross models the child penalty is larger in the reduction in the OLS model is even smaller. OLS model. Moreover, half of the reduction in the fixed-

Table 2. Unstandardized Coefficients for the Effect of Total Number of Children (Continuous Variable) on Women's Hourly Wage (In), from Fixed-Effects Models and OLS Models: NLSY, 1982
to 1993
Control Variables     Fixed-Effects     OLS
in Model     Model     Model
Gross (no controls)     -.068"*     
Marital status     -.073"*     


Marital status and human -.047** capital variables a (.004)

Marital status, human -.037** capital variables, and (.004) job characteristics

Notes: OLS models include age and year, each in linear, squared, and cubed form. Numbers in pa- rentheses are standard errors. Standard errors in OLS models were corrected using the Huber-White method.

a Measures of human capital include education, full-time seniority, part-time seniority, full-time ex- perience, part-time experience, number of breaks in employment, and whether currently enrolled in school.

Job characteristics include the QES and DOT measures listed in Table 1, whether the current job is part-time, percent female of the respondents' oc- cupation by industry category, dummies for whether the job is in government, unionized, in a child care occupation, or self-employment, and in- dustry dummies.

*p< .05 **p < .O1 (two-tailed tests)

effects model is achieved by simply includ- ing a single job characteristic: whether the woman is working part-time. Working part- time reduces hourly pay, either directly or through forcing women into less desirable jobs that offer part-time hours.

No other job characteristic, when added alone to the human capital model, changes the child penalty to any nontrivial extent (results not shown). Mothers are less likely to be in jobs involving authority and more likely to work in jobs involving child care (Table 1). But neither of these variables, when added to the model, reduces the child penalty by even one percentage point. Con- trolling for the sex composition of the woman's job had no effect on the child pen- alty (results not shown). Although "female" jobs pay less (Appendix Table B), mothers are no more likely than non-mothers to be in them (Table 1). In fact, the zero-order correlation between number of children and the percent female of one's job is slightly negative (results not shown). Thus, there is no evidence that women select female jobs because they are more mother-friendly. Oc- cupational sex segregation and the wage penalty for working in a female job appear orthogonal to having children and the wage penalty for children.

We also experimented with adding groups of job variables, but no group of related vari- ables had a nontrivial effect on the child penalty. The five QES measures of effort re- quired by the occupation do not change the penalty by even one percentage point. Inter- estingly, Table 1 shows similar means for mothers and non-mothers on these variables, and Appendix Table B shows that not all the effort measures have the predicted effects on earnings. Similarly, if all the dummy vari- ables for industry are added to the human capital model, the child penalty is reduced by less than one percentage point.

Two supplementary analyses added job characteristics that were available only for certain years of the NLSY panel (results not shown). Limiting the analysis to those years for which we had data on firm size (1986- 1993) revealed no change in the size of the child penalty with the inclusion of firm size. A second supplementary analysis assessed whether the child penalty arises be- cause mothers sacrifice higher pay for a shorter commute. Because commuting time was measured only in 1989 and 1993, we ran fixed-effects models including those women employed in both 1989 and 1993. Adding commuting time to the human capi- tal model had no effect on the estimated motherhood penalty.

Given that job characteristics do not sub- stantially mediate the effect of motherhood on wages, we need not worry about whether any such indirect effect comes from volun- tary selection of mother-friendly jobs, employer discrimination relegating mothers to worse jobs, or some other process. Mother- hood does not seem to have its effects through the kinds of jobs women hold, with the important exception of working part- time.

Table 3. Unstandardized Coefficients for the Effect of Number of Children (Dummy Variables) on Women's Hourly Wage (In), from Fixed-Effects Models and OLS Models: NLSY, 1982 to


Control Variables in Model

Gross (no controls)

Marital status

Marital status and human capital variables

Marital status, human capital variables, and job characteristics

Fixed-Effects Models OLS Models

One Two Three or More One Two Three or More Child Children Children Child Children Children

-.020* -.125" -.217*' (.008) (.011) (.015)

-.038 " -. 142*' -.232** (.008) (.011) (.015)

-.045" -. 1 12*' -.151** (.008) (.010) (.015)

-.032'* -.089*' -.121'* (.008) (.010) (.014)

Note: Numbers in parentheses are standard errors. Standard errors in OLS models were corrected using the Huber-White method. For descriptions of models and variables, see notes to Table 2.

'p < .05 < .O1 (two-tailed tests)


Table 3 presents a check on whether mea- suring "motherhood" with a continuous variable counting total number of children obscured nonlinear or nonmonotonic rela- tionships. We measured the presence of children with three dummy variables (one child, two children, and three or more chil- dren), each relative to a reference category of no children. Table 3 shows that the gross penalty is 2 percent for one child, 13 per- cent for two children (i.e., an additional 11 percent for the second child), and 22 percent for three or more children. Controlling for marital status and all of the human capi- tal variables, the penalties are 5 percent, 11 percent (an additional 6 percent for the sec- ond child), and 15 percent. As with the models entering number of children as a continuous variable, the addition of job variables reduces the penalty little. The penalty for having one child is small and none of it is explained by lost experience (the penalty goes up slightly in the model including the human capital variables). Having a second child has a much larger in- cremental effect than does having the first child. Women may be more likely to take a break from employment when there are two children at home because the difference be- tween their earnings and the cost of care for two children makes employment no longer compelling. But, this is not the whole story, because most of the incremental loss in wage after the second child is present in the human capital model, which controls for experience. Given that effects are at least monotonic, if not perfectly linear, our judg- ment is that the imprecision introduced by measuring number of children as a continu- ous variable in our analyses is worth the gain in simplicity.


Next we consider interactions to investigate what characteristics of women or their jobs increase the size of motherhood penalties. Table 4 shows results from interacting dummy variables for marital status with number of children. The left column of Table 4 presents coefficients for total number of children, which, in this model including in- teractions, tell us the effect of each child on wages for never-married women.7 The col- umns to the right present effects for married

Coefficients for "additive" terms in models including an interaction involving that variable give the effect of that variable when all other variables with which it has been interacted equal

0. When both marital status dummy variables equal 0, this indicates never-married status since "never-married" is the reference category.

Table 4. Effect of Number of Children (Continuous Variable) on Women's Hourly Wage (In) from Fixed-Effects Models and OLS Models, by Marital Status

Control Variables in Model Gross (no controls)

Human capital variables

Human capital variables and job characteristics
Fixed-Effects Models         OLS Models
Never-    Never-    
Married     Married     Divorceda     Married     Married     Divorced
-.035'*     -.079*'     -.079'*     -.lo1 "     -.073*     -.08 1 '*
(.007)     (.006)     (.007)     (.004)     (.005)     (.006)
-.026*'     -.051**     -.046"     -.025*'     -.015'*     -.025"
(.007)     (.006)     (.007)     (.004)     (.005)     (.006)
-.019*'     -.O4Ow*     -.038*'     -.014'*     -.014'*     -.014"
(.007)     (.006)     (.007)     (.004)     (.004)     (.005)

Note: Numbers in parentheses are standard errors. Standard errors in OLS models were corrected using the Huber-White method. Effects are calculated from unstandardized coefficients in models containing in- teractions between marital status and number of children (using a continuous measure).

For descriptions of models and variables, see notes to Table 2.

Includes separated, divorced, or widowed. *pi.05 "p < .O1 (two-tailed tests)

and divorced or separated women obtained by adding the coefficient for number of chil- dren and the coefficient for the relevant in- teraction. The fixed-effects models show that women who have never been married experience lower child penalties than do married or divorced women, both before and after adding controls for human capital vari- ables and job characteristic^.^ This result holds if we combine married and previously married women into one category (results not shown). In the OLS models, never-mar- ried women show child penalties as high as or higher than those in the fixed-effects models. We are more confident in the fixed- effects models for drawing conclusions con- cerning causation, particularly because in re- cent cohorts, women with more earning power are also more likely to marry. Thus, fixed-effects modeling is needed to net out the selectivity into marriage.

The fact that marriage increases the child penalty suggests that at least some part of the penalty arises because the ratio of time and energy mothers allocate to children ver- sus jobs is affected by whether they have a

Appendix Table B shows results from mod- els that interact dummy variables for number of children with marital status. It shows greater pen- alties for married and divorced women as in Table 4, but these are largely limited to second and higher order births.

source of financial support other than their own earnings. Without assuming a sex-based division of labor, the direction we would predict for this interaction would not be clear. Husbands could, in principle, provide money that allows married mothers to focus more on their children than single women can; or they could simply be a second per- son to share child-care responsibilities, al- lowing married mothers to focus more on their jobs than single mothers. The higher child penalty for married mothers suggests that the first scenario is more common.

The higher penalty for married mothers also suggests that child penalties are not entirely a matter of discrimination against mothers, unless we believe that employers discriminate more against married mothers than against single mothers.

It is puzzling that married and divorced women have similarly high child penalties. After all, divorced women do not have hus- bands to provide financial support and they usually get relatively little child support. The similarity implies that the larger penalties experienced by married women are long- lasting, enduring even if the marriage ends. Perhaps the penalties operate through missed promotions, or cumulative impacts of im- pressions made, or small raises earned early in one's employment history.

The fact that marriage increases the pen- alties for children does not mean there is a marriage penalty. In fact, on average there is a marriage premium: Marriage has positive effects in all models that do not interact mar- riage with children (see Appendix Table A). The interaction between marital status and the presence of children implies not only that the child penalty varies by marital sta- tus but also that the effect of marriage varies by child status. Calculations from Table 4's (gross or human capital) models with inter- actions show that marriage has a wage pre- mium for women with one child or no chil- dren, and no effect for women with two chil- dren. But for women with more than two children, marriage has a net wage penalty. Thus, marriage increases the child penalty, while children reduce the marriage premium, turning it into a penalty for mothers with more than two children.

To test whether more skilled women ex- perience higher penalties, we interacted hu- man capital with number of children (results not shown). There was no interaction be- tween years of education and number of children. We found that women with more full-time experience suffer larger child pen- alties, but the opposite was true for full- or part-time seniority. Thus, there is no clear evidence that more skilled or committed women experience higher penalties.

Do women with higher level jobs incur a larger motherhood penalty? This might be true if such jobs are organized on a "male" model that penalizes any behavior that ap- pears to be less than a full commitment, whether or not the behavior affects produc- tivity. To test this, we interacted number of children with job characteristics (models also included human capital measures). The child penalty is higher for women in full- time jobs than for those in part-time jobs. The ~enaltv is slightly lower for women in


more heavily male jobs. Penalties were no higher in jobs requiring more on-the-job or vocational/professional training or more cognitive skill (they were trivially but sig- nificantly lower). Finally, we created a vari- able intended to capture high-level male jobs. We coded this dummy variable 1 if the job was classified as professional or man- agement in the census's broad occupational categories and the job's percent female (of the occupation-by-industry category) was no more than 35 percent. We interacted this dummy variable with number of children in a model that also controlled for marital sta- tus and the human capital variables. Women in these heavily male professional and mana- gerial jobs actually had smaller (1 to 2 percentage point) child penalties. Thus, it ap- pears that high-level, "male" jobs penalize women a bit less for having children.

Finally, we considered whether child pen- alties differ by race. Limiting this analysis to Latinas, non-Hispanic blacks, and non- Hispanic whites, we interacted race with number of children (results not shown). For the gross model, the penalties for number of children did not differ by race. After adjust- ing for the human capital variables, black/ white penalties still did not differ, but Latinas had smaller penalties. When we used dummy variables for number of children, al- lowing nonlinear effects, it was only for mothers with three and more children that we found smaller penalties for blacks and Latinas (whether or not the human capital variables were controlled). Of course, most women have fewer than three children. There were no three-way interactions be- tween marital status, children, and race in any model, implying that the lower penalties that women of color experience for third or higher parity births are not explained by the fact that more of their births occur outside marriage. Waldfogel (1997) and Neumark and Korenman (1994), using models that control for human capital variables, also re- port a smaller penalty for black women com- pared with white women in an earlier NLS data set. Our findings show that this differ- ence exists only for women with more than two children.


We find a wage penalty for motherhood of approximately 7 percent per child among young American women. Roughly one-third of the penalty is explained by years of past job experience and seniority, including whether past work was part-time. That is, for some women, motherhood leads to employ- ment breaks, part-time employment, and the accumulation of fewer years of experience and seniority, all of which diminish future earnings. However, it is striking that about two-thirds of the child penalty remains after controlling for elaborate measures of work experience.

We added numerous job characteristics to models to assess whether mothers earn less because their jobs are less demanding or be- cause they offer mother-friendly character- istics. These factors had only a small effect in explaining the child penalty, and about half of the effect came from a single job characteristic-whether the current job is part-time. Most job characteristics had no effect on the motherhood penalty-either because the characteristics don't affect pay or because motherhood does not affect whether women hold these jobs.

In what social locations are motherhood penalties the steepest? Black women and Latinas have smaller penalties, but only for the third and subsequent births. Never-mar- ried women have lower child penalties than married or divorced women. Second chil- dren reduce wages more than a first child, especially for married women. There is no evidence that penalties are proportionately greater for women in more demanding or high-level jobs, or "male" jobs, or for more educated women, although the penalties are higher for women who work full-time and already have more work experience.

Our use of fixed-effects modeling gives us some confidence that the effects of mother- hood identified here are causal rather than spurious. Further, our detailed measures of work experience assure us that no more than one-third of the motherhood penalty arises because motherhood interrupts women's em- ployment, leading to breaks, more part-time work, and fewer years of experience and se- niority. Finally, we find that little of the child penalty is explained by mothers' placement in jobs with characteristics associated with low pay. However, we did not have direct measures of many job characteristics that would make jobs easier to combine with parenting. Thus, we may have underesti- mated the importance of this particular fac- tor. For future research to be able to answer this question and generalize to the nation as a whole, we need the inclusion of questions about job characteristics that accommodate parenting on national surveys using prob- ability sampling, preferably panels.

What explains the approximately two-thirds of the 7-percent-per-child penalty not

explained by the reductions motherhood makes in women's job experience, if little of it is from working in less demanding or mother-friendly jobs? The remaining moth- erhood penalty of about 4 percent per child may arise from effects of motherhood on productivity andlor from employer discrimi- nation. A weaknesses of social science re- search is that direct measures of either pro- ductivity or discrimination are rarely avail- able. Thus, new approaches to measuring productivity or discrimination would be a welcome contribution. In the meantime, our analyses provide indirect evidence that at least part of the child penalty may result from mothers being less productive in a given hour of paid work because they are more exhausted or distracted. Net of human capital variables, women earn less with each subsequent child, and children reduce women's pay more if the mothers are mar- ried or divorced than if they are never-mar- ried. Employers may discriminate against all women by treating them all like mothers, or they may discriminate against all mothers relative to other women. But is it plausible that employers discriminate by number of children, and discriminate more against mar- ried mothers than single mothers (but give a premium for marriage when women have no child or one child)? This seems far fetched. This does not mean that none of the child penalty is discriminatory. It may be that a base amount is discriminatory, and that the portion that is related to productivity is the portion that varies by number of children and marital status, because those factors af- fect decisions about how time and energy is allocated between child rearing and jobs.

How should public policy respond to wage penalties for motherhood? Because distin- guishing between discriminatory and non- discriminatory differences by race and sex is institutionalized in our legal system, it is tempting to conclude that a motherhood pen- alty is not of public concern unless it results from employers' discrimination. We don't know how much of the penalty arises from discrimination in the form of "differential treatment" of equivalently qualified and pro- ductive mothers and non-mothers. Nor do we know how many policies that have a dis- parate impact on mothers would fail the le- gal standard of being a "business necessity."

But we think there is a serious equity prob- lem, even ifthe penalty were found to be en- tirely explained by mothers having less work experience, lower productivity, and choosing mother-friendly jobs, and even if employers' policies had the intent and effect only of maximizing output relative to costs. In short, we think there is a serious equity problem when we all free ride on the benefits of mothers' labor, while mothers bear much of the costs of rearing children. At this point we depart from the narrow scientific analysis, and articulate our findings with a norma- tively based notion of equity.

Reducing the extent to which mothers bear the costs of rearing children is a worthy goal, in our view. Broadening the concept of dis- crimination to include anything about how jobs are structured or what is rewarded that has a disparate impact on mothers, and mak- ing employers change such policies, would be one way to approach this. But should em- ployers have to get rid of any policy that pe- nalizes mothers? We suspect that this would reduce the net output of organizations be- cause policies that reward experienced workers and workers who can work long hours when needed by the employer would need to be changed. Of course, the net effect on output is an empirical question; in some cases the productivity gains resulting from increased morale and continuity of mothers' employment would offset costs.

But if there are costs to employers of re- structuring work to eliminate the mother- hood penalty, deciding who should pay them is part of the larger question of who should bear the costs of raising the next generation. A general equity principle is that those who receive benefits should share in the costs. As Marxist feminists pointed out in the 1970s, capitalist employers benefit from the unpaid work of mothers, who raise the next genera- tion of workers. But employers are not the only ones who benefit when children are well reared-we all free ride on mothers' la- bor. Thus, mandating that employers share in these costs makes sense only as part of a broader redistribution of the costs of child rearing.

Those who rear children deserve public support precisely because the benefits of child rearing diffuse to other members of so- ciety. Indeed, child rearing (whether unpaid or paid), broadly construed, creates more diffuse social benefits do than most kinds of work. In our view, the equitable solution would be to collectivize the costs of child rearing broadly-to be paid not just by em- ployers but by all citizens-because the ben- efits diffuse broadly. While most U.S. mothers today are employed, mothers continue also to bear the lion's share of the costs of rearing children. Yet other industrial democ- racies have collectivized the costs to a much greater extent than has the United States (al- beit often with other, pronatalist, motiva- tions). Costs can be socialized through fam- ily allowances, child care, and medical care that are financed by progressive taxes. Adopting such policies in the United States would not eliminate the fact that motherhood lowers wages, although it might reduce some of the gross effect if the presence of subsi- dized child care increased women's employ- ment. Such policies would put a floor under the poverty of families with mothers, and would redistribute resources toward those who now pay a disproportionate share of the costs of rearing children. In a period when most mothers are employed, when welfare mothers are being required to take jobs, and when the economy is generating budget sur- pluses unthinkable a decade ago, there may be a political opening for creative proposals that would increase equity for mothers while also helping children.

Michelle J. Budig is a Ph.D, candidate in soci- ology at the University of Arizona and is writing her dissertation at the University of Pennsylva- nia. Her general research interests lie in gender stratification and inequality, and focus specifically on gender, occupations, labor markets, family, and nonstandard work arrangements. With Richard Arum and Donald S. Grant, she is co-author of "Labor Market Regulation and the Growth of Self-Employment" (International Jour- nal of Sociology, 2001, vol. 30, no. 4, pp. 3-27). In Fall 2001, she will join the Department of So-

ciology at the University of Massachusetts, Amherst as Assistant Professor.

Paula England is Professor of Sociology and Di- rector of Women's Studies and the Alice Paul Re- search Center at the University of Pennsylvania. Her current research focuses on gender, altru- ism, and self-interest in family dynamics. In 1999 she won the American Sociological Association's Jessie Bernard award for career contributions to the study of gender.

Appendix Table A. Unstandardized Coefficients from the Regression of Women's Hourly Wage (In)
on Selected Independent Variables: NLSY, 1982 to 1993

Independent Variable Intercept

Family Characteristics

Total number of children


Divorced/separated/ widowed

Human Capital

Education (in years)

Enrolled in school

Number of breaks in employment Full-time seniority (in years) Part-time seniority (in years) Full-time experience (in years) Part-time experience (in years)

Job Characteristics

Part-time job Work effort ratio Work efforta Extra work effort" Percent of time

waiting on job Percent of time goofing off on job Hazardous conditions Strength requirement Specific vocational training Cognitive skill Authority

Fixed-Effects OLS Fixed-Effects OLS Coefficient Coefficient Independent Varaible Coefficient Coefficient


Percent female in -.063** -.058** occupation/industry (.012) (.011)

Government job .004 -.029** (.008) (.008)

Unionized job .085** .126** (.006) (.006)

Child-care occupation -.397** -.493** (.014) (.014)

Self-employed -.041** ,011 (.009) (.009)

Industry (Reference Category is Agriculture, Mining, Forestry)

Public administration -. 149** .071** (.020) (.019)

Finance, insurance, and -. 142** ,049"" real estate services (.019) (.018)

Professional services -.178** ,002 (.017) (.017)

Personal services -.357** -. 173** (.019) (.018)

Business and repair -.178** ,037" services (.018) (.018)

Communications -. 1 19** .132** (.027) (.023)

Wholesale trade -. 128** .054* durables (.024) (.024)

Wholesale trade -. 172** ,013 non-durables (.024) (.025)

Retail trade -.273** -. 106** (.017) (.017)

Entertainment and -.215** -.022 recreation services (.022) (.022)

Utilities -.05 1 .162** (.032) (.028)

Transportation -. 114** .198** (.022) (.021)

Construction -.082** .09 1 ** (.025) (.024)

Food, tobacco, textile -. 132** -.030 manufacturing (.018) (.017)

Chemical, petroleum, rubber, and leather -.091** .08 1 ** manufacturing (.023) (.022)

Lumber, furniture, stone, glass -. 130** ,018 manufacturing (.027) (.026)

Metal industries -.081** .090** manufacturing (.028) (.027)

(Appendix Table A continued on next page) (Appendix Table A continued from previous page)
Independent Variable     Fixed-Effects OLS Coefficient Coefficient     1
    Independent Varaible    Fixed-Effects OLS Coefficient Coefficient
Industry (Continued)                 (Age)'     -    -.0Ola*
Machinery     -. 1OO*^     .126*"                 (.OOO)
manufacturing     (.020)     (.019)         Interview vear     -    .036**
Equipment     -.048     ,179''                 (.007)
manufacturing OLS Control Variables Age     (.027) -    (.026) ,061.. (.006)     1
    (Interview year)' (Interview year)3    --    ,012'' (.001) -.001"" (.OOO)

Note: Numbers in parentheses are standard errors. Standard errors in OLS models were corrected using the Huber-White method. Age is not included in fixed-effects models, but is implicitly controlled because period is controlled, the person fixed-effects cancel out cohort, and period and cohort together uniquely determine age.

a Variable was coded such that high scores indicate a low average effort reported by those in the occupation. Signs on regression coefficients are reversed so that a positive coefficient indicates a positive effect on earnings. *p < .05 < .O1 (two-tailed tests)

Appendix Table B. Effects of Number of Children (Dummy Variables) on Women's Hourly Wage (In) from Fixed-Effects Models and OLS Models, by Marital Status: NLSY, 1982 to 1993

Fixed-Effects Models OLS Models

Control Variables in One Two Three or More One Two Three or More Model and Marital Status Child Children Children Child Children Children

Gross (No Controls)

Never-married -.023 -.056"
(.014) (.019)

Married -.023 -. 162**
(.014) (.019)

Divorced a -.082'* -. 180**
(.022) (.025)

Human Capital Variables

Human Capital Variables and Job Characteristics

Never-marrled -026' -.039*
(.012) (.017)

Married -.026* -. 102'
(.012) (.018)

Divorceda -.026' -.101**
(.012) (.023)

Note: Numbers in parentheses are standard errors. Standard errors in OLS models were corrected using the Huber-White method. Effect sizes are calculated from unstandardized coefficients in models containing interac- tions between marital status and dummy variables for number of children. For descriptions of models and vari- ables, see note to Table 2.

"ncludes separated, divorced, or wldowed.

"p< .05 '"p < .O1 (two-tailed tests)

Acker, Jean. 1990. "Hierarchies, Jobs, and Bod- ies: A Theory of Gendered Organizations." Gender and Society 4: 139-58.

Aigner, Dennis J. and Glenn G. Cain. 1977. "Sta- tistical Theories of Discrimination in Labor Markets." Industrial and Labor Relations Re- view 30:1975-87.

Becker, Gary S. 1991. A Treatise on the Family. Cambridge, MA: Harvard University Press.

Bielby, Denise and William Bielby. 1988. "She Works Hard for the Money: Household Re- sponsibilities and the Allocation of Work Ef- fort." American Journal of Sociology 93: 1031-


Blumstein, Philip and Pepper Schwartz. 1983. American Couples: Money, Work, Sex. New York: William Morrow.

Coleman, James S. 1993. "The Rational Recon- struction of Society." American Sociological Review 58:l-15.

Colwill, N. 1982. The New Partnership: Women and Men in Organizations. Palo Alto, CA: Mayfield.

Connelly, Rachel. 1992. "Self-Employment and Providing Child Care." Demography 29:17


Corcoran, Mary, Greg Duncan, and Michael Ponza. 1984. "Work Experience, Job Segrega- tion, and Wages." Pp.171-91 in Sex Segrega- tion in the Workplace: Trends, Explanations, Remedies, edited by B. F. Reskin. Washington, DC: National Academy Press.

Davies, Hugh and Heather Joshi. 1995. "Social and Family Security in the Redress of Unequal Opportunities." Pp. 313-34 in The Economics of Equal Opportunities, edited by J. Humphries and J. Rubery. Manchester, En- gland: Manchester Equal Opportunities Com- mission.

Desai, Sonalde and Linda Waite. 1991. "Women's Employment During Pregnancy and After the First Birth: Occupational Character- istics and Work Commitment." American So- ciological Review 56:551-66.

England, Paula. 1992. Comparable Worth: Theo- ries and Evidence. Hawthorne, NY: Aldine de Gruyter.

England, Paula and Nancy Folbre. 1999. "Who Should Pay for the Kids?" The Annals of the American Academy of Political Science 563: 194-207.

England, Paula and Barbara Kilbourne. 1990. "Markets, Marriages, and Other Mates: The Problem of Power." Pp. 163-89 in Beyond the Marketplace: Rethinking Economy and Soci- ety, edited by R. Friedland and A. F. Robertson. Hawthorne, NY: Aldine de Gruyter.

Felmlee, Diane. 1995. "Causes and Conse- quences of Women's Employment Discontinn- ity, 1967-1973." Work and Occupations 22: 167-87.

Folbre, Nancy. 1994a. "Children as Public Goods." American Ecor~omic Review 84:86-90.

. 1994b. Who Pays for the Kids? Gender and the Structures of Constraint. New York: Routledge.

Glass, Jennifer. 1990. "The Impact of Occupa- tional Segregation on Working Conditions." Social Forces 68:779-96.

Glass, Jennifer and Valerie Camarigg. 1992. "Gender, Parenthood, and Job-Family Compat- ibility." American Journal of Sociology 98: 131-51.

Glass, Jennifer and Sarah Beth Estes. 1997. "The Family Responsive Workplace." Annual Re- view of Sociology 23:289-314.

Glass, Jennifer and Tetsnshi Fnjimoto. 1995. "Employer Characteristics and the Provision of Family Responsive Policies." Work and Occu- pations 22: 380-41 1.

Glass, Jennifer and Lisa Riley. 1998. "Family Responsive Policies and Employee Retention Following Childbirth." Social Forces 76: 1401-35.

Harkness, Susan and Jane Waldfogel. 1999. "The Family Gap in Pay: Evidence from Seven In- dustrialized Countries." CASE paper 30, Cen- tre for Analysis of Social Exclusion, London School of Economics, London, England. Un- published manuscript.

Hill, Martha. 1979. "The Wage Effects of Mari- tal Status and Children." Journal of Human Resources 14579-94.

Hochschild, Arlie. 1989. The Second Shift. New York: Avon Books.

Joshi, Heather. 1990. "The Cash Opportunity Cost of Childbearing: An Approach to Estima- tion Using British Evidence." Population Stud- ies 44:41-60.

Joshi, Heather and Marie-Louise Newell. 1989. Pay Differentials and Parenthood: Analysis of Men and Women Born in 1946. Coventry, En- gland: University of Warwick Institute for Employment Research.

Kilbourne, Barbara Stanek, Paula England, George Farkas, Kurt Beron, and Dorothea Weir. 1994. "Returns to Skill, Compensating Differentials, and Gender Bias: Effects of Oc- cupational Characteristics on the Wages of White Women and Men." American Journal of Sociology 100:689-7 19.

Klerman, Jacob Alex and Arleen Leibowitz. 1999. "Job Continuity among New Mothers." Demography 36: 145-55.

Korenman, Sanders and David Nenmark. 1992. "Marriage, Motherhood, and Wages." Journal of Human Resources 27:233-55.

Loh, Eng Seng. 1996. "Productivity Differences and the Marriage Wage Premium for White Males." Journal of Human Resources 31:566


Lundberg, Shelly and Elaina Rose. 2000. "Par- enthood and the Earnings of Married Men and Women." Labour Economics 7:689-710.

Lundberg, Shelly and Richard Startz. 1983. "Pri- vate Discrimination and Social Intervention in Competitive Labor Markets." American Eco- nomic Review 73:340-47.

McLanahan, Sara and Erin L. Kelly. 1999. "The Feminization of Poverty: Past and Future." Pp. 127-46 in Handbook of the Sociology of Gen- der, edited by J. Chafetz. New York: Plenum.

Neumark, David and Sanders Korenman. 1994. "Sources of Bias in Women's Wage Equations: Results Using Sibling Data." Journal of Hu- man Resources 29:379-405.

Okamoto, Dina and Paula England. 1999. "Is There a Supply Side to Occupational Sex Seg- regation?" Sociological Perspectives 42:557


Presser, Harriet. 1994. "Employment Schedules, Gender, and Household Labor." American So- ciological Review 59:348-64.

Quinn, Robert P., and Graham L. Staines. 1979.

Quality of Employment Survey, 1977: Cross Section. Ann Arbor, MI: Inter-university Con-


sortium for Political and Social Research.

Risman, Barbara and Myra Marx Ferree. 1995. "Making Gender Visible." Comment on Coleman, ASR, February 1993. American So- ciological Review 60:775-82.

    Bureau of the Census. 1993. "1990 Census of the Population Supplementary Reports: De- tailed Occupation and Other Characteristics from the EEO File for the United States." Washington, DC: Government Printing Office.
    S. Department of Labor. 1977. Dictionary of Occupational Titles. 4th ed. Washington, DC: Government Printing Office.

Waldfogel, Jane. 1997. "The Effects of Children on Women's Wages." American Sociological Review 62:209-17.

. 1998a. "The Family Gap for Young Women in the United States and Britain: Can Maternity Leave Make a Difference?" Journal of Labor Economics 16:505-45.

. 1998b. "Understanding the 'Family Gap' in Pay for Women with Children." Journal of Economic Perspectives 12: 137-56.

Williams, Christine. 1995. Still a Man's World. Berkeley, CA: University of California Press.

  • Recommend Us