The Motherhood Wage Penalty Revisited: Experience, Heterogeneity, Work Effort, and Work-Schedule Flexibility

The motherhood wage penalty revisited:

Experience, heterogeneity, work effort and work-schedule flexibility

Deborah AndersonMelissa Binder

Kate Krause

Department of EconomicsUniversity of New Mexico

Albuquerque, NM 87131-1101(505) 277-5304 (505) 277-9445 (fax)

[email protected] (Anderson)[email protected] (Binder)[email protected] (Krause)

mailto:[email protected]



1

The motherhood wage penalty revisited:


ABSTRACT: A wage penalty is regularly observed for working mothers. In this paper we use the

NLSYW to revisit the motherhood wage penalty. Like other studies, we analyze the extent to

which readily controlled human capital measures and individual heterogeneity reduce the

measured penalty. We go further by using simple sample decomposition techniques to consider

the role of sample composition, which changes as mothers of older children enter the work force.

We then assess whether the work-effort hypothesis can account for the portion of the penalty not

explained by human capital and heterogeneity. The incidence of the penalty among mothers is

not wholly consistent with a work-effort argument for the unexplained wage penalty. Instead, we

offer an explanation based on the high costs of flexible work schedules for medium skill �office-

hour� jobs.

2

The Motherhood Wage Penalty Revisited:


I. Introduction

It is an empirical regularity that mothers earn less than women who have no children. A

variety of factors could explain this wage penalty, including reduced investment in wage-

enhancing human capital, unobserved heterogeneity among women (for example, mothers may

have less dedication or career ambition than non-mothers), and lower effort at work due to

absenteeism, exhaustion or distraction. Many empirical studies have considered the role of

human capital, a few have tackled the mother versus non-mother heterogeneity problem, but

none, to our knowledge, has investigated heterogeneity among working mothers or tested Gary

Becker�s (1985) hypothesis that reduced work effort of mothers contributes to their lower

wages.1

In this paper, we follow a �best practice� methodology derived from previous studies that

include controls for human capital and mother versus non-mother heterogeneity. In addition, we

consider heterogeneity among mothers in the timing of their return to the labor force. We expect

that mothers of infants and toddlers who return to the labor force may be more career-oriented

and thus will earn higher wages than the mothers who return when their children are older. An

estimated wage penalty calculated for a pooled sample of these mothers will mask the true

1 Hersch and Stratton (1997) find that married women�s wages are negatively correlated with the amountof time they spend on housework, but they do not explicitly consider mothers. Bielby and Bielby (1988)consider effort at work by men and women by parental status. They find that mothers of pre-schoolchildren report less effort, controlling education and workforce attachment, but they do not test whether thereduced effort leads to a wage penalty. Moreover, their estimates suggest that mothers of pre-schoolchildren who also bear primary responsibility for the child during working hours (i.e., they, rather than thefather, stay home with a sick child) do not reduce their work effort.

3

penalty, as well as its source.

We also investigate several implications of Becker�s work-effort explanation by

decomposing the sample in several ways. First, since young children require more physical care

(lifting, holding, diapering, chasing, etc.) and are more likely to wake up and scream at night,

mothers of young children should be more exhausted and thus less productive at work than

mothers of older children. We therefore test whether the wage penalty declines as children grow

older. Second, we expect that effort is relatively more important in some jobs than in others. In

particular, since most low-skill jobs held by women require little physical or mental effort, we

test whether the wage penalty rises with skill (i.e., education). Third, some researchers (Hill

1979, Waldfogel 1997) report in passing that black women bear a smaller motherhood penalty.

If this is the case, then the work-effort explanation becomes less plausible because it is difficult

to imagine how race would systematically affect energy availability at work. We therefore test

whether black and white mothers experience different wage penalties.

Using data from the National Longitudinal Survey of Labor Market Experience of Young

Women (NLSYW), we find that about 35 percent of the �raw� wage gap can be attributed to

precisely measured labor market experience, time out of the labor force, occupation and

household resources such as husband�s income and own unearned income. Another 25 percent

derives from unobserved heterogeneity. These adjustments still leave unexplained three and six

percent penalties for one and two or more children, respectively. These estimates lie between

estimates reported in previous literature (see, for example, Korenman and Neumark 1992,

Waldfogel 1997).

The unexplained wage penalty is largest for working mothers who have infants and

4

toddlers at home. However, the penalty appears to persist for older children as well. Moreover,

when we divide the sample by education and timing of mothers� return, we find that it is

primarily medium-skill mothers (those with a high school education but without a college

degree) who postponed their return to the labor force who suffer a significant penalty. For these

mothers, the penalty persists until their children have entered high school. This result casts doubt

on the work-effort explanation for the motherhood wage penalty. Instead, we propose that a

simple model of heterogeneous jobs with different elasticities of substitution among effort, time

and work-schedules provides a better explanation for our empirical results.

The paper proceeds as follows. Section II reviews motherhood wage gap findings from

the literature. Section III reviews Becker�s work-effort model, proposes several testable

implications, and discusses flexibility in working hours as an alternative view of the working

mother problem. Section IV describes the data and our empirical approach. Section V

determines the amount of the total motherhood gap that is explained by differences in human

capital and other controls, heterogeneity between mothers and non-mothers, and heterogeneity

among mothers in timing of return to work. Section VI reports the motherhood wage gap by

mothers� race and education level and concludes that the patterns that emerge are not consistent

with the work-effort hypothesis. Section VII provides an alternative explanation in which

inflexible work-schedules, rather than work-effort, give rise to the unexplained portion of the

motherhood wage gap. Section VIII concludes.

II. Measuring the motherhood wage gap

Cross-sectional analyses have typically identified significant wage penalties associated

5

with having children even when controlling differences in human capital. Waldfogel (1997) uses

the NLSYW, 1968-1988 to measure the penalty, controlling precise measures of human capital

including actual work experience and education. When data for the entire time range are pooled,

she finds that mothers of one child experience a penalty of approximately five percent while

mothers with two or more children experience a penalty of approximately 13 percent. Estimates

for 1988 alone are slightly smaller at approximately four and ten percent, respectively.

Waldfogel (1998) reports similar results for the penalty in cross-sectional analyses using the

NLSYW for 1980, and higher (ten and eleven percent, respectively) penalties in the 1991

National Longitudinal Survey of Youth.2

Hill (1979) finds support for the idea that the wage penalty is explained by reduced labor

force attachment. 3 In her cross-sectional analysis using the Panel Study of Income Dynamics,

she asks whether the presence of children is a good proxy for more direct measures of labor force

attachment such as absenteeism, limited job hours and mobility, and intention to stop working in

the future. For the 1975 - 1976 time period that she studied, she finds that white mothers� penalty

disappears when these actual measures of work attachment are included in regressions. Hill does

not observe a penalty among black mothers.

An obvious drawback to the use of cross-sectional analysis is the possibility that mothers

are different from non-mothers in ways that are not observed in the data. Waldfogel (1997)

addresses this concern by estimating several fixed effects models. She finds that the motherhood

2 Waldfogel (1995) estimates higher penalties (approximately ten percent and twenty percent for one andtwo or more children, respectively) in her cross-sectional analysis of women in Great Britain.3 Using 1984 the Survey of Income and Program Participation, Jacobsen and Levin (1995) estimate theeffect of labor market interruptions on wages. They find that wages of women who leave the labor marketnever fully rebound. They do not directly address the effect of children on women�s wages.

6

wage penalty is nearly the same as that estimated in cross-sectional analyses. 4 Thus Waldfogel

concludes that the motherhood penalty cannot be explained by unobserved heterogeneity.

Korenman and Neumark (1992) reach a different conclusion in their study of married

white women. Applying cross-sectional analysis to NLSYW (1982) data, and controlling

residence (South and Urban) and education, they identify the same motherhood penalty as

Waldfogel. However, when measures of experience and job tenure are included, the penalty for

one child is insignificant and the penalty for two or more children is reduced by more than half,

from 18 to seven percent. In fixed effects estimations using a two-year time span, they find no

significant motherhood penalty. Waldfogel (1997) argues that short differences understate the

penalty because new mothers may not have returned to work yet and because the penalty may be

subject to cumulative effects. She estimates fixed effects models using short (two-year)

differences and longer differences (up to nine years) and finds that the magnitude of the

estimated wage penalty increases with the number of years spanned.

The variable of interest in the studies cited above is the presence of children under the age

of 18. However, there is some evidence that children affect the allocation of household work

differently at different ages. For example, using PSID data (1979 � 1987), Hersch and Stratton

(1994) find that the presence of younger children increases housework time for both spouses.

However, when children older than 12 are present, husbands� share of housework falls. Hill

(1979) noted a concentration of wage effects in families with older children, but did not report

the regressions in which this was found. These findings suggest the possibility that the wage

penalty is not solely a problem of mothers of infants and pre-school children. In addition, the

4Using data from Great Britain, Waldfogel (1995) finds that neither first differencing nor estimating a

7

penalty appears to vary by the race and education level of the mother. Hill (1979) does not

observe a motherhood penalty for black women, and Waldfogel (1997) finds a smaller penalty

for black mothers than she does for white mothers. Waldfogel also mentions that the

motherhood penalty increases with education, although she provides no estimates for models that

include education interactions. Our study fills these gaps in the literature by explicitly testing

how the motherhood wage penalty varies by age of child, mothers� education and race, and

timing of mothers� return to the labor force.

III. Wages and work effort

The crux of Becker�s (1985) work-effort hypothesis is that mothers dissipate their energy

caring for their children and therefore bring less energy to work. Less energy translates into less

work effort and lower productivity, which in turn lead to lower wages. Mothers are also likely to

have higher rates of absenteeism. Finally, mothers tend to choose less energy-demanding

occupations. Despite the widespread availability of occupation information in most data sets

used to study mothers� wages, none of the studies reviewed above controls occupation.

Certainly, controlling occupation should help to account for that part of the motherhood gap that

is attributable to choice of a low-effort occupation.

Becker�s discussion also invites some other testable implications. First, since younger

children require more physical energy, wage penalties should fall as children grow older. Indeed,

Bielby and Bielby (1988) find a significant reduction of reported work effort for mothers of pre-

schoolers, but no significant reduction for mothers of older children. If the work-effort

fixed effects model substantially reduces the penalty estimated using OLS.

8

hypothesis holds, then these effort patterns should affect wages systematically. Second, racial

differences in the size of the penalty would be inconsistent with the productivity story. 5 Finally,

higher skilled jobs typically require more effort, so wage penalties should rise with mothers�

years of schooling.

Discussions of the difficult �balancing act� required of working mothers in both the

academic and popular press reflect the relevance of Becker�s model. The following quote from

Spain and Bianchi (1996) typifies the literature:

�Man�s work ends at set of sun. Women�s work is never done.� This is acliché most girls hear while growing up, but few appreciate it until they tryto combine employment and family responsibilities. Women find that atwenty- or forty-hour workweek outside the home does not excuse themfrom the tasks inside the home. (p. 169)

Contrary to Becker�s explanation, the �balancing act� analysis suggests that women are

working hard both at work and at home.6 If they were just serenely distributing a fixed amount

of energy, after all, they wouldn�t be so tired. When asked directly, mothers insist that flexibility

with work schedules is the main issue. In a 1997 Pew Research Center survey, 73 percent of the

457 mothers interviewed rated a flexible work schedule as �very important� in choosing a job.

Moreover, although 51 percent of mothers work full-time, only 30 percent said they prefer this

option (Pew Research Center 2000). Thus anecdotal evidence suggests that it is time itself, and

not energy, that is the central problem for working mothers. Our analyses of the data also

provide some evidence for this view.

5 It is also true that racial differences in the unexplained wage penalty may reflect omitted variables (otherthan energy dissipation) such as family structure and age of children. We include both of these controls inour models.6 Bielby and Bielby (1988) report that women and even mothers of young children report work efforts thatequal or exceed men�s reported effort.

9

IV. Data and empirical approach

When the NLSYW surveys began in 1968, a nationally representative sample of 5159

women, aged 14-24 at that time, was interviewed. In each subsequent round, data were gathered

on each respondent�s educational attainment, employment and fertility since the preceding round,

and current labor force status. As of the 1988 survey, 3508 women (currently aged 34-44) were

still being interviewed, representing 68% of the original sample.7 For our analyses, we restrict

the sample to non-Hispanic white women and black women who are currently (as of the

interview date) employed and not currently enrolled in school.8 Further, we restrict the sample to

person-year observations for which information is available regarding education, actual labor

market experience, and other regression variables, and in which the hourly wage is between $1

and $150 in 1997 dollars. The final sample includes an unbalanced panel of 4246 women (2993

whites and 1253 blacks) observed up to 15 times between 1968 and 1988, resulting in 27,204

woman-year observations.9

Table 1 provides summary statistics for the pooled cross section (27,204 woman-year

observations) and a sample including the last observation per woman (4246 observations). For

each sample, variable means are presented for all women and separately for mothers and non-

mothers. A few differences are worthy of note. Mothers are more likely to be black, married,

and part-time workers than non-mothers. On average, mothers are about four years older and

have completed almost one year less education than non-mothers. Focusing on the last

7 There are 62,736 person-year observations for which an interview was completed.8 We eliminate 158 Hispanic women (920 woman-year observations) from the sample since they are toosmall a group to examine separately.

10

observation per woman, we see that by the end of the survey period, women who were ever

mothers earned wages that were nine percent lower than those earned by women who were never

mothers.

[Table 1 Approximately Here]

To show the effect on the motherhood wage penalty of adding different controls, we

begin by regressing the log wage on race, marital status, and motherhood status using the pooled

cross-section. This provides the total motherhood gap. We then add human capital measures

which always include years of schooling and incrementally include quadratics in potential

experience, actual experience, and actual experience and age. Including age in the regression

effectively controls for years absent from the labor market.10 Finally, we control current part-

time status, occupation and household resources available to working mothers. The most

complete wage equation is:

(1) ln Wageit = α + β1Blacki + β2Marriedit + β3Childrenit + β4Educationit + β5Experienceit

+ β6Experience2it + β7Ageit + β8Age2

it + β9Part-timeit + β10Occupationit

+ β11Resourcesit + υit

where i indexes individual women (i = 1 ... 4246 in the full sample), t indexes time (t = 1968 ...

1988), and υit is an error term.11 LnWage is the natural log of the hourly wage in real 1997

dollars; Black and Married are indicator variables for race and marital status (married, spouse

present), respectively. Children is a vector of two dummy variables for one child and two or

9 On average, each woman is observed 6.4 times throughout the survey period.10 Time out of the labor force is equal to (age - schooling - 6 - time in the labor force), so the inclusion ofage effectively controls for time not working.11 In all pooled cross section models, we account for the unbalanced panel and multiple observations perwoman by weighting the regression (by the inverse of the probability that the woman is in the sample for

11

more children (under the age of 18) living in the household. This format follows Korenman and

Neumark (1992) and Waldfogel (1997). Measures of human capital investment include

Education (years of completed education), Occupation (a series of indicator variables for the

following categories: professional, technical and kindred; managers, officials, and proprietors;

clerical and kindred (the omitted category); sales workers; crafts, foremen and kindred;

operatives and kindred; service workers including private household workers; and other

occupations including farmers and farm managers, farm laborers and foremen, laborers, and

armed forces), and a quadratic in labor market experience (Experience and Experience2). Part-

time is an indicator variable for usually working less than 35 hours per week. Finally, Resources

is a vector of variables that measure each woman�s access to resources in the household that may

mitigate (or exacerbate) the motherhood wage penalty: number of adult children (age 18 and

older) in the household, number of other (related or unrelated) adults (age 18 and older) in the

household, non-labor income,12 and husband�s income.13 On the one hand, we might expect

greater household resources to reduce the motherhood penalty by providing working mothers

with more adults to help in the care of children and more income with which to purchase

childcare, restaurant meals and other substitutes for home production. On the other hand, these

same resources may, in fact, increase the motherhood penalty if the other adults in the household

also need care (for example, aged parents) or if husbands with greater income contribute less to

all 15 years) and reporting robust clustered standard errors.12 This includes all income reported by the respondent except earnings from wage and salary, own businessor farm, or unemployment insurance. If the respondent is married, this variable includes half of incomethat was reported joint with her spouse.13 This variable is equal to all income (labor and non-labor) of the respondent�s husband; it includes half ofincome that was reported jointly for wife and husband. This variable is set to zero for unmarried women.

12

home production.14

We then proceed to a fixed effects analysis in order to control unobserved heterogeneity

between mothers and non-mothers. We use a specification similar to equation (1), except that all

variables pertain to mean-differenced values across years for each woman, and the error term is

composed of a fixed component (αi) and time-varying component (µit). In addition, for some

models, we include a more detailed vector of children variables equal to the number of children

in each of five age groups, corresponding to five distinct stages in children�s lives: infants and

toddlers (birth through two years old), pre-school children (three through five years old),

elementary school children (six through ten years old), middle school children (eleven through

thirteen years old), and high school children (14 - 17 years old), in order to test whether the

penalty is largest for working mothers of young children.

Finally, we address heterogeneity among working mothers and test other implications of

the work-effort hypothesis by estimating the wage penalties for different sub-groups of mothers.

Mothers who return to work when their children are infants may differ in unobservable, wage-

enhancing capital from mothers who stay home until their child is older. These differences

would be only partly captured in measures of time out of the labor force. To control this

heterogeneity, we interact all model variables by a discrete return-timing variable which groups

mothers by the age of their youngest child over all years they worked. We also estimate penalties

by race and education in models that fully interact by these characteristics. This allows us to test

whether the penalty varies by race, and if it is higher for more educated mothers.

14 There is some evidence to support the latter. Hersch and Stratton (1994) find that husbands� share ofhousework declines with increases in his labor income. This is consistent with a household bargainingmodel in which greater own income leads to greater bargaining power within the relationship.

13

V. Human capital, heterogeneity and sample composition: Results

The first six columns of Table 2 show the effect on the motherhood wage penalty of

controlling various factors. Column (1) shows OLS results for the pooled cross-section,

controlling only race, marital status and the presence of one child or two or more children. We

interpret these coefficients of the children variables as the total motherhood penalty: the

presence of one child reduces a mother�s wage by seven percent; the presence of two or more

children reduces her wage by 13 percent. These estimates are similar in magnitude to those

found by Waldfogel (1997) in regressions that included measures of human capital. Column (2)

shows that the penalties rise by two and three percentage points, respectively, when we include

education and a quadratic in potential experience, the standard arguments of the Mincerian wage

equation.15 The increase probably arises because mothers tend to be older than non-mothers, and

thus have greater potential experience.

It is well known that potential experience overestimates women�s actual work experience

if women take any time off to bear and raise children. This is apparent in the data: when we

measure experience directly16 (see column (3) of Table 2), the effects of one child and two or

15 Potential experience is equal to the minimum of (age � education � 6) or (age � 16). This helps to reducemeasurement error, in terms of too much potential work experience, for individuals with very littleeducational attainment.16 Note that �actual labor market experience� is measured using information on actual weeks worked ineach year as reported in the fifteen surveys. If a woman works 50 or more weeks in a given year, she iscredited with a full year of labor market experience; if a woman works less than 50 weeks, her labormarket experience for that year is equal to actual weeks worked divided by 50. This is the standarddefinition of a full-year worker according to the Bureau of Labor Statistics (see, for example, Hayghe andBianchi 1994, Cohen and Bianchi 1999).

Total labor market experience is equal to the sum of actual labor market experience in every yearbetween 1968-1988, plus �potential labor market experience� prior to 1968. (Given the young age ofwomen in 1968 � ages 14 to 24 � the use of potential experience for this period is unlikely to induce much

14

more children fall by 1.3 and 3.9 percentage points, respectively. The addition of age and age

squared in column (4) controls both numbers of years in the work force and number of years out

of the work force, and reduces the motherhood wage penalties by another two and four

percentage points for one and two-or-more children, respectively. Further controlling part-time

work, occupation, and measures of household resources in columns (5) and (6) has a net effect of

reducing the penalty for one child by 0.4 percentage point and the penalty for two children by

close to one percentage point. This results in an estimated penalty for one child that is similar to

that found by Korenman and Neumark (1992) (in their cross-sectional model) and Waldfogel

(1997). However, the estimated penalty for two or more children is smaller than is found by

others.


Taken together, human capital, occupational and household resources variables account

for 25 percent of the total wage penalty for one child, and 46 percent of the total wage penalty for

two or more children. Actual experience and years out of the labor force alone account for about

30 percent of the observed total motherhood penalties.

Column (7) of Table 2 reports estimates for a fixed effects model using the same control

variables as column (6). The fixed effects estimation reduces the penalty by an additional 30

percent for one child, and an additional 13 percent for two or more children. Thus we find that

measurement error.)

Limitations of the data set (that is, gaps in the available information on work experience) make itnecessary to impute actual experience in some cases. We use a very conservative method in order toensure the best possible accuracy of this important human capital variable. Specifically, if information onweeks worked is missing in year t but not missing in years t-1 and t+1, then weeks worked in year t isestimated to equal the average of weeks worked in t-1 and t+1. Thus, we impute labor market experiencefor any given year only if we have valid information on actual labor market experience in the twosurrounding years.

15

unobserved heterogeneity between mothers and non-mothers accounts for a significant portion �

up to 30 percent � of the motherhood wage gap. Controlling both observed and unobserved

differences between mothers and non-mothers, we find a three percent wage penalty for having

one child and a six percent penalty for having two or more children. Our estimates of the

unexplained penalty lie between those reported by Korenman and Neumark (1992), who report

no unexplained penalty in a short differences model, and Waldfogel (1997), who reports

penalties of four percent for one child and 12 percent for two or more children.17

Our estimates thus far represent the average penalty across all working mothers.

Columns (8)-(10) of Table 2 show results for models estimated separately for mothers who

returned to work when their youngest child was an infant or toddler (0-2 years old), pre-school

age (3-5 years old), and school-aged (6-17 years old), respectively. Mothers who returned to

work with infants or toddlers comprise the vast majority (74 percent) of working mothers: their

sub-sample penalties are just about the same as those calculated for the entire sample. Mothers

who returned to work with children three through five years of age (17 percent of working

mothers) experience a slightly higher wage penalty for one child, but that penalty is only

marginally distinguishable from zero. The wage penalty for those women with two or more

children and for mothers who returned to the work force when their children were six or older are

not statistically significant. Although these results suggest that mothers who return most quickly

to the work force bear the brunt of the motherhood penalty, the decompositions in the following

section show that this is not, in fact, the case.

17 We find lower penalties than Waldfogel (1997) even when we use a near-identical model specification. We suspect that the difference lies in our more conservative rule for imputing work experience, sinceWaldfogel retains more observations in her sample of working women.

16

VI. Who bears the motherhood wage penalty?

We begin our decompositions by evaluating the wage effect of children at different ages.

As discussed above, if the work-effort hypothesis holds, then the motherhood wage penalty will

diminish as children grow older. Column (1) of table 3 shows a 2.7 wage penalty per child for

infants and toddlers (0-2 years old) for all women. The wage penalty for older children ranges

between 1.7 and 1.1 percent. F-tests reject equality of the wage effects of all age groups. As

expected, infants and toddlers have the largest estimated penalty, and F-tests reject equality of

this penalty with the penalties for older children at the 15 percent significance level or better.

However, it is also clear that the wage penalties, although small in magnitude, persist even as

children age.


Columns (2)-(4) of Table 3 show estimates for a fixed effects model fully interacted by

timing of mother�s return to work. This decomposition shows that at least some of the

persistence in the penalty is due to sample composition. For the majority of women who return

to work with a child five years or younger, the wage penalty is greatest when they first return.

Note that mothers returning to work with a preschooler (three through five years old) experience

a wage penalty of 3.8 percent per pre-school child, compared with a 2.6 percent penalty (per

infant or toddler) for mothers who work when their child is less than three years old. This

difference is not, however, statistically significant. The penalties in the first years of return to

work also suggest that mothers experience a period of adjustment in managing child and job

responsibilities. For example, mothers who return to work when their children are infants

17

experience only a 1.0 percent penalty for each pre-school child, compared with a 3.8 percent

penalty for mothers who recently returned to work, a difference that is significantly different.

Finally, we observe that penalties associated with children do diminish as they grow older; these

patterns are consistent with the work-effort hypothesis.

As mentioned earlier, some researchers report that black women do not bear a

motherhood wage penalty. This result would be inconsistent with the work-effort hypothesis as

long as other household resources (that might affect the motherhood penalty and that might differ

by race) are adequately controlled. In order to compare wage penalties over both return timing

and race, we estimate a fixed effects model that is fully interacted by these variables. The top

panel of table 4 shows main wage effects for the biggest sub-group: white women who returned

to work with an infant or toddler. The second and third panels show wage effects relative to the

main group for white and black women, respectively.18


The penalty incurred by white early-returning mothers for having an infant or toddler, at

2.4 percent, is similar to the estimate for all early-returning mothers in column (2) of Table 3.

The penalty for elementary and middle school children, though, appears to be larger. Early-

returning black mothers are statistically indistinguishable from their white counterparts. Both

black and white mothers who return when their youngest child is pre-school aged face a large

penalty upon re-entry. The difference by race is not, however, significant. In sum, our analysis

shows no systematic difference between white and black women, a finding that again accords

18 Total effects for each group are equal to the sum of the main and relative coefficients.

18

with the work-effort hypothesis.19

Finally, we estimate the motherhood wage penalty for women with different educational

levels. According to our discussion in section III, the effort requirement of work should rise with

skill level. To clarify the analysis, we restrict the sample to women whose educational group

does not change over her work history. Eighty-five percent of the sample (2518 of 2958 mothers)

meets this criterion. We compare four distinct educational groups: high school dropouts (less

than 12 years of schooling), high school graduates (exactly 12 years of schooling), those with

some college (13-15 years) and college graduates (16 or more years).

The first panel of Table 5 reports the main effects for the largest education-timing sub-

group: high school graduates who returned to work when their youngest child was an infant or

toddler. All other figures represent effects relative to the base group. The base group

experienced a three percent motherhood penalty for each infant and toddler when they returned to

work. The penalty, however, falls to only one percent for older children. Moreover, the

estimates of older-children penalties are below conventional confidence levels, with p-values of

30 percent.


The experience of early-returning high school dropouts is not statistically distinguishable

from that of the base group, except for a marginally significant wage premium associated with

elementary school age children. This suggests that high school dropouts also experience a wage

penalty when their children are very young. Nevertheless, all of the relative effects for this group

19 Our results by race are robust to the exclusion of occupation and household resources from theregression. When we estimate a similar model using the �one child� and �two or more children�specification, late-returning black women appear to avoid the penalty. This group is probably the source of

19

are positive and, beginning with pre-school age children, are large enough to offset the negative

base group effects. High school dropouts who return to work when their youngest child is 3-5

years of age appear to experience no wage penalty initially, and, like earlier returning dropouts,

receive a wage premium for their school-age children, relative to the base group. The premium

for this group is statistically significant at the 10 percent level.

By contrast, high school graduates who delay their return to work experience a

statistically significant penalty of more than six percent per child when they first return, and,

moreover, the penalty persists even as their children grow older. Compared to the marginally

significant one percent penalty associated with school-age children of high school graduates who

returned to work within three years of the birth of a child, those who returned to work three to

five years after a birth experienced wage penalties of five to six percent for elementary and

middle-school children. These effects are also statistically significant in models that restrict the

sample to later-returning high school graduates. Note that the pattern for those with some

college is similar, with no discernible difference from the base group for early-returners, and

higher penalties for late-returners.

Finally, we consider the wage penalty experienced by college graduates relative to the

base group. Mothers who return right away actually experience a one percent wage premium for

their infants and toddlers. For the majority of mothers who return before their children enter

school, the wage penalty is not statistically distinct from the base group. College graduates who

return to work when their children are pre-schoolers exhibit no significant penalty. Those

postponing their return until their children are school-age exhibit a strong penalty for middle

the diminished penalties reported in Hill (1979) and Waldfolgel (1997).

20

school children only. This aberration, although statistically significant, is probably due to the

small size of this sub-group of mothers.

The results for mothers of different educational levels are striking. Specifically, it is

medium-skilled mothers�those who graduated from high school but did not complete four years

of college--who bear the brunt of the wage penalty, especially if they delay their return to work

until their children are pre-schoolers. Mothers who are high school dropouts experience a wage

penalty only for infants and toddlers; almost all college graduates experience no wage penalty at

all. The non-monotonic effect of education and the persistence of the penalty for later-returning

medium-skilled mothers are particularly troublesome for the work-effort explanation.

VII. Flexible schedules and medium-skilled workers

Bielby and Bielby (1988) find that women with less labor market continuity (in our

sample, the late-returners) report less effort at work and that more educated women report more

effort. By assigning these different effort levels to different groups of mothers, and by assuming

that high school dropouts also expend more effort on the job, at least once their children are

older, one could argue that our findings do not contradict the work-effort hypothesis. We are,

however, uncomfortable with a model that relies on extreme and unsystematic heterogeneity

among sub-groups of mothers in expending work-effort. We therefore propose an alternative,

and perhaps complementary, hypothesis based on the work-schedule inflexibility of some types

of jobs.

Mothers are more likely to experience a wage penalty in jobs where flexibility is costly

for the employer. Consider, for example, the case of an economics department. If the

21

department receptionist arrives late to open the office, she interferes with the functioning of the

department. If the professors are late, no one notices. By the same reasoning, less-skilled

workers who punch a clock are also more likely to suffer wage penalties associated with

absenteeism. Thus explained, a flexibility model predicts a falling wage penalty with skill. A

model consistent with our findings, however, must account for the lack of an unexplained wage

penalty for both the least and the most skilled workers.

Suppose that a mother�s wage is a function of the time she spends at work and the amount

of effort she devotes to her work when there. This is the basis of Becker�s (1985) model in

which a worker will allocate her fixed amount of effort between household production and

market labor. Becker acknowledges the high effort needed to care for young children and

proposes that, given finite energy, mothers will choose jobs that require relatively less energy.

Because wages are a function of time and effort, mothers who choose jobs requiring lower effort

will earn a lower wage. We modify Becker�s basic model to explain the motherhood wage

penalty of high school graduates and to explain the persistence of the penalty among mothers

who delay their return to work until their youngest child is older than two. While Becker�s

analysis focuses on finite energy as the primary constraint, we believe that time, and in particular

time spent at work during the middle of the day, presents a more restrictive constraint. After all,

effort is difficult to measure and will vary in unpredictable and unobservable ways across

workers and across time for the same worker. A poor night�s sleep, for whatever reason, will

erode energy, and thus effort. In contrast, time is strictly finite, easily measured, and, as we

describe below, may have few substitutes in some jobs.

We assume that a worker�s productivity, and thus her wage, is determined by a

22

combination of three inputs: regular office hour time at work (OH); all other time spent working,

whether at the workplace or at home (T); and effective effort (E). Further, effective effort is a

function of education (ed), such that ∂E/∂ed is strictly greater than zero. 20 Thus we describe

wages of a mother, i, for job, j as:

(1) wij = wij (OHj, Tj, Ej(edi))

and define the partial derivatives with respect to each argument as

(2) =∂∂OHwj αj ≥ 0;

(3) Twj

∂∂

= βj ≥ 0; and

(4) ( )

j j

i i

w w EE ed E ed∂ ∂ ∂=

∂ ∂ ∂= γji ≥ 0.

Suppressing subscripts, we differentiate the function and write

(5) dw = α dOH + β dT + γ dE(ed).

A worker who is time- or effort-constrained may be able to avoid a wage penalty (i.e., can set

dw = 0) by making offsetting adjustments among the three inputs. Whether this is possible will

depend on the degree of substitutability among these factors. We define θm,n as the magnitude of

the elasticity of substitution of factor m with respect to factor n. 21 Thus, for example, θOH, T

20 This would be true under either the theory that education increases productive human capital (Becker1964, Mincer 1974) or under a signaling theory by which more ambitious workers obtain more educationbecause it is less costly for them to do so (Spence, 1973).21 We define these as magnitudes to make comparisons more intuitive. Defined as magnitudes, a largervalue of θ implies a greater degree of substitutability.

23

measures the elasticity of substitution of Office Hours with respect to Other Time.

Setting dw = 0 and disregarding any indirect effect of effort,

(6) θOH, T = dO H T TdT O H O H

βα

= − .

Jobs vary by the extent to which productivity depends on each input and by the extent to which

one input is substitutable for another. For example, telecommuters can freely substitute between

office hours and other time, implying a high θOH, T. For these workers, αj≈βj . Similarly the

saying �work smarter, not longer� suggests that, in some jobs, putting forth greater effective

effort can reduce time spent working (i.e. θT, E > 0). Specifically, we propose that the relative

importance of each input and the elasticity of substitution among them will vary systematically

with the kind of jobs that are available to women with different levels of education. Let

JE denote jobs for which γji > αj and γji > βj;

JOH denote jobs for which αj > βj and αj > γji, and

JT denote all other jobs, that is βj ≥ γji and βj ≥ αj.

Jobs that require a college degree are more likely to be jobs of type JE. This follows from

the relationship between effort and education specified above: γji is an increasing function of

education. Further, college graduates are more likely to have greater autonomy in their jobs both

in their working hours (θOH, T > 0) and in their working methods (θOH, E > 0; θT, E > 0). High

school graduates are those most likely to have jobs that require their presence during office hours

(JOH), but are less likely than college graduates to have sufficient autonomy in those jobs to allow

them to substitute either effort or other time for actual office hour time (θOH, E ≈ 0; θOH, T ≈ 0).

Finally, many of the jobs that are available to high school dropouts, including food service,

housekeeping or manufacturing, entail shift work and are of type JT . For these jobs, time in

24

general matters more than effort, and time outside of office hours is no less valuable than regular

office hours (θT,E ≈ 0; θOH, E ≈ 0; θ,OH, T > 0).

To complete this explanation, assume that the same factors that determine a mother�s

productivity generate disutility for her, again at different rates and for different reasons. Just as

spending more time at work reduces leisure time, spending more effort at work will eventually

undermine a mother�s energy level at home. Working during office hours may be particularly

difficult both because those are the hours that children are most active and thus need the most

supervision and because those are the hours that a helping husband is most likely to be working,

too. Mothers who have the strongest distaste for office-hour jobs, either because they lack

support at home or because they more strongly prefer spending those hours with their children,

are the most likely to delay re-entry into the work force. When they do return to the work force,

particularly if they are high school graduates, they will be most likely to find jobs that require

their presence during regular office hours. If they have higher absenteeism (for the same reasons

they delayed re-entry) and are unable to substitute other time or increased effort, they will suffer

a wage penalty relative to early-returning mothers who have evidenced a milder disutility for

office hour work. College graduates and low-skilled workers avoid the penalty by substituting

effort and flexible time.22

Although this model is plausible, it relies on the assumption that jobs available to high

school graduates have different work-schedule requirements. Medium skill workers in our data

set are much more likely to report clerical occupations: 47 and 44 percent of high school

22 For separate, education-level based labor markets to co-exist, the high school graduate�s wage, with thepenalty, must still be greater than the non-penalized high school dropout�s wage. If not, the high schoolgraduates with strong disutility for working during office hours would take the lower-skilled, but more

25

graduates and those with some college, respectively, work as clericals, compared with 20 percent

of high school dropouts and 15 percent of college graduates. If workers in the same broadly

defined occupation have different work-schedules, then occupational controls alone cannot

resolve the issue, especially if work-schedules are correlated with skill. A test of our model

would require detailed information on work schedules and not just hours. The NLSYW does not,

unfortunately, provide this information.

VIII. Conclusions and Policy Implications

Others have identified a wage penalty among working mothers and have investigated

reasons for its existence. In this paper we use a detailed panel data set to consider each of the

proffered explanations. Like researchers before us, we control for human capital inputs and for

unobserved heterogeneity. These controls explain about 60 percent of the wage gap between

mothers and non-mothers. A wage gap of three to six percent persists. We also address

heterogeneity among mothers identified by the timing of their return to the work force. We find

that mothers tend to face the highest wage penalty when they first return to work, even if their

children are older. This may reflect learning to manage the dual responsibilities of work and

children, or the greater incidence of illnesses among children when they first enter a daycare or

school environment.

Finally, we investigate the incidence of the penalty by further decomposing the sample by

age of children and mothers� race and education. We find that younger children impose a higher

penalty than older children in general and that black and white mothers face the same penalty,

flexible, jobs.

26

patterns that are consistent with a work-effort explanation. But the largest differences in the

penalty arise among education groups. Although more educated women are likely to have jobs in

which effort is relatively important, we find that college-educated mothers do not, in fact, face

any penalty for having children. And while high school dropouts face a three per cent penalty if

they work when their children are infants and toddlers, they do not bear any penalty for older

children. Thus high school dropouts who delay their return to the work force until their children

are older bear no motherhood wage penalty at all. By contrast, high school graduates --

especially those who return to work when their children are older -- face persistent penalties of

three to six percent up until the time their children enter high school.

The results for wage gap differentials by education cast doubt on the work-effort

hypothesis as a complete explanation for the wage penalty. It is unlikely that high school

graduates reduce work effort in response to childcare duties at home when mothers with less and

more education do not. Instead, we propose a model that emphasizes the importance of time, and

in particular, time during the middle of the day, rather than effort while at work in explaining the

motherhood penalty. If high school graduates are the most likely to have jobs that require their

presence during regular office hours and the least likely to gain flexibility either by finding work

at other hours, by working harder when at work or by taking work home in the evening, then our

time-flexibility model provides a compelling explanation for education patterns in the

motherhood wage penalty.

Becker (1985) includes both absenteeism (i.e., reduced time at work) and reduced energy

for work as components of the proposed lesser work effort of mothers. Our findings are easily

reconciled with Becker�s if we restrict the definition of work effort to time at work. An emphasis

27

on time accomplishes several things. First, it provides a better explanation for the data. Second,

it avoids potentially damaging, and unfounded, claims that mothers put forth less effort while at

work. Finally, it provides constructive direction for addressing working mothers� needs,

specifically flexibility (rather than absolute reduction) in time spent at work.

28

References

Becker, Gary. Human Capital. (Chicago: University of Chicago Press 1964).

Becker, Gary. Human Capital, Effort, and the Sexual Division of Labor. Journal of LaborEconomics 3(1), (1985) 533 - 558.

Bielby, Denise and William Bielby. She Works Hard for the Money: HouseholdResponsibilities and the Allocation of Work Effort. American Journal of Sociology 93(5),(1988) 1031-59.

Cohen, Philip N. and Suzanne M. Bianchi. Marriage, children and women�s employment: Whatdo we know? Monthly Labor Review 122 (12), (December 1999) 22 � 31.

Hayghe, Howard V. and Suzanne M. Bianchi. Married mothers� work patterns: The job-familycompromise. Monthly Labor Review 117 (6), (June 1994) 24 � 30.

Hersch, Joni and Leslie S. Stratton. Housework, Fixed Effects, and Wages of Married WorkersThe Journal of Human Resources 32, (1997) 285-307.

______________________. Housework Wages, and the Division of Time for Employed SpousesAEA Papers and Proceedings 84, (1994) 120 -125.

Hill, Martha S. The Wage Effects of Marital Status and Children The Journal of HumanResources 14 , (1979) 579 - 94.

Jacobsen, Joyce P. and Laurence M. Levin. Effects of intermittent labor force attachment onwomen�s earnings. Monthly Labor Review (Sept 1995) 14 - 19.

Korenman, Sanders and David Neumark. Marriage, Motherhood, and Wages. The Journal ofHuman Resources 27 (1992) 233-55.

Mincer, Jacob. Schooling, Experience and Earnings. (Boston: NBER 1974).

The Pew Research Center. �Motherhood Today�A Tougher Job, Less Ably Done.� (accessed2000) www.people-press.org/mommor.html.

Spain, Daphne and Suzanne Bianchi. Balancing Act: Motherhood, Marriage, and EmploymentAmong American Women. (New York: Russell Sage 1996).

Spence, Michael. Job Market Signaling. Quarterly Journal of Economics 87, (August, 1973)355 � 374.

29

Waldfogel, Jane. The Effect of Children on Women�s Wages. American Sociological Review 62, (April 1997) 209-17.

Waldfogel, Jane. The Price of Motherhood: Family Status and Women�s Pay In a Young BritishCohort. Oxford Economic Papers 47(4), (1995) 584-610.

_____________. Understanding the Family Gap in Pay for Women with Children. Journal ofEconomic Perspectives 12 (winter 1998) 137 - 156.

Table 1: Summary Statistics (standard deviations for continuous variables in parentheses)

Pooled cross section Last observation per woman

Variable All women MothersNon-

mothers All womenEver

MothersNever

MothersWage

(1997 $)10.430(5.415)

10.017(5.367)

10.958(5.431)

9.852(5.248)

9.580(5.152)

10.476(5.414)

Black 0.294 0.357 0.213 0.295 0.333 0.207

Married,Spouse Present 0.585 0.725 0.406 0.570 0.673 0.333

Children < 18in Home

1.037(1.171)

1.849(0.973)

1.370(1.172)

Education12.655(2.311)

12.315(2.232)

13.090(2.337)

12.584(2.408)

12.375(2.350)

13.064(2.470)

PotentialExperience

9.331(6.323)

11.436(5.646)

6.640(6.119)

7.796(6.078)

9.035(6.188)

4.950(4.716)

ActualExperience

7.553(4.845)

8.469(4.574)

6.382(4.930)

6.133(4.375)

6.631(4.466)

4.990(3.926)

Age28.121(6.395)

29.911(5.871)

25.831(6.306)

26.544(6.223)

27.587(6.350)

24.148(5.181)

Part-time 0.184 0.239 0.114 0.210 0.237 0.147Professional 0.191 0.160 0.230 0.185 0.166 0.227

Managerial 0.051 0.049 0.053 0.038 0.041 0.033Sales 0.042 0.044 0.039 0.053 0.050 0.058

Clerical 0.390 0.363 0.424 0.377 0.362 0.411Craft 0.014 0.016 0.012 0.011 0.012 0.010

Operative 0.146 0.179 0.105 0.143 0.160 0.105Service 0.157 0.178 0.130 0.182 0.197 0.148

Children 18+in Home

0.078(0.351)

0.095(0.371)

0.056(0.321)

0.060(0.310)

0.085(0.368)

0.002(0.039)

Other Adultsin Home

1.074(0.924)

0.941(0.705)

1.245(1.122)

1.162(0.983)

1.061(0.838)

1.393(1.223)

Husband�sIncome in 1000s

of 1997 $

17.462(21.147)

22.360(21.871)

11.196(18.370)

16.610(21.020)

19.943(21.642)

8.957(17.236)

Own Non-labor Income in1000s of 1997 $

0.592(2.391)

0.810(2.601)

0.314(2.058)

0.536(2.363)

0.639(2.419)

0.300(2.214)

Observations 27204 15268 11936 4246 2958 1288

Notes: All differences in means (mothers versus non-mothers) are statistically significant at the 5 percentlevel except the following: managerial and sales in the pooled cross section; managerial, sales, craft, other

31

occupations in the individual women sample. For the pooled cross section, �Mother� is defined as havingat least one child under the age of 18 at home during a given year. For the individual woman sample,�Ever Mother� is defined as ever having at least one child under the age of 18 at home.

Table 2: The Effect of Children on Women�s Log Wages: Pooled Cross Section and Fixed Effects Models (robust clustered standarderrors in parentheses, weighted regressions)

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Fixed effects by age of child at

return to work forceIndependentVariable

Pooled cross sectionFixedeffects

0-2 years 3-5 years 6-17 yearsOne child -0.069 ***

(0.013)-0.089 ***(0.012)

-0.076 ***(0.011)

-0.056 ***(0.012)

-0.049 ***(0.012)

-0.052 ***(0.073)

-.030***(.007)

-0.029*** (0.008)

-0.037H

(0.023) -0.018(0.032)

2 or morechildren

-0.134 ***(0.013)

-0.161 ***(0.014)

-0.122 ***(0.012)

-0.082 ***(0.014)

-0.071 ***(0.014)

-0.073 ***(0.013)

-.055***(.008)

-0.053***(0.010)

-0.012 (0.027)

-0.040(0.035)

Education X X X X X X X X X

Experience X X X X X X X X X

Experience 2 X X X X X X X X X

ExperienceType Potential Actual Actual Actual Actual Actual Actual Actual Actual

Age X X X X X X X

Age2 X X X X X X X

Part - time X X X X X X

Occupation& householdresources

X X X X X

R-squared .029 0.247 0.281 0.297 0.302 0.375

Observations 27204 woman-years 4246 2197 490 271

Point difference in motherhood gap from previous columnTotal cols.

2-6(% of col 1)

Point difference in motherhood gap fromprevious column (% of col. 1)

Kid1 +2.0 -1.3 -2.0 -0.7 +.3 -1.7 (25%) -2.2 (57%) -0.1 (58%) +0.7 (46%) -1.2 (74%)

Kid2+ +2.7 -3.9 -4.0 -1.1 +.2 -6.1 (46%) -1.8 (59%) -0.2 (60%) -4.3 (91%) -1.5 (63%)

*** p<.01 ** p<.05 * p<.10 H p<.15 X indicates variables that were included in regression.Notes: Sample includes 27204 woman-year observations (4246 women observed between 1-15 times, or 6.4 times on average). To control for anunbalanced panel and multiple observations per woman, we weight the pooled cross section regression (by the inverse of the probability that thewoman is in the sample for all 15 years) and we estimate robust clustered standard errors. All models also include controls for race and marital

34

status.

35Table 3: Wage Effects of the Number of Children of Different Ages by Timing of Return to Work Force, FixedEffects Models (Standard error in parentheses)

(1) (2) (3) (4)All women Age of youngest child at return to work

0-2 3-5 6-17Infants and Toddlers (0 - 2 years old)

-0.027 ***(0.006)

-0.026***(0.006)

-- --

Pre-school(3 � 5 years old)

-0.016 ***(0.005)

-0.010*(.005)

-.038***(.015)

--

Elementary School (6 � 10 years old)

-0.012 ***(0.004)

-.009*(.005)

.009(.013)

-.026H

(.017)Middle School(11 � 13 years old)

-0.017 ***(0.005)

-.006(.006)

-.016(.016)

-.021(.018)

High School( 14 � 17 years old)

-0.011 **(0.005)

-.008(.007)

.015(.015)

.006(.017)

Number of woman-years 27204 162559 2872 1439Number of women 4246 2197 490 271

Null Hypothesis: F-Value (p-value)Equal effects for all agegroups

1.83(.120)

4.55 3.15 1.58

Ages 0 � 2 = Ages 3 � 5 2.53(.112)

6.08 -- --

Ages 3 � 5 = Ages 6 � 10 .66(.418)

.03 8.90 --

Ages 6 � 10 = Ages 11 � 13 .68(.410)

.16 2.70 .09

Ages 11 � 13 = Ages 14 �17

.88(.349)

.07 3.35 1.82

Ages 0 � 2 = Ages 6 � 10 6.21(.013)

8.13 -- --

Ages 0 � 2 = Ages 11 - 13 2.26(.133)

7.63 -- --

Ages 0 � 2 = Ages 14 � 17 5.74(.017)

5.82 -- --

Ages 3 � 5 = Ages 14 � 17 .93(.335)

.05 9.52 --

Ages 6 � 10 = Ages 14 � 17 .08(.777)

.01 .19 3.05

*** p<.01 ** p<.05 * p<.10 H p<.15

Notes: Sample includes 27204 woman-year observations (4246 women observed between 1-15 times, or 6.4 times onaverage). There are 2958 mothers in the sample: 2197 mothers returned to work with a 0-2 year old child, 490mothers returned to work with a 3-5 year old child, and 271 mothers returned to work with a 6-17 year old child. Allmodels are estimated using fixed effects; additional control variables include marital status, education, actual labormarket experience (quadratic), age (quadratic), part-time status, occupation, and household resources.

36Table 4. Wage Effects of Children by Mother's Race and Return TimingFully-interacted fixed effects models (Standard errors in parentheses)

Main Effect: White women who returned to work when their child was 0 � 2 years oldAge of children

0 � 2 -0.024**

(0.007)

3 � 5-0.007(0.007)

6 � 10-0.012*(0.006)

11 � 13-0.015*(0.009)

14 � 17-0.003(0.009)

Number of women years 11027Number of women 1448

Relative Effect: Other white womenAge group of youngest child over mother�s work history

Age of children 0 � 2 3 � 5 6 � 17

3 � 5 -0.037**(a)

(0.018)

6 � 100.016

(0.016)-0.020(a)

(0.021)

11 � 13-0.007(0.020)

-0.037*(a)

(0.023)

14 � 17

SeeMain

EffectsAbove

0.027(0.020)

0.013(0.022)

Number of women years 11027 1965 1134Number of women 1448 312 212

Relative Effect: Black womenAge group of youngest child over mother�s work history

Age of children 0 � 2 3 � 5 6 � 17

0 � 2-0.005(a)

(0.013)

3 � 5-0.005(b)

(0.011) -0.052**(b)

(0.026)

6 � 100.004

(0.010)0.008

(0.024)-0.005(0.049)

11 � 130.019

(0.013)-0.002(0.028)

0.080*(a)

(0.048)

14 � 17-0.011(c)

(0.014)-0.018(0.028)

-0.013(0.051)


**Significant main or relative effect, 5% level. * Significant main or relative effect, 10% level.

(a), (b) and (c) denote significant total effect in models run separately for each sub-group at the 5, 10

37and 15% levels, respectively.

See notes to Table 3.

38Table 5. Wage Effects of Children by Mother's Education and Return TimingFully-interacted fixed effects models (Standard errors in parentheses)

Main Effect: High school grads who returned to work when their child was 0 � 2 years oldAge of children

0 � 2 -0.032(a) (0.009)3 � 5 -0.009 (0.009)

6 � 10 -0.009 (0.008)11 � 13 -0.010 (0.010)

14 � 17 -0.011 (0.011)

Number of women years 7312Number of women 909

Relative Effect: Mothers with less than 12 years of schoolingAge group of youngest child over mother�s work history

Age of children 0 � 2 3 � 5 6 � 170 � 2 0.002(a) (0.017)3 � 5 0.011 (0.016) 0.008 (0.031)

6 � 10 0.022H (0.014) 0.050*(b) (0.027) -0.028 (0.038)11 � 13 0.012 (0.017) 0.034 (0.030) 0.039 (0.039)14 � 17 0.009 (0.017) 0.030 (0.029) 0.015 (0.040)


Relative Effect: Mothers with 12 years of schoolingAge group of youngest child over mother�s work history

Age of children 0 � 2 3 � 5 6 � 173 � 5 -0.058**(a) (0.024)

6 � 10 -0.033H(c) (0.023) -0.012 (0.038)11 � 13 -0.058**(a) (0.027) -0.057H (0.038)14 � 17

SeeMain

EffectsAbove 0.001 (0.027) -0.021 (0.034)

Number of women years 1291 535Number of women 220 101

Relative Effect: Mothers with more than 12 years of schooling, but without college degreeAge group of youngest child over mother�s work history

Age of children 0 � 2 3 � 5 6 � 170 � 2 0.018 (0.019)3 � 5 0.003 (0.019) -0.058(b) (0.057)

6 � 10 -0.004 (0.019) -0.027 (0.053) -0.102 (b) (0.083)11 � 13 0.014 (0.024) -0.033 (0.056) -0.059 (0.089)14 � 17 0.031 (0.025) -0.061(c) (0.057) -0.039 (0.070)


Relative Effect: Mothers who are college graduatesAge group of youngest child over mother�s work history

Age of children 0 � 2 3 � 5 6 � 170 � 2 0.040** (0.019)3 � 5 0.003 (0.020) -0.010 (0.068)

6 � 10 0.003 (0.021) 0.022 (0.065) 0.021 (0.067)11 � 13 -0.019 (0.029) 0.079 (0.089) -0.118*(a) (0.065) 14 � 17 -0.005 (0.035) 0.100 (0.101) 0.045 (0.072)


**Significant main or relative effect, 5% level. *Significant main or relative effect, 10% level.

H Significant main or relative effect, 15% level. (a), (b) and (c) denote significant total effect in

39models run separately for each sub-group at the 5, 10 and 15% levels, respectively.

See notes to Table 3.

Documents

The Motherhood Wage Penalty Revisited: Experience, Heterogeneity, Work Effort, and Work-Schedule Flexibility