Does Early Maternal Employment Harm Child Development? An Analysis of the Potential Benefits of Leave Taking

by Charles L. Baum II
Citation
Title:
Does Early Maternal Employment Harm Child Development? An Analysis of the Potential Benefits of Leave Taking
Author:
Charles L. Baum II
Year: 
2003
Publication: 
Journal of Labor Economics
Volume: 
21
Issue: 
2
Start Page: 
409
End Page: 
448
Publisher: 
Language: 
English
URL: 
Select license: 
Select License
DOI: 
PMID: 
ISSN: 
Abstract:

 

Does Early Maternal Employment Harm Child Development? An Analysis of the Potential Benefits of Leave Taking

Charles L. Baum II, Middle Tennessee State University

More mothers engage in marketplace work today than ever before, with over 33% returning to work by the time their child is 3 months old. This article identifies the effects of maternal marketplace work in the initial months of an infant’s life on the child’s cognitive development. Results suggest that such work in the first year of a child’s life has detrimental effects. Where significant, the results also indicate negative effects of maternal employment in the child’s first quarter of life. However, the negative effects of maternal marketplace work are partially offset by positive effects of increased family income.

I. Introduction

More mothers with young children engage in marketplace work today than ever before. In 1960, less than 20% of mothers with preschool-aged children (under age 6) were working, but by the 1990s, this portion had increased to about 60% (Leibowitz and Klerman 1995; Barrow 1999; Ruhm 2000). This trend is similar for mothers with infants: in 1995, over 66% of mothers who were working before giving childbirth returned to work by the time their infant was 2 years of age (Klerman and Leibowitz 1999). Actually, over one-third of all mothers who were working during the pregnancy returned to work by the time their infant was 3 months

I would like to thank David M. Blau for helpful comments and guidance. I also thank the Faculty Research and Creative Activity Committee (FRCAC) and Donald L. Curry at Middle Tennessee State University for financial support.

[ Journal of Labor Economics, 2003, vol. 21, no. 2]

� 2003 by The University of Chicago. All rights reserved. 0734-306X/2103-0007$10.00

409

old (Klerman and Leibowitz 1990; Klerman and Leibowitz 1994). Thus, many of these mothers take only a brief period of time off before returning to work. Returning to work shortly after giving childbirth may have a detrimental effect on the development of infants. Psychologists have certainly suggested that the first year of a child’s life is crucial to development because it is in this period that an infant develops a sophisticated cognitive conception of objects and people (Lewis and Brooks-Gunn 1979; Harris 1983). On the other hand, if an infant does not benefit from time spent with the mother, then maternal marketplace work in the first few months may not affect an infant’s development.

Given the importance of the infant’s first year on cognitive development, there are two explanations why maternal marketplace work could potentially be detrimental to an infant. First, such work decreases the quantity of maternal time spent with the infant. Spending less time in maternal care will be detrimental if nonmaternal childcare arrangements are of inferior quality. Second, marketplace work may decrease the quality of maternal time spent with the infant if mothers who work long hours are subject to exhaustion, emotional distress, and overload. Therefore, working mothers may be more contentious in the home.

Conversely, if stay-at-home mothers are more likely to be depressed and to withdraw from their children, then marketplace work may increase the quality of maternal time spent with the infant (Parcel and Menaghan 1994b). Further, maternal marketplace work may have additional positive side effects on children by increasing family income and decreasing fertility rates (Stafford 1987).1 Increased family income would enhance child development by allowing the family greater financial resources with which to purchase child development inputs such as books and educational trips (Blau 1999). Decreased fertility would benefit child development by reducing the number of siblings over which family resources and parental time must be divided (Hanushek 1992).

In this article, I examine the effects of maternal marketplace work during the first months of an infant’s life on three measures of cognitive development—the Peabody Picture Vocabulary Test (PPVT) and the Peabody Individual Achievement Tests of Mathematics (PIAT-M) and Reading Recognition (PIAT-R). I focus on the mother’s work behavior during the infant’s initial months, because it is during this period that a sizable portion of mothers first work after giving childbirth. Part of my analysis focuses exclusively on mothers who were working (within 3 months) before giving birth, because it these mothers who must decide whether to maintain ties to their prechildbirth employer by returning to work shortly after giving childbirth. If working mothers systematically differ

1 Maternal work may also have positive side effects on a mother by reducing depression (Moore and Driscoll 1997).

in unobserved ways from mothers who show no attachment to the labor force, then examining this subsample will reduce potential bias from unobserved heterogeneity. By determining the effects of the mother’s marketplace work shortly after giving birth, I can project the benefits, if any, of the mother taking time off from work to spend with her child. My results give the total effect of an exogenous change in maternal labor supply, as well as the partial effect holding the mother’s earnings constant. These results suggest that maternal work in the first year of a child’s life has detrimental effects. Where significant, the results also indicate negative effects of maternal work in the child’s first quarter of life. Thus, an infant’s cognitive development may benefit from the mother taking more time off from work after giving childbirth. However, results also show that the negative effects of maternal work are partially offset by positive effects of increased family income.

The remainder of the article is organized as follows. In Section II, I present a review of the literature and explain how this study improves upon existing results. In Section III, I describe the data, and in Section IV I describe the model and estimation technique. I present the results in Section V. In Section VI, I discuss the results and conclude the article.

II. Existing Evidence

The effects of maternal employment on child development have been widely studied in the psychology, sociology, and economics literature. Many of these studies examine the Peabody Picture Vocabulary Test (PPVT), and most of these studies use National Longitudinal Survey of Youth (NLSY) data.2 However, it is not clear whether maternal employment has a detrimental effect on child development since the literature often reaches conflicting conclusions.

A few researchers have found that maternal employment has detrimental effects on a child’s cognitive development.3 In particular, Parcel and Menaghan (1990) and Hill and O’Neill (1994) find that PPVT scores

2 Some psychologists have used Ainsworth’s Strange Situation setting to examine how maternal employment affects the attachment of the child to its mother (Ainsworth 1973). Some find that infants who have spent a substantial amount of time in nonmaternal care are more likely to exhibit avoidance upon reunion with the mother and are more insecure (Schwartz 1983; Barglow, Vaughn, and Molitor 1987; Belsky and Rovine 1988). If the mother does not spend a sufficient amount of time with her child, then the child may develop an insecure attachment to its mother, hindering the development of the child-mother attachment. Others have found that maternal employment has no effect on the child’s attachment to its mother (Hock 1980; Chase-Lansdale and Owen 1987).

3 Others have studied socioemotional development and have found that maternal employment has a negative effect on measures of a child’s adjustment, compliance, and attachment to its mother (Belsky 1988; Belsky and Rovine 1988; Belsky 1990; Belsky and Eggebeen 1991).

are decreasing in maternal hours of work. In a more recent study, Ruhm (2000) examines the PPVT, as well as reading and math scores, and finds that maternal employment has a negative effect on all three measures. Stafford (1987) examines older children and finds that maternal market-work has a negative effect on seven teacher-evaluated measures of a child’s school performance.

Many have found mixed results. Blau and Grossberg (1992) find that the portion of weeks worked by the mother during her child’s first year has a detrimental effect on the PPVT but that weeks worked in the second year have a positive effect. Similarly, Baydar and Brooks-Gunn (1991) show that maternal employment during the last 2 quarters of a white child’s first year has a detrimental effect on the PPVT and Behavior Problems Index (BPI) outcomes but that maternal employment begun in the child’s second or third years does not. Mott (1991) finds that unhealthy boys in maternal care had higher test scores measuring socioemotional development than did unhealthy boys in nonmaternal care. Mott also finds, however, that healthy girls in nonmaternal care had higher test scores measuring cognitive development than did healthy girls in maternal care.4 Desai, Chase-Lansdale, and Michael (1989) examine the effect of continuous and intermittent maternal employment during a child’s first 4 years and find that maternal employment had a significant negative effect on high-income boys but did not have a significant effect on low-income boys or either high-income or low-income girls.5

Others have found that maternal employment does not have detrimental effects. For example, Leibowitz (1977) finds that neither full-time nor part-time maternal employment have significant effects on PPVT scores, and a recent study by Harvey (1999) also finds that various measures of maternal employment do not have consistent effects on children.6 Similarly, Parcel and Menaghan (1994a) examine the effects of maternal and paternal employment on cognitive and social outcomes, and they, too, find that maternal employment has virtually no effect on cognitive development or social behavior. Greenstein (1995) focuses on “advantaged” children and finds that neither they nor their nonadvantaged counterparts are harmed by maternal employment.

Fleisher (1977) and Datcher-Loury (1988) examine longer-term effects. Fleisher’s analysis indicates that the number of years (during the first 14

4 Belsky and Rovine (1988) also find nonmaternal care to be particularly detrimental for boys. However, Baydar and Brooks-Gunn (1991) do not find differences across genders.

5 Blau and Grossberg (1992) also find that maternal employment in a child’s first year has a larger negative effect in high-income families than in low-income families.

6 Harvey does find that intensive early maternal work may decrease cognitive development at young ages.

years of a child’s life) in which a mother works less than 6 months has little effect on the child’s ultimate level of schooling and IQ. Similarly, Datcher-Loury’s results show that a child’s ultimate education level is not significantly affected by the number of hours that a mother works during a 10-year span of her child’s life.

One study has found that maternal employment has a positive effect. Vandell and Ramanan (1992) examine low-income families, and their results show that maternal employment actually has a significantly positive effect on reading and math scores.

The existing literature has a primary shortcoming: none of these studies accurately identifies the effect of maternal marketplace work in the initial months after giving birth. Instead, previous research has examined the effect of the mother’s employment and assumed that employed mothers are actually working. However, many employed mothers are not actually working shortly after giving birth because they are on maternity leave (Klerman and Leibowitz 1994). In fact, Klerman and Leibowitz find that 27.6% of all mothers and 72.6% of employed mothers were on leave from work initially after giving birth.7 Thus, many of the employed mothers that the literature assumes are working are actually with their infants on maternity leave.8 The work-employment distinction is important because employed mothers who are on leave may be more likely to provide maternal care for their infant, while employed mothers who are working must use nonmaternal childcare arrangements for hours spent working.

The literature has additional shortcomings. First, much of the previous research has examined only the PPVT. This provides a limited picture of the effect of maternal employment on cognitive development. Second, most of the literature uses small and unrepresentative samples. Many studies that use NLSY data include children only from the 1986 wave and consequently contain children from unusually young mothers.9 Finally, most of the estimates found in the literature fail to control for unobserved heterogeneity. Unobserved heterogeneity could bias the re

7 This information comes from the Current Population Survey (CPS) from 1986–88.

8 Researchers who use the National Longitudinal Survey of Youth (NLSY) data set (e.g., Desai et al. 1989; Baydar and Brooks-Gunn 1991; Mott 1991; Blau and Grossberg 1992; Hill and O’Neill 1994; Parcel and Menaghan 1994a) are unable to determine the effects of a mother’s marketplace work on children because the NLSY prior to the 1988 survey did not identify whether employed mothers were working or on leave.

9 The NLSY respondents were aged 21–28 in 1986, and the NLSY mothers had given birth to PPVT-tested children at least 3 years prior to 1986. Baker and Mott (1989) estimate that the children evaluated by the 1986 NLSY represent roughly the first 40% of the NLSY cohort’s childbearing. For more on this, see Harvey (1999).

sults if the same unmeasured traits that determine maternal marketplace work simultaneously affect the child outcomes.

I seek to improve upon the existing literature by identifying, in order to more accurately measure a mother’s work behavior, whether employed mothers are working or on leave in each week after giving birth. I also make the distinction between the total effect of maternal work, which includes the effect on family income, and the partial effect of maternal work, which holds family income constant. Further, I examine the effects of maternal marketplace work on three measures of child development that evaluate a child’s cognitive achievement, verbal skills, and math skills. This will provide a more complete assessment of the effects of maternal work on cognitive development. I use NLSY data and include infants from more recent waves, which produces a larger sample of children from older mothers. Finally, I attempt to control for potential unobserved heterogeneity bias by including an unusually large number of exogenous background characteristics, by instrumenting maternal labor supply using local labor market conditions, and by estimating a portion of the models with a more homogeneous group of mothers who were employed prior to giving birth.

III. Data

I use the National Longitudinal Survey of Youth (NLSY) data set because it collects extensive information on child outcomes and also identifies each mother’s work status. The NLSY began annually interviewing 12,686 individuals who were between the ages of 14 and 21 in 1979, and the survey remains in progress. The original NLSY sample contained 6,283 women and an oversample of blacks, Hispanics, low-income whites, and military personnel. The military sample was dropped in 1984 and the low-income white sample was dropped in 1990, leaving 4,944 women to be assessed in 1996.10

The NLSY collects extensive information on background characteristics and labor market experiences. Beginning in 1986 and continuing in every even-numbered year, the NLSY has collected information on the children

10 The NLSY contains sampling weights which, when used, make this a nationally representative data set. The sampling weights have been found to be appropriate only if the low-income white oversample is included (MaCurdy, Mroz, and Gritz 1998). Because my sample does not include the low-income white oversample, I do not report results from models that used the sample weights. However, I did reestimate the models using the original 1979 sampling weights and the revised 1990 sampling weights. The results with and without sampling weights are very similar. There is no discernible pattern in the changes of the effects of the maternal labor supply variables when sampling weights are used. This may be because I include variables that control for the characteristics upon which the NLSY is stratified.

of the mothers in the original sample. This information includes measures of each child’s cognitive development and verbal and math skills. As of 1996, 10,507 children of mothers in the NLSY had been interviewed at least once.11 The low-income white oversample and the military sample are excluded from my analysis. I also exclude children who did not live with their mother during their first 3 years.12 My sample includes children born between 1988 and 1993, which, when weighted, will be a nationally representative sample of children born to mothers who were between the ages of 23 and 30 in 1988. However, results from this sample will not necessarily be representative of other cohorts or of children born to this cohort before 1988 or after 1993.

Because the NLSY collects extensive information on each mother’s employment status, I am able to construct a work history for each mother that identifies her employment status in each week after giving birth. However, many mothers who are employed shortly after giving birth are not actually working because they are on leave. The NLSY identifies unpaid gaps in employment spells when the individual was employed but not working. Thus, it is possible to identify unpaid leave from work. However, prior to the 1988 survey, it was not possible to identify whether employed respondents were actually working or on paid leave. Again, this distinction is important because employed mothers who are on leave are assumed to provide maternal childcare for their infant, while employed mothers who are working are assumed to purchase work-related marketplace childcare arrangements. Fortunately, beginning with the 1988 survey, the main NLSY questionnaire began identifying whether employed mothers were working or on paid leave.13 Thus, I am able to further disaggregate each mother’s employment status in each week after giving birth into the following mutually exclusive and exhaustive categories: working and not working. Using the work history just described, I then identify the number of hours worked during the weeks in which each mother was actually working. Therefore, I include only children who were born after 1987 in my sample, so that I can accurately identify the hours and weeks in which their mothers actually worked after their births.

11 For more information on the children of the NLSY youth, see the Center for Human Resource Research’s NLSY79 1996 Child and Young Adult Users Guide (1996).

12 About 5% of the NLSY79 children did not live with their mothers during their first 3 years.

13 The main NLSY questionnaire began collecting information on the starting and stopping dates of any period of paid leave from work due to giving birth or pregnancy in 1988. The NLSY work history file collects information on starting and stopping dates of any unpaid periods of leave from work due to pregnancy. For more information on identifying paid and unpaid maternity leave in the NLSY data set, see Klerman (1995).

I create multiple variable specifications to measure the effects of maternal labor supply. In some models, I include a set of maternal work variables that measure the number of hours that the mother was actually working. Specifically, these variables equal the total number of hours worked in each week divided by 40. Thus, increasing these variables from 0 to 1 shows the effect of working the equivalent of an additional 40hour week. In other models, I include a different set of variables that identifies the number of weeks in which the mother worked (regardless of the number of hours worked). Changing these variables from 0 to 1 simulates the effect of working in an additional week. I also create a version of this set of variables that measures the number of weeks in which the mother worked full time, where full time is defined as working at least 35 hours. Additionally, I include a set of dummy variables that indicates whether the mother engaged in any marketplace work and a set that identifies the period in which the mother first started working after giving birth.

Table 1 gives descriptive statistics for each set of maternal labor supply variables for the full sample and a “working” subsample of mothers who were at work within 3 months prior to giving birth. According to table 1, the number of 40-hour week equivalents worked by the full sample in the first quarter after giving birth is 2.6. Maternal work dramatically increases between the first and second quarter, with a more gradual increase occurring thereafter. Working mothers average more hours in each period, working the equivalent of 3.8 40-hour weeks in the first quarter and over 8 40-hour week equivalents in each of the remaining quarters. Thus, those who were attached to the labor force prior to giving birth are more likely to work after the pregnancy. A similar pattern is shown when measuring the number of weeks in which the mother worked, with working mothers averaging more weeks in each period. Working mothers also work a larger number of full-time weeks.

When considering whether a mother works at all, 44.5% of the full sample and 70.2% percent of working mothers work at some point during the first quarter. Ultimately, over 75% of both samples of mothers will work at some point during their child’s first 3 years, which is similar to what was found by Klerman and Leibowitz (1994). Additionally, many mothers first start working in the first quarter after giving birth. This number is larger for the working subsample than for the full sample, with a majority (70.2%) of working mothers returning to work in the first quarter. Over 63% of all mothers and 86% of working mothers start work in their infant’s first year.

To determine the effects of maternal marketplace work on a child’s cognitive development, I use three measures. These measures are the scores on the following tests: the Peabody Picture Vocabulary Test (PPVT), the Peabody Individual Achievement Test of Mathematics (PIAT-M), and the

Table 1 Employment Patterns and Child Outcomes of All Mothers and Working Mothers

All Mothers Working Mothers Mean SD Mean SD

No. of hours worked (divided by 40): Quarter 1 2.619 (3.742) 3.851 (4.012) Quarter 2 5.805 (6.182) 8.307 (5.864) Quarter 3 6.357 (6.240) 8.580 (5.766) Quarter 4 6.658 (6.301) 9.019 (5.793) Years 2 and 3 55.723 (44.610) 72.808 (38.631)

No. of weeks worked: Quarter 1 3.012 (4.012) 4.451 (4.210) Quarter 2 6.497 (6.189) 9.282 (5.864) Quarter 3 7.088 (6.230) 9.817 (5.325) Quarter 4 7.414 (6.218) 9.965 (5.290) Years 2 and 3 62.361 (43.170) 80.832 (33.789)

No. of weeks worked full time: Quarter 1 1.997 (3.412) 2.952 (3.815) Quarter 2 4.402 (5.976) 6.369 (6.282) Quarter 3 4.818 (6.072) 6.748 (6.276) Quarter 4 5.037 (6.186) 6.888 (6.349) Years 2 and 3 40.723 (44.301) 53.361 (44.769)

Worked at all: Quarter 1 .445 (.497) .702 (.457) Quarter 2 .527 (.499) .774 (.419) Quarter 3 .551 (.497) .783 (.413) Quarter 4 .565 (.496) .784 (.412) Years 2 and 3 .762 (.426) .920 (.271)

Started work: Quarter 1 .445 (.497) .702 (.457) Quarter 2 .107 (.309) .113 (.317) Quarter 3 .049 (.216) .039 (.193) Quarter 4 .030 (.171) .015 (.121) Years 2 and 3 .156 (.363) .077 (.267)

Dependent child outcome variables: PPVT 89.193 (18.955) 93.261 (17.829) PIAT-M 100.420 (13.038) 102.584 (12.309) PIAT-R 105.501 (13.276) 107.290 (13.146)

Note.—There are 3,103 mothers in the full sample and 1,767 in the working sample.

Peabody Individual Achievement Test of Reading Recognition (PIAT-R). The PPVT is a broad test measuring vocabulary knowledge and verbal aptitude based on a child’s hearing vocabulary. The PIAT-M measures cognitive achievement in mathematics and analytical ability. The PIAT-R is designed to measure reading comprehension based on a child’s ability to recognize and pronounce words. These tests have been found to be correlated with alternative measures of cognitive development, and each has a completion rate of about 90% (see Baker et al. 1993). The PPVT is administered to children aged 3 and 4, and the PIAT-M and the PIATR are given to children who are 5 years of age or older. Some children were assessed some of the tests more than once. When a particular test was taken multiple times, I follow Harvey (1999) and average the test scores and include only one test observation per child.14 I do this in order to increase the accuracy of the dependent variables.15 My sample has 2,022 PPVT scores, 1,658 PIAT-M scores, and 1,655 PIAT-R scores. Each test score has been standardized from a nationally representative sample.

Table 1 lists the test score averages and standard deviations for the full sample of mothers and for the subsample of working mothers. According to table 1, my samples score below the national average on the PPVT but above the average on the PIAT-M and PIAT-R. Also, children with working mothers score higher on the PPVT, PIAT-M, and PIAT-R. Thus, a child’s cognitive development is positively correlated with maternal work. However, these simple correlations do not necessarily imply a causal effect. Rather, mothers who are more firmly attached to the labor market may be mothers who have more ability, and this ability may be a family trait that is genetically passed to children.

I use multivariate regression analysis to determine the causal effects of maternal work on child development. Each of the models contains a “standard” set of explanatory variables that have been widely used in the literature. These variables are designed to control for the child’s characteristics and endowments and the mother’s preferences and productivity. Specifically, to control for differences between boys and girls, a child gender variable is included. I also include the child’s birth order to control for differences between first-borns and children with older siblings.16 Race is controlled for with a dummy variable for being black and a dummy variable for being of Hispanic origin. Blacks and Hispanics may perform

14 Earlier work that used NLSY data did not have access to multiple test scores per child because the only child data used (or available at the time) came from the 1986 survey (Desai et al. 1989; Parcel and Menaghan 1990; Baydar and Brooks-Gunn 1991; Belsky and Eggebeen 1991; Mott 1991; Blau and Grossberg 1992; Vandell and Ramanan 1992). Two others used the first test score available (Hill and O’Neill 1994; Greenstein 1995). Harvey (1999) takes an average of the scores, and Blau (1999) includes all test scores from all children regardless of whether a test had been taken previously.

15 To determine the importance of this decision, I reestimated the models twice, first using the first test score available and then including all test scores from all children. However, the results were very similar to the ones reported here. Thus, the decision of how to handle multiple test scores would not seem to be an important one. Perhaps this is not surprising since the methods use essentially the same data. However, the standard errors for average test scores are somewhat smaller than the standard errors for the “first test score” measure. This is also found by Currie and Thomas (1995) and may indicate that average test scores are a better measure. However, the “average test score” standard errors are not smaller than the standard errors when including each test score as a separate observation.

16 Hanushek (1992) finds that children born earlier in the birth order have enhanced child development.

poorly on these tests due to economic disadvantage or cultural bias. I include the child’s age in months to control for age differentials. Although these test scores are age-standardized, others have found that the child’s age has a significant effect (Hill and O’Neill 1994). I control for the child’s health with a dummy variable for low birth weight (birth weight of less than 5.5 pounds) and with a variable measuring the mother’s hospital stay in days after giving birth.17 These variables are included because children with health-related problems may be more likely to perform poorly due to delayed development. Continuing, the mother’s education level is included because maternal care by highly educated mothers may be of higher quality resulting in better child outcomes.18 The mother’s education level also measures the cognitive achievement or skills of the mother and may pick up attitudes about scholastic achievement. I include the mother’s age because other studies have shown that younger mothers are more likely to be of lower socioeconomic status (Harvey 1999), and socioeconomic status may affect child development (Desai et al. 1989; Vandell and Ramanan 1992). The mother’s marital status is included because the presence of the father or a spouse may increase parental time with the child. The “standard” set of variables also includes region dummy variables to control for regional differences and year dummy variables to control for time trends (see Hill and O’Neill 1994).

Table 2 explains how each standard explanatory variable is measured, giving each variable’s mean and standard deviation for the full sample of mothers and the working subsample. According to table 2, working mothers have slightly fewer children and are less likely to be black or Hispanic. Working mothers have shorter hospital stays, and their infants are less likely to be of low birth weight. As we would expect, working mothers have more education, are older, and are more likely to be married. Further, working mothers are somewhat more likely to reside in the south or the northeast. Thus, it appears that working mothers systematically differ from mothers who are not firmly attached to the labor force. Controlling for these observed differences will remove some of the potential for heterogeneity bias.

To further limit the potential for unobserved heterogeneity bias, I include a set of “supplemental” regressors that controls for additional background characteristics and the mother’s attitudes.19 This set includes a measure of the mother’s Armed Forces Qualifications Test (AFQT) score to control for the mother’s ability, which could be passed between generations. In particular, this measure is the residual from a regression of

17 Many others use a cut-off of 5.5 pounds (e.g., see Belsky and Eggebeen 1991).

18 Fleisher (1977) discovers that the benefit of maternal time on child outcomes is increasing in the mother’s education level.

19 This strategy is used by Ruhm (2000).

Table 2 Descriptive Statistics of Explanatory Variables for All Mothers and Working Mothers

All Mothers Working Mothers
Mean SD Mean SD
Standard variables:  
Child gender (p 1 if male, 0 if female) .520 (.500) .529 (.499)
Child’s birth order (parity) 2.097 (1.076) 1.796 (.895)
Black dummy variable (p 1 if child is    
black, 0 otherwise) .272 (.445) .218 (.413)
Hispanic dummy variable (p 1 if child    
is Hispanic, 0 otherwise) .195 (.396) .181 (.385)
Child’s age (in months) 68.113 (19.699) 67.501 (19.475)
Low birth weight (p 1if ! 5.5 pounds,    
0 otherwise) .076 (.266) .066 (.249)
Mother’s hospital stay (days after giving    
birth) 3.243 (4.451) 3.112 (3.606)
Mother’s education (level in years) 13.079 (2.177) 13.592 (2.138)
Mother’s age (years) 33.132 (2.555) 33.289 (2.544)
Mother’s marital status (p 1 if married,    
0 otherwise) .705 (.456) .787 (.410)
Northeast dummy variable (p 1if    
mother resides in northeast, 0    
otherwise) .158 (.365) .163 (.370)
South dummy variable (p 1 if mother    
resides in south, 0 otherwise) .361 (.480) .367 (.482)
West dummy variable (p 1 if mother    
resides in west, 0 otherwise) .192 (.394) .174 (.379)
Test taken in 1994 (p 1 if child took    
test in 1994, 0 otherwise) .371 (.483) .371 (.483)
Test taken in 1996 (p 1 if child took    
test in 1996, 0 otherwise) .428 (.495) .432 (.495)
Supplemental background variables:    
Mother’s AFQT residual test score    
(controlling for age) 1.409 (26.793) 7.802 (26.537)
Mother’s mother native born (p1if    
mother’s mother born outside the    
United States, 0 otherwise) .042 (.201) .040 (.195)
Mother spoke a foreign language (p1if    
mother spoke a foreign language, 0    
otherwise) .232 (.422) .223 (.416)
Mother lived with both parents (p 1if    
mother lived with both parents at age    
14, 0 otherwise) .584 (.493) .673 (.469)
Mother’s household received magazines    
(p 1 if mother’s household received    
magazines when mother was age 14, 0    
otherwise) .602 (.489) .674 (.469)
Mother’s household received newspapers    
(p 1 if mother’s household received    
newspapers when mother was age 14,    
0 otherwise) .754 (.431) .802 (.399)
Mother’s household had library card (p    
1 if mother’s household had a library    
card when mother was age 14, 0    
otherwise) .749 (.434) .782 (.413)
South dummy variable for age 14 (p 1    
if mother lived in the south at age 14,    
0 otherwise) .332 (.471) .337 (.473)
Early Maternal Employment and Child Development 421
Table 2 (Continued)  
All Mothers Working Mothers
Mean SD Mean SD
Urban dummy variable for age 14 (p 1  
if mother lived in an urban area at age  
14, 0 otherwise) .818 (.386) .812 (.391)
Grandmother’s education (education  
level in years of mother’s mother) 10.484 (3.619) 10.969 (3.389)
Grandmother’s education missing (p1if  
mother’s mother’s education level is  
missing, 0 otherwise) .037 (.190) .031 (.173)
Grandmother worked part time when  
mother was age 14 (p 1 if mother’s  
mother worked part time when  
mother was age 14, 0 otherwise) .202 (.402) .211 (.408)
Grandmother worked full time when  
mother was age 14 (p 1 if mother’s  
mother worked full time when  
mother was age 14, 0 otherwise) .385 (.487) .419 (.494)
Grandfather’s education (education level  
in years of mother’s father) 9.643 (5.177) 10.468 (4.892)
Grandfather’s education missing (p1if  
mother’s father’s education level is  
missing, 0 otherwise) .134 (.340) .097 (.296)
Mother’s siblings (no. of mother’s  
siblings) 3.972 (2.622) 3.674 (2.450)
Mother’s Rotter test score (higher scores  
p less efficacy) 8.820 (2.358) 8.616 (2.325)
Mother’s attitudes about family roles  
(higher scores p traditional views) 17.112 (3.483) 16.693 (3.445)
Income regressor:  
Family income (in $1,000s) 26.536 (13.455) 30.677 (13.222)
Childcare regressors:  
Care by relative (p 1 if child received  
this care, 0 otherwise) .250 (.433) .317 (.465)
Care by nonrelative (p1 if child re  
ceived this care, 0 otherwise) .249 (.432) .359 (.479)
Care in center, such as group, nursery,  
or preschool care (p 1 if child re  
ceived this care, 0 otherwise) .209 (.405) .277 (.447)

Note.—There are 3,103 mothers in the full sample and 1,767 in the working sample.

the mother’s AFQT score on age dummy variables. I use this residual in place of the actual AFQT score to control for pure age effects.20 This set also includes a dummy variable that measures whether the mother’s mother is native born and characteristics of the mother’s household when she was 14 years of age, such as dummy variables measuring whether a foreign language was spoken, whether the mother lived with both parents,

20 More than 90% of the original NLSY sample were administered the AFQT test in 1980 when the respondents were between the ages of 15 and 23. The AFQT residual scores are obtained by regressing AFQT scores on age dummy variables. Each mother’s predicted AFQT score is then differenced from the actual AFQT score to get the residual. For those in my sample who were not administered the test, a value is assigned using the race-specific mean from the age regression.

whether the mother’s household took a magazine subscription and a newspaper, whether someone in the mother’s household had a library card, whether the mother lived in the south and an urban area, and whether the mother’s mother worked full time and part time. These background variables also include the education level of the mother’s parents and corresponding dummy variables indicating that each parent’s education level was unknown.21 This set also controls for the mother’s number of siblings, as well as for her efficacy and attitudes. The mother’s Rotter test score proxies for the mother’s motivation and efficacy.22Higher Rotter scores indicate greater efficacy or control over life events. The “attitude” variable summarizes the degree to which the mother agrees with statements such as “the woman’s place is in the home” and “traditional husband/wife roles work best.”23 Higher values indicate more traditional views about family roles.

Table 2 gives descriptive statistics for the supplemental background variables for the full sample and for the subsample of working mothers. As we would expect, working mothers have higher AFQT test scores, indicating more ability. Working mothers are less likely to be foreign born or to have spoken a foreign language in their childhood household. Further, working mothers are more likely to have lived with both parents and to have had access to a magazine, newspaper, and library card. The parents of working mothers tend to have more education, and their mothers were more likely to have worked both part time and full time. Working mothers also have fewer siblings. Finally, working mothers have a stronger sense of efficacy and less traditional views about family roles.

Models that include the “standard” and “supplemental” variables are designed to provide the total effect of maternal work. To get the total effect, explanatory variables that will change in response to a change in

21 This method of controlling for missing background information has been used by Blau and Grossberg (1992), Hill and O’Neill (1994), and Ruhm (2000).

22 I follow Hill and O’Neill (1994) in summing the respondent’s answers to four questions about efficacy from the original Rotter test score to arrive at a composite Rotter index score. These questions address whether the respondent feels he or she is in control of life’s events, whether planning ahead is effective, the degree to which luck affects outcomes, and whether the respondent can infuence things that happen. The Rotter index scores range from 4 to 16.

23 There are eight questions in this series, which include, e.g., “a wife with a family has no time for other employment,” “working wives feel most useful,” “employment of wives leads to juvenile delinquency,” “inflation necessitates the employment of both parents,” “men should share housework,” and “women are happiest in the home.” Mothers respond with the degree to which they agree with each statement. I then code each response according to the degree to which it indicates a preference toward traditional roles. To arrive at a composite “attitude” variable, I sum the mother’s responses to the eight questions, where higher scores are associated with more traditional attitudes about a mother’s family role.

maternal labor supply should not be included. Therefore, family income, which may increase with maternal labor supply, is absent from this list of variables.

However, family income will have an effect on child outcomes if greater income increases the amount of money that can be spent on inputs that contribute to the child’s well-being. For example, families with more money can purchase more books and take more family vacations, which may enhance child development.24Maternal marketplace work should increase family income. For comparison purposes, I include family income in a portion of the models to see how the effect of maternal marketplace work changes. Again, models that include family income will show only the partial effect of maternal labor supply because the effect of maternal work will be found holding income constant.

When family income is included, it consists of alimony, child support, wages and salary earned by the mother and/or her spouse, income from farm and/or business after expenses, and income from savings accounts and assets. Invalid missing values for any of the five components of family income are replaced with the sample average for that component from the appropriate year. Valid missing information is replaced with a zero. All income figures are deflated using the Consumer Price Index (1987 p 100). The final measure of family income used in the estimation equals the average annual family income during the child’s first 3 years in $1,000s. According to table 2, average family income for the full sample is about $26,000. However, the working mothers’ average family income is only slightly higher: $30,677.

Also, for comparison purposes, I include controls for the types of nonmaternal childcare used.25 This is done because the effect of maternal marketplace work will be partially determined by the quality of non-maternal childcare arrangements. For example, if a high-quality childcare arrangement is used in place of maternal childcare, then the potential detrimental effects of maternal labor supply may be reduced. However, results that control for nonmaternal childcare arrangements must be interpreted with caution: selected childcare arrangements may be partially determined by maternal labor supply decisions. If this is true, then es

24 Blau (1999) examines the effect of income on child outcomes. He finds that current income generally has a small effect on child development. Permanent income has a larger positive effect, but its effect is still small.

25 Others have examined the effects of childcare arrangements. In particular, recent studies by the National Institute of Child Health and Human Development (NICHD) Early Child Care Research Network have examined the factors that determine the characteristics of nonmaternal childcare selected (NICHD 1997b) and the effects of nonmaternal childcare on child outcomes such as the infant-mother attachment (NICHD 1997a).

timates from these models will not give the total effect of maternal labor supply.

The controls for nonmaternal childcare arrangements take the form of dummy variables that equal 1 if the child spent at least 1 month in that particular type of care during the first 3 years. The types of childcare include care by a relative, care by a nonrelative, and formal care in a center (such as a preschool, nursery, or day care center). If the mother reports that no nonmaternal childcare arrangements were used, then the childcare mode dummy variables for that child are set to 0. Therefore, my models do not eliminate any children with valid missing childcare mode information.26 Table 2 indicates the percent of children who spent at least 1 month in each type of care. Between 20% and 30% of all children in the full sample receive each type of care. Care by a relative is used most frequently, and care in a center is used the least. More working mothers use each mode of care, and the type of care selected most frequently by working mothers is care by a nonrelative.

IV. The Model

I assume that households maximize utility over consumption goods, leisure time, and child development subject to a budget constraint, a time constraint, and a child development production function. The child development production function is indexed by the child’s standardized test scores of cognitive development, and the arguments of this production function are inputs such as the time of the mother, consumption goods (food, health care, educational trips, etc.), endowments, and background characteristics. The decision-making unit is assumed to be a household that includes at least a mother and one child, and expectations are a function of current and past values of variables.

The optimization problem can be solved for the maternal labor supply and child outcome functions in terms of exogenous variables such as background characteristics. The empirical analysis will estimate these functions in reduced form in that the relationship between the functions’ determinants and preferences, productivity, and expectations will not be identified. It would be difficult to separately identify these effects because strong identifying assumptions would be required.

The child development production function will be estimated with maternal marketplace work as an argument. Maternal marketplace work should be included if child development depends on past values of such

26 Very few of the mothers who report that their child was in a regular childcare arrangement fail to characterize the arrangements. Instead, in most cases, if no childcare modes are reported, then the mother previously responded that no regular childcare arrangements were used.

work. More formally, assume that the development outcomes for child i ( ) are specified as

Yi

Yik p Xbk � H ak � eik,

ik ik

for k p PPVT , PIAT-M, and PIAT-R, where X is a vector of exogenous explanatory variables (such as background characteristics), H is a vector of maternal labor supply variables, b and a are vectors of coefficients to be estimated, and e is the disturbance. Expectations about future values of exogenous variables should also be included in X but will instead be represented by current and past values used as proxies for future values. Thus, the variables in X capture the direct effects of these variables and their effects via expectations.

Ideally, X would include all factors that affect child development. However, if some of these factors are unobserved (or unmeasurable) by the researcher, then the error term will capture these unobserved variables. Too, unobserved heterogeneity will produce biased results if the error term and any included explanatory variable are correlated with the same unobserved characteristics. This will be true of the maternal labor supply variables if working mothers systematically differ in unobserved ways from mothers who do not work.27 For example, suppose that mothers who work have high ability and high ability is a family trait that is passed along to children, resulting in better child outcomes. If these family traits are unobserved, then the error term in the child outcome equation will be correlated with maternal marketplace work, which is an explanatory variable in that equation. If this is the case, then maternal labor supply and the child outcomes are correlated with the same unobserved family traits, which will produce bias results. If these unobserved factors are not adequately controlled for, then maternal work may spuriously appear to affect child development when, in fact, there may not be a causal relationship.

I control for unobserved heterogeneity in three ways. First, I include an unusually large number of explanatory variables to control more extensively for background characteristics that reflect differences between working and nonworking mothers. Second, a portion of the models that I estimate contain only mothers who were working before giving birth. This sample should be less susceptible to unobserved heterogeneity bias because these mothers should be more homogeneous. Finally, I employ

27 Empirical evidence suggests that employed mothers do systematically differ from mothers who are not employed. For example, McCartney (1984) finds that the child-rearing views of employed mothers are “less traditional” than those of nonemployed mothers. Hock, DeMeis, and McBride (1988) find that employed mothers are more willing to be separated from their infant during a portion of the day. Finally, Greenberger and Goldberg (1989) discover that employed mothers are more attached to the labor force and their career.

the standard instrumental variables (IV) procedure to predict maternal labor supply in a portion of the models.28 Following Hill and O’Neill (1994), I identify maternal labor supply with variables representing local labor market conditions that may affect labor supply but not child development. These variables include the local unemployment rate; potential earnings (proxied by the local per capita income); the percentage of the local labor force that is female; the percentage of the local population that is urban and female; local population; percentage of the local population with a high school education and a college education; percentage of the local population employed; the percentage of the local labor force in manufacturing, in wholesale/retail trade, in services, in government work, and self-employed; and “per capita government transfer payments” such as Aid to Families with Dependent Children (AFDC).29 The F-tests fail to reject the null hypothesis that the local labor market conditions have no effect on the child development test scores, and Hoynes (2000) shows that local labor market conditions are valid determinants of labor supply.

Tables 3 and 4 present the results from the first-stage models that predict maternal labor supply. The first-stage results are similar to those typically found in the literature. For example, maternal labor supply is significantly decreasing in the number of children present, nonwage income, and traditional attitudes (traditional views about family roles) and is significantly increasing in education, marital status, and ability (as measured by AFQT residuals). The mother’s labor supply is also significantly higher if her mother worked part time or full time. Some of the local labor demand conditions are also significant: maternal labor supply is increasing in the percentage of the local population with a high school education and the percentage employed, and maternal labor supply is decreasing in the percentage with a college education. Further, R2 values range from 0.132 to 0.225, and the average predicted values of the maternal labor supply variables are within 2%–3% of the actual sample averages.30

28 I also considered a fixed-effects estimator, which would require either selecting mothers who gave birth to two children between 1988 and 1993 or mothers with a sister in the NLSY. I chose not to take this approach because selecting a sample based on that criteria might induce bias and would result in a significantly reduced size.

29 These equations also include the number of children present, dummy variables for being black and Hispanic, a low birth weight dummy variable, the length of the mother’s hospital stay, the mother’s education, the mother’s age, the mother’s marital status, nonwage income, region dummy variables, and all of the supplemental background characteristics described in Sec. III.

30 For example, the predicted 40-hour week equivalent worked in the first quarter is 2.598 (1.452); in the second, third, and fourth quarters, it is 18.778 (8.490); in the first year, it is 21.376 (9.640); in the second and third years, it is 54.881 (21.749).

V. Results

I estimate various specifications of the model with the full sample of mothers. Results from the maternal labor supply variables are presented in tables 5, 6, and 7. I then reestimate the model specifications with the subsample of working mothers. The maternal labor supply coefficients from these models are presented in table 8. Appendix tables A1 and A2 present the sample results from the nonmaternal employment explanatory variables on PPVT, PIAT-M, and PIAT-R.

A. The Full Sample of Mothers

My first specification includes only the “standard” variables with no controls for unobserved heterogeneity. Model 1 includes two maternal labor supply variables, which equal the number of hours worked (divided by 40) in the child’s first year and in the child’s second and third years. The number of hours worked has statistically insignificant effects on the measures of cognitive development, though these effects are consistently negative in the first year and positive in the second and third years. In model 2, I separately identify the effects of maternal labor supply in the first quarter after the child’s birth.31 Hours worked in the first quarter have a significant negative effect in the PPVT model but insignificant positive effects in the PIAT-M and PIAT-R models. In fact, switching from no work to working 40 hours in the first quarter lowers PPVT scores by about 4.2 points, which is 22% of a standard deviation. Hours worked in the remaining 3 quarters of the first year have statistically insignificant effects, and the other effects are similar to model 1.

Next, I examine model specifications that include the supplemental background characteristics in addition to the standard variables.32 Model 3 shows the effects of hours worked in the first year and the second and third years, and model 4 separately identifies the effects of such work in the first quarter. In model 3, the effects of hours worked in the first year on cognitive development become more negative in the PIAT-M and PIAT-R models (and slightly less negative in the PPVT model) when supplemental background characteristics are held constant. In fact, such work in the first year now has a marginally significant negative effect in the PIAT-M model, with switching from no work to working 40 hours in each of the 52 weeks lowering this score by about three points. This reduction represents about 23% of a standard deviation in PIAT-M scores. However, the positive coefficients on hours worked in the second and

31 I also estimated models with a maternal labor supply variable for each of the first 4 and 8 quarters. The results were generally similar to the ones reported here, though estimated less precisely.

32 The F-tests reject the null hypothesis that the set of supplemental background characteristic variables has no effect at the 5% level in each model.

Table 3 First-Stage Maternal Labor Supply Equations

Quarter 1 Quarters 2, 3, and 4
Coefficient SE Coefficient SE
Intercept �1.430 (6.566) �78.342** (30.390)
No. of children mother has �.260* (.111) �2.960** (.515)
Black dummy variable (p 1 if child is    
black, 0 otherwise) �.084 (.361) 1.733 (1.672)
Hispanic (p 1 if child is Hispanic, 0    
otherwise) �.389 (.497) 2.294 (2.300)
Low birth weight (p 1if ! 5.5. pounds, 0    
otherwise) �.367 (.391) �1.515 (1.807)
Mother’s hospital stay (days after giving    
birth) �.011 (.030) .095 (.138)
Mother’s education (level in years) .187** (.062) 1.316** (.287)
Mother’s age (years) .044 (.044) .586** (.206)
Mother’s marital status (p1 if married, 0    
otherwise) .549* (.289) 5.721** (1.336)
Nonwage income ($1,000) �22.443** (6.851) �199.880** (31.708)
Northeast dummy variable (p 1 if mother    
resides in northeast, 0 otherwise) �.427 (.270) �3.234** (1.235)
South dummy variable (p 1 if mother re    
sides in south, 0 otherwise) .764* (.364) 5.449** (1.662)
West dummy variable (p 1 if mother re    
sides in west, 0 otherwise) �.067 (.296) �1.323 (1.351)
Mother’s AFQT residual test score (con    
trolling for age) .012* (.005) .062* (.025)
Mother’s mother native born (p1if    
mother’s mother born outside the    
United States, 0 otherwise) �.636 (.581) �2.812 (2.688)
Mother spoke a foreign language (p1if    
mother spoke a foreign language, 0    
otherwise) .218 (.445) 1.356 (2.059)
Mother lived with both parents (p 1if    
mother lived with both parents at age    
14, 0 otherwise) .339 (.266) .531 (1.229)
Mother’s household received magazines (p    
1 if mother’s household received maga    
zines when mother was age 14, 0    
otherwise) �.664** (.244) �3.395** (1.129)
Mother’s household received newspapers (p    
1 if mother’s household received news    
papers when mother was age 14, 0    
otherwise) �.011 (.271) 1.437 (1.256)
Mother’s household had library card (p 1    
if mother’s household had a library card    
when mother was age 14, 0 otherwise) �.145 (.263) �2.153� (1.219)
South dummy variable for age 14 (p 1if    
mother lived in the south at age 14, 0    
otherwise) .103 (.372) 2.304 (1.724)
Urban dummy variable for age 14 (p 1if    
mother lived in an urban area at age 14,    
0 otherwise) �.469� (.262) �.506 (1.214)

428

Table 3 (Continued)

Quarter 1 Quarters 2, 3, and 4
Coefficient SE Coefficient SE
Grandmother’s education (education level  
in years of mother’s mother) �.098* (.047) �.372� (.219)
Grandmother’s education missing (p1if    
mother’s mother’s education level is    
missing, 0 otherwise) �.718 (.786) �2.786 (3.636)
Grandmother worked part time when    
mother was age 14 (p 1 if mother’s    
mother worked part time when mother    
was age 14, 0 otherwise) .539* (.271) 2.389� (1.256)
Grandmother worked full time when    
mother was age 14 (p 1 if mother’s    
mother worked full time when mother    
was age 14, 0 otherwise) .573* (.228) 1.993� (1.054)
Grandfather’s education (education level in    
years of mother’s father) .002 (.038) �.084 (.176)
Grandfather’s education missing (p1if    
mother’s father’s education level is miss    
ing, 0 otherwise) .744 (.502) �2.117 (2.323)
Mother’s siblings (no. of mother’s siblings) .018 (.043) �.081 (.199)
Mother’s Rotter test score (higher scores p    
less efficacy) �.016 (.043) �.057 (.199)
Mother’s attitudes about family roles    
(higher scores p traditional views) �.051� (.032) �.428** (.147)
Local unemployment rate �.627 (.516) 1.646 (2.388)
Local per capita income .038 (.040) �.032 (.184)
Percent of labor force female �2.716 (4.469) �34.941� (20.686)
Percent of population urban .652 (.906) �2.719 (4.195)
Percent of population female 7.007 (11.710) 88.420� (54.196)
Local population (/10,000) .006 (.005) �.023 (.022)
Percent of local population with high    
school education .088 (2.902) 31.095* (13.433)
Percent of local population with college    
education �8.645� (4.500) �51.932* (20.829)
Percent of local population employed 4.637 (4.056) 64.439** (18.774)
Percent of local labor force in    
manufacturing �.815 (2.170) 4.286 (10.045)
Percent of local labor force in trade �2.491 (5.892) 18.601 (27.270)
Percent of local labor force in services �3.432 (5.247) 37.848 (24.287)
Percent of local labor force in government 5.035 (3.650) 23.732 (16.895)
Percent of local labor force self-employed �3.648 (5.544) �16.323 (25.659)
Per capita local government transfer    
payments �.367� (.232) 1.258 (1.075)

Note.—For the first quarter equation, R2 is .132; for the second, third, and fourth quarters equation, it is .210. The dependent variables are the number of 40-hour week equivalents worked in each period by the mother.

� p ! .10.

* p ! .05. ** p ! .01.

429

Table 4 First-Stage Maternal Labor Supply Equations

Year 1 Years 2 and 3
Coefficient SE Coefficient SE
Intercept �79.772* (35.084) �102.812 (75.126)
No. children mother has �3.219** (.594) �6.780** (1.273)
Black dummy variable (p 1 if child is    
black, 0 otherwise) 1.650 (1.930) 7.621� (4.132)
Hispanic (p 1 if child is Hispanic, 0    
otherwise) 1.905 (2.655) 5.786 (5.685)
Low birth weight (p 1if ! 5.5. pounds, 0    
otherwise) �1.882 (2.087) .037 (4.468)
Mother’s hospital stay (days after giving    
birth) .084 (.159) .008 (.341)
Mother’s education (level in years) 1.504** (.331) 3.456** (.709)
Mother’s age (years) .631** (.237) .642 (.508)
Mother’s marital status (p1 if married, 0    
otherwise) 6.270** (1.542) 15.821** (3.302)
Nonwage income ($1,000) �222.323** (36.604) �309.440** (78.382)
Northeast dummy variable (p 1if    
mother resides in northeast, 0    
otherwise) �3.661** (1.428) �5.005� (3.067)
South dummy variable (p 1 if mother re    
sides in south, 0 otherwise) 6.212** (1.922) 16.587** (4.127)
West dummy variable (p 1 if mother re    
sides in west, 0 otherwise) �1.390 (1.563) 3.413 (3.356)
Mother’s AFQT residual test score (con    
trolling for age) .074** (.029) .103* (.061)
Mother’s mother native born (p1if    
mother’s mother born outside the    
United States, 0 otherwise) �3.448 (3.103) �8.929 (6.645)
Mother spoke a foreign language (p1if    
mother spoke a foreign language, 0    
otherwise) 1.574 (2.377) �.627 (5.090)
Mother lived with both parents (p 1if    
mother lived with both parents at age    
14, 0 otherwise) .869 (1.419) �.035 (3.039)
Mother’s household received magazines (p    
1 if mother’s household received maga    
zines when mother was age 14, 0    
otherwise) �4.059** (1.303) �7.869** (2.790)
Mother’s householdreceivednewspapers(p    
1 if mother’s household received news    
papers when mother was age 14, 0    
otherwise) 1.425 (1.450) 1.499 (3.104)
Mother’s household had library card (p 1    
if mother’s household had a library    
card when mother was age 14, 0    
otherwise) �2.299� (1.407) �1.314 (3.014)
South dummy variable for age 14 (p 1if    
mother lived in the south at age 14, 0    
otherwise) 2.407 (1.990) 1.957 (4.261)

430

Table 4 (Continued)

Year 1 Years 2 and 3
Coefficient SE Coefficient SE
Urban dummy variable for age 14 (p 1if  
mother lived in an urban area at age 14,  
0 otherwise) �.975 (1.402) �6.655* (3.002)
Grandmother’s education (education level    
in years of mother’s mother) �.470� (.253) �1.454** (.541)
Grandmother’s education missing (p1if    
mother’s mother’s education level is    
missing, 0 otherwise) �3.504 (4.198) �8.327 (8.989)
Grandmother worked part time when    
mother was age 14 (p 1 if mother’s    
mother worked part time when mother    
was age 14, 0 otherwise) 2.928* (1.450) 5.708� (3.105)
Grandmother worked full time when    
mother was age 14 (p 1 if mother’s    
mother worked full time when mother    
was age 14, 0 otherwise) 2.566* (1.217) 8.478* (2.606)
Grandfather’s education (education level    
in years of mother’s father) �.082 (.203) �.390 (.435)
Grandfather’s education missing (p1if    
mother’s father’s education level is    
missing, 0 otherwise) �1.373 (2.682) �11.867� (5.743)
Mother’s siblings (no. of mother’s    
siblings) �.063 (.230) �.438 (.492)
Mother’s Rotter test score (higher scores p    
less efficacy) �.073 (.230) .472 (.493)
Mother’s attitudes about family roles    
(higher scores p traditional views) �.479** (.170) �2.161** (.364)
Local unemployment rate 1.019 (2.757) �4.655 (5.903)
Local per capita income .007 (.212) .302 (.454)
Percent of labor force female �37.657 (23.880) �95.537� (51.136)
Percent of population urban �2.067 (4.843) �7.693 (10.371)
Percent of population female 95.427 (62.565) 212.392 (133.972)
Local population (/10,000) �.017 (.025) �.012 (.053)
Percent of local population with high    
school education 31.183* (15.508) 54.469� (33.207)
Percent of local population with college    
education �60.576* (24.046) �153.094** (51.491)
Percent of local population employed 69.075** (21.674) 140.216** (46.411)
Percent of local labor force in    
manufacturing 3.470 (11.596) 2.527 (24.832)
Percent of local labor force in trade 16.110 (31.482) �24.565 (67.412)
Percent of local labor force in services 34.416 (28.037) 162.945** (60.037)
Percent of local labor force in government 28.767 (19.504) 34.023 (41.765)
Percent of local labor force self-employed �19.971 (29.622) �68.051 (63.430)
Per capita local government transfer    
payments .890 (1.241) .448 (2.657)

Note.—For the first quarter equation, R2 is .203; for the second, third, and fourth quarters equation, it is .225. The dependent variables are the number of 40-hour week equivalents worked in each period by the mother.

� p ! .10.

* p ! .05. ** p ! .01.

431

Table 5 The Effect of Maternal Work on Child Development (Full Sample): Models 1–6

PPVT PIAT-M PIAT-R Coefficient SE Coefficient SE Coefficient SE

Model 1—hours worked with standard regressors (OLS): Year 1 �.028 (.035) �.046 (.032) �.017 (.030) Years 2 and 3 .011 (.016) .013 (.014) .011 (.014)

Model 2—hours worked with standard regressors (OLS): Quarter 1 �.328* (.169) .018 (.157) .048 (.150) Quarters 2, 3, and 4 .032 (.048) �.058 (.043) �.029 (.041) Years 2 and 3 .005 (.016) .014 (.015) .012 (.014)

Model 3—hours worked with standard � supplemental regressors (OLS): Year 1 �.024 (.034) �.057� (.031) �.027 (.029) Years 2 and 3 .007 (.016) .019 (.014) .015 (.013)

Model 4—hours worked with standard � supplemental regressors (OLS): Quarter 1 �.316� (.165) �.003 (.154) .072 (.144) Quarters 2, 3, and 4 .035 (.047) �.068� (.040) �.046 (.039) Years 2 and 3 .001 (.016) .020 (.014) .016 (.014)

Model 5—hours worked with standard � supplemental regressors (IV): Year 1 �.043 (.165) �.126 (.149) �.053 (.140) Years 2 and 3 �.041 (.079) �.025 (.072) .001 (.068)

Model 6—hours worked with standard � supplemental regressors (IV): Quarter 1 �.182 (.669) .355 (.614) �.355 (.574) Quarters 2, 3, and 4 �.026 (.184) �.190 (.169) �.013 (.158) Years 2 and 3 �.039 (.080) �.027 (.072) .001 (.068)

Note.—There are 2,022 observations in the PPVT model, 1,658 in the PIAT-M model, and 1,655 in the PIAT-R model. The R2 values range from .14 in PIAT-R model 1 to .36 in PPVT model 4.

� p ! .10.

* p ! .05.

third years become larger in PIAT-M and PIAT-R models 3 and 4 (and smaller in PPVT models 3 and 4). In model 4, the effects of hours worked in the first quarter do not change substantially, with the negative effect of maternal labor supply in the first quarter in the PPVT remaining marginally significant. However, in PIAT-M model 4, the effect of hours worked in the second, third, and fourth quarters now has a marginally significant negative effect. In general, these results suggest that controlling for additional background characteristics has a relatively small effect on the maternal labor supply coefficients.

Models 5 and 6 include the standard and supplemental variables and control for unobserved heterogeneity by instrumenting maternal labor supply. The negative coefficients on hours worked in the first year increase

Table 6 The Effect of Maternal Work on Child Development (Full Sample): Models 7–10

PPVT PIAT-M PIAT-R Coefficient SE Coefficient SE Coefficient SE

Model 7—weeks worked (OLS): Year 1 �.029 (.033) �.060* (.030) �.012 (.028) Years 2 and 3 .013 (.016) .025� (.014) .008 (.013)

Model 8—weeks worked (OLS): Quarter 1 �.126 (.144) .008 (.133) .145 (.124) Quarters 2, 3, and 4 �.009 (.044) �.074* (.039) �.044 (.037) Years 2 and 3 .011 (.016) .026� (.015) .010 (.014)

Model 9—nonlinear effects (OLS): Worked dummy variables (p1

if worked in the period): Worked (year 1) 1.317 (1.588) �.083 (1.481) .162 (1.391) Worked (years 2 and 3) 1.792 (1.578) 1.733 (1.487) .748 (1.396)

Hours worked (in the period): Hours (year 1) �.044 (.040) �.059� (.037) �.030 (.035) Hours (years 2 and 3) �.004 (.018) .010 (.016) .010 (.015)

Model 10—nonlinear effects (OLS): Worked dummy variables (p 1

if worked in the period) Worked (quarter 1) .486 (1.520) �1.279 (1.435) .795 (1.349) Worked (quarters 2, 3, and

4) 1.691 (1.731) 1.178 (1.604) .111 (1.504) Worked (years 2 and 3) 1.546 (1.598) 1.445 (1.503) .704 (1.411)

Hours worked (in the period): Hours (quarter 1) �.336� (.191) .084 (.179) .024 (.168) Hours (quarters 2, 3, and 4) �.003 (.057) �.082� (.051) �.057 (.048) Hours (years 2 and 3) �.008 (.018) .012 (.016) .012 (.015)

Note.—There are 2,022 observations in the PPVT model, 1,658 in the PIAT-M model, and 1,655 in the PIAT-R model.

� p ! .10.

* p ! .05.

in magnitude in model 5. Further, the coefficients on hours worked in the second and third years are now negative in the PPVT and PIAT-M models (and are less positive in the PIAT-R model). This suggests that cognitive development and maternal labor supply are positively correlated with unobserved factors. However, these effects are measured imprecisely.

Next, I identify the effects of maternal labor supply in the first quarter. Model 6 shows that hours worked in the first quarter now has a larger negative coefficient in the PIAT-R model but a smaller negative coefficient in the PPVT model that is no longer statistically significant. Further, this coefficient in the PIAT-M model becomes positive. However, the instrumental variables effects are again imprecisely measured.

Because the instrumental variables (IV) estimator is inefficient relative to OLS, it is useful to test whether the IV and OLS estimates are statistically different. If the estimates are not statistically different from one another, then we can conclude that there exists no heterogeneity bias. To do this, I performed Hausman tests between models 3 and 5 and between models 4 and 6 for each test score. In each case, the null hypothesis that the estimates are the same cannot be rejected. Specifically, the Hausman x 2 test statistics range from 0.99 (probability that the Hausman test statistic is greater than the critical value p 0.60) to 3.20 (probability that the Hausman test statistic is greater than the critical value p 0.35). Because the Hausman tests suggest that the OLS estimates are consistent, I assume that no heterogeneity bias exists and use OLS to estimate the remaining models. However, it should be noted that the OLS and IV estimates may appear to be similar because the IV standard errors are estimated imprecisely. This could result from collinearity between the first-stage instruments and the second-stage regressors. However, this problem should be minimized by the exclusion of 15 instruments (the local labor market conditions) from the second-stage test score regressions.

Next, in order to determine the robustness of these results, I estimate additional models that contain alternative specifications of maternal labor supply. These models also contain the standard and supplemental variables. Model 7 specifies maternal labor supply to equal the number of weeks worked in the first year and in the second and third years, and the results are largely consistent with the previous models. Weeks worked in the first year have negative effects on the measures of cognitive development, with this effect being significant in the PIAT-M model. There is some evidence that weeks worked in the second and third years increase cognitive development, with the effect in the PIAT-M model being marginally significant. Next, I separately identify weeks worked in the first quarter in model 8. However, such work in the first quarter does not have a significant effect in any of the models. There is some evidence that weeks worked in the remainder of the first year reduce these scores, with this effect being statistically significant in the PIAT-M model.

I also examine the number of weeks worked full time (over 35 hours). The results (not shown) are similar to those of models 7 and 8. Weeks worked full time tend to reduce scores of cognitive development in the first year and increase such scores in the second and third years, with significant effects in some of the models. Full-time weeks worked in the first quarter have detrimental effects on the PPVT and PIAT-R, but none of these effects are statistically significant. Additionally, I specify the maternal labor supply variables to be dummy variables that equal 1 if any work was done (results not shown). However, none of these effects are statistically significant. The effects of working in the first quarter were also identified, but no consistent patterns emerged.33

Next, I determine whether maternal labor supply has nonlinear effects

33 I also examined models that specify the maternal labor supply to be dummy variables that equal 1 if the mother started working in that period. However, these models produced few significant results.

on child development. To do this, I estimate a model that includes dummy variables indicating whether any work was done and the “hours worked” variables. Model 9 identifies these effects in the first year and in the second and third years. The results suggest that the dummy variables do not have detrimental effects but the hours of work variables for the first year reduce the measures of cognitive development. This effect is statistically significant in the PIAT-M model. Model 10 examines whether maternal labor supply has nonlinear effects in the first quarter. These results show that each hour worked in the first quarter has a detrimental effect that is statistically significant in the PPVT model. Each work hour in the remainder of the first year decreases all three test scores, with this effect being significant in the PIAT-M model.

Models 1–10 show the total effect of maternal labor supply. However, maternal labor supply may enhance cognitive development by increasing family income. For comparison purposes, I reestimate the models with family income included. I present a representative set of the estimates in table 7. Models 11 and 12 are reestimates of models 3 and 4 from table 5 with family income included. These models show that family income has a statistically significant effect. For example, increasing family income by 20 ($20,000) enhances child development test scores by an average of about .9, which is about 6% of a standard deviation. Further, including family income causes the maternal labor supply coefficients to become more negative (or less positive), which indicates that maternal marketplace work has a positive side effect on child development by increasing family income. For example, hours worked in the first year now have a negative effect that is statistically significant at the 5% level in the PIAT-M model, and the negative effect of hours worked in the second, third, and fourth quarters of the first year becomes significant in PIAT-M model 12. Increasing maternal labor supply by 40 hours in each of the 52 weeks in the first year now decreases the PIAT-M score by 3.5 points (as compared with about 3 points in model 3), which is 0.5 or about 4% of a standard deviation more. This change in the maternal labor supply coefficients is about what we would expect given the size of family income’s direct impact on child development.

These results suggest that the negative effects of maternal employment are partially attenuated by increased family income. For example, assume that changing from no work to working 20 hours (in each of the 52 weeks of the child’s first year) increases family income by $20,000. According to model 11, increasing maternal weekly work by 20 hours (in the first year) decreases the measures of cognitive development by an average of about 8% of a standard deviation, while the $20,000 increase in family income increases the cognitive development scores by an average of about 6% of a standard deviation. Therefore, whether there is a net detrimental effect from maternal employment depends on the amount of additional family income that each mother’s work generates.

For comparison purposes, I also include controls for the mode of non-maternal childcare used. These results show whether controlling for non-maternal childcare arrangements changes the effects of maternal work. I tried various specifications for the childcare mode variables but the results were similar.34Models 13 and 14 in table 7 (which reestimate models 11 and 12 including the childcare mode variables) are representative of the results with the childcare mode variables included. Relative care tends to have a positive effect on cognitive development, and nonrelative care and center care have mixed results, but none of the childcare mode variables are statistically significant. However, the effects of some of the maternal labor supply variables become more negative. For example, in model 13, maternal labor supply in the first year now reduces PIAT-M scores by over 4 points, and this negative effect is now statistically significant in the PIAT-R model. Further, in model 14, the effect of maternal labor supply in the second, third, and fourth quarters significantly decreases PIAT-M and PIAT-R scores by an average of 3.2 points, which is about 25% of a standard deviation. However, the negative effect of hours worked in the first quarter is smaller in the PPVT model. Although controlling for childcare mode appears to have an effect on some of the maternal labor supply coefficients, it should be noted that the childcare mode data are unable to thoroughly explore the linkages between childcare and maternal labor supply because childcare mode does not necessarily reflect childcare quality.

B. The Subsample of Working Mothers

I next examine the sample of “working” mothers, that is, those mothers who were working at some point during the 3 months prior to giving birth. Each model estimated in this section includes the standard and supplemental regressors, and the results are presented in table 8. I first examine the effects of the number of hours worked in model 1. The results from this model are surprisingly similar to results with the full sample of mothers. This indicates that there is little systematic difference between the full sample and the subsample of mothers who are more firmly attached to the labor force. There is some evidence that hours worked have a detrimental effect on cognitive development in the first year, with this effect being statistically significant in the PPVT model at the 5% level. There is little evidence that hours worked have a significant effect in the second and third years. In model 2, I identify the effects of hours worked

34 These specifications included separate childcare mode dummy variables for each of the child’s first 3 years.

Table 7 The Effect of Maternal Marketplace Work on Child Development (Full Sample): Models 11–14

PPVT PIAT-M PIAT-R Coefficient SE Coefficient SE Coefficient SE

Model 11—hours worked with family income (OLS): Year 1 �.035 (.034) �.068* (.031) �.035 (.029) Years 2 and 3 .004 (.016) .018 (.014) .014 (.013) Family income .066** (.024) .045* (.021) .033� (.019)

Model 12—hours worked with family income (OLS): Quarter 1 �.310� (.165) �.019 (.154) .062 (.144) Quarters 2, 3, and 4 .020 (.047) �.077� (.041) �.053 (.039) Years 2 and 3 �.001 (.016) .019 (.014) .015 (.014) Family income .064* (.024) .045* (.021) .033� (.019)

Model 13—hours worked with family income and childcare mode (OLS): Year 1 �.038 (.035) �.080* (.032) �.048� (.029) Years 2 and 3 .004 (.016) .020 (.015) .010 (.014) Relative care .565 (1.131) 1.462 (1.039) .976 (.975) Nonrelative care �.145 (1.136) .264 (1.061) .487 (.994) Center care �.279 (1.168) �.816 (1.095) 1.283 (1.033) Family income .066* (.024) .048* (.022) .036� (.020)

Model 14—hours worked with family income and childcare mode (OLS): Quarter 1 �.244� (.160) �.013 (.157) .082 (.147) Quarters 2, 3, and 4 .004 (.049) �.093* (.043) �.073� (.041) Years 2 and 3 .001 (.017) .021 (.015) .011 (.014) Relative care .445 (1.134) 1.505 (1.044) 1.055 (.980) Nonrelative care �.177 (1.136) .278 (1.062) .513 (.995) Center care �.353 (1.169) �.798 (1.096) 1.315 (1.034) Family income .065* (.024) .048* (.022) .038� (.020)

Note.—There are 2,022 observations in the PPVT model, 1,658 in the PIAT-M model, and 1,655 in the PIAT-R model.

� p ! .10.

* p ! .05. ** p ! .01.

in the first quarter. As with the full sample of mothers, such work has a significant negative effect in the PPVT model.

To determine the robustness of these results, I include maternal labor supply specifications that equal the number of weeks worked. The results (not shown) are similar to models 1 and 2, with weeks worked in the first year still having a significant negative effect on PPVT scores. I also identify the effects of weeks worked in the first quarter (not shown). However, the results are mixed: weeks worked have detrimental effects in the PPVT model that are statistically insignificant, but they have a positive effect on PIAT-R scores.

Table 8 The Effect of Maternal Work on Child Development (Working Mothers)

PPVT PIAT-M PIAT-R Coefficient SE Coefficient SE Coefficient SE

Model 1—hours
worked (OLS):
Year 1 �.082* (.039) �.033 (.036) �.021 (.035)
Years 2 and 3 .006 (.019) .001 (.017) .006 (.017)
Model 2—hours            
worked (OLS):            
Quarter 1 �.313* (.169) .063 (.155) .181 (.149)
Quarters 2, 3, and 4 �.031 (.054) �.053 (.148) �.064 (.046)
Years 2 and 3 �.001 (.020) .004 (.018) .010 (.017)
Model 3—hours            
worked with fam            
ily income and            
childcare mode            
(OLS):            
Year 1 �.092* (.041) �.050 (.038) �.038 (.036)
Years 2 and 3 .004 (.020) .009 (.019) .007 (.018)
Relative care .607 (1.279) �.034 (1.157) .337 (1.107)
Nonrelative care �.560 (1.226) �.619 (1.120) �.640 (1.069)
Center care �.233 (1.298) �.747 (1.178) 1.624 (1.128)
Family income .067* (.027) .048* (.024) .040* (.023)
Model 4—hours            
worked with fam            
ily income and            
childcare mode            
(OLS):            
Quarter 1 �.236 (.173) .049 (.160) .193 (.152)
Quarters 2, 3, and 4 �.059 (.056) �.071 (.050) �.086* (.047)
Years 2 and 3 .001 (.026) .011 (.019) .011 (.018)
Relative care .499 (1.286) .039 (1.163) .541 (.110)
Nonrelative care �.584 (1.266) �.606 (1.121) �.613 (1.068)
Center care �.294 (1.300) �.721 (1.180) 1.686 (1.270)
Family income .065* (.027) .048* (.024) .040* (.023)

Note.—There are 1,133 observations in the PPVT model, 917 in the PIAT-M model, and 916 in the PIAT-R model.

� p ! .10.

* p ! .05.

I also specify the maternal labor supply variables as follows: dummy variables equal to 1 if the mother worked in the period (not shown), dummy variables equal to 1 in the period when the mother first returned to work after giving birth (not shown), and the number of weeks worked full time (not shown). I also estimate models that determine whether maternal labor supply has nonlinear effects (not shown). Results from these models are generally similar, indicating that maternal labor supply has some negative effects on cognitive development in year 1 and insignificant effects in years 2 and 3.

For comparison purposes, I again include family income in the models to determine the degree to which maternal labor supply has positive effects on child development via increased family income. These results (not shown) are similar to those found in the full sample of mothers, where family income has statistically significant positive effects. When family income is included, the effects of maternal labor supply become more negative (or less positive). The statistical significance of the maternal labor supply variables also increases. In the PIAT-R model, the negative effect of hours worked in the latter three quarters of the first year becomes statistically significant. Increased family income again appears to play a role in mitigating the negative effects of maternal work on child development.

I also reestimate the models including various combinations of childcare mode variables. Representative sets of results are displayed in models 3 and 4 in table 8. As with the full sample of mothers, including the childcare mode variables does not systematically change the effects of the maternal labor supply variables. The effect of hours worked in the first year continues to significantly reduce PPVT scores, and such work in the second, third, and fourth quarters of the first year significantly reduces PIAT-R scores. However, the negative effect of hours worked in the first quarter is no longer statistically significant in PPVT model 4. Continuing, relative care tends to enhance the measures of cognitive development, but these effects are not significant. Nonrelative care has detrimental effects that are insignificant, and center care has mixed results.

VI. Discussion and Conclusion

The preferred results from this analysis, where statistically significant, suggest that maternal marketplace work in the child’s first year has detrimental total effects on cognitive development. In the full sample, switching from no work to full-time work in the first year generally reduces the PPVT, PIAT-M, and PIAT-R scores by an average of about 2.5 points, which is about 17% of a standard deviation, depending on the model. This evidence contrasts with results from previous studies (see Leibowitz 1977; Desai et al. 1989; Mott 1991) that also examine “full” samples (rather than low-income samples or samples of only boys or only whites), as well as with results from more recent work by Parcel and Meneghan (1994a), Greenstein (1995), and Harvey (1999), all of whom find no detrimental effects. One factor that may explain this difference is that these studies identify the effects of maternal employment rather than maternal work. By identifying the amount that mothers actually work, the effects of maternal labor supply on cognitive development in this study are stronger. Thus, more care may be needed in specifying the maternal labor supply. Another reason why these results differ from the literature is because these results carefully distinguish between the total effect of maternal labor supply and the partial effect holding family income constant. Another potentially important difference is that many of these previous studies examined children from young mothers, while my sample contains children from mothers who are older. Since some of these studies examine only “first-born” children, results may differ because first-born children do not initially face competition from siblings for family resources.

This study uniquely identifies the effects of maternal labor supply during the first quarter of a child’s life—the period in which many mothers first return to work after giving childbirth. Although some of the evidence is mixed, hours worked in the first quarter seem to have a negative effect on PPVT scores in the full sample. Additional evidence suggests that hours worked in the first quarter by the subsample of “working” mothers decrease PPVT scores. This is an important result because my preferred estimates suggest that, if mothers were to take additional weeks off from work after giving childbirth, this would enhance PPVT scores.

The PIAT-M and PIAT-R scores, however, are not significantly affected by maternal labor in the first quarter. Rather, these test scores seem to be determined more by maternal labor supply in the latter three quarters of a child’s first year. This may be because math and reading skills begin developing slightly later than the picture-vocabulary associations tested by the PPVT. Alternatively, the effects of early maternal employment during the child’s first quarter may be temporary and “fade out.” Certainly the PPVT is measured at younger ages (ages 3 and 4) than the PIAT scores (ages 5 and older). Regardless, results also suggest that maternal labor supply in the remainder of the first year detrimentally affects these scores. If mothers were to take additional weeks off from work (into the second and later quarters of the infant’s life), then cognitive development might be further enhanced.

The results from the full sample are often similar to results from the subsample of working mothers, and they do not differ in any systematic way. For example, the detrimental effect of maternal labor supply in the first year in the PIAT-M model is greater in the full sample, but the negative effect of such work in the PIAT-R model is greater in the working subsample. In general, it is surprising how little some of the results changed when examining mothers who are more firmly attached to the labor force. This is what we might expect since the results did not change substantially when including supplemental background characteristic variables or predicting maternal labor supply using instrumental variables (IV). We can conclude that the effects of maternal labor supply are similar whether or not the mother exhibits a prior attachment to the labor force.

This study has also shown that maternal labor supply partially affects child development via increased family income. When family income is held constant, the partial effects of maternal labor supply almost always become more negative (or less positive) and statistical significance generally increases. Further, increasing family income by $20,000 directly enhances child development by an average of about 6% of a standard deviation, which is similar to Blau’s (1999) findings. Thus, results suggest that increased family income from maternal work partially offsets the negative effects of maternal labor supply. Of course, the degree to which this occurs depends on the amount of increased family income that maternal employment generates.

Additionally, the results suggest that moderating effects of nonmaternal childcare arrangements are relatively small. Although maternal labor supply effects in the PIAT-R model become statistically significant, many of the effects of maternal labor supply remain unchanged when controlling for childcare mode. This may be because family income proxies for the quality of nonmaternal childcare arrangements. Alternatively, it may be that childcare mode is a poor proxy for childcare quality. Therefore, it is not clear from these results that the policy recommendation should be that “mothers should stay home with their infants” rather than “government should work to improve the quality of childcare.”

Although these results have implied that child development may benefit from mothers taking time off from work in the initial months after giving birth, these results examine only a small number of child outcomes. Maternal work may also have effects on child health or on the child-mother attachment. Further, early maternal labor supply may have additional longer-term effects on children, such as effects on future grades, educational attainment, or wages. For example, it would be interesting to determine whether the effects of early maternal employment are temporary or permanent and, if temporary, how long they last. Thus, more research is clearly needed to determine the importance of maternal labor supply (and taking maternity leave from work) in the period immediately after giving birth on children.

Appendix

Table A1 The Effects of Nonmaternal Employment Explanatory Variables on PPVT, PIAT-M, and PIAT-R (Full Sample)

PPVT PIAT-M PIAT-R

Coeffi-Coeffi-Coefficient SE cient SE cient SE

Intercept 70.616** (10.196) 86.721** (10.160) 126.817** (9.437)Child gender (p1 if male,

0 if female) �2.730** (.917) �1.237 (.849) �3.693** (.797)Child’s birth order (parity) �3.850** (.518) �1.845** (.479) �1.871** (.451)Black dummy variable (p

1 if child is black, 0

otherwise) �8.782** (1.561) �2.090 (1.449) 5.270** (1.354)Hispanic (p 1 if child is

Hispanic, 0 otherwise) �2.464 (2.297) �1.909 (2.026) 1.594 (1.921)Child’s age (in months) �.022 (.047) .025 (.064) �.413** (.057)Low birth weight (p 1if

! 5.5. pounds, 0

otherwise) �1.580 (1.855) �1.382 (1.732) �1.448 (1.622)Mother’s hospital stay

(days after giving birth) �.132 (.130) .013 (.124) �.062 (.116)

Table A1 (Continued)

PPVT PIAT-M PIAT-R

Coeffi-Coeffi-Coefficient SE cient SE cient SE

Mother’s education (level

in years) .450� (.283) .188 (.258) �.081 (.243)Mother’s age (years) .537* (.238) .204 (.214) .204 (.200)Mother’s marital status

(p1 if married, 0

otherwise) 3.411** (1.200) 1.383 (1.099) .097 (1.032)Northeast dummy variable

(p 1 if mother resides in

northeast, 0 otherwise) 1.170 (1.400) �.966 (1.291) 2.036� (1.211)South dummy variable (p

1 if mother resides in

south, 0 otherwise) �4.764** (1.796) �3.085* (1.583) �1.452 (1.484)West dummy variable (p 1

if mother resides in west,

0 otherwise) �.048 (1.439) �1.271 (1.348) �.944 (1.272)Test taken in 1994 (p 1if

child took test in 1994, 0

otherwise) 3.899* (1.613) 2.422 (1.590) �3.594* (1.497)Test taken in 1996 (p 1if

child took test in 1996, 0

otherwise) �.014 (1.489) 1.630� (.996) �1.262 (.937)Mother’s AFQT residual

test score (controlling

for age) .131** (.026) .141** (.023) .173** (.022)Mother’s mother native

born (p1 if mother’s

mother born outside the

United States, 0

otherwise) �.108 (2.771) �.665 (2.313) .927 (2.165)Mother spoke a foreign

language (p1 if mother

spoke a foreign language,

0 otherwise) �2.697 (2.026) �.120 (1.834) .334 (1.738)Mother lived with both

parents (p 1 if mother

lived with both parents

at age 14, 0 otherwise) �.821 (1.043) �.936 (.949) .172 (.890)Mother’s household re

ceived magazines (p 1if

mother’s household re

ceived magazines when

mother was age 14, 0

otherwise) 1.752 (1.143) �.192 (1.040) .699 (.975)Mother’s household re

ceived newspapers (p 1

if mother’s household

received newspapers

when mother was age

14, 0 otherwise) �.669 (1.292) �2.869* (1.172) �1.649 (1.094)Mother’s household had li

brary card (p 1if

mother’s household had

a library card when

mother was age 14, 0

otherwise) 2.325� (1.232) 1.555 (1.138) 2.740* (1.067)South dummy variable for

age 14(p 1 if mother

lived in the south at age

14, 0 otherwise) 2.978� (1.764) 1.782 (1.554) 1.044 (1.458)Urban dummy variable for

age 14(p 1 if mother

lived in an urban area at

age 14, 0 otherwise) �2.056 (1.233) 1.915� (1.113) .589 (1.045)

442

Table A1 (Continued)

PPVT PIAT-M PIAT-R

Coeffi-Coeffi-Coefficient SE cient SE cient SE

Grandmother’s education

(education level in years

of mother’s mother) .260 (.221) .383* (.208) .293 (.196)Grandmother’s education

missing (p1 if mother’s

mother’s education level

is missing, 0 otherwise) 3.731 (3.659) 7.546* (3.444) 4.032 (3.227)Grandmother worked part

time when mother was

age 14 (p 1 if mother’s

mother worked part time

when mother was age

14, 0 otherwise) 2.975* (1.276) .782 (1.195) �.113 (1.122)Grandmother worked full

time when mother was

age 14(p 1 if mother’s

mother worked full time

when mother was age

14, 0 otherwise) 1.298 (1.078) 1.677� (1.000) .251 (.938)Grandfather’s education

(education level in years

of mother’s father) .350* (.177) .083 (.164) .086 (.154)Grandfather’s education

missing (p1 if mother’s

father’s education level is

missing, 0 otherwise) 2.499 (2.394) .329 (2.187) �1.656 (2.055)Mother’s siblings (no. of

mother’s siblings) .450* (.204) .390* (.189) .264 (.178)Mother’s Rotter test score

(higher scores p less

efficacy) .063 (.206) �.002 (.189) .076 (.177)Mother’s attitudes about

family roles (higher

scores p traditional

views) �.149 (.150) .027 (.133) .004 (.125)

Note.—These results are from model 5 in table 3. There are 2,022 observations in the PPVT model, 1,658 in the PIAT-M model, and 1,655 in the PIAT-R model. The R2 values for these models range from .21 to .36.

� p ! .10.

* p ! .05. ** p ! .01.

Table A2 The Effects of Nonmaternal Employment Explanatory Variables on PPVT, PIAT-M, and PIAT-R (Working Mothers)

PPVT PIAT-M PIAT-R

Coeffi-Coeffi-Coefficient SE cient SE cient SE

Intercept 78.776** (12.067) 81.712** (11.916) 140.204** (11.395)Child gender (p1 if male,

0 if female) �1.152 (1.113) �.265 (1.006) �3.123** (.970)Child’s birth order (parity) �4.779** (.725) �1.979** (.671) �2.509** (.647)Black dummy variable (p

1 if child is black, 0

otherwise) �9.137** (1.939) �4.305* (1.749) 5.080** (1.679)Hispanic (p 1 if child is

Hispanic, 0 otherwise) �3.098 (2.833) �2.498 (2.480) .692 (2.399)

443

Table A2 (Continued)

PPVT PIAT-M PIAT-R

Coeffi-Coeffi-Coefficient SE cient SE cient SE

Child’s age (in months) .014 (.058) .106 (.075) �.426** (.070)

Low birth weight (p 1if

! 5.5. pounds, 0

otherwise) �2.781 (2.323) .861 (2.153) �1.293 (2.080)Mother’s hospital stay

(days after giving birth) �.125 (.164) �.283 (.328) �.551� (.318)Mother’s education (level

in years) .268 (.322) .447 (.287) �.130 (.278)Mother’s age (years) .499� (.286) .167 (.253) �.056 (.244)Mother’s marital status

(p1 if married, 0

otherwise) 3.022* (1.524) 1.141 (1.395) �.634 (1.342)Northeast dummy variable

(p 1 if mother resides in

northeast, 0 otherwise) �.083 (1.666) �.410 (1.497) 2.843** (1.448)South dummy variable (p

1 if mother resides in

south, 0 otherwise) �5.972** (2.267) �2.357 (1.950) 1.169 (1.887)West dummy variable (p 1

if mother resides in west,

0 otherwise) .700 (1.755) �1.532 (1.611) .053 (1.559)Test taken in 1994 (p 1if

child took test in 1994, 0

otherwise) 1.502 (1.896) 3.128� (1.933) �4.770* (1.881)Test taken in 1996 (p 1if

child took test in 1996, 0

otherwise) �.327 (1.738) 1.921� (1.196) �2.059� (1.162)Mother’s AFQT residual

test score (controlling

for age) .129** (.030) .122** (.027) .128** (.026)Mother’s mother native

born (p1 if mother’s

mother born outside the

United States, 0

otherwise) �4.137 (3.370) �2.551 (2.783) �1.823 (2.694)Mother spoke a foreign

language (p1 if mother

spoke a foreign language,

0 otherwise) �1.354 (2.448) .514 (2.210) 1.587 (2.140)Mother lived with both

parents (p 1 if mother

lived with both parents

at age 14, 0 otherwise) .066 (1.281) .122 (1.138) 1.696 (1.102)Mother’s household re

ceived magazines (p 1if

mother’s household re

ceived magazines when

mother was age 14, 0

otherwise) �.056 (1.420) �1.015 (1.251) 1.620 (1.204)Mother’s household re

ceived newspapers (p 1

if mother’s household

received newspapers

when mother was age

14, 0 otherwise) .366 (1.654) �1.295 (1.483) .257 (1.423)Mother’s household had li

brary card (p 1if

mother’s household had

a library card when

mother was age 14, 0

otherwise) 2.653� (1.490) 1.645 (1.337) 2.055 (1.290)

Table A2 (Continued)

PPVT PIAT-M PIAT-R

Coeffi-Coeffi-Coefficient SE cient SE cient SE

South dummy variable for

age 14 (p 1 if mother

lived in the south at age

14, 0 otherwise) 2.805 (2.289) 1.404 (1.921) �.967 (1.861)Urban dummy variable for

age 14 (p 1 if mother

lived in an urban area at

age 14, 0 otherwise) �2.905* (1.463) 2.840* (1.319) .065 (1.279)Grandmother’s education

(education level in years

of mother’s mother) .123 (.271) .226 (.244) .308 (.236)Grandmother’s education

missing (p1 if mother’s

mother’s education level

is missing, 0 otherwise) 1.561 (4.580) 7.668� (4.299) 7.040� (4.164)Grandmother worked part

time when mother was

age 14 (p 1 if mother’s

mother worked part time

when mother was age

14, 0 otherwise) 2.916� (1.537) .619 (1.405) �1.360 (1.362)Grandmother worked full

time when mother was

age 14 (p 1 if mother’s

mother worked full time

when mother was age

14, 0 otherwise) 2.239� (1.292) 1.263 (1.184) �.401 (1.146)Grandfather’s education

(education level in years

of mother’s father) .486* (.213) .067 (.190) .055 (.183)Grandfather’s education

missing (p1 if mother’s

father’s education level is

missing, 0 otherwise) 5.492� (3.102) 2.331 (2.753) �.745 (2.662)Mother’s siblings (no. of

mother’s siblings) .789** (.253) .345 (.236) .088 (.228)Mother’s Rotter test score

(higher scores p less

efficacy) �.198 (.249) .066 (.225) .206 (.217)Mother’s attitudes about

family roles (higher

scores p traditional

views) �.163 (.180) �.100 (.158) �.075 (.153)

Note.—These results are from model 5 in table 3. There are 1,133 observations in the PPVT model, 917 in the PIAT-M model, and 916 in the PIAT-R model. TheR2 values for these models range from .26 to .40.

� p ! .10.

* p ! .05. ** p ! .01.

References

Ainsworth, Mary D. “The Development of Infant-Mother Attachment.” In Review of Child Development, edited by B. M. Caldwell and H. N. Ricciuti, 3:1–94. Chicago: University of Chicago Press, 1973.

Baker, Paula; Keck, Canada K.; Mott, Frank L.; and Quinlan, Stephen V. NLSY Child Handbook, 1986–1990. Columbus, OH: Center for Human Resource Research, 1993.

Baker, Paula, and Mott, Frank. NLSY Child Handbook, 1989. Columbus: Ohio State University, Center for Human Resources Research, 1989.

Barglow, Peter; Vaughn, Brian E.; and Molitor, Nancy. “Effects of Maternal Absence Due to Employment on the Quality of Infant-Mother Attachment in a Low-Risk Sample.” Child Development 58, no. 4 (August 1987): 945–54.

Barrow, Lisa. “An Analysis of Women’s Return-to-Work Decisions following First Birth.” Economic Inquiry 37, no. 3 (July 1999): 432–51.

Baydar, Nazli, and Brooks-Gunn, Jeanne. “Effects of Maternal Employment and Child-Care Arrangements on Preschoolers’ Cognitive and Behavioral Outcomes: Evidence from the Children of the National Longitudinal Survey of Youth.” Developmental Psychology 27, no. 6 (November 1991): 932–45.

Belsky, Jay. “The ‘Effects’ of Infant Day Care Reconsidered.” Early Childhood Research Quarterly 3, no. 3 (September 1988): 235–72.

———. “Developmental Risks Associated with Infant Day Care: Insecurity, Aggression, and Noncompliance?” In Balancing Working and Parenting:Psychological and Developmental Implications of Day Care, edited by S. Chehrazi, pp. 36–68. New York: American Psychiatric Press, 1990.

Belsky, Jay, and Eggebeen, David. “Early and Extensive Maternal Employment and Young Children’s Socioemotional Development: Children of the National Longitudinal Survey of Youth.” Journal of Marriage and the Family 53, no. 4 (November 1991): 1083–98.

Belsky, Jay, and Rovine, Michael J. “Nonmaternal Care in the First Year of Life and the Security of Infant-Parent Attachment.” Child Development 59, no. 1 (February 1988): 157–67.

Blau, David M. “The Effect of Income on Child Development.” Review of Economics and Statistics 81, no. 2 (May 1999): 261–76.

Blau, Francine D., and Grossberg, Adam J. “Maternal Labor Supply and Children’s Cognitive Development.” Review of Economics and Statistics 74, no. 3 (August 1992): 474–81.

Center for Human Resource Research. NSLY79 1996 Child and Young Adult Users’ Guide. Columbus: Ohio State University, 1996.

Chase-Lansdale, P. Lindsay, and Owen, Margaret Tresch. “Maternal Employment in a Family Context: Effects on Infant-Mother and Infant-Father Attachments.” Child Development 58, no. 6 (December 1987): 1505–12.

Currie, Janet, and Thomas, Duncan. “Does Head Start Make a Difference?” American Economic Review 85, no. 3 (June 1995): 341–64.

Datcher-Loury, Linda. “Effects of Mother’s Home Time on Children’s Schooling.” Review of Economics and Statistics 70, no. 3 (August 1988): 376–73.

Desai, Sonald; Chase-Lansdale, P. Lindsay; and Michael, Robert T. “Mother or Market? Effects of Maternal Employment on Intellectual Ability of 4-Year-Old Children.” Demography 26, no. 4 (November 1989): 545–61.

Fleisher, Belton M. “Mothers’ Home Time and the Production of Child Quality.” Demography 14, no. 2 (May 1977): 197–212.

Greenberger, Ellen, and Goldberg, Wendy A. “Work, Parenting, and the Socialization of Children.” Developmental Psychology 25, no. 1 (January 1989): 22–35.

Greenstein, Theodore N. “Are the ‘Most Advantaged’ Children Truly Disadvantaged by Early Maternal Employment?” Journal of Family Issues 16, no. 2 (March 1995): 149–69.

Hanushek, Eric A. “The Trade-Off between Child Quantity and Quality.” Journal of Political Economy 100, no. 1 (February 1992): 84–117.

Harris, P. L. “Infant Cognition.” In Handbook of Child Psychology, Socialization, Personality, and Social Development, edited by P. H. Mussen, 4:689–782. New York: Wiley & Sons, 1983.

Harvey, Elizabeth. “Short-Term and Long-Term Effects of Early Parental Employment on Children of the National Longitudinal Survey of Youth.”Developmental Psychology 35, no. 2 (March 1999): 445–59.

Hill, M. Anne, and O’Neill, June E. “Family Endowments and the Achievement of Young Children with Special Reference to the Underclass.” Journal of HumanResources 29, no. 4 (Fall 1994): 1064–1101.

Hock, Ellen. “Working and Nonworking Mothers and Their Infants: A Comparative Study of Maternal Caregiving Characteristics and Infants’ Social Behavior.” Merrill-Palmer Quarterly 26, no. 2 (April 1980): 79–101.

Hock, Ellen; DeMeis, Debra; and McBride, Susan. “Maternal Separation Anxiety: Its Role in the Balance of Employment and Motherhood in Mothers of Infants.” In Maternal Employment and Children’s Development: Longitudinal Research, edited by A. E. Gottfried and A. W. Gottfried, pp. 191–230. New York: Plenum, 1988.

Hoynes, Hilary Williamson. “Local Labor Markets and Welfare Spells: Do Demand Conditions Matter?” Review of Economics and Statistics 82, no. 3 (August 2000): 351–68.

Klerman, Jacob Alex. “Characterizing Leave for Maternity.” Unpublished manuscript. Santa Monica, CA: RAND Corporation, Economics and Statistics Department, 1995.

Klerman, Jacob Alex, and Leibowitz, Arleen. “Child Care and Women’s Return to Work after Childbirth.” American Economic Review 80, no. 2 (May 1990): 284–88.

———. “The Work-Employment Distinction among New Mothers.” Journal of Human Resources 29, no. 2 (Spring 1994): 277–303.

———. “Job Continuity among New Mothers.” Demography 36, no. 2 (May 1999): 145–55.

Leibowitz, Arleen. “Parental Inputs and Children’s Achievement.” Journal of Human Resources 12, no. 2 (Spring 1977): 242–51.

Leibowitz, Arleen, and Klerman, Jacob Alex. “Explaining Changes in Married Mothers’ Employment over Time.” Demography 32, no. 3 (August 1995): 365–78.

Lewis, M., and Brooks-Gunn, Jeanne. Social Cognition and the Acquisition of Self. New York: Plenum, 1979.

MaCurdy, Thomas; Mroz, Thomas; and Gritz, R. Mark. “An Evaluation of the NLSY.” Journal of Human Resources 33, no. 2 (Spring 1998): 35–436.

McCartney, Kathleen. “Effects of Quality of Day Care Environment on Children’s Language Development.” Developmental Psychology 20, no. 2 (March 1984): 244–60.

Moore, Kristin A., and Driscoll, Anne K. “Low-Wage Maternal Employment and Outcomes for Children: A Study.” Future of Children 7, no. 1 (1997): 122–27.

Mott, Frank L. “Developmental Effects of Infant Care: The Mediating Role of Gender and Health.” Journal of Social Issues 47, no. 2 (Summer 1991): 139–58.

National Institute of Child Health and Human Development (NICHD) Early Child Care Research Network. “The Effects of Infant Child Care on Infant-Mother Attachment Security: Results of the NICHD Study of Early Child Care.” Child Development 68, no. 5 (October 1997): 860–79. (a)

———. “Familial Factors Associated with the Characteristics of Non-maternal Care for Infants.” Journal of Marriage and the Family 59, no. 2 (May 1997): 389–408. (b)

Parcel, Toby L., and Menaghan, Elizabeth G. “Maternal Working Conditions and Children’s Verbal Facility: Studying the Intergenerational Transmission of Inequality from Mothers to Young Children.” Social Psychology Quarterly 53, no. 2 (June 1990): 132–47.

———. “Early Parental Work, Family Social Capital, and Early Childhood Outcomes.” American Journal of Sociology 99, no. 4 (January 1994): 972–1009. (a)

———. Parents’ Jobs and Children’s Lives. Sociology and Economics Controversy and Integration Series. New York: Aldine De Gruyter, 1994. (b)

Ruhm, Christopher J. “Parental Employment and Child Cognitive Development.” Working Paper no. 7666. Cambridge: MA: National Bureau of Economic Research, 2000.

Schwartz, Pamela. “Length of Day-Care Attendance and Attachment Behavior in 18-Month-Old Infants.” Child Development 54, no. 4 (August 1983): 1073–78.

Stafford, Frank P. “Women’s Work, Sibling Competition, and Children’s School Performance.” American Economic Review 77, no. 5 (December 1987): 972–80.

Vandell, Deborah Lowe, and Ramanan, Janaki. “Effects of Early and Recent Maternal Employment on Children from Low-Income Families.” ChildDevelopment 63, no. 4 (August 1992): 938–49.

Comments
  • Recommend Us