Developmental Psychology© 1999 by the American Psychological Association
March 1999 Vol. 35, No. 2, 445-459
For personal use only--not for distribution.
Short-Term and Long-Term Effects of Early Parental Employment on
Children of the National Longitudinal Survey of Youth
Elizabeth Harvey
Department of Psychology
University of Connecticut
ABSTRACT
This study examined the effects of early parental employment on children
in the National Longitudinal Survey of Youth. Minimal effects on
children's later functioning were found. Early maternal employment
status and the timing and continuity of early maternal employment were
not consistently related to children's development. Working more hours
was associated with slightly lower cognitive development through age 9
and slightly lower academic achievement scores before age 7 but had no
significant relation to children's behavior problems, compliance, or
self-esteem. Early parental employment appeared to be somewhat more
beneficial for single mothers and lower income families. There was some
support for the hypothesis that early parental employment positively
affects children's development by increasing family income.
------------------------------------------------------------------------
I thank Conrad Schwarz and Len Milling for their comments on earlier
drafts of this article.
Correspondence may be addressed to Elizabeth Harvey, Department of
Psychology, University of Massachusetts at Amherst, Tobin Hall, Amherst,
Massachusetts, 01003-7710.
Electronic mail may be sent to eharvey@psych.umass.edu
Received: June 30, 1997
Revised: June 6, 1998
Accepted: June 6, 1998
------------------------------------------------------------------------
The past several decades have seen an increase in the number of employed
mothers, with a particularly large increase in the frequency with which
mothers of young children are employed ( Bureau of the Census, 1994 ).
Researchers who have reviewed the literature on the effects of maternal
employment on children's development have agreed that there is little
evidence of negative effects when children are older ( Belsky, 1988 ,
1990 ; Hoffman, 1961 , 1974 , 1989 ). However, there remains debate
about the effects of early maternal employment on children. After
reviewing the literature, Belsky (1988) concluded that maternal
employment during infancy had ill effects on children's well-being.
Specifically, he concluded that infants who were in nonmaternal care for
more than 20 hr per week were at elevated risk for being insecurely
attached at age 1 and were more disobedient and aggressive between ages
3 and 8. However, his conclusions have been criticized on several
counts. For example, it has been argued that the studies Belsky reviewed
failed to take into account background variables that may have been
confounded with maternal employment and children's well-being (e.g.,
Clarke-Stewart, 1988 , 1989 ). In addition, Clarke-Stewart noted that
many of these studies were based on nonrepresentative samples. Finally,
she argued that the measures of attachment used in these studies may not
have predictive validity for children of working mothers and that more
longitudinal studies are needed to determine effects on children's later
functioning.
In recent years, several studies have addressed some of these criticisms
using the National Longitudinal Survey of Youth (NLSY). The NLSY is a
survey of women who have been interviewed annually since 1979 when they
were 14 to 22 years old. Beginning in 1986, children of these women were
also assessed. Six published studies have used this sample to examine
the longitudinal effects of early maternal employment on children's
development, controlling for various family background variables. The
results of these studies have been surprisingly mixed considering they
used the same data set. The sample size and longitudinal design of the
NLSY make it potentially valuable for illuminating the effects of early
maternal employment. However, rather than shedding light on this issue,
conflicting results of studies based on these data have added further
confusion.
Although these six studies all used the same data set, they each used
quite different methodological approaches with respect to sample
selection, construction of independent and dependent variables, and
selection of control variables. The present study sought to resolve some
of these differences through a reanalysis using an updated version of
the NLSY that contained a much larger and more representative sample. In
this article, I examined the six studies, explored how their
methodological differences might have yielded discrepant results,
evaluated their methodological strengths and limitations, and reanalyzed
the effects of early parental employment by drawing on the strengths and
addressing the limitations of these six previous studies.
A Comparison of Methodological Approaches and Findings 1
Of the six previous studies, none used the exact same sample. Slightly
different age ranges were selected, and three studies did not include
all races. Four studies included the Peabody Picture Vocabulary
Test-Revised (PPVT-R) as a dependent variable, and four studies used the
Behavior Problems Index (BPI). Vandell and Ramanan (1992) conducted the
only study that examined school achievement. There was surprisingly
little overlap across studies in the construction of maternal employment
variables. There was also considerable variation in the background
variables used as controls. The number of control variables varied from
4 to over 25. Mother's IQ, race, and family income or poverty status
were the only variables that were controlled across all six studies.
Child gender was the only moderating variable examined in all six
studies. Only one study directly evaluated potential mediators of early
maternal employment.
It is therefore less surprising that these studies yielded very
different findings. In general, the Vandell and Ramanan (1992) , Parcel
and Menaghan (1994) , and Greenstein (1995) studies found no adverse
effects of early maternal employment on PPVT-R or behavior problems, and
Vandell and Ramanan (1992) found some positive effects on children's
Peabody Individual Achievement Test (PIAT) scores. Desai,
Chase-Lansdale, and Michael (1989) found negative effects on PPVT-R
scores only for boys from high-income families, and Bayder and
Brooks-Gunn (1991) found negative effects on PPVT-R and behavior
problems for White families only. Belsky and Eggebeen (1991) found
adverse effects on their variable called ADJUST, a composite of BPI
scores and temperament variables, but when they examined BPI scores
alone, they did not find significant effects.
Because the studies were based on the same data set, the conflicting
results must be due to one or more of the methodological variations,
although it is not clear which. Possible ways in which these
methodological differences could have caused these discrepancies and
affected results are explored next.
An Examination of Various Methodological Approaches and Their Possible
Effects on Findings
Sample selection. Differences in sample selection might be one possible
reason for the discrepancies. The children studied by Vandell and
Ramanan (1992) were much older and their mothers were much younger than
in any other study. Thus, it is possible that any negative effects of
early maternal employment disappear by second grade or that maternal
employment is not harmful for younger, lower income mothers. Similarly,
as Bayder and Brooks-Gunn (1991) pointed out, the discrepancy between
their results and those reported by Desai et al. (1989) may have been
because the former study included only White children.
There are theoretical reasons why the effects of early maternal
employment might vary depending on the age of the child. The effects
might fade over time as more proximal factors play a larger role in a
child's development. On the other hand, sleeper effects might be
observed as minor, early negative effects spiral into larger problems. A
cross-study comparison of these six studies supports the former
hypothesis. However, a within-study comparison of children across a wide
age range would better address this issue. This was not feasible using
the 1986 NLSY but is an option with the more recent version.
Each of the studies recognized the importance of addressing race but did
so in different ways. Some limited their sample to certain racial
groups, arguing that other racial groups were not adequately represented
and that the assessments were conducted in English. Others included all
children but controlled for race. Using the former approach limits the
generalizability of the results, whereas using the latter approach may
mask important racial differences. Greenstein (1995) included all races
and performed separate analyses by race. This avoids both of these
problems, but it does not directly test whether race moderates the
effects of early employment, and it increases the number of analyses and
the chance of Type I error. A modification of the approach used by
Vandell and Ramanan (1992) would avoid these problems: include all races
and evaluate the moderating effect of race. In the updated NLSY,
adequate sample size for each racial group is no longer an issue. In
addition, beginning in 1988 the PPVT-R was administered in Spanish to
those children who so preferred, decreasing somewhat the cultural bias
of the assessment. It is important to note that results would still need
to be interpreted with caution given questions regarding the validity of
standardized cognitive assessments in non-White children.
Constructing maternal employment variables. The different approaches to
maternal employment variable construction could be another reason for
the discrepancies. For example, Parcel and Menaghan (1994) , who found
no effects, examined the number of hours worked among employed mothers,
whereas each of the other studies involved comparisons between employed
mothers and nonemployed mothers. It may be that employment status rather
than the intensity of employment is related to child development.
Consistent with this hypothesis, Desai et al. (1989) found no effects on
PPVT-R scores when they compared mothers working full time during the
first year with mothers who did not, whereas Bayder and Brooks-Gunn
(1991) found an effect on PPVT-R scores when they compared mothers who
worked at least 1 hr per week during the first year with mothers who did
not work during the first year.
In each study, the primary approach to constructing maternal employment
variables was to create categories of early employment characteristics
and dummy-code variables to examine each category. Creating categorical
maternal employment variables can be useful in interpreting results
because it is sometimes easier to use categories to organize
information. Researchers have also justified using categories to detect
nonlinear effects ( Parcel & Menaghan, 1994 ). However, this approach
has been criticized ( Scarr, 1991 ) and presents the following problems.
For the most part, these categories were formed from continuous
variables, so there are infinite ways one could form categories. This
approach can also exclude some participants if the categories do not
include every employment pattern. For example, Belsky and Eggebeen's
(1991) approach to variable construction excluded almost half of the
children in their study. This approach also results in arbitrary
boundaries between categories. Participants with similar employment
patterns will often be assigned to two different categories.
Furthermore, using continuous variables does not preclude one from
examining nonlinear effects of early employment and in fact provides
more sensitivity in detecting such relationships. Another problem with
previous studies' use of categories is that they often combined
different dimensions. For example, comparing mothers who worked full
time during the first year with mothers who did not combines information
about the timing and intensity of early employment. It is not clear
whether results reflect one or both employment dimension.
Selecting control variables. The six studies varied in their approaches
to selection of control variables, with some studies controlling for
many more variables than others did. Some studies used theory to guide
variable selection, others used empirical methods, and others used a
combination. In addition, some studies controlled for family variables
assessed at the same time that the dependent variables were measured (in
1986), whereas others did not. Three of the studies that used theory
only used a larger number of control variables and tended to control for
1986 variables. These three studies were also the only ones that failed
to find any adverse effects of maternal employment.
Controlling for relevant background family characteristics reduces
third-variable effects and removes error from the dependent variable.
However, if a background variable does not correlate with child outcome,
then it cannot act as a third variable and including it will not remove
error. In fact, including irrelevant variables in regression models will
on average increase standard errors and make it more likely that Type II
error will occur. Therefore, the three studies may have failed to find
adverse effects because they included irrelevant variables. Thus, in
addition to using theory in control variable selection, it is important
to empirically evaluate whether each background variable should be
included in the regression equations.
In selecting control variables, one must distinguish between selection
factors and mediators of the effect of early employment on children.
Selection factors are variables that might affect, but are not affected
by, early employment. These might include parents' education and
mothers' IQ. Mediators are variables that are affected by patterns of
early employment and in turn affect children. Thus, many variables
assessed in 1986 should not be considered selection factors. Selection
factors and mediating effects must be examined using different analytic
approaches. For example, each of the six studies controlled for family
income or poverty status measured after the birth of the child. One of
the ways in which early maternal employment may positively affect
children may be through its financial benefits. Controlling for income
after the child is born involves controlling for an important benefit of
maternal employment. It is not surprising, then, that some of these
studies found detrimental effects of maternal employment when they
statistically removed its positive effects. Family income before the
birth of the child should be controlled, because the wealth of the
family is likely to affect whether a mother chooses to be employed.
However, income after the child's birth should be examined as a
mediating variable of positive effects of early employment.
Improvements in the Sample
The sample available in the more recent NLSY is stronger than the 1986
sample used by five of the six studies in several ways. First, the 1986
NLSY data on which these studies were based were from unusually young,
low-socioeconomic status (SES) mothers. The NLSY has continued to follow
these women and their children, and the data set now contains more
older, higher SES mothers. Thus, the most recent version of the NLSY
allows for an examination of the effects of early maternal employment in
a more representative sample. Whereas the children in the 1986 data were
estimated to represent the first 40% of the offspring that these women
would bear, the recent data set is estimated to represent the first 70%
to 75% of these women's offspring ( Center for Human Resource Research,
1997 ). In addition, the sample size is now much larger. This should
allow for more stable, robust results and should provide more confidence
that null effects truly represent no or negligible effects rather than
Type II error. Furthermore, the more recent data allow for an
examination of much longer term effects, because there are now a sizable
number of elementary school age children in the sample. This also means
that two additional outcome variables specific to older children are
available: self-esteem and academic achievement.
Paternal Employment
In contrast to the focus on the effects of maternal employment on child
development, few studies have examined early paternal employment. McHale
and Huston (1984) found that fathers' time in paid work affected the
amount but not the quality of interaction with their infants. However,
Parcel and Menaghan's (1994) analysis of the NLSY indicated that fathers
working fewer hours during their children's first few years was
associated with more behavior problems in children. The small body of
literature on fathers' employment focuses on job quality or unemployment
rather than on how the time demands of fathers' jobs affect the family
(e.g., Barling, 1986 ).
The Present Study
The more recent NLSY provides the opportunity to reexamine the effects
of early parental employment on children's development, addressing a
number of important limitations of these six previous studies as
follows: (a) A larger, more representative sample was used that should
significantly increase the generalizability of results; (b) selection
factors were distinguished from mediating variables; (c) control
variables were selected using both theory and empirical decision rules;
(d) the sample contained a wider age range of children allowing for an
evaluation of short-term and long-term effects of early employment; (e)
children from all races were included; and (f) whenever possible,
continuous variables of early parental employment were used rather than
artificially creating categorical variables. Thus, by drawing on the
methodological strengths and addressing the methodological weaknesses of
previous studies of early parental employment, it is hoped that this
study will more fully realize the potential of the NLSY to address this
topic.
Past research and theory have been conflicting regarding the effects of
early parental employment, making it difficult to make specific
predictions regarding the presence or direction of effects. Although
previous inconsistent results are likely caused by methodological
differences, the studies differ in too many ways to determine which
findings are correct. Thus, the goal of this study is to examine whether
and when early parental employment affects children's emotional,
cognitive, and academic development.
Method
The NLSY Sample
The NLSY is a survey of approximately 12,600 individuals who have been
interviewed annually since 1979 when they were 14 to 22 years old. This
survey oversampled African American, Hispanic, and economically
disadvantaged White individuals. The economically disadvantaged White
oversample was dropped in 1990 for financial reasons. In 1986, 1988,
1990, 1992, and 1994, the survey conducted child assessments on
offspring of the female participants.
Children of all races who were between 3 and 12 years of age at any of
the five child assessment dates and who were born in 1980 or later
(several background variables were unavailable for children born before
1980) were included. To assess possible developmental differences in the
effects of early employment, I examined four different age groups
separately: 3- to 4-year-olds, 5- to 6-year-olds, 7- to 9-year-olds, and
10- to 12-year-olds. See Table 1 for descriptive statistics on this
sample.
It should be noted that there is partial overlap of participants across
these four age groups; each age group contains some participants who
were also included in another age group. Of the 4,924 children who were
3- to 4-year-olds, 3,371 were also in the 5- to 6-year-old group. There
were 4,486 children in the 5- to 6-year-old group, 3,203 of whom were
also in the 7- to 9-year-old group. There were 3,711 children in the 7-
to 9-year-old group, 1,951 of whom were also in the 10- to 12-year-old
group. There were 2,095 children in the 10- to 12-year-old group. Thus,
some of the data in this study are longitudinal and some are
cross-sectional. (These figures are based on the number of children in
each age group who had scores on at least one of the child outcome
measures. The actual number of participants for each analysis varies
because not all children were administered all measures.) In addition,
although this sample is more representative than samples used in
previous studies of early maternal employment ( Bayder & Brooks-Gunn,
1991 ; Belsky & Eggebeen, 1991 ), it still does not represent women who
bear children after age 34 and is still somewhat socioeconomically
disadvantaged.
Measurement of Variables
Early maternal employment variables. Five indexes of early maternal
employment were used. They were constructed on the basis of mothers'
reports of how many weeks after their children's birth they returned to
work and their estimates of the average number of hours they worked per
week during each quarter-year of the first 3 years of their children's
lives.
For an evaluation of whether early employment status affects children,
the first variable indicated whether or not the mother was employed at
some time during the child's first 3 years ( employment during the first
3 years ); mothers who reported returning to work by the 156th week
after their child's birth were coded 1 and mothers who did not return to
work in the first 3 years were coded 0. The second, third, and fourth
variables were applicable only for women who were employed at some point
during the first 3 years. The second variable was the number of weeks
after the child's birth before the mother returned to work ( timing of
early employment ), the third variable was the average number of hours
per week that she worked when she returned to work ( early employment
hours ), and the fourth variable was the number of quarters the mother
did not work after she had returned to work ( discontinuous employment
). Early employment hours was constructed by identifying the
quarter-year during which the mother returned to work and calculating
the average number of hours per week she worked from this quarter-year
until the child's third birthday (not including quarters during which
she reported working 0 hr). Many of the previous studies on the NLSY
have included a variable representing whether or not mothers worked
during the first year of the child's life. Although the timing of early
maternal employment should capture the effects of employment during the
first year, a variable employment during the first year was also
included to facilitate comparison of the results of this study with the
findings of previous studies.
Note that employment during the first 3 years and early employment hours
could have been combined into one variable by assigning mothers who were
never employed during the first 3 years a score of 0 hr. However, doing
so assumes that the difference between not being employed and being
employed 5 hr per week is the same as the difference between working 35
hr per week and working 40 hr per week. That is, it could not be assumed
that not being employed is simply one end of the employment intensity
dimension. Early employment status (comparing being employed with not
being employed) might have a different effect on children than the
intensity of employment and was therefore examined separately.
Early parental employment variable. Mothers reported the average number
of hours per week that their spouses spent at their jobs every year from
1979 to 1994. The variable for early paternal employment hours consisted
of an average of the spouse's job hours over the first 3 yearly
assessments following the birth of the child.
Child outcome measures. Five child outcome variables were examined:
compliance, behavior problems, cognitive development, self-esteem, and
academic achievement. For children who were in one of the four specified
age ranges at more than one assessment date, their multiple scores were
averaged to obtain the best possible estimate of their functioning.
Compliance was assessed using a six-item subscale from the Temperament
scale, which was developed for the NLSY. This scale correlates modestly
but significantly with later behavior problems ( Center for Human
Resource Research, 1995 ). Higher scores indicate greater compliance.
Children's behavior problems were assessed using the BPI, which was
developed to measure behavior problems in children age 4 and older. Many
items were derived from the Child Behavior Checklist ( Achenbach &
Edelbrock, 1981 ) and other child behavior scales ( Graham & Rutter,
1968 ; Kellam, Branch, Agrawal, & Ensminger, 1975 ; Peterson & Zill,
1986 ; Rutter, 1970 ). The parent reports the frequency with which the
child exhibited each of 28 specific problems (1 = often true, 2 =
sometimes true, 3 = not true ). Standard scores (based on all children,
not same sex) were used, with higher scores indicating more behavior
problems. This scale has demonstrated good construct validity ( Center
for Human Resource Research, 1993 ).
Children's self-esteem was measured using the global self-worth subscale
of the Self-Perception Profile for Children. This is a self-report
measure that assesses children's sense of self-competence in the domain
of academic skills and general self-worth ( Harter, 1985 ). It
correlates highly with teacher ratings and has good reliability (
Harter, 1985 ). This measure was administered to children age 8 years
and older.
Children's cognitive development was assessed using the PPVT-R ( Dunn &
Dunn, 1981 ), a widely used test of receptive language. This measure was
administered to all children age 3 years and older. It was administered
to all children in 1986 and 1992, and only to children without a
previous valid score in 1988, 1990, and 1994. Children's academic
achievement was measured using the PIAT. Three subtests from the PIAT
were administered to children age 5 years and older: mathematics,
reading recognition, and reading comprehension. Standard scores were
used for both the PPVT-R and PIAT. Both cognitive measures have good
reliability and validity ( Dunn & Dunn, 1981 ; Dunn & Markwardt, 1970 ).
Selection factors. Theory and previous research suggested a number of
family background variables that might act as selection factors for
early employment. The following background variables were created for
examining the effects of early maternal employment: family income,
mother's education, mother's age at the child's birth, child gender,
mother's IQ, child's race, birth order of the child, and marital status.
For the effects of parental employment, the following background
characteristics were used: family income, father's education, father's
age at the child's birth, child's gender, child's race, and birth order
of the child. Fathers' IQs were not assessed in the NLSY. Mother's
reports of these variables at the first assessment following the birth
of the child were used, except for income, which was based on reports
regarding the year before the child's birth. Mother's intelligence was
assessed using the Armed Forces Qualification Test (AFQT), which was
administered to all NLSY participants in 1980. The AFQT consists of the
sum of scores on four subtests of the Armed Services Vocational Aptitude
Battery: word knowledge, numeric operations, paragraph comprehension,
and arithmetic reasoning.
Moderating variables. On the basis of theory and research, the following
variables were examined as possible moderators of the effects of early
parental employment: marital status, race, gender, family income, and
job satisfaction. The six NLSY studies yielded conflicting findings
regarding the moderating effects of these variables, but each variable
was supported by at least one study. Job satisfaction was assessed by
asking individuals to indicate how they felt about their current jobs on
a 4-point scale from 1 = like it very much to 4 = dislike it very much.
Previous st
