Review of Economics of the Household

, 7:227

The connection between maternal employment and childhood obesity: inspecting the mechanisms

Authors

    • University of Georgia
  • Gerhard Glomm
    • Indiana University
  • Rusty Tchernis
    • Indiana University and NBER
Article

DOI: 10.1007/s11150-009-9052-y

Cite this article as:
Fertig, A., Glomm, G. & Tchernis, R. Rev Econ Household (2009) 7: 227. doi:10.1007/s11150-009-9052-y

Abstract

This paper investigates the channels through which maternal employment affects childhood obesity. We use time diaries and interview responses from the Child Development Supplement of the Panel Study of Income Dynamics which combines information on children’s time allocation, children’s BMI, and mother’s labor force participation. We find some evidence that supervision and nutrition play significant but small roles in the relationship between maternal employment and childhood obesity. Although the difference in the effect of maternal employment varies by mother’s education, we find few differences in the mechanisms by mother’s education.

Keywords

Maternal employmentChildhood obesityTime use data

JEL Classicification

I12J13J22

1 Introduction

Over the past several decades, obesity has swept across the US and other industrialized countries, affecting all age groups. The fraction of overweight children between the ages of 6 and 11 increased from 4% in the 1960s to 19% by 2003–2004. The problem of childhood obesity has already triggered a federal policy response—a recent law (Public Law 108–265) required schools to have a local wellness program in place starting by the beginning of school year 2006–2007. The program must address both nutritional and physical activity goals. The immediate cause of the increase in obesity is clear: calories taken in persistently exceed calories burned. The more fundamental reasons are less clear: why would so many people in these particular years choose to systematically take in more calories than they expend? According to Cutler et al. (2003) and Philipson and Posner (2003), technological progress is responsible for cheaper fattening foods and a more sedentary lifestyle, while Chou et al. (2004) claim that a decrease in smoking and an increase in the availability of restaurants, especially fast food restaurants, are responsible.

Any potential explanation for the phenomenal increase in childhood obesity must also involve changes in parental behavior, lifestyle, or attitudes (Patrick and Nicklas 2005; Golan and Crow 2004; Ebbling et al. 2002). One important change over this period that has touched family life in many ways is the increase in employment among mothers. Recently several papers have documented a positive relationship between maternal employment and the bodyweight of her children (Anderson et al. 2003; Lamerz et al. 2005; Classen and Hokayem 2005; Liu et al. 2005; Phipps et al. 2006; Courtemanche 2007; Cawley and Liu 2007; Chia 2008; Ruhm 2008). Interestingly, this connection seems to be especially pronounced for more educated, well off, white families. Taking the connection between mothers’ employment and childhood obesity as given, this paper aims to identify the mechanisms by which mothers’ labor supply affects children’s weight, and to explore why the effect of maternal employment is more pronounced for children from higher socioeconomic backgrounds. In addition, following the previous literature we concentrate our analysis on the intensive margin (hours of work) and not the extensive margin (labor force participation).

Specifically, our empirical strategy is to estimate the effect of maternal work hours on child’s weight controlling for a wide variety of potential channels. We obtain data on the potential channels through time diaries of the child’s day. Our approach treats the potential channels as omitted variables and tests whether the inclusion of the channel results in a significant difference in the effect of maternal work hours on child’s weight. Because the relationships between a mother’s work hours and her child’s activities, and between the child’s activities and the child’s weight status, may be due not only to a direct causal link, but also to some unobservable characteristic of the family or mother which results in these correlations, we must interpret the findings from this approach as highly suggestive and not necessarily causal. However, empirical techniques like instrumental variables or fixed effects, which may reduce some of this endogeneity, are not possible because of the small sample size available with the data used here.

The overarching theoretical principle guiding the empirical investigation is the concept of a health production function for children, where child’s health (inversely measured by obesity) is the output and mother’s time at home with the child is the input. Given a low level of maternal education, the child’s health production function is depicted by locus L in Fig. 1. Each additional hour of mother’s time spent with the child increases the child’s health but there are diminishing returns to mother’s time. The production function for a mother with a high level of education would lie above L because mothers with more schooling have superior information which allows the same input level to produce a better health outcome. However, it is not clear from economic theory whether the slope of the production function is affected by maternal education. Thus, the production function for a more educated mother could look like H1, with the same shape as L, or could look like H2, where the slope is steeper at every input level. The steeper slope implies that children benefit more in terms of health from an additional hour of their mother’s time if she is more educated than if she is not, at every input level. Klohe-Lehman et al. (2006) and Wardle et al. (2000) document that nutrition knowledge is an important determinant of dietary intake even after controlling for socio-economic characteristics such as income, race, schooling. The findings in Variyam et al. (1999) that nutrition knowledge of mothers gleaned from questionnaires is associated with substantially better diet of children is consistent with our specification of how the health production functions may vary by mother’s educational status and health knowledge.
https://static-content.springer.com/image/art%3A10.1007%2Fs11150-009-9052-y/MediaObjects/11150_2009_9052_Fig1_HTML.gif
Fig. 1

Maternal Education and Child’s Health Production. Notes: Line L represents the production function of child health for less educated mothers, while line H1 and H2 represent two possible production functions for more educated mothers of different levels of efficiency

Thus, we might expect the effect of mothers’ employment on children’s health, represented by the slope of the production function, to be different by mother’s education for two reasons. First, mother’s education may be related to the average input level. If more educated mothers work more hours on average, then even if the production functions have the same shape—as depicted by L and H1 in Fig. 1, more educated mothers are going to be on a steeper portion of the curve (point B) than the average less educated mother (point A). Second, mother’s education may increase the slope of the production function, as depicted by H2, such that given the same input level, more educated mothers are on a steeper slope (point C) than less educated mothers (point A). In either case, we would observe that an additional hour worked by a more educated mother will have a more detrimental effect on her child’s health than an additional hour worked by a less educated mother.

Economic theory suggests that there are various channels through which maternal employment can influence childhood bodyweight. First, a working mother has less time available to cook and prepare meals. Working mothers may decide to cook fewer meals at home, opting instead for more restaurant meals, or preparing more ready to eat meals, take out or delivered meals. Restaurant meals, especially from fast food restaurants, and ready to eat meals are generally more densely packed with calories than meals prepared at home. In addition, children of mothers with less time to devote to household activities might skip some meals, e.g. breakfast. There is ample evidence (Stauton and Keast 1989; Morgan et al. 1986) that skipping breakfast is associated with overall higher calorie consumption. Moreover, a low meal frequency may lead to higher concentrations of 24 h insulin, which, in turn, can lead to increased fat deposition and higher body weight (see Ma et al. 2003). This channel suggests that higher maternal employment results in higher children’s bodyweight.

Similarly, working mothers may have less time and energy available to supervise and participate in their children’s activities. This may mean that children are more autonomous in choosing their own activities or that the children spend more time in the care of others—either in school or in child care. Since parents presumably care more about the future health of their children than do other caretakers or children themselves, this may result in more time in front of the television, less time in outside activities, and a greater quantity of unhealthy snacks. Anderson and Butcher (2006) argue that schools “have given students greater access to ‘junk’ foods and soda pop,” and find that access to junk food in schools increases students’ weight. However, using time use data, Bianchi (2000) finds that working mothers do not sacrifice time with their children but instead reduce their leisure time. In addition, other caretakers may be able to offer a more structured routine involving physical activities with other children and healthier snacks than parents might provide. For example, von Hippel et al. (2007) show that children gain weight during summer months more rapidly than during the school year. Thus, it is not clear a priori whether this channel suggests that more maternal working hours results in higher or lower child weight.

Third, increased hours worked by the mother results in higher household income. There is a large empirical literature which finds a negative relationship between obesity and socioeconomic status (e.g. Gordon-Larsen et al. 2003; Zhang and Wang 2004a, b). The reasons for this linkage are debatable. Higher disposable income may allow households to provide better quality food or enroll children in organized activities which would reduce children’s weight. However, the linkage might be entirely due to selection; people with low discount rates invest in education, which brings them higher earnings, and invest in their health, which keeps their weight in the normal range. While income in general is believed to have a negative effect on obesity, higher household income results in more restaurant meals (if restaurant meals are normal goods) which could result in a higher bodyweight for reasons elaborated above. Thus, economic theory does not unambiguously predict whether this channel results in higher or lower bodyweight for children.

Finally, we expect that currently working mothers returned to work sooner after birth and thus were less able to breastfeed or stopped breastfeeding earlier. There is evidence that bottle fed infants are more likely to be overweight as children and adults than breastfed infants (Lucas et al. 1980, 1981). Thus, it may be that a mother’s average work hours are correlated with her child’s BMI because they are a good indicator of the probability that her child was bottle fed.

To quantify the importance of the various channels through which maternal employment may affect children’s body weight, we use the Child Development Supplement (CDS) of the Panel Study of Income Dynamics (PSID). The CDS is well suited for this analysis because it includes the height and weight of the child, time diaries of the child’s activities during one weekday and one weekend day, and a great deal of information about the child’s household through the linkage with the main PSID survey.

The data and sample are described in Sect. 2. There, using our dataset, we also replicate the empirical finding that maternal employment has a positive and significant effect on children’s BMI which is stronger for more educated mothers. In Sect. 3, we detail our empirical strategy, which involves estimating a series of equations to determine whether a variety of potential mediators significantly change the estimated effect of maternal employment on children’s BMI. The results are presented in Sect. 4. We find that TV watching and the number of meals eaten in a day appear to play significant but small roles in the relationship between maternal employment and children’s weight. In addition, we find only one difference in the potential channels by the mother’s educational status. Among mothers with no more than a high school diploma, maternal employment significantly increases the amount of time a child spends in school, which in turn, significantly decreases a child’s BMI. For more educated mothers, it is child care that increases when mothers work more, but the effect on children’s BMI is similarly negative. While we don’t claim that our findings are causal we see them as thought provoking and suggesting directions for further research. Finally, we offer a summary and conclusions in Sect. 5.

2 The data

The data used in this study come from the CDS of the PSID. The PSID has followed approximately 5,000 families since 1968. This original sample includes an equal probability, nationally representative sample of about 3,000 households called the Survey Research Center (SRC) sample, and a sample of about 2,000 low-income families called the Survey of Economic Opportunity (SEO) sample. Over time, the study has added the ‘split-off’ households of children and other members of the original PSID households after they leave and start their own families, such that in 1996 there were over 8,700 families involved in the survey.

Currently, the CDS consists of two waves. The first wave involves a sample of approximately 3,500 children under the age of 13 who are members of PSID families in 1997. Because the sample of children is drawn from both the SRC and the SEO samples, the children’s sample has unequal selection probabilities. The second wave involves re-interviewing about 2,900 children in 2002, when they were between the ages of 5 and 17. In this analysis, we use approximately 3,400 observations of 2,500 children from 1,100 PSID families. We have two observations on many children and there are some siblings as the CDS included at most two children from a family. Of the approximately 3,000 observations (3,500 + 2,900 − 3,400 = 3,000) that we omit, 700 are of children under the age of 3 at the first interview; the remainder are omitted because of missing information on height, weight, or mother’s work hours, or there was no complete time diary for the child in a given wave.1 Because we omit a large fraction of the total observations available in the CDS due to these sample restrictions, we provide comparison statistics on the sample without these restrictions in Appendix Table 8 to demonstrate that the sample used in this analysis is not selected on any important observable characteristics.

Table 1 presents some descriptive statistics by mother’s education (Appendix Table 8 presents statistics on the full CDS sample for comparison). The primary variables of interest in this analysis are the child’s percentile BMI and whether the child is overweight. We do not use the BMI scale itself because its connection to overweight is highly dependent on the child’s age and sex, unlike for adults. BMI is calculated as weight in kilograms divided by height in meters squared (kg/m2). The Centers for Disease Control (CDC) has produced a chart of percentiles describing the BMI distribution by the age (in months) and sex of children based on early waves (from the 1960s, 1970s, and 1980s) of the nationally representative National Health and Nutrition Examination Survey (NHANES). Thus, we use these percentiles as our continuous dependent variable instead of the actual BMI because we believe that it may better capture the child’s weight status. Because we are most interested in changes in weight near the overweight threshold, following the CDC, we define children to be overweight if the child’s BMI is above the 95th percentile for their age and sex, and at risk of being overweight if their BMI is between the 85th and 95th percentiles, and create limited dependent variables with these cutoffs. Because of the growing numbers of overweight children and because this sample includes an oversample of black children, who have a higher rate of overweight than the national average, substantially more than 5% of children are classified as overweight in our sample measured in 1997 and 2002: the percentage of children overweight is 23.4 among less educated mothers and 20.3 among more educated mothers. More strikingly, 39.4% of children with less educated mothers and 35.0% of children with more educated mothers are either overweight or at risk of being overweight. Finally, the children’s average percentile in the CDC’s BMI distribution ranges between 64 and 67%, which again reflects the high proportion of overweight children in this sample.
Table 1

Descriptive statistics by mother’s education

 

Mother’s Ed ≤ 12 years

Mother’s Ed > 12 years

Mean

N

Mean

N

Overweight (BMI > 95th percentile for age and sex)

23.4%

1,734

20.3%

1,539

Risk of overweight (85th percentile–95th percentile)

16.0%

1,734

14.7%

1,539

BMI of children

20.59

1,734

19.85

1,539

BMI percentile of childrena

66.5

1,734

64.3

1,539

Hours per week worked by mother in last 2 years (if any)

25.88

1,377

27.88

1,342

Hours per week worked by mother over child’s life (if any)

17.27

1,519

20.53

1,439

Mother didn’t work in last 2 years

20.6%

1,734

12.8%

1,539

Mother never worked during child’s life

12.4%

1,734

6.5%

1,539

Age of child

9.82

1,734

9.52

1,539

Black

43.9%

1,732

30.4%

1,534

Hispanic

10.9%

1,732

2.2%

1,534

Female

51.2%

1,734

48.0%

1,539

First born child

43.4%

1,734

58.8%

1,539

Birth weight (pounds)

7.19

1,714

7.47

1,525

Number of children in household

2.41

1,734

2.23

1,539

Age of mother at child’s birth

25.52

1,622

28.72

1,528

Education of mother in 1997 (years)

10.53

1,734

14.81

1,539

Mother is obese (BMI > 30)

27.3%

1,690

18.9%

1,518

Father is obese (BMI > 30)

23.8%

1,397

18.3%

1,386

Parents always married over child’s life

47.2%

1,734

70.0%

1,539

Annual labor income over child’s life (‘000)

$26.30

1,390

$52.16

1,363

Hours per week worked by father over child’s life

34.38

1,384

39.21

1,341

Mother received foodstamps in last year

19.8%

1,676

5.4%

1,498

Northeast

10.5%

1,734

18.7%

1,536

North Central

22.1%

1,734

24.5%

1,536

South

45.7%

1,734

38.5%

1,536

West

21.7%

1,734

18.3%

1,536

Urban

51.8%

1,547

57.5%

1,444

aPercentiles based on 2000 CDC Growth Charts by gender and child’s age in months

The other key variable in this study is the hours per week worked by the mother. For this measure, we average the mothers’ weekly working hours for a representative week over the 2 years prior to the interview. If the mother does not work, zero hours are assigned. On average, less educated mothers worked about 26 h per week while more educated mothers worked closer to 28 h per week on average. The fraction of mothers who did not work at all during these 2 years is almost 21% for less educated mothers and about 13% for more educated mothers. We use only the last 2 years because most of the channels investigated here would only be affected by recent employment patterns. However, because a high BMI is typically the result of persistent and long-run over-consumption of calories, we also examine the effect of mother’s hours averaged over the child’s entire lifetime (which ranges between 3 and 17 years). Over the longer period, mothers’ average hours are about 8 h lower and fewer mothers have never worked during this child’s life. Again, the hours and labor force participation rates are higher for the more educated mothers.

The remainder of Table 1 describes the sample with a list of our main demographic controls. On average, children were just under 10 years old. Because the CDS draws from both the SRC and the SEO samples, there are a large proportion of black families—44% of children with less educated mothers are black and 30% of children with more educated mothers are black. On the other hand, the Hispanic sample is disproportionately small because of the design of the PSID. The Latino sample added to the PSID in 1990 was dropped in 1995 and a new immigrant sample was added in 1997. The CDS includes about 250 immigrant children from this new sample, some of whom are Hispanic.

Of particular importance to this analysis, we can calculate the mother’s BMI from the main household interview which asks the height and weight of the head of the household and the spouse. These questions are only available in the 1986, 1999, 2001, and 2003 interviews. We use the mother’s BMI in 1999 for the 1997 CDS wave and her BMI in 2001 for the 2002 CDS wave. The definition for obesity among adults accepted by the CDC is a BMI above 30. In our sample, over 27% of less educated mothers are obese while almost 19% of more educated mothers are obese. The difference in father’s BMI between the two groups is slightly less pronounced. We include a control for the number of hours worked by the father because father’s time may affect child’s weight and, given assortative mating and specialization in marriage, is likely related to mother’s work hours (although the coefficient on father’s hours is not significant in any specification). We include whether the mother receives food stamps as there is some evidence of a positive effect of food stamps on obesity (Gibson 2003) (although the coefficient on food stamp receipt is not significant in any specification).

2.1 Replication

Before describing the time diaries which allow us to investigate the mechanisms by which mother’s employment can affect a child’s BMI, we want to confirm that the empirical relationship between mother’s employment and child’s BMI exists in the PSID. Most of the previous American studies on this issue used the NLSY. We replicate the previous analysis using PSID data in Table 2. For comparison, we construct control variables similar to those used by Anderson et al. (2003). We find that, in the full sample, mother’s work hours over the child’s lifetime are positively correlated with the probability that the child is overweight. In addition, consistent with Anderson et al. (2003), Lamerz et al. (2005), and Ruhm 2008), the effect of maternal employment is greater for more advantaged children (those with a more educated mother). Thus, this relationship appears to exist across several data sources.
Table 2

Replication of result that mother’s employment affects probability of being overweight

Dependent variable: overweight

Sample

Full sample

Mother’s Ed ≤ 12

Mother’s Ed > 12

Log hours/week worked by mother over child’s life

0.016*

0.006

0.021*

(0.006)

(0.009)

(0.010)

Child’s age

−0.019*

−0.017

−0.021+

(0.009)

(0.013)

(0.012)

Child’s age squared

0.001

0.000

0.000

(0.000)

(0.001)

(0.001)

Black

0.068**

0.056+

0.086**

(0.021)

(0.029)

(0.032)

Hispanic

0.145**

0.174**

0.152*

(0.038)

(0.051)

(0.064)

Female

−0.050**

−0.040*

−0.052**

(0.013)

(0.020)

(0.019)

First born child

0.014

0.009

0.018

(0.015)

(0.022)

(0.023)

Birth weight (pounds)

0.016**

0.016*

0.019*

(0.006)

(0.008)

(0.008)

Number of children in household

−0.011

−0.011

−0.004

(0.008)

(0.010)

(0.012)

Age of mother at birth

0.002

0.001

0.002

(0.001)

(0.002)

(0.002)

Education of mother in 1997 (years)

−0.005

−0.000

−0.021*

(0.003)

(0.005)

(0.009)

Mother is obese (BMI>30)

0.149**

0.116**

0.184**

(0.020)

(0.026)

(0.032)

Breastfed

−0.042*

−0.048*

−0.027

(0.017)

(0.023)

(0.024)

Fraction of child’s life parents married

−0.031

−0.001

−0.095*

(0.025)

(0.034)

(0.040)

Log labor income/1000 over child’s life

−0.008

−0.002

0.006

(0.010)

(0.014)

(0.015)

Log hours/week worked by father over child’s life

−0.008

−0.013

−0.007

(0.010)

(0.015)

(0.017)

Northeast

−0.027

−0.021

−0.019

(0.022)

(0.035)

(0.030)

North Central

−0.013

−0.030

0.000

(0.019)

(0.027)

(0.029)

West

−0.033

−0.035

−0.033

(0.021)

(0.031)

(0.031)

Urban

−0.021

−0.047*

0.006

(0.016)

(0.023)

(0.022)

2002 interview

0.057**

0.058**

0.064**

(0.015)

(0.022)

(0.023)

Observations

4290

2188

1911

Robust standard errors in parentheses. Standard errors adjusted for intra−cluster correlations at the family level. + significant at 10%; * significant at 5%; ** significant at 1%

2.2 Time diaries

The time diaries are a unique feature of these data. Time diaries have been shown in various studies to provide valid information about one’s activities.2 In the CDS, the primary caregiver or the child was asked to write down what the child was doing at every point in time over 2 days—1 week day and 1 weekend day. We have taken this information and divided the child’s time into 16 categories: sleeping, eating, attending school, being baby-sat, attending to the personal care of oneself or others, reading/talking/listening to music, watching TV, playing indoor games, socializing, shopping, traveling, playing on the computer or with video games, doing homework, playing sports, doing chores around the house or working for pay, and other miscellaneous passive activities, e.g. hobbies and watching others do activities. For a detailed list of the specific activities included in each category, see the Appendix.

Since a child can be engaged in multiple activities simultaneously, the time diary permits two activities to be assigned to any given time—a primary and a secondary activity. For example, a child could be watching television while being in daycare. Either one of these could be listed as the primary or secondary activities. We use all of the available information and, as a result the total number of hours accounted for over the 2 days is greater than 48 h.

Table 3 provides the number of hours that a child spends on these activities by the child’s age and mother’s education. These activities are sorted by time use in the first column. By far, sleeping takes the most amount of time. After sleeping, children spend most of their time reading/talking/listening to music, in school, and watching TV.
Table 3

Average time use over 2 days by mother’s education and child’s age (in hours)

 

Mother’s Ed ≤ 12

Mother’s Ed > 12

Child’s age < 10

Child’s age ≥ 10

Child’s age < 10

Child’s age ≥ 10

Sleeping

22.0

20.4

21.7

19.8

Reading/talking/listening music

6.6

7.8

6.9

7.9

Attending school

5.3

5.5

5.1

5.9

TV watching

5.3

5.8

4.5

5.3

Eating

3.1

2.9

3.1

2.9

Playing indoor games

2.9

0.8

3.1

0.8

Sports

2.7

2.1

2.5

2.0

Personal care of self/others

2.2

2.5

2.0

2.4

Socializing

1.6

1.8

1.7

2.1

Misc. passive activities

1.4

1.2

1.9

1.3

Traveling

1.3

1.5

1.4

1.7

Shopping

1.0

1.1

1.2

1.0

Chores/work

0.8

1.8

0.9

1.6

Computer/video games

0.8

1.7

0.9

1.8

In child care

0.8

0.1

0.9

0.1

Homework

0.5

0.9

0.5

1.4

Totala

58.4

57.7

58.5

57.9

Observations

825

908

788

752

Note: A description of each of the categories is provided in the appendix

aTotal is greater than 48 h because at any given time, two activities can be reported

One potential problem for our analysis is the possibility that the data quality of the time diary entries may be worse for mothers who work. That is, mothers who work may know less about their child’s activities and thus report those activities with more measurement error. It is true that children are more likely to fill out the diary themselves if their mother works more hours. However, on average, the children of mothers who work long hours are older and the age of the child is the strongest predictor of how involved the child was in filling out the diary. This measurement error argument assumes that the mother is a more accurate reporter of their children’s activities than the children themselves. However, mothers are more likely to be influenced by social norms in their responses than children, so one could make the argument that measurement error is smaller when children report their own activities. In any case, we argue that any bias from this type of measurement error is negligible because when we control for whether the mother filled out the diary without the child’s help, the results presented in this paper are unchanged.

Finally, these data also provide a few diet-related aspects of the household. These are shown in Table 4. We know from the time diary whether meals take place in a restaurant or at home. However, we cannot distinguish whether the meal eaten at home is from a restaurant (like take-out or delivery pizza). On average, fewer than six meals were eaten over the 2 days, and less than one on average was eaten in a restaurant. We are also interested in breastfeeding and allowances which can be affected by maternal employment and may impact a child’s nutritional intake. These variables are available from the CDS parent interview. Although we find a negligible difference in the probability of a child being breastfed among mothers by working hours (not shown), more educated mothers are almost twice as likely to have breastfed. Between 50 and 60% of children over the age of 5 receive an allowance.
Table 4

Determinants of diet by mother’s work hours

 

Mother’s Ed ≤ 12

Mother’s Ed > 12

Child’s age < 10

Child’s age ≥ 10

Child’s age < 10

Child’s age ≥ 10

Total number of meals

5.8

4.8

6.4

5.1

Percent of meals in a restaurant

9.7%

11.9%

12.2%

14.6%

Child breastfed as infant

36.4%

34.7%

62.8%

57.3%

Percent with an allowance (age > 5)

61.4%

59.3%

57.6%

53.1%

Observations

825

908

788

752

3 The empirical strategy

The goal of this paper is to investigate the channels through which maternal employment affects children’s BMI. We assume that maternal employment affects the number and composition of meals and the nature of her children’s activities, which in turn influence calorie intake and expenditure, thereby affecting the child’s BMI. Thus, our empirical strategy is to estimate the effect of maternal work hours on child’s BMI controlling for a wide variety of potential channels.

More formally, we estimate several regression equations, one for each of the potential channels for which we have data. This estimation strategy has been used in other contexts by Hoxby (2000); Levine and Rothman (2006), and Baum and Ruhm (2007) and we follow their example. As a baseline, we first measure the effect of maternal employment, MWH, on a child’s percentile BMI, pBMI, using OLS estimation. We use the log of mother’s work hours to reduce the effect of a few mothers who work a large number of hours per week.3 In this regression we do not include any possible channels, but only control variables for characteristics of the child and family, X. This regression equation is given by
$$ pBMI_{i} = \alpha_{0} + \alpha_{1} \ln \left( {MWH} \right)_{i}\,+\,\alpha_{2} X_{i} + \varepsilon_{i} . $$
(1)
Each additional equation also includes one potential channel. In Eq. 2, for example, time spent watching TV is added. Thus we estimate
$$ pBMI_{i} = \beta_{0} + \beta_{1} \ln \left( {MWH} \right)_{i}\,+\,\beta_{2} TV_{i} + \beta_{3} X_{i} + \mu_{i,} $$
(2)
where εi and μi are idiosyncratic error terms with mean zero. Thus a reduction in the marginal effect of maternal employment on the child’s percentile BMI in the presence of a potential channel, such as watching television, can be interpreted as the part of the effect of maternal employment operating via that channel. In addition, we test whether α1 β1, and report the difference (α1 − β1) and the corresponding robust standard errors. We adjust the standard errors in all regressions for intra-cluster correlations at the family level.
Thinking of each potential channel as an omitted variable is useful in interpreting the values of the coefficients we present. The omitted variable equation is given by:
$$ \alpha_{1} \approx \beta_{1} + \beta_{2} \rho , $$
(3)
where α1, β1, and β2 are the coefficients from Eqs. 1 and 2, and ρ represents the correlation between mother’s working hours and the potential channel, in this case, the amount of time a child spends watching TV. Thus, a positive coefficient on the difference (α1 − β1) indicates that β2ρ > 0, suggesting that the effect of the potential channel on the child’s percentile BMI (β2) and the correlation between MWH and the potential channel (ρ) are the same sign. For example, if α1 − β1 > 0 when watching TV is included in the specification, then this suggests that the effect of TV watching on child’s BMI percentile is positive and TV watching is positively correlated with mother’s work hours. Alternatively, both β2 and ρ could be negative, meaning that an activity decreases a child’s BMI and higher maternal employment decreases the amount of time a child spends on this activity. Given this, we could interpret this significant difference as suggesting that the direct effect of mother’s working hours on the child’s BMI is reduced when TV watching is included in the specification. On the other hand, a negative value indicates that β2 and ρ have opposite signs. For example, if α1 − β1 < 0 when “attending school” is included in the specification, then this suggests that, since we expect that mother’s working hours and the time spent attending school are positively correlated, the effect of attending school on child’s BMI is negative. Thus, if this coefficient was significant, it would suggest that mothers’ working could reduce their child’s BMI because the child spends more time attending school.

Besides using percentile BMI as a dependent variable, we also estimate two sets of Probit regressions with the dependent variables representing whether the child is overweight or at risk of overweight (BMI percentile above 85) or simply overweight (BMI percentile above 95). The goal for the Probit regressions is to evaluate the effects of potential channels at the right tail of the weight distribution. We report marginal effects in all tables.

In all of the regressions, we control for the variables listed in Table 1 plus child’s age squared, whether the time diary was taken in the winter months, winter interacted with the North Central region, winter interacted with the Northeast region, whether the weekend day of the diary was a Saturday (vs. a Sunday), which wave of the CDS the observation comes from, and whether the mother answered the time diary without the assistance of the child. South is the omitted region category. We also include missing indicators for all of the control variables included.

4 The results

Table 5 presents the results of our analysis using the full sample. The first column uses percentile BMI and the second and third columns use the risk of overweight and overweight cutoff categories described above. The first row presents α1 from Eq. 1, the effect of log mother’s working hours on the dependent variable with only the family and child characteristics included in the specification. The rest of the table reports the difference between α1 and β1j, where j indicates which one of the twenty potential channels is included in the specification. For example, including the number of meals reduces the coefficient on maternal employment on BMI percentile from 1.339 to 1.216 (1.339–0.123). The potential channels are listed to the left of the coefficient and the significance levels of the difference are indicated with asterisks after the coefficient.
Table 5

The change in the effect of mother’s employment on child’s weight for the full sample, by channel

Dependent variable

percentile BMI

>85th percentile

>95th percentile

Ln MWH

1.339**

0.021**

0.014*

(0.489)

(0.008)

(0.007)

Number of meals

0.123*

0.0040*

0.0027

(0.044)

(0.0017)

(0.0017)

% meals in restaurant

0.023

0.0010

0.0002

(0.018)

(0.0008)

(0.0006)

Breastfed

0.021

−0.0001

−0.0005

(0.026)

(0.0011)

(0.0014)

Receives allowance

−0.003

0.0002

0.0001

(0.009)

(0.0004)

(0.0004)

Fraction of time spent

Sleeping

0.004

0.0004

0.0005

(0.009)

(0.0006)

(0.0007)

Reading/talking/listening music

0.049+

0.0013

0.0021

(0.030)

(0.0010)

(0.0014)

Attending school

−0.010

−0.0006

−0.0002

(0.019)

(0.0009)

(0.0010)

TV watching

0.042+

0.0022+

0.0015

(0.026)

(0.0013)

(0.0012)

Eating

0.018

0.0007

0.0011

(0.017)

(0.0008)

(0.0010)

Playing indoor games

0.005

0.0001

0.0005

(0.012)

(0.0005)

(0.0007)

Sports

0.000

0.0002

0.0000

(0.004)

(0.0007)

(0.0001)

Personal care of self/others

0.007

0.0000

−0.0006

(0.010)

(0.0004)

(0.0006)

Socializing

0.011

0.0002

0.0002

(0.016)

(0.0004)

(0.0005)

Misc. passive activities

−0.002

0.0002

0.0000

(0.010)

(0.0004)

(0.0004)

Traveling

−0.009

−0.0007

−0.0005

(0.012)

(0.0007)

(0.0006)

Shopping

0.003

0.0001

0.0000

(0.007)

(0.0003)

(0.0005)

Chores/work

0.009

0.0003

0.0010

(0.015)

(0.0007)

(0.0009)

Computer/video games

0.000

0.0000

0.0000

(0.006)

(0.0002)

(0.0003)

In child care

−0.040

0.0002

−0.0012

(0.038)

(0.0015)

(0.0017)

Homework

0.017

0.0004

0.0001

(0.018)

(0.0005)

(0.0003)

Observations

3424

3424

3424

The first row presents the effect of ln MWH on the dependent variable. All of the following rows present the difference between the coefficient in the first row and the coefficient on ln MWH if the variable listed to the left is included in the specification. Each coefficient is from a separate regression. OLS estimation is used in the first column and probit in the second and third columns. Marginal effects are reported in all columns. Robust standard errors are in parentheses. Standard errors are adjusted for intra-cluster correlations at the family level. + significant at 10%; * significant at 5%; ** significant at 1%

We can see in Table 5 that maternal working hours are associated with a significant increase in all three measures of child’s weight for the full sample of mothers. The only three channels whose inclusion results in a significant difference are the number of meals, the time spent reading/talking/listening to music, and the time spent watching TV. The decline associated with including the number of meals explains about 9% (0.123/1.339 = 0.09) of the effect of maternal working hours on child’s percentile BMI, 19% (0.004/0.021 = 0.19) of the effect on the probability of being at risk of overweight with no significant effect for probability of overweight. The decline associated with including reading/talking/listening to music explains about 4% of the effect on BMI percentiles. Similarly, the decline associated with including watching TV explains about 3% of the effect on BMI percentiles, and 10% on probability of being at risk of overweight. The TV watching result has been found in other research (Proctor et al. 2003; Hancox et al. 2004). We find evidence in these data that TV watching and reading/talking/listening to music are substitutes. In particular, holding all else constant, children who watch TV more spend significantly less time reading/talking/listening to music. While reading/talking/listening to music is a passive set of activities like TV watching, they may be healthier activities in that they are less complementary to eating junk food than is TV watching. Thus, these results parallel each other; consistent with this, their magnitudes are similar. The result that requires further discussion is the effect of the number of meals.

Research has found an inverse relationship between frequency of eating and BMI (Ma et al. 2003) and the importance of breakfast (Stauton and Keast 1989; Morgan et al. 1986), both of which are consistent with this finding. In some additional analysis that we conducted (not shown), we found that the inclusion of the number of breakfasts (defined as a meal that occurs between 5 AM and 11 AM), whose value can be 0, 1, or 2, renders the coefficient on the number of meals insignificant. The coefficient on the number of breakfasts is significant so we interpret this pattern as suggesting that the effect of the number of meals on child’s percentile BMI through mother’s work hours is explained by skipping breakfasts. The probability of having 2 breakfasts over the 2 diary days is 67% for mothers who work <10 h/week but only 46% for mothers who work fulltime or more (34+ hours/week). Other variables which are also related to the number of meals were inspected but none of them had significant coefficients or altered the coefficient on the number of meals. In particular, we examined the effect of the number of breakfasts, where breakfast was defined as a meal within 2 h of waking up after a sleep of 5 h or more; the effect of whether the primary caregiver says that the child regularly has breakfast; and the effect of whether the child participated in a federal breakfast/lunch program. Thus, we argue that the evidence points toward skipping breakfast as an important mechanism. Our evidence is consistent with findings in a large body of literature that finds a pronounced correlation between skipping breakfast and overweight and obesity (Gibson and O’Sullivan 1995; Pastore et al. 1996; Ortega et al. 1998; Summerbell et al. 1996; Wolfe et al. 1994).

However, it may be that children who never miss breakfast also eat more nutritiously in general and since neither of our measures of the number of meals nor the number of breakfasts captures the calorie content or size of the meals, we cannot rule out other explanations. In particular, there is evidence that working married mothers spend a smaller share of their food budgets on vegetables, fruits, milk, and meat and beans than non-working married mothers (Ziol-Guest et al. 2006), suggesting that the content of the meals may be the important factor.

In any case, it is surprising that the effect of number of meals is stronger than the effect of TV watching or reading/talking/listening to music. Even more surprising is the small number of significant effects observed overall. In particular, we expected that time in the care of others and playing sports would have an effect; however, the zero effect of sports is consistent with Cawley et al. (2007), who find that an increase in mandated time for physical activity in school does not have a significant impact on children’s BMI.

To determine if the effect varies by mother’s education, we repeat this strategy on two subsets of the sample divided by mother’s education. These results are presented in Table 6. For mothers with a high school education or less, the effects of mother’s working hours on percentile BMI and the probability of being (at risk of) overweight are not significant in the baseline specifications. For more educated mothers, the total effect of the mother’s working hours is large and highly significant. This effect is roughly 50% larger than in the full sample and many times larger than in the sample of mothers with less education. This difference between education groups that we find is consistent with Anderson et al. (2003).
Table 6

The change in the effect of mother’s employment on child’s weight, by mother’s education and channel

Sample

Mother’s Ed ≤ 12

Mother’s Ed > 12

Dependent variable

percentile BMI

>85th percentile

>95th percentile

percentile BMI

>85th percentile

>95th percentile

Ln MWH

0.356

0.009

0.004

2.013**

0.025*

0.017+

(0.645)

(0.011)

(0.009)

(0.758)

(0.012)

(0.010)

Number of meals

0.052

0.0020

0.0009

0.1503*

0.0044

0.0024

(0.041)

(0.0018)

(0.0017)

(0.075)

(0.0030)

(0.0033)

% meals in restaurant

0.006

0.0004

0.0000

0.047

0.0017

0.0000

(0.013)

(0.0009)

(0.0003)

(0.039)

(0.0018)

(0.0017)

Breastfed

0.010

−0.0023

−0.0028

0.002

0.0002

0.0010

(0.046)

(0.0017)

(0.0018)

(0.019)

(0.0013)

(0.0019)

Receives allowance

−0.005

0.0003

0.0010

−0.002

0.0000

0.0002

(0.022)

(0.0011)

(0.0014)

(0.008)

(0.0004)

(0.0008)

Fraction of time spent

Sleeping

−0.004

0.0005

0.0001

0.008

0.0005

0.0011

(0.025)

(0.0011)

(0.0012)

(0.022)

(0.0014)

(0.0023)

Reading/talking/listening music

0.058

0.0007

0.0015

0.022

0.0019

0.0027

(0.046)

(0.0012)

(0.0020)

(0.038)

(0.0019)

(0.0026)

Attending school

−0.103+

−0.0041+

−0.0009

0.020

0.0007

0.0001

(0.054)

(0.0023)

(0.0019)

(0.027)

(0.0011)

(0.0009)

TV watching

0.007

0.0004

0.0000

0.036

0.0044

0.0046

(0.024)

(0.0012)

(0.0009)

(0.060)

(0.0030)

(0.0035)

Eating

−0.010

0.0000

0.0008

−0.004

0.0000

−0.0001

(0.032)

(0.0016)

(0.0019)

(0.044)

(0.0014)

(0.0017)

Playing indoor games

0.004

0.0003

0.0003

0.012

−0.0003

0.0019

(0.011)

(0.0006)

(0.0008)

(0.032)

(0.0015)

(0.0020)

Sports

−0.004

−0.0003

−0.0001

−0.005

0.0001

0.0000

(0.023)

(0.0017)

(0.0006)

(0.027)

(0.0003)

(0.0001)

Personal care of self/others

−0.013

−0.0008

−0.0019

−0.034

−0.0006

−0.0003

(0.029)

(0.0013)

(0.0016)

(0.040)

(0.0010)

(0.0011)

Socializing

0.005

0.0000

0.0001

0.007

0.0003

0.0004

(0.023)

(0.0002)

(0.0004)

(0.020)

(0.0008)

(0.0011)

Misc. passive activities

−0.013

−0.0001

0.0004

0.001

0.0000

−0.0001

(0.019)

(0.0006)

(0.0009)

(0.006)

(0.0002)

(0.0011)

Traveling

−0.025

−0.0013

−0.0004

−0.002

−0.0003

−0.0007

(0.029)

(0.0015)

(0.0008)

(0.012)

(0.0009)

(0.0012)

Shopping

0.002

0.0001

0.0000

0.000

0.0001

−0.0001

(0.011)

(0.0009)

(0.0011)

(0.001)

(0.0006)

(0.0005)

Chores/work

0.026

0.0009

0.0011

−0.003

−0.0002

0.0003

(0.026)

(0.0011)

(0.0012)

(0.013)

(0.0007)

(0.0012)

Computer/video games

−0.004

0.0001

−0.0002

0.009

0.0003

−0.0001

(0.019)

(0.0008)

(0.0006)

(0.022)

(0.0010)

(0.0005)

In child care

0.045

0.0009

0.0010

−0.153+

0.0003

−0.0053+

(0.039)

(0.0017)

(0.0023)

(0.081)

(0.0032)

(0.0031)

Homework

−0.001

0.0000

0.0001

0.049

0.0014

0.0004

(0.016)

(0.0008)

(0.0011)

(0.043)

(0.0016)

(0.0011)

Observations

1,733

1,733

1,733

1,537

1,537

1,537

The first row presents the effect of ln MWH on the dependent variable. All of the following rows present the difference between the coefficient in the first row and the coefficient on ln MWH if the variable listed to the left is included in the specification. Each coefficient is from a separate regression. OLS estimation is used in the first column and probit in the second and third columns. Marginal effects are reported in all columns. Robust standard errors are in parentheses. Standard errors are adjusted for intra-cluster correlations at the family level. + significant at 10%; * significant at 5%; ** significant at 1%.

We observe that the number of meals significantly changes the coefficient on MWH for more educated mothers only. The fact that we find a stronger effect for more educated mothers may imply that missing meals is more important for these children because the mothers know more about nutritious meals and transfer the knowledge of healthy eating to their children. However, the difference between the education groups in the difference in the effects is not significant (not shown on the table). One interesting difference by education does appear however. For less educated mothers, the effect of mother’s work hours is reduced when we control for the time spent in school. In contrast, for more educated mothers, the effect of mother’s work hours is reduced when we control for the time spent in child care. These differences across education groups are both significant.

The interpretation of this finding may be causal, a correlation or due to reverse causality. In additional analysis (not shown), we find that the effect of school for less educated mothers is only significant when school-age children (age 5–16) who did not attend school either of the time diary days are included. This suggests that school absences, rather than attending before or after school programs, for example, are driving this result. In addition, we find that the child care result is driven by children in fair or poor health. If we restrict the sample to only those children whose primary caregiver rates their health as excellent, very good or good (97% of the sample), the child care result disappears.4 Thus, mothers who work more hours may ensure that their children do not miss school or child care, consistent with Nomaguchi (2006), and this reduces their BMI. However, it may be that parents who ensure that their children do not miss school or child care are also less likely to have overweight children. Or, alternatively, it may be that overweight children are more likely to miss school or child care for health reasons and this affects their mother’s ability to work more hours consistently. The latter two interpretations would imply that a change in mothers’ hours would not change their children’s weight status. We are not able to determine which of the interpretations are accurate without additional analysis beyond the scope of this study.

There is some concern that our main measure of maternal work hours, which captures only the current pattern of employment, may limit the effect of maternal work since weight status is the cumulative result of many years of behaviors. That is, past maternal work may have an effect on current weight which we are not capturing. To address this concern, we use mother’s work hours since birth, not just the last 2 years. These results are presented in Table 7 and suggest that long run and short run maternal employment affects child weight similarly. Specifically, for the full sample, only the inclusion of the number of meals and TV watching lead to significant changes in the effect of mother’s work. Reading/talking/listening to music is positive and similar to TV watching in magnitude as in Table 5, but the difference is not significant in this case. As in Table 6, when the sample is divided by education, three potential channels have a significant effect on child weight.
Table 7

The change in the effect of mother’s employment over child’s life on child’s weight, by channel

Sample

Full sample

Mother’s Ed ≤ 12

Mother’s Ed > 12

Dependent variable

percentile BMI

>85th percentile

percentile BMI

>85th percentile

percentile BMI

>85th percentile

Ln MWH

1.323*

0.020*

0.435

0.010

1.961*

0.023+

(0.560)

(0.009)

(0.772)

(0.012)

(0.857)

(0.014)

Number of meals

0.1316**

0.0042*

0.0593

0.0023

0.1448*

0.0043

(0.0464)

(0.0019)

(0.0476)

(0.0021)

(0.0699)

(0.0029)

% meals in restaurant

0.0255

0.0012

0.0096

0.0008

0.0452

0.0016

(0.0197)

(0.0010)

(0.0177)

(0.0011)

(0.0400)

(0.0018)

Breastfed

0.0364

0.0009

0.0742

0.0000

−0.0066

0.0006

(0.0332)

(0.0014)

(0.0607)

(0.0022)

(0.0265)

(0.0016)

Receives allowance

−0.0051

0.0003

−0.0077

0.0005

−0.0061

0.0002

(0.0135)

(0.0007)

(0.0309)

(0.0015)

(0.0195)

(0.0008)

Fraction of time spent

Sleeping

0.0078

0.0008

−0.0063

0.0007

0.0275

0.0016

(0.0159)

(0.0008)

(0.0378)

(0.0015)

(0.0326)

(0.0018)

Reading/talking/listening music

0.0461

0.0014

0.0637

0.0009

0.0207

0.0018

(0.0294)

(0.0011)

(0.0505)

(0.0014)

(0.0337)

(0.0019)

Attending school

−0.0067

−0.0004

−0.1239*

−0.0049*

−0.0042

0.0001

(0.0143)

(0.0007)

(0.0595)

(0.0025)

(0.0268)

(0.0008)

TV watching

0.0530+

0.0028+

0.0091

0.0003

0.0454

0.0060

(0.0319)

(0.0015)

(0.0255)

(0.0013)

(0.0789)

(0.0039)

Eating

0.0144

0.0005

−0.0093

0.0000

−0.0187

−0.0003

(0.0147)

(0.0006)

(0.0297)

(0.0014)

(0.0477)

(0.0015)

Playing indoor games

0.0061

0.0002

0.0133

0.0009

0.0107

−0.0002

(0.0151)

(0.0007)

(0.0272)

(0.0013)

(0.0248)

(0.0014)

Sports

0.0002

0.0007

0.0132

0.0008

−0.0151

0.0001

(0.0106)

(0.0009)

(0.0279)

(0.0019)

(0.0297)

(0.0006)

Personal care of self/others

−0.0002

0.0000

−0.0079

−0.0005

−0.0732

−0.0013

(0.0073)

(0.0001)

(0.0196)

(0.0009)

(0.0510)

(0.0016)

Socializing

0.0019

0.0000

−0.0210

−0.0002

0.0081

0.0004

(0.0133)

(0.0003)

(0.0280)

(0.0010)

(0.0222)

(0.0010)

Misc. passive activities

−0.0048

0.0003

−0.0256

−0.0002

0.0024

0.0001

(0.0166)

(0.0007)

(0.0328)

(0.0013)

(0.0112)

(0.0006)

Traveling

−0.0177

−0.0015

−0.0509

−0.0030

−0.0036

−0.0005

(0.0210)

(0.0011)

(0.0457)

(0.0021)

(0.0213)

(0.0013)

Shopping

0.0044

0.0001

−0.0018

−0.0001

0.0004

−0.0001

(0.0081)

(0.0003)

(0.0128)

(0.0010)

(0.0087)

(0.0007)

Chores/work

0.0116

0.0003

0.0330

0.0011

−0.0041

−0.0001

(0.0183)

(0.0008)

(0.0302)

(0.0013)

(0.0177)

(0.0006)

Computer/video games

−0.0002

0.0000

−0.0061

0.0000

0.0058

0.0002

(0.0054)

(0.0002)

(0.0200)

(0.0008)

(0.0144)

(0.0005)

In child care

−0.0414

0.0003

0.0552

0.0011

−0.1400+

0.0005

(0.0403)

(0.0016)

(0.0437)

(0.0019)

(0.0777)

(0.0030)

Homework

0.0125

0.0002

−0.0007

0.0000

0.0346

0.0009

(0.0164)

(0.0004)

(0.0149)

(0.0006)

(0.0384)

(0.0012)

Observations

3,424

3,424

1,733

1,733

1,537

1,537

The first row presents the effect of ln MWH over the child’s life on the dependent variable. All of the following rows present the difference between the coefficient in the first row and the coefficient on ln MWH if the variable listed to the left is included in the specification. Each coefficient is from a separate regression. OLS estimation is used when the dependent variable is BMI percentile and probit is used when the dependent variable is risk of overweight. Marginal effects are reported. Robust standard errors are in parentheses. Standard errors are adjusted for intra-cluster correlations at the family level. + significant at 10%; * significant at 5%; ** significant at 1%

As a final consideration, we address the possibility that our findings are dependent on the particular estimation strategy used here. As one check, we decompose the effects by each potential channel and find similar results. In particular, we estimate the effect of children’s activities and meal routines on BMI, and separately estimate the effect of maternal employment on these activities and routines. These results are shown in Appendix Table 9.

Combining these two decomposed effects supports our interpretations of the findings discussed above. First, we find that the number of meals is significantly and negatively associated with percentile BMI, and mother’s work hours are significantly and negatively associated with the number of meals. The combined effect is consistent with the positive value found on Tables 5 and 6 related to the number of meals, suggesting that more hours working increases children’s BMI through the mechanism of fewer meals. Second, reading/talking/listening to music is significantly and negatively associated with percentile BMI, and mother’s work hours are significantly and negatively associated with the time spent reading/talking/listening to music. The combined effect is consistent with the positive value found on Table 5, suggesting that more hours working increases children’s BMI through the mechanism of reading/talking/listening to music. Third, we find that for less educated mothers, the time in school is significantly and negatively associated with percentile BMI, and mother’s work hours are significantly and positively associated with time spent in school. The combined effect suggests that among these mothers, more hours working is associated with their children having a lower BMI through the mechanism of school attendance. Fourth, for the full sample, we find that TV watching is significantly and positively associated with high percentile BMI, and that mother’s work hours are significantly and positively associated with more time spent watching TV. The combined effect suggests that more work hours is associated with a higher child’s BMI through the mechanism of TV watching. Finally, we find that for all mothers, more work hours are significantly associated with more time spent in child care, but for the more educated mothers, more time spent in child care is associated with lower child BMI (although not significantly). This combines to produce the effect that more work hours is associated with lower child BMI through the mechanism of child care.

5 Conclusions

In this paper, we have replicated the empirical connection found in the NLSY between mother’s employment and childhood BMI percentile/obesity for the PSID. We then inspect the mechanisms which connect hours worked by the mother to BMI percentile/obesity of the child. We find that maternal employment is related to children’s BMI through the average number of meals consumed in a day, through reading/talking/listening to music, and through TV watching, although the magnitudes are small. In addition, while the direct effect of the first two activities on children’s BMI is negative, TV watching is associated with higher BMI. Some of these results complement the findings from Cawley and Liu (2007) who examine the time use of mothers using the American Time Use Survey and find that employed women spend less time cooking and eating with their children. We also find some evidence that mother’s work hours are associated with their children having a lower BMI through an increased amount of time spent in school (in the case of less educated mothers) or child care (in the case of more educated mothers). Overall, we examine a large number of potential channels which theory and intuition would predict to be important and are able to explain a relatively small percentage of the total relationship.

Two important limitations of this study are the small sample size and the lack of detail available about meals. A larger sample would allow us to employ empirical techniques which address the endogeneity issues, such as fixed effects and instrumental variables estimation. A larger sample would also allow us to disaggregate by child’s age which would sharpen the analysis since the activities of 3 year-olds are quite different from the activities of teenagers and the effects of maternal employment on childhood obesity are likely age specific. Despite the small sample sizes, we did find some evidence (not shown) that maternal work hours only affect TV watching for children over age 9.

We believe that a possible reason that we do not get stronger results on restaurant meals, as opposed to meals at home, for example, is that we do not have information on take-out meals. The pizza delivered from the hut to the home and eaten at home is as fattening as the pizza eaten in the hut. In our data set we can also not distinguish between a meal at a fast food restaurant and a salad in a conventional restaurant. We suspect that families who frequently eat greasy pizzas and fatty burgers in restaurants also use more fatty and calorie rich foods in meals that are cooked at home. Answering the question of how mother’s employment affects childhood obesity via the channel of the number and variety of meals cooked requires a much more detailed data set.

Because of these limitations, we believe it would be premature to conclude that the majority of the mechanisms evaluated in this analysis are not relevant based on the results of this single study. Prior to making this conclusion, it is necessary to replicate these findings with other data and research strategies.

Footnotes
1

In addition, we omitted 132 observations for children with physical measures that would put them in the first tenth of the first percentile of the CDC Growth charts.

 
2

For a review, see Robinson and Bostrum (1994). More recent studies include Yaroch et al. (2006), and Chodick et al. (2008).

 
3

If MWH < 1, the value is bumped up to 1 so that ln MWH = 0. The results are qualitatively the same whether maternal work hours are logged or not.

 
4

The attending school result is robust to this sample restriction.

 

Acknowledgements

We thank John Cawley, Michael Grossman, Robert Kaestner, Sara Markowitz, Una Osili, Robert Sandy, Pravin Trivedi, Diane Whitmore Schanzenbach, and participants in seminars at Indiana University, IUPUI, University of Georgia, University of Arkansas, Florida State University, Southern Methodist University, University of Chicago, and the NBER Summer Institute in Health Economics for helpful comments. All the remaining errors are our own.

Copyright information

© Springer Science+Business Media, LLC 2009