Manajemen | Fakultas Ekonomi Universitas Maritim Raja Ali Haji 643.full

Mothers’ Time Choices
Caregiving, Leisure, Home Production, and
Paid Work
Jean Kimmel
Rachel Connelly
abstract
Using data from the 2003 and 2004 American Time Use Survey, we study
the role that socioeconomic factors play in mothers’ time choices. We
estimate a four-equation system in which the dependent variables are the
minutes used in home production, active leisure, market work, and child
caregiving. Our results show that mothers’ caregiving time increases with
the number of children, decreases with age of the child, and increases
with the price of child care. We also find a substantial positive wage
elasticity for caregiving time, while both leisure and home production
time declines with increased wages.

I. Introduction
The last half of the twentieth century was a period of dramatic transformation in the role that women play in society, highlighted most clearly by the
rapid rise in paid employment of mothers with young children. Currently, about
60 percent of mothers with children younger than six participate in the paid work
force. As mothers have increased their paid work efforts, conflicts between employment and family responsibilities have grown, leading researchers to explore more

Jean Kimmel is an associate professor of economics at Western Michigan University and Research
Fellow of IZA, Bonn. Rachel Connelly is professor of economics at Bowdoin College. The authors thank
the W.E. Upjohn Institute for Employment Research for financial support for this research. They also
thank Dawit Senbet and Fei Tan for research assistance and Erdal Tekin, Dorinda Allard, and two
anonymous reviewers for providing feedback that substantially improved the paper. Finally, they thank
their children for enlightening them regarding the distinction between the process and outcome utility
components of caregiving time. Earlier drafts of this manuscript were presented at the November 2005
SEA meeting, December 2005 ATUS Early Results Conference, and the January 2006 ASSA meeting.
The data used in this article can be obtained beginning January 2008 through December 2010 from
Rachel Connelly, Department of Economics, Bowdoin College, Brunswick ME 04011,
connelly@bowdoin.edu.
[Submitted March 2006; accepted September 2006]
ISSN 022-166X E-ISSN 1548-8004 Ó 2007 by the Board of Regents of the University of Wisconsin System
T H E JO U R NAL O F H U M A N R E S O U R C E S

d

XLII

d

3

644

The Journal of Human Resources
fully the role that caregiving responsibilities play in mothers’ time choices and the
effects that these choices may be expected to have on their children.
Our paper relies on the recently released American Time Use Survey (ATUS) data
to describe the time-use choices of mothers in the United States. Using these data, we
describe how mothers allocate time to four aggregated categories of time: home production, caregiving time, leisure, and paid market work, with a focus on identifying
differential responses to demographics and prices for these different time uses. Our
four time-use categories, which expands the analysis beyond the traditional three
time-use categories of paid work, leisure and home production by explicitly separating caregiving time from home production, permits us to identify the factors relevant
to caregiving time choices. Additionally, this more detailed stratification enables us
to gain a better understanding of how mothers’ caregiving time choices compare to
their choices regarding other unpaid uses of time, household production and leisure
time. If caregiving time responds differently in any substantive way to economic and
demographic factors, then aggregating caregiving time into household production or
leisure time might yield mistaken empirical conclusions.

The first goal of the paper is to describe mothers’ time responsiveness to economic
factors. Specifically, we estimate market wage and childcare price elasticities for
each of four general categories of time utilization, thereby providing for both absolute and relative interpretations. We find that all four time uses of mothers in the
United States are sensitive to wages and childcare time is sensitive to childcare prices
of preschoolers but less so to childcare prices for school-age children. Most interestingly, we find that higher-wage mothers devote more time to caregiving both on
weekdays and weekend days, ceteris paribus. Additionally, paid work time on weekdays also responds positively to higher wages, while leisure time and home production weekday time are reduced by higher wages.1 On weekends, only leisure time
and childcare time are impacted by higher wages, with leisure time decreased and
childcare increased for higher-wage mothers.
The second motivation for this research is to gain a better understanding of the
importance that marital status, race, and other demographic factors play in time choices,
once economic factors are controlled, and to determine whether these factors affect
competing time choices differently. We expect to find that single mothers make timeuse decisions very differently from their married counterparts, in part due to the
lesser availability of adults in the household to engage in home production, resulting
in a greater need to purchase products in the market. With regard to race, previous
research on the use of nonparental childcare has revealed different childcare utilization patterns by race, and we wonder whether these differences carry over to maternal time use as well. Difference in time use by race may help explain differences
observed in other policy-relevant variables such as differences by race in the gender
wage gap or in wealth acquisition. Examining the role of race in four different time
uses will allow us to identify the different roles that race could play in these very
different activities.
Finally, our third goal with this research is reflected in methodological improvements, most importantly our implementation of a structural model of time choice

in which the correlation across time choices is incorporated into the econometric
1. Note that this is employment time on a given day, not total employment time.

Kimmel and Connelly
methodology. Additionally, we implement a Tobit model to permit incorporation of
individuals observed allocating their time into fewer than our four categories (for example, mothers who do not work for pay on diary day or who have no home production minutes on diary day).
The rest of the paper is as follows: Section II reviews the economic literature on
time allocation from the Robbins labor/leisure tradeoff model through the work of
Becker and Gronau, concluding with the empirical literature that looks explicitly
at childcare time. Section III outlines a behavioral model of four distinct time uses
for mothers of young children, while Section IV discusses the ATUS data and our estimation strategy. Section V presents our results and Section VI summarizes the findings.

II. Time Allocation Models and Mothers’ Time Use
Economists have long understood that the standard Robbins (1930)
labor/leisure model is inadequate for understanding the time-allocation process of
mothers. The microeconomic theory underlying the labor/leisure framework posits
that paid work yields no direct utility thus paid work enhances the worker’s utility
only via the goods purchased with the earned income (outcome utility). In addition,
the theory posits that all time devoted to leisure produces utility directly (process
utility). The very foundation of the theory of neoclassical labor supply relies on

the credibility of this stratification of time into time-yielding process utility and
time-yielding outcome utility. This theoretical underpinning, while perhaps a reasonable stretch for men, is not appropriate for women who still do the majority of housework and childcare even as their paid work hours have increased.2 The New Home
Economics models of the early 1960s acknowledged that a substantial portion of
time not spent in paid employment is home production time, not leisure.3 Becker’s
(1965) methodology omits all time categorizations and focuses on the production
of final consumption commodities, but this approach has been difficult to implement
empirically due to difficulty identifying the final commodities. Since then, alternative
approaches have focused on expanding the traditional two-dimensional time allocation model to three or more uses of time with the hope of disentangling activities that
are unpaid but produce substantial outcome utility (and can be traded across members of a household ) from the leisure time category.
Gronau (1977) and Graham and Green (1984) stratified time outside the labor market into home production and pure leisure. Gronau (1977) establishes two criteria for
aggregating time uses and concludes that leisure time and home production time
should not be combined. Gronau writes: ‘‘From the theoretical point of view, the justification of aggregating leisure and work at home into one entity, nonmarket time (or
home time) can rest on two assumptions: (a) the two elements react similarly to
changes in the socioeconomic environment and therefore nothing is gained by studying them separately; and (b) the two elements satisfy the condition of a composite
2. For a discussion of the difficulty associated with stratifying time use into separate categories to distinguish those generating outcome utility from those generating process utility see Juster (1985), Dow and
Juster (1985), and Oi (1992).
3. See, most importantly for our purpose here, Becker (1965), Gronau (1977), Graham and Green (1984).

645

646

The Journal of Human Resources
input, that is, their relative price is constant and there is no interest in investigating
the composition of the aggregate since it has no bearing on production and the price
of the output’’ (p. 1100).4
But in fact, Gronau’s two criteria explain why, particularly for mothers, his three
uses of time are still not enough. In his model, unpaid ‘‘home work’’ is defined as
time spent producing a good that is a perfect substitute for one that also could be
purchased in the market. In addition to home-produced goods and market-produced
goods being indistinguishable, the home-production process in Gronau’s model provides no utility per se. However, home-produced childcare (henceforth referred to as
parental childcare or caregiving) is usually considered an imperfect substitute for
market childcare in the production of child services and certainly most parents receive utility from some of the time they spend on caregiving.5 For studying the time
use of mothers, the best solution to this problem is to expand the Gronau trinity into a
model with five aggregated uses of time: (paid) market work, (unpaid) home work,
child care, leisure and other.6 The other category includes sleep, personal care time,
education, and job-seeking endeavors and can be loosely thought of as personal investment time.
This expanded Gronau model is, in our view, a credible means of assigning activities into composite categories that produces a manageable number of categories
while disentangling a complex group of unpaid caregiving activities we refer to as
caregiving from leisure and household production. But it must be made clear that

any attempt to take an extensive list of human activities and categorize the activities
into five distinctly different groups with total homogeneity within groups is impossible.7 Thus, the categorization we choose is ad hoc at best, but does seem sensible in
light of Gronau’s two criteria and yields a number of time uses that can be modeled
with appropriate econometric sophistication.8 Additionally, isolating caregivingrelated activities into their own category permits us to compare caregiving time choices
to other unpaid time choices.
What is the existing state of knowledge regarding the determinants of mothers’
child caregiving time? The bulk of the previous literature that examines caregiving
time focuses on couples, often dual earner households. Kooreman and Kapteyn
(1987) looked exclusively at married couples and found that higher wages of the
father increased the time their wives spent in child care, but that women’s own wages
affected neither her childcare time nor her husband’s childcare time. Nock and
4. Note that neither of Gronau’s two criteria rely on the distinction between process and outcome utility.
5. See, for example, Aguiar and Hurst (2006).
6. Van den Brink, Maassen, and Groot (1997) use four categories of time use, leisure, home production,
childcare, and employment. Kooreman and Kapteyn’s (1987) model includes eight categories.
7. In fact, we do not estimate the determinants of the ‘‘other’’ time category as it is far too heterogenous in
terms of activities included.
8. Note that empirical application of any aggregated time-use model requires something akin to a leap of
faith because it is not possible to distinguish time uses that provide process utility from those providing
only outcome utility. Many activities that would typically be categorized as household production or caregiving may produce process utility (as does paid work). For example, time spent preparing a meal may be a

chore for some, while others may enjoy the process of creating a meal. Our four time aggregates, while
limited by the same problems associated with distinguishing process utility from outcome utility that have
always hindered neoclassical time-use models, is a best attempt to categorize comparable activities into distinct groups that respond in similar ways to economic and demographic factors.

Kimmel and Connelly
Kingston (1988) found that mother’s employment reduced their childcare time, but
that the reductions were mostly in secondary activities with children.9 Using data
from the Netherlands for married mothers currently employed, Van den Brink, Maassen,
and Groot (1997) found no effect of husband’s earnings on the time allocation of his
wife in employment, home production, or child care. Closest to our research is the
recent paper by Hallberg and Klevmarken (2003). They examine the determinants
of parents’ time allocated to childcare in Sweden and their structural model incorporates instruments for both parents’ wages and parents’ employment time. They find
that own wages do not affect childcare time of their sample of Swedish parents.
Studies using more recent time diary data from the United States have found that
employment hours are a predictor of hours spent with children; however, mothers appear to shield their children from the full impact of their employment by cutting back
on personal time, sleep, leisure, and home production rather than reducing child care.
Thus, there is some evidence that caregiving time is treated differently than either
home production or leisure.10 The differences by gender in the time trend data described by Sayer (2005) provide more support for this notion. Sayer notes that over
time, men and women have adjusted their out of market time substantially, concentrated mainly in their movement from unpaid home production into family time.
Thus, the disaggregation of unpaid activities is becoming more important over time.11

Kalenkoski, Ribar, and Stratton (2005a) use British data to estimate a reduced form
model, in which wages are not included directly. They find that women with an advanced degree spend more time on primary child care, secondary child care, and market
work. What they are spending less time on is not clear since these are the only three uses
of time included in their analysis. Kalenkoski, Ribar, and Stratton (2005b) show a similar result for U.S. mothers using the ATUS, namely that mothers with a bachelor’s
degree or a graduate degree spend more time on primary childcare and in market
employment.

III. Underlying Behavioral Model
The behavioral model underlying our empirical specification is the
standard neoclassical individual-based utility maximizing problem in which mother’s
utility is expressed as a function of leisure, tL, child services, CS, and aggregated
adult consumption of final goods and services excluding child services, G.
ð1Þ max U ¼UðtL ; CS; GÞ

9. Both papers use data from U.S time diaries from the 1975-81 Time Use Longitudinal Panel.
10. For examples of this research, see Howie et al. (2006), Bianchi et al. (2005), Reimers (2002), Sandberg
and Hofferth (2001), Bianchi (2000), and Bryant and Zick (1996).
11. Sayer (2005) and Craig (2006) both note that while men are adjusting their unpaid time in response to
mothers’ increased paid work time, the result thus far is not one of gender equity in all time uses. Craig
expresses the concern that this the movement of women into the paid work force is only half a ‘‘revolution,’’

and has resulted in an increased total work load for women. One positive outcome of the time-use evolution
(as noted by Sayer et al. 2004) is that parental time investments in their children has increased, contrary to
much media reporting.

647

648

The Journal of Human Resources
Adult consumption goods, G, are home produced with a combination of household production time, thp, and purchased intermediate goods; G¼G(thp, X;u). u is an
efficiency parameter which is affected by differences in ability, but also by differences in personal investment including sleep time and educational endeavors. Child
services, CS, are also home produced, but with a combination of the mother’s caregiving time, tmcc, nonmaternal childcare time (including that provided by the father),
tcc, and market-produced child goods, CX; CS¼CS( tmcc, tcc, CX;f). Like u, f is an
efficiency parameter. On the constraint side of the model, there is a mother’s time
constraint, Equation 2, and a budget constraint, Equation 3.
ð2Þ

T ¼ tem + thp + tmcc + tL + ts

ð3Þ

PX X + Pcc tcc + PCX CX ¼ wtem + V

The mother’s total time can be divided into market-paid time, tem, home-production
time, thp, caregiving time, tmcc, leisure, tL, and investment time, ts.12 In addition, implicitly there is a child’s time constraint in that parents with young children (which
we define as parents with children younger than 13) must have someone watching
their children at all times.
ð4Þ

CT ¼ tmcc + tcc + tscc

CT is the total amount of time available to children. We have already defined tmcc as
maternal childcare time and tcc as nonmaternal time, both of which contribute to
increased levels of child services. The final term, tscc, is secondary childcare time
in which children are being supervised but not actively engaged. For the sake of
modeling ease, this secondary care time is assumed to not contribute to child services and to be provided without opportunity cost by the mother or without money
cost by anyone other than the mother. If we think of tscc as including sleep time
then CT ¼ T.
The above three constraints result in three distinct costs of a mother’s time. First,
there is the cost of time in the labor market, w 2 Pcc, if the alternative to employment
is primary care of one’s own children.13 The second cost of time is when the mother
is engaged in leisure or home-production activities while children are not present because the children are actively engaged by an alternative caregiver. The opportunity
cost of that activity’s time is the price of earnings forgone plus the price of nonparental child care, w + Pcc, At other times, when children are being supervised in the
background, when children are in public school, or when children are old enough for
self care, the opportunity cost of that activity’s time is simply the wage, w. Since the
cost of time is sometimes the wage, sometimes the wage plus the price of child care,
12. Note that this theoretical model reflects a Gronau like assumption that only leisure yields process
utility. Caregiving time, household production time, and market work yield outcome utility only.
13. See Connelly (1992) and Ribar (1992, 1995) for models of nonmaternal childcare that derive the w-Pcc
cost of time.

Kimmel and Connelly
and sometimes the wage minus the price of child care, the wage and the price of
childcare must appear separately in any estimation model.14
The behavioral model described above results in time-demand functions for the
five different uses of time, as well as the more standard consumption good demand
functions for G and CX. Equation 5 presents a general functional representation of
the time-demand functions.
ð5Þ tj ¼ f ðw; Pcc ; VjZ; H; DÞ

for j ¼ em; hp; mcc; L; s

Time use is related to factors reflecting the value of time, including the wage and the
hourly prices of nonparental childcare for women with preschool-aged children and
school-aged children, the amount of nonlabor income available to the mother, preferences, and institutional structure, all of which are expected to be related to personal
characteristics of the mother, Z, characteristics of the household in which she resides,
H, and characteristics of the diary day, D.

IV. Data and Estimation Strategy
A. The ATUS
As described above, our primary goal in this paper is to examine the determinants of
time spent in four activities: leisure, child care, home production, and paid employment for mothers, with a focus on caregiving. We exclude the ‘‘other’’ category as
too heterogeneous to aggregate. We aim to contrast the factors important in caregiving choices with other unpaid activity, namely household production and leisure. We
admit that in both our theoretical discussions and empirical implementation we have
largely ignored the role played by spouses or partners. The ATUS collected one time
diary per household. Because one’s spouse’s time data are unavailable, we make the
assumption that the only role of the spouse is in his production of earned income
which is treated as unearned income by his wife. Time diary designs in other countries allow a more nuanced look across couple time tradeoffs.
We draw our data from the 2003 and 2004 samples of the American Time Use Survey. Our estimating sample is comprised of female ATUS diary respondents between
the ages of 18 and 65 who have children younger than13 living in the household, are
not part of a multifamily household, and are not currently in the active duty military,
in school, or unemployed. These criteria, in addition to requiring each observation to
have information on the husband’s wage if married with spouse present, leads to a
sample size of 4,552 mothers.15 Because weekdays and weekends represent very
14. Kimmel (1998) provides empirical justification for including separate measures for the wage and the
price of care in employment equations.
15. Only women who are mothers are included in our empirical analyses. This sample stratification is
based on the outcome of motherhood, which is modeled within the choice framework by economists.
We do not address this endogeneity because the complexity of our model renders such econometric extensions infeasible and such methodologies require data typically unavailable (e.g., variables thought to explain motherhood that do not also explain time use). However, our focus here just on mothers makes
this endogeneity correction less important. It would be more important in a time-use comparison between
mothers and nonmothers.

649

650

The Journal of Human Resources
different time-use patterns for many families and because the ATUS oversamples weekend days, we separate the data throughout the analysis into weekends and weekdays.16 Ultimately, we have usable time diary information from 2,156 women for
weekdays and 2,396 for weekend days.
As described earlier, we categorized the time activities included in the ATUS data
into four categories for estimation purposes, plus an excluded fifth category.17 Appendix Table 1 records the full set of decisions we made in assigning the six-digit
time categories provided by the ATUS into our five category framework. In terms
of the caregiving time, we mainly use time categorized by the ATUS as ‘‘Caring
and Helping Household Children,’’ but have added time spent engaging nonparental
caregivers and transportation time to nonparental care arrangements to the main category of child caregiving time. Similarly, telephone and travel time related to securing professional help in housework was added to home-production time.
The delineation of leisure time requires some elaboration. According to the New
Home Economic models, all time that produces utility directly ought to be considered leisure time. Application of this theory to an actual list of human activities is
not straightforward. One example of an activity that can be problematic to categorize
is sleeping. While some fraction of sleeping can be considered an activity producing
process utility, a sizable portion is a necessity for human existence and would best be
categorized as a variation of human capital investment. (See, for example, Biddle and
Hamermesh 1990). Aguiar and Hurst (2006) discuss in detail the difficulties inherent
in ‘‘defining’’ leisure and examine leisure trends using four different definitions of
leisure time. They too express concern with the investment component and necessity
of sleep. To address this concern, we omit time spent sleeping or engaging in personal care from our aggregate measure of leisure, resulting in a measure that can best
be thought of as active leisure.
The group category of caregiving also can be difficult to conceptualize, particularly when one considers the distinction between process utility and outcome utility.
Only a portion of mothers’ time with their children is exclusively devoted to caregiving. This time would include feeding, bathing, playing games, talking, and reading to
children. We expect that some of this time provides process utility but surely not all
of the time on all of the days. Yet, these are the sorts of activities that the ATUS picks
up in primary caregiving time, time when the respondent herself categorizes the time
as primarily childcare activities. Other activities in which children are involved are
16. In fact, weekend employment typically is referred to as nonstandard employment. Also, note that we
include weekday holidays in the weekend sample so the sample is really weekends and holidays. We do this
because there are very few weekday holidays in the data and but those holidays look very different from
weekdays in time-use patterns. Kalenkoski, Ribar and Stratton (2005b) divide their data in the same way.
17. This fifth time category is all other time and includes time spent sleeping, in personal care, time spent
in educational pursuits, unpaid time that contributes to success in one’s current employment, and job search
time. It is not essential for our estimation procedure that all 24 hours in a day be accounted for in these four
time uses, and in fact, we account for considerably fewer than 24 hours. The correlation among the equations is addressed through cross-equation covariances, not in cross-equation coefficient restrictions as
would be necessary if we were accounting for all time within a 24-hour period. Solberg and Wong
(1992) and Kim and Zepeda (2004) do account for all time while Kalenkoski, Ribar, and Stratton
(2005a, 2005b) do not. Concern for the hetergeneity of the fifth category and the complexity of the estimation model even without trying to account for cross-equation restrictions led us to estimate the Tobit equivalent of Seemingly Unrelated Regressions (see Prowse 2004).

Kimmel and Connelly
more likely categorized as time spent with children, such as the always-pleasurable
trip to the supermarket with squabbling children in tow, which is likely to have been
categorized as home-production time.18 Again some of this time may produce process utility, though most would be expected to produce outcome utility (healthy children). Finally, there is passive care that the ATUS collects as time when children are
recorded as ‘‘in the room’’ which excludes time when all household children are
sleeping and the time when the respondent is sleeping. Passive care includes time
at home when one is listening for children calling or overseeing children engaged
in play. Folbre et al. (2005, p. 374) encourage moving beyond the simple categorization of time into activities to incorporate even broader notions of responsibilities
or constraints. Thus, their description of passive care would be broader than that
available in the ATUS, and would include being responsible for sleeping children,
even if the parents are sleeping too. Unfortunately, the data produced by existing
time-use surveys do not reflect the broad characterization suggested by Folbre et al.19
Table 1 presents the average minutes spent in the four time categories for the
mothers in our sample. Looking at Table 1, we see substantial differences between
weekdays and weekends in the time spent in our four categories. Leisure and
home-production times are higher on weekends while the opposite is true for employment and childcare time. It is a bit surprising that less time is spent in active
caregiving on the weekend than weekday as the number of hours of employment also
is substantially lower on the weekend. These differences provide suggestive evidence
that time spent in childcare is distinct from home production and leisure. Additionally, the dramatic differences in time use between weekdays and weekends serves to
support our decision to estimate our time-use models separately for those two diary
day groups.
Table 1 also shows substantial differences in the means of childcare, home production, and employment time depending on whether the zeroes are included. The substantial number of respondents with zero time use in these categories means our
estimation strategy should explicitly take account of the left censoring of the time
responses at zero. On the other hand, very few in the sample have zero minutes of
active leisure, and no one has a full day of leisure so we need not be concerned about
censoring of the leisure-time equation.20
Further descriptive information is presented in Table 2. The top half of the table
shows the expected relationship between age of the youngest child and the number
of minutes spent on child care. Married mothers whose youngest child is younger
than five spend 198 minutes in active childcare on a weekday and 145 minutes on
a weekend day. The number of minutes decreases substantially when the youngest
18. Kalenkoski, Ribar, and Stratton (2005a) examine childcare time reported both as primary and secondary time usages.
19. ATUS permits respondents to record childcare as a secondary activity (having children ‘‘in your care’’)
when it is not the reported primary activity. Two recent papers by Kalenkoski, Ribar, and Stratton (2005a,
2005b) address the childcare measurement issue by focusing on different measures of nonprimary caregiving. Kalenkoski, Ribar, and Stratton (2005a) use data from the United Kingdom to examine factors important to primary and secondary childcare (as well as market) time use, while Kalenkoski, Ribar, and Stratton
(2005b) use data from both the United Kingdom and the United States and a broader definition of secondary
care that includes all time spent with children not reported as the primary activity.
20. Fifteen respondents have zero minutes of active leisure on weekday and five report zero minutes of
active leisure on weekends.

651

652

The Journal of Human Resources
Table 1
Average Minutes Spent per Day on Leisure, Childcare, Home Production, and
Employment for Women with Children Younger than 13

Dependent Variables
Minutes of childcare
(including zeros)
Minutes of childcare
(excluding zeros)
Minutes of employment
(including zeros)
Minutes of employment
(excluding zeros)
Minutes of home production
(including zeros)
Minutes of home production
(excluding zeros)
Minutes of leisure
(including zeros)

Weekdays

Weekends or Holidays

Mean
(Standard Deviation)
Sample Size

Mean
(Standard Deviation)
Sample Size

127.278
(133.21)
2,630
157.153
(131.20)
2,175
245.167
(250.40)
2,630
444.248
(158.56)
1,509
207.386
(167.19)
2,630
220.809
(163.70)
2,491
280.439
(163.81)
2,630

90.804
(122.48)
2,866
140.247
(127.43)
1,918
63.572
(162.37)
2,866
331.333
(220.73)
560
254.921
(175.46)
2,866
269.531
(169.15)
2,724
404.325
(194.16)
2,866

Note: Each cell contains the variable mean, standard deviation, and number of observations

child is elementary-school aged, to 75 minutes on weekdays, and 46 minutes on a
weekend day. The pattern is similar for unmarried mothers though the number of
minutes is less, particularly for unmarried mothers with preschool children.21 Married women also spent more time on home production than unmarried women, suggesting that a husband’s presence increases requisite work in the home or that
unmarried mothers ‘‘contract out’’ more of the home production. Standards may
be higher in married households and there is another person in the house for whom
to produce goods. This finding is consistent with previous research into gender
21. We define those who are married with spouse present as married and all other mothers are defined
as not married. Thus, unmarried mothers with partners residing in the household are treated the same as
mothers without partners. This is a weakness we aim to address in future research.

Table 2
Average Minutes Of Time Used by Age of the Youngest Child, Marital Status, Wage Category for both Weekdays and Weekend Days
Married Spouse Present
Youngest
Child: 6 to 12

Youngest
Child: 0 to 5

Youngest
Child: 6 to 12

Youngest
Child: 0 to 5

in
in
in
in

paid work
childcare
home production
leisure

284.69
75.44
224.10
422.47

188.98
197.72
227.75
383.18

314.55
67.79
164.00
287.60

179.46
174.48
157.50
389.10

in
in
in
in

paid work
childcare
home production
leisure

68.79
46.13
280.25
422.47

40.71
145.21
252.08
383.18

86.71
44.91
251.23
423.20

80.13
121.74
202.39
389.10
(continued)

Kimmel and Connelly

Weekday
Minutes
Minutes
Minutes
Minutes
Weekend
Minutes
Minutes
Minutes
Minutes

Not Married Spouse Present

653

654

Married Spouse Present

Weekday
Minutes
Minutes
Minutes
Minutes
Weekend
Minutes
Minutes
Minutes
Minutes

Not Married Spouse Present

Low Wage

Mid Wage

High Wage

Low Wage

Mid Wage

High Wage

in
in
in
in

paid work
childcare
home production
leisure

320.54
85.75
171.56
254.14

368.21
98.59
168.01
233.24

369.81
111.46
161.46
248.47

287.37
124.49
150.61
253.88

254.06
96.72
142.26
243.10

400.06
86.88
126.14
249.51

in
in
in
in

paid work
childcare
home production
leisure

128.80
59.85
254.14
360.46

79.68
74.42
276.82
382.95

69.50
95.25
275.18
393.87

128.95
61.76
221.75
374.66

130.77
76.06
229.88
374.22

86.21
73.09
242.93
412.61

Note: Data from 2003 and 2004 ATUS. Reported results are weighted to reflect population averages. The Mid-wage category was calculated the mean predicted wage plus
or minus one standard deviation from the mean.

The Journal of Human Resources

Table 2 (continued)

Kimmel and Connelly
differences in home-production time, including that of South and Spitze (1994) and
Stratton (2003).
The bottom half of Table 2 shows the distribution of average time use by marital
status and wage rate categories. The middle wage category is defined as the approximate mean wage in the full sample ($10) plus and minus one standard deviation
($2.00). Thus, a low wage is a wage less than $8 an hour and a high wage is a wage
greater than $12 an hour.22 As expected, women with high wages are employed in
the labor market on the diary day for more minutes but only on weekdays. On weekends, higher wages are correlated with fewer minutes of employment time. For the
most part, unmarried mothers are employed more minutes than married women, especially on weekdays for unmarried mothers in the highest wage category.
Table 2 also shows that childcare time, like employment time, increases with married women’s wages, but the same is not true on weekdays for unmarried mothers.
Those unmarried mothers with the highest wages devote less time for childcare than
middle wage unmarried mothers and less still than low-wage unmarried mothers. For
married women, the result is consistent with Bryant and Zick’s (1996) finding that
more highly educated mothers spend more time in direct child care.
B. Empirical Model
The evidence presented in Table 2 is merely a cross-tabulation. A fuller understanding of the relationship among time use, marital status, age of children, and wages
requires a multivariate analysis. Our basic estimation model is a system of four linear
time-use equations based on the time-demand equations shown in Equation 5. The
estimation version of Equation 5 can be characterized as:
ð6Þ tj ¼ b0j + b#j X + ej

for j ¼ em; hp; mcc; L

where tj* is the latent number of minutes a mother would choose to spend in activity
j. The actual observed minutes, tj will equal zero when tj* is less than zero. In our
sample, active leisure was so seldom zero that we can ignore the censoring problem
for the leisure equation.
A Tobit model is used often to account for situations like Equation 6 if one is willing to assume that the ej’s are normally distributed. In our case, in addition to accounting for the lower limit constraint, we also must account for the fact that four
observed time uses come from the same sample respondent. More time used in
one activity likely means less time used in some other activity. We assume that
the ej terms are correlated across equations in the following way:
00 1 0
1
s2L
0
eL
BB C B r
B emcc C
BB 0 C B Lmcc sL smcc
C
B
;B
@ ehp A;N B
@@ 0 A @ rLhp sL shp
0
eem
rLem sL sem
0

rLmcc sL smcc
s2mcc
rmcchp smcc shp
rmccem smcc sem

rLhp sL shp
rmcchp smcc shp
s2hp
rhpem shp sem

11
rLem sL sem
C
rmccem smcc sem C
CC
CC
rhpem shp sem AA
s2em

This situation is analogous to Seemingly Unrelated Regression except we must use a
nonlinear estimation technique to account for the lower limit constraints in three out
22. The wage measure used here is actually a wage measure generated from preliminary estimation. This
predicted wage is created using a standard two-step Heckman correction. For details, see Section IV4.

655

656

The Journal of Human Resources
of the four equations. Accounting for both the constraint at zero minutes and the correlation among error terms, leads to our choice to estimate a system of four correlated equations, three of which are estimated as Tobit equations.23
The X’s of Equation 6 include standard demographic characteristics of the mother,
Zi, characteristics of the household, Hi, characteristics of the diary day, Di, and three
predicted price of time measures. We discuss each set of variables briefly in turn. The
vector Zi includes variables such as age, education, urban/rural residence, Southern/
non-Southern residence, Nonwhite/White, and Hispanic/non-Hispanic. These variables may reflect differences in time preferences and also, in the case of residence variables, differences in the price of commodities and structural demands on one’s time.
For example, urban dwellers may spend more or less time commuting to work and
traveling related to shopping. Southern dwellers may have more yard work or other
differences in time demands.
We do not have strong theoretical predictions concerning the pattern of the effect
of these demographic variables, nor are there many previous empirical studies of
caregiving versus home-production time to inform our expectations. Studies of nonparental childcare use have shown that nonwhite use more relative care than whites,
so that we could hypothesize that nonwhite mothers will spent less time in caregiving (Capizzano et al. 2000 and NCES 2004).24 Finally, studies of hours of housework have reported a substantial drop in time spent on housework. If standards
of housecleaning have declined, we might expect an age-cohort effect such that
older women spend more time on home production than younger women. (Bianchi
2000).
The variables included in characteristics of the household, H, include Married
Spouse Present/ Not Married or No Spouse Present, Husband’s Earnings (if Married
Spouse Present, zero otherwise), the presence of other adults in the household (persons older than 17 who are not the woman or her spouse), and five counts of the number of children in the household aged zero to two, three to five, six to nine, ten to 12,
and 13 to 17. Other studies simply have included the total number of children and the
age of the youngest child, but we expect that children of different ages contribute
differently to the demands on mothers’ time. Certainly, Table 2 shows substantial
time-use differences by age of the youngest child. Studies of the effect of the presence of children on mother’s employment have found differences between having a
zero to two-year-old versus having a three to five-year-old. One reason for this difference is that many families view preschool as an educational investment in their
children, not just as supervised time that facilitates women’s employment, but still,
utilizing preschool does free up the mother’s time while the children are at school.
Six- to nine-year-olds are in school much of the day but they are usually not left
alone before and after school, while ten- to 12-year-olds are left alone often. Teenagers bring their own set of time and money demands which may affect mothers’
23. Our estimation procedure is very similar to that used by Kalenkoski, Ribar, and Stratton (2005a, 2005b)
except we have four uses of time and they model three. Our fourth equation, for leisure time, is estimated
via OLS. The estimation was done as a system of equations using the statistical software package aML.
24. Even if nonwhite mothers use more relative care it might not affect their own caregiving time if relative
care simply substitutes for other employment enabling childcare such as center or family daycare homebased care. On the other hand, a relative who is available for employment enabling care also may provide
home-production-enabling or leisure-enabling care.

Kimmel and Connelly
employment. (This has sometimes been shown for school-aged children as well.)
Teenagers also could contribute time (at least potentially) to household childcare and
home production, freeing up maternal time for employment and leisure.25 Connelly,
DeGraff, and Levison (1996) found strong evidence of this in urban Brazil, but the
results for the United States have been weaker. However, sibling care may be used
to facilitate mothers’ leisure or home-production activities rather than employment,
which would not show up in employment-based studies. The presence of other adults
in the household may affect mothers’ time use if household members truly behave as
a cohesive unit, but even in that case, the direction of the effect is not clear. A coresiding adult could contribute income to the household, thus freeing up the mother
to do more of the home production and caregiving, or could contribute childcare and
home production time, freeing up the mother for more employment time. The coresiding adult also may increase home-production time, especially if the coresiding
adult is an elderly relative who requires care.26
Economic theory does not provide us with strong predictions about the effect of
marriage and husband’s earnings on time-use decision-making. The presence of
the spouse should reduce childcare and home-production time to the extent that he
participates in these tasks, but the demand for home-production tasks also increases.
We saw in Table 2 that the presence of the spouse is correlated with greater homeproduction time. In addition, controlling for husband’s earnings, married women
have higher rates of employment, which would lead to increased employment time
and reduced child caregiving time. The higher rates of employment could be due
to positive assortative mating or the husband facilitating maternal paid employment,
even if he is not directly providing child care. His presence may provide another set
of relatives to draw on for emergency care or his time simply may add flexibility to
the woman’s time choices, such that employment is easier to sustain. Assuming husband employment time is exogeneous (which is, in the dawn of the 21st century still
not a bad assumption), husband’s earnings play the role of nonlabor income in our
model. Higher levels of nonlabor income are expected to reduce all ‘‘work’’ time
(employment and home production), and should increase leisure time, but the effect
on caregiving depends on the weighting of the ‘‘work’’ versus the ‘‘consumption’’
components of childcare time. However, higher nonlabor income also may mean a
bigger house or more ‘‘stuff’’ to take care of, so even the effect on home production
is ambiguous.
We control for the season of the year the diary information was collected with a
dichotomous variable indicating if the diary month was June, July, or August. In preliminary work, we included a dummy for each season but found that the only significant differences in time use were observed in summer. We expect that summer
matters for mothers of young children because of school vacation and changes in
the activities and even sleep patterns of children with the increased daylight hours
and warm temperatures.
The last three regressors represent components of the price of time. The price of
time is expected to affect all uses of time and so is included as a determinant of
25. Mothers who have only teenagers at home are excluded from this analysis so the teenagers in our sample are all potential babysitters as they all have younger (12 years of age or younger) siblings.
26. We have included care of other household members in the home-production category.

657

658

The Journal of Human Resources
leisure, caregiving, home production, and employment time. As discussed in Section
III, women with children younger than13 sometimes face an additional cost of their
time beyond their hourly wage which is represented by the price of child care. For
very young children, supervision is necessary at all times so that an hour of mother’s
time in the labor market means she receives her wage, but she may have to pay for an
hour of child care. On the other hand, an hour of childcare provided by the mother
means that she does not have to pay for childcare that hour. Leisure and home production are sometimes engaged in simultaneously with supervisory child care. For
these uses of time, the price of childcare is not relevant, but there are uses of adult
leisure time for which nonparental childcare is used and the same can be expected for
some home-production time (try painting the bedroom with an awake two-year-old at
home). Thus, the price of childcare is included in all four time-use equations.
C. Estimations to Produce Instruments for Economic Factors
All the variables used as determinants of time use described above except for the three
time cost variables come directly from the ATUS. The three price of time variables, on
the other hand, are predicted values obtained from initial stage estimation. The predicted wage is obtained, as is typical, by estimating a sample-selection-corrected wage
equation using ATUS data. We would have liked to generate the price of nonparental
childcare the same way but the ATUS data do not include childcare expenditure information.27 Instead, to model childcare costs we used the fourth wave of the 2001 Panel
of the Survey of Income and Program Participation, which was administered between
September and December 2002. Employed women with children younger than five
were asked about their expenditure on childcare for their youngest child. In addition,
employed women with children between the ages of six and 14 were asked about their
expenditure on childcare for their youngest child in that age range. We eliminated those
whose youngest child was 13 or 14 and those who were either currently in the military,
in school, or unemployed. We used the resulting sample to estimate the price of childcare for children age five or under and separately for children between the ages of six
and 12. We could have then averaged the zero- to five-year-old price of childcare and
the six- to 12-year-old price of childcare, but we have chosen to keep them separate
since the availability of five or six hours of school time which doubles as nonparental
childcare time makes childcare for six- to 12-year-olds very different from that for children five and under. The procedure we used to estimate the hourly price of childcare is a
standard bivariate selection correction model described by Tunali (1986) and used by
Connelly and Kimmel (2003a, 2003b). Using this procedure, we predicted the weekly
expenditure on childcare correcting for the self-selection of being both employed and
paying for care. The childcare price equation is
ð7Þ

Pcci ¼ a0 + aZi + u1 l1i + u2 l2i + ui

where l1 and l2 correct for the sample selection of only employed mothers who are
paying for care being included in the estimating sample. Having estimated the model
27. Another option is to use exogenous local childcare price measures but those data are unavailable
nationally for our time period.

Kimmel and Connelly
using SIPP data, we then used the dot product of the predicted coefficients a and the
Zi’s from the mothers in the ATUS sample. This dot product can be interpreted as
the predicted weekly expenditures, unconditional on paying of care and being
employed.28
D. Identification
Estimating multistep models such as these requires strict attention to equation identification. We confront these issues at two levels. At the first level, the wage and the
two prices of childcare are estimated in two