The Effects of Female Employment Status on the Presence and Number of Children

Author(s): Adriaan S. Kalwij

Source: Journal of Population Economics, Vol. 13, No. 2 (Jul., 2000), pp. 221-239
Published by: Springer
Stable URL: http://www.jstor.org/stable/20007712 .
Accessed: 05/12/2014 17:53
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

Springer is collaborating with JSTOR to digitize, preserve and extend access to Journal of Population
Economics.

http://www.jstor.org

This content downloaded from 136.145.174.50 on Fri, 5 Dec 2014 17:53:40 PM

All use subject to JSTOR Terms and Conditions

J Popul

Econ

13: 221-239

(2000)

--Journal

of

Population
Economics
?

Springer-Verlag

2000

The effects of female employment status on the presence

and number of children
Adriaan
Institute
Oxford

S. Kalwij
of Economics

OX1

(Fax: +44-1865-271094;
Received:

and Statistics,

3UL, United

3 November

University

of Oxford,

St. Cross

Building,

Manor

Road,

Kingdom
e-mail: adriaan.kalwij@economics.ox.ac.uk)

1998/Accepted:

22 September

1999

The main concern of this paper is to analyze the effects of female

status on the presence and number of children in households
in
employment
the Netherlands.
For this purpose a hurdle count data model
is formulated
and estimated by the generalized method
of moments.
The hurdle takes ex?
account
into
the
between
female
status
plicitly
interrelationship
employment
and timing of first birth. The number of children, once children are present in
on female employment
the household,
is modeled
status. The
conditional
status is a major determinant
empirical results show that female employment
of the presence and number of children in households:
employed women
schedule children
later in life and have fewer children compared
to non
attainment constant. After controlling
employed women, holding educational
for female employment
attainment of both the woman
status, the educational
are found to have relatively small effects on the
and the man in the households
presence and number of children.

Abstract.

JEL

classification:

Key words: Hurdle

C35,

J13, J20

count data model,

fertility,

female

employment

I would
like to thank Jo?o Santos Silva, Frank Windmeijer
and three anonymous
referees for
and suggestions.
I gratefully
the hospitality
of the International
helpful comments
acknowledge
Research
Centre for Economic
in Torino
Research
(ICER) where most of the writing up has been
done. The empirical
Institute Tilburg while being
analysis has been carried out at the Economics
there. The data was provided
in
employed
by Statistics Netherlands
(CBS). The views expressed
this paper are those of the author and do not necessarily
reflect the views of the CBS. Responsible
editor: Rainer Winkelmann.

This content downloaded from 136.145.174.50 on Fri, 5 Dec 2014 17:53:40 PM

All use subject to JSTOR Terms and Conditions

222

A.S.

Kalwij

1. Introduction
The main concern of this paper is to analyze the effects of female employment
status on the presence and number of children in households
in the Nether?
lands. For this purpose a hurdle count data model
is formulated
and esti?
mated. The hurdle takes into account
the interrelationship
between female
status and timing of first birth. Once children are present in the
employment
ismodeled
the number of children, i.e. the count variable,
condi?
household,
status. This approach takes the endogeneity
tional on female employment
of
status explicitly
to
into account and makes
it possible
female employment
status
effects
of
female
and
the
educational
attain?
employment
disentangle
on the presence and the
ment of both the man and woman
in the household
number of children. In the empirical analysis, only the conditional
expectation
of the number of children is specified and a generalized method
of moments
estimator
is employed. This approach
relaxes the distribution
assumptions
made when estimating count data models using fertil?
that are conventionally
ity data.
The outline of the paper is as follows: Section 2 discusses the relevant lit?
erature and the main contribution
of this paper to the literature. Section 3
discusses the data. Section 4 formulates
the hurdle count data model,
specifies
conditions and describes the estimation procedure. Section 5 dis?
the moment
cusses the estimation results and Sect. 6 concludes.

2. Previous

empirical

studies

variables affect the fertility decisions

Becker (1960) argues that socioeconomic
of households
and that fertility decisions can be analyzed within an economic
income and the
framework. He emphasizes the relationship between household
that there is a positive correlation between
number of children and concludes
for contra?
household
income and the number of children, after controlling
on
the
number
of children
Most
the
of
research
analysis
ceptive knowledge.
builds on this pioneering
study of Becker and is concerned with the determi?
nants of the number of children in a household
1973,
(see for instance, Willis
and more recently Siegers 1985; O'Malley Borg 1989; Cigno 1991). During the
last two decades the emphasis has been on life-cycle fertility behavior and re?
the timing of births rather than com?
search has shifted towards investigating
and
and McCulloch
1984; Heckman
pleted fertility (see for instance, Newman
Walker
1990; Groot and Pott-Buter
1992). These studies on fertility dynamics
to analyze the timing and spacing of births. One of
employ hazard rate models
is that women with
the empirical findings in the literature on fertility dynamics
schedule
costs of having children (e.g. high wage women)
high opportunity
to women with low op?
births later in life and have fewer children compared
portunity costs. An important implication of the empirical findings in these
is that the presence and number of
studies on household
fertility decisions
decision
constraints
children are not exogenous
imposed on the household
decisions
and are affected by socio
of household
but are outcomes
making
the female labor supply literature provides
economic variables. Furthermore,
more
than sufficient evidence that the presence of children has a significant
probability
negative effect on the female employment
(see for instance, Heck?
one may consider
man and Macurdy
1980; Mroz
1987). As a consequence,

This content downloaded from 136.145.174.50 on Fri, 5 Dec 2014 17:53:40 PM

All use subject to JSTOR Terms and Conditions

Female

employment

and number

of children

223

to be closely interrelated. In order

and fertility decisions
female employment
to get a better understanding
of life-cycle fertility behavior and its determi?
nants, one cannot ignore this interrelationship with life-cycle female employ?
ment,

and

vice

versa.

studies investigating
the interrelationship
be?
empirical
Conventionally,
tween female employment
and fertility decisions at the household
level em?
Studies such as Willis
equations model.
ploy a simultaneous
(1973) and Sieg?
ers (1985) use a static framework
and investigate jointly the female labor
study of
supply decisions and completed fertility, in line with the pioneering
of
Hotz
Later
studies
Blau
and
Robins
Becker
and
Miller
(1960).
(1989);

(1988);Moffitt (1984);Walker (1995) and Bloemen and Kalwij (1996)model

birth decisions rather than completed fertility jointly with the female employ?
ment decisions. The empirical analysis of Hotz and Miller
is restricted to
one
at
who
have
least
child.
This
be
in their
rationalized
may
couples
birth
of
but
around
the
the
first
child
the
especially
approach,
interrelationship
between fertility and female labor supply is observed to be strongest, hence it
to model
would be desirable
this. Moffitt
the
(1984) skillfully demonstrates
into account but his results also
importance of taking the interrelationship
suggest that timing issues cannot be investigated
properly using a static
econometric
framework.
Blau and Robins
(1989) have taken a dynamic
the
they
approach. Although
acknowledge
importance of the interrelationship
between fertility and labor supply decisions,
the econometric
framework uti?
lized (a competing
risks model)
does not allow for this. Basically
they im?
assume
between
labor
female
market
transitions
and the
plicitly
independence
a dy?
of
births.
Walker
and
Bloemen
and
utilize
timing
Kalwij
(1995)
(1996)
a multiple
namic econometric model,
state transition model, which explicitly
takes the interrelationship
between the female employment
and fertility deci?
sions into account. Such an approach makes
it possible to analyze the effects
of socioeconomic
of the household on the timing of births and
characteristics
the number of
and, consequently,
lifecycle female employment
simultaneously
children at the end of a woman's
fertile period.
Given the main concern of this paper, this latter approach of employing a
seems most appropriate1. However,
state transition model
the data
multiple
requirements for estimating such a dynamic model are high: panel data with a
or retrospective
data on the complete female employ?
large time dimension
a researcher has available only cross
ment and fertility histories. Usually
section data or panel data with a short time dimension. Fertility history may
on the basis of the age of the children in the household
be reconstructed
but
the complete
labor market history of the woman
in the household will be
more difficult or even impossible
one observes for
to reconstruct. Typically
each household
the number of children present and the employment
status of
in the household
at the time of interview. The observed values of
the woman
these two variables are the outcomes of a sequential decision-making
process
of the household
this paper adopts a
up to the time of interview. Therefore
count data model
to analyze the effects of female employment
status on the
number of children. Conceptually
such a model takes the underlying dynamic
nature of the stochastic process into account and can be estimated on a single
cross-section.

Count data models have already been applied in the demographic

litera?
ture. Typically,
the numbers of children observed are assumed to be realiza?
tions of a (generalized) Poisson process (e.g.Winkelmann
1995 and Wang and

This content downloaded from 136.145.174.50 on Fri, 5 Dec 2014 17:53:40 PM

All use subject to JSTOR Terms and Conditions

224

A.S.

Kalwij

Famoye,
1997) or of a more complex process taking hurdles into account
Silva
and Covas
1998). (An excellent discussion on hurdle count data
(Santos
can be found in Mullahy
models
and Trivedi
(1986, 1998) and Cameron
female employment
status or in?
(1998).) These studies either do not model
clude it as an exogenous explanatory variable. If female employment
status is
included as an explanatory
variable then it is almost always found to be a
in both the relative impact on the number of children
variable,
dominating
a discussion
and the level of significance. However,
regarding the possible
a possible
of
is
If one acknowledges
this
variable
absent.
endogeneity
usually
status
between
female
and
the
presence and
interrelationship
employment
number of children, then one important reason for not explicitly modeling
status is that one cannot estimate a simultaneous
female employment
equa?
for
female
and the number of children using a Pois?
tions model
employment
son based count data model.
and Santos Silva
See, for instance, Windmeijer
a
on
for
internal
discussion
this.
for
reasons, one
Basically,
consistency
(1997)
needs to assume that the presence and number of children does not affect the
it extremely hard, if not impos?
female employment
probability. This makes
sible, to come up with an instrument to identify the effect of female employ?
ment status on the number of children.
Santos Silva and Covas (1998) demonstrate
the importance of taking hur?
dles into account when modeling
argue
fertility and convincingly
completed
characteriz?
that hurdles may be the reason for the observed underdispersion
the discussion above, a woman
ing completed fertility data. Given
(or house?
at
faces
hurdle
the
time
of
first
the
birth, largely be?
largest
hold) presumably
cause of the interrelationship
status and the timing
with female employment
to be a more
is not only considered
of first birth. A hurdle count data model
a standard
to
household
of
way
modeling
fertility, relatively
appropriate
count data model,
it also makes
it possible to take into account the simulta?
status. This
the presence of children and female employment
neity between
of modeling
female
simultaneously
partly solves the limitation
approach
a
status
and
the
number
of
children
Poisson
using
generalized
employment
to
regression model. This is the route followed in this paper and is considered
to the literature. As a consequence
of using such
be the main contribution
an approach,
status and the educational
the effects of female employment
attainment on the number of children can be disentangled.

3. Data:

the Dutch

SocioEconomic

Panel

from the SocioEconomic

Panel
empirical analysis is based on micro-data
waves
At
of
this
research
all
from
the
time
of
the
Netherlands.
starting
(SEP)
re?
1994 were available. About
5000 households
1986 up to and including
more
one
wave.
can
There
be
than
to
in
each
the
survey
respondent per
spond
is asked questions
about his socioeconomic
and each respondent
household
is a person at least 16 years old. In
situation. (A respondent
and demographic
over 15 should complete the question?
principle each person in the household
naire.) Up to 1990 the survey has been conducted twice a year, a wave inApril
for this paper are asked in
in October. The relevant questions
and a wave
October. From 1990 onwards the survey has been conducted only once a year
in May. Although
the empirical analysis of
is collected
and all information
in Sect. 4
this paper is based on panel data, the econometric model proposed
The

This content downloaded from 136.145.174.50 on Fri, 5 Dec 2014 17:53:40 PM

All use subject to JSTOR Terms and Conditions

Female

employment

and number

225

of children

can be estimated on a single cross-section.

are no calendar time effects, panel data
cohort and age effects (see Sect. 5).

Under
the assumption
that there
is required to identify both birth

selection

3.1. Sample

sample is restricted to married and cohabiting women who are at most 40

is not observed
years of age. The total number of children born to a woman
directly and has to be inferred from the observed number of children in the
at the time of the interview. Households
in which the woman
is
household
40 years of age are observed to reduce in size because of
over, approximately,
children leaving the parental home. For this reason the age of 40 is chosen as
the upper bound in order to reduce the potential problem of underestimating
the number of children of a household. Household
is not modeled
formation
and for this reason the sample is restricted to married and cohabiting women.
on 2416 households
over the period
The resulting sample has information
in total). (About 900 observations were deleted
1986-1994
(11391 observations
from the sample because of missing
values on the educational
attainment
in the household.) All results in this paper are
variables of the man or woman
conditional on this selection. Addressing
possible selection problems is beyond
the scope of this paper.

The

3.2. Descriptive

statistics

are observed. This shows that

Table 1 reports the number of years households
about 20% of the households
remain in the panel for 8 or 9 years. These
households
attribute to almost 50% of the observations.
Table 2 reports the
statistics
of
in
the
variables
used
the
per
year
sample
empirical analysis. A
woman
is on average about 33 years of age and the man in the household
is
on average 2 to 3 years older. There is slight increase over time in the per?
The educational
attainment
of men
is
centage of higher educated women.
more or less stable over time. Employment
is defined as having a paid job (full
or part-time). A woman who has a job but works zero hours because of ma?
rate is about
ternity leave is registered as being employed.2 The employment
leave. About 80% of the households have
38%, including women on maternity
children and the average number of children for households who have chil?
dren is just over 2. Table 3 reports the employment
behavior before and after
birth of the first child for each level of education. This table is based on a
in which the woman has given birth to her first
subsample of 339 households
child during the observation
period. Table 3 shows that the average age at
which a woman gives birth to her first child increases with the level of educa?
tion: from 27 years for women with education
level 1 to 31 years for women

Table
Number

1. The

number

of years households

of years

are observed

123456789

Number of households

441

288

in the panel

Total

301

163

247

195

172

271

This content downloaded from 136.145.174.50 on Fri, 5 Dec 2014 17:53:40 PM

All use subject to JSTOR Terms and Conditions

338

2416

226 A.S.

Table

2. Number

of observations

and

(NOB)

Year 1986
NOB 1567

1987
1462

sample means

of all relevant

1988
1283

1989
1357

1990
1372

1991
949

variables

Kalwij

per year

1992
1221

1993
1129

1994
1051

'60

Variable

Sample means

Age of thewoman
Year of birth

31
'55

32
'56

32
'56

32
'57

32
'58

33
'59

33
34
'59

34
'60

Educational
attainment

of the

woman3
1

0.33

0.33

0.32

0.29

0.31

0.34

0.29

0.28

level 2

0.48

0.48

0.48

0.49

0.47

0.45

0.50

0.50

0.50

level 3

0.19

0.19

0.20

0.22

0.22

0.21

0.21

0.22

0.23

34

34

35

35

35

36

36
37

37

level

Age of theman

0.27

Educational
of the

attainment
mana
1

level

level 2
level 3

Employment status0
Presence
Number

0.19

0.18

0.18

0.15

0.17

0.20

0.18

0.19

0.18

0.52

0.52

0.53

0.54

0.52

0.50

0.52

0.52

0.51

0.29

0.29

0.29

0.31

0.31

0.30

0.30

0.30

0.31

0.38

0.39

0.38

0.40

0.36

0.34

0.38

0.39

0.33

of children

0.76

0.75

0.77

0.77

0.79

0.83

0.81

0.81

0.81

of children0

2.00

2.02

2.06

2.06

2.08

2.09

2.12

2.18

2.15

1 is at most primary or secondary

Level
level 2 is intermediate
vocational
education
education,
a university
education
or higher.
(MBO) and level 3 is higher vocational
degree
(HBO),
b
to 1 if the woman
is employed,
0 otherwise.
Equal
c
with children.
Only for households

Table

3. The

average
age of the women
(Age), the average number of children
(Kids) and the
rate (ER) before and after the birth of the first child for each level of educa?
employment
tion. YB is the year from first birth (for instance, YB = ? 5 is defined as 5 years before the birth of
the first child) and n denotes
the number of observations

female

Education

YB

Age

-8

-7

1
7

-4

1 level

Kids

26

-6
18
25 -5

Education

ER
1.00
-

Age

level 2

Kids

23

17

23

Education

ER
1.00
0.88

Age

level 3

Kids

ER

22

1.00

25

1.00

0.00

26

24

0.92

14

27

0.86

0.86

39

25

0.95

15

27

0.07

0.87

11

25

0.82

58

26

0.02

0.95

33

28

0.06

0.91

18

25

0.83

82

26

0.01

0.93

45

29

0.09

0.87

-2

34
43

25
27

0
0

0.71
0.63

114
149

26
27

0.04
0

0.88
0.91

62
67

29
29

0.06
0

0.91
0.87

76

27

1.01

0.17

175

28

1.05

0.39

88

31

1.07

0.66

67

27

1.07

0.09

161

29

1.10

0.23

87

32

1.13

0.62

63

28

1.33

0.06

145

29

1.52

0.21

74

32

1.45

0.42

56

29

1.55

0.02

134

30

1.83

0.23

56

33

1.80

46

29

1.83

0.07

120

31

1.95

0.18

49

33

2.00

0.43

1.95

-3

-1

0.48

39

30

0.03

93

32

2.06

0.18

38

33

2.08

0.42

34

31

2.06

0.03

66

32

2.18

0.14

29

34

2.31

0.41

25

32

2.12

0.08

44

33

2.34

0.16

15

35

2.40

0.53

33

2.29

0.12

34

2.53

0.16

36

2.40

0.60

17

19

This content downloaded from 136.145.174.50 on Fri, 5 Dec 2014 17:53:40 PM

All use subject to JSTOR Terms and Conditions

Female

employment

and number

227

of children

rate drops for lower educated

level 3. The female employment
with education
women
from 63% in the year before to 9% in the year after the birth of the
first child and for higher educated women from 87% in the year before to 62%
in the year after the birth of the first child. The differences in the employment
to the
rate before the birth of the first child are relatively small compared
difference after the birth of the first child. More
importantly,
relatively to
lower educated women, higher education women are less likely to leave em?
ployment after the birth of their first child. The average number of children is
over 2 for all women
8 years after the birth of the first child. There is some
in a relatively
that higher educated women
have their children
evidence
smaller time span, compared to lower educated women.
em?
in the empirical analysis
is that a woman's
One assumption made
status remains unchanged
after the birth of the first child. Table 3
ployment
shows that the percentage of women who continue in employment
after the
first birth remains roughly constant. Or at least, there is very little evidence
to give birth return to work shortly after the
that women who stop working
status
birth of their first child. Women
appear to determine their employment
for the period after the birth of the first child around the birth of the first child.
Bloemen and Kalwij
make a
(1996) using different data from the Netherlands
similar observation.
are more
1 shows that higher educated women
likely to be em?
Figure
ployed than lower educated women, at all ages. Figures 2 and 3 show that the
schedule children later in life and, perhaps less con?
higher educated women
have
fewer
than the lower educated women. These figures
children
vincingly,
show that observed employment
and fertility behavior already confirms most
in the introduction. Figure 4 shows that fe?
findings of the studies discussed
status has a large impact on both the timing of children and
male employment
the number of children. Figure 4 also shows that, after controlling for female
attainment of the woman has relatively little
status, educational
employment

Fig.

1. Female

employment

rate by age and education

level

This content downloaded from 136.145.174.50 on Fri, 5 Dec 2014 17:53:40 PM

All use subject to JSTOR Terms and Conditions

228

A.S.

Educationlevel 1
Educationlevel2
Educationlevel3
k .LJ"
A-

Fig.

2. The

fraction

of households

with

children

by age and education

Kalwij

-A- -A. . ,Ai

level

' Educationlevel 1
Educationlevel2
Educationlevel3

"'
&*
A. .A'
, a'

&
18

20

.A
22

24

26

28

30

32

34

36

38

40

Age
Fig.

3. The

average

number

of children

by age and education

level

impact on the number of children of 40 year old women. Of course, educa?

tional attainment of the woman
is highly correlated with educational
attain?
ment of the man. This and possible birth-cohort
effects make any inferences
based on these figures ambiguous. Therefore
the main purpose of the econo?
metric analysis in the next sections is to disentangling
the effects of female
status
woman
on the presence
and
educational
attainment
of
the
employment
and number of children, controlling for educational
attainment of the man in
the household
and the birth-cohort.

This content downloaded from 136.145.174.50 on Fri, 5 Dec 2014 17:53:40 PM

All use subject to JSTOR Terms and Conditions

Female

Fig.

employment

4. The

average

4. Empirical
4.1. Model

and number

number

229

of children

of children

by age, employment

status and education

level

framework
outline: a hurdle count data model

on her employment
status and
is assumed to decide simultaneously
A woman
the timing of births. After the birth of the first child she is assumed to no
status. Hence,
she is assumed to decide simul?
longer change her employment
to
children
with
work or not. The observed
whether
combine
having
taneously
status at any given period in time
number of children and female employment
are assumed to be the outcome of this decision making
process. To model
a hurdle count data model
in which the inter?
is employed
these outcomes
status is
female
of
children
and
between
the
presence
employment
relationship
taken into account in the hurdle. The number of children, once children are
on
is modeled
conditional
i.e. the count variable,
present in the household,
status after the birth of
status. Hence, female employment
female employment
variable for the
the first child is assumed to be a predetermined
explanatory
on the presence of children. Furthermore,
number of children, conditional
conditional
independence between the hurdle and the count process is assumed.
can be used to analyze the observed be?
The proposed
reduced form model
in the sample under 40 and is not restricted to women
havior of all women
the childbearing period.
who completed

4.2. The moment

conditions

and estimation procedure

and is de?
The dependent variable is the number of children in the household
index. The ex?
noted by Yth, where t is the time index and h is the household
pectation of Yth conditional on some exogenous household characteristics
(z^)

This content downloaded from 136.145.174.50 on Fri, 5 Dec 2014 17:53:40 PM

All use subject to JSTOR Terms and Conditions

230

A.S.

Kalwij

is denoted by E[Yth\zth]6], where 9 denotes the parameters

of interest. Typi?
one
some
is
in
the
interested
effects
of
characteristic
cally
exogenous
marginal
on the expected number of children:
one
If
is
prepared to
dE{Yty\zth\ 6)/dzth.
of Yth the maximum
likelihood estimator can be used
specify the distribution
of a correctly specified distribution,
leads to the
and, under the assumption
most efficient consistent estimates of 6 (see for instance, Mullahy
1986; Pohl
one
assumes
meier and Ulrich
If
one can
is
Poisson
distributed
Yth
1995).
likelihood estimator which results in consistent
employ a pseudo-maximum
estimates even if the true distribution
is not Poisson
(see for instance, Cameron
and Trivedi 1986). Unfortunately,
for a hurdle type of distribution no pseudo
likelihood
of the distribution
type result is known. Hence, misspecification
results in inconsistent
estimates. For this reason a generalized methods
of
moments
estimator
(e.g. Hansen
1982), is employed as proposed by Santos
Silva and Windmeijer
of
(1998) and Mullahy
(1998). Only the first moment
the distribution
is specified, hence the marginal
effects dE(Yth\zth',0) / dzth are
identified.
Given the model outlined above and using the law of iterative expectation,
the conditional
expectation of Yth can be written as follows:
E[Yth\zth',0]

EI{Yth>^Wth[E\Yth\I(Yth>o), Wth,zth\9\

\zth;0\, (1)

= 1 if the woman
where Wth denotes female employment
status Wth
is em?
(
an
is
indicator
function
for
the
of
ployed and 0 otherwise).
presence
I(Yth>o)
children. Equation
between children and
(1) shows that the interrelationship
female employment
status is only allowed for in the first step, i.e. the hurdle.
In the second step, once children are present in the household,
female em?
status is assumed to be predetermined.
Given
the binary nature of
ployment
status and the indicator function, Eq. (1) can be written as
female employment
follows:
E[Yth\zth,6]
=

P(I(Yth>0)

IN,

a)

=
x {P(Wth =
0\I{Yth>0)
+ P(Wth =

l\I{Yth>0)

x E[Yth\Wth =

l,/(yA>o)

l;zth,a)E{Yth\Wth

0,I{Yth>0)

1;**,^]

l;zth,a)
=

1;W2]}. (2)

where 6T = (a7,/?7,/?7).
Note
that in the case where there are no children in
=
the household,
i.e.
the
conditional
expectation of Yth is equal to 0,
I(Yth>o) 0>
status.
of
female
irrespective
employment
Once the functional forms of the choice probabilities
and the conditional
of the number of children are specified, one can obtain estimates
expectation
of all parameters of interest using the empirical moment
condition
implied by
in
this
is
and
for this reason the
difficult
However,
practice
extremely
(2).
moment
conditions
condition
implied by the conditional moment
(2) are used
to estimate all parameters of interest stepwise. In total, six moment
conditions
are formulated:

This content downloaded from 136.145.174.50 on Fri, 5 Dec 2014 17:53:40 PM

All use subject to JSTOR Terms and Conditions

Female

E[wth(\-I{yth>{)))-Vx(Wth

=
El(l-wth)I{yth>0)-?r{Wth
E[wthI{yth>0)
-

Pr( Wth

E[Yth\ Wo,

E[(yth

if wth= 0

231

of children

and number

employment

\,I{Yth>0)=0\zth,cc)\zth\
0J{Yth>0)

\,I(Yth>o)

\\zth,a)\zth\

(3)

=0;

(4)

1 \zth,a) IZth\= 0; (5)

0,/(rA>o)

=0;

\;zth,?x]\zth)

0,

and yth > 0; (6)

=

E[{yth-E[Yth\Wth

l,/(rA>o)

1;W2]

=
\zth] 0,

if wi?= 1 and yth > 0; (7)

E[yth

P{Wth

-P{Wth

=
=

X ^[r^|^

0,/(rrt>0)
l,/(rrt>o)
=

=
=

l\zth,0L)E[Yth\Wth

0,I{Yth>0)

\',zth,?x]

IN,?)

l,/(yA>0)

hZth,?2]\Zth}

= 0.

(8)

number of children is denoted by yth and the observed female

status
is denoted by w^. The estimation of the parameters of in?
employment
terest can be done in two stages. In the first stage GMM
estimates of a are
conditions
the moment
the binary
obtained by exploiting
(3) to (5). Given
nature of the dependent
variables entering these three moment
conditions,
a maximum
function and employing
distribution
the probability
specifying
likelihood estimator yields identical estimates. In the second stage the sample
is restricted to those couples who are observed to have children (yth > 0) and
the resulting sample is split on the basis of the observed female employment
status (wth). Estimates
the moment
of ?x and ?2 are obtained by exploiting
moment
The
additional
condition
and
conditions
(8) yields over
(7).
(6)
test. For this purpose, a condi?
and is used for a specification
identification
test statistic is constructed based on the estimates obtained from
tional moment
moment
and MacKinnon,
five
conditions
the
first
using
(see e.g., Davidson

The

observed

Chapt.
16).
For the empirical analysis one needs to specify the probability distribution
status and the presence of children and the
function of female employment
of
of
the
number
of children. The joint distribution
conditional
expectation
status and the presence of children is specified as follows:
female employment

P(Wth

i,I{Yth>0)=j\zth,x)= ^
E

exP(^a(/j))
,,(9)
FI:V;f
exp(z^a(/J))5

(iJ)eS

where S is the set of feasible alternatives, S = {(0,0), (1,0), (0,1), (1,1)}. This
function is the well-known
distribution
multinomial
distribution
probability
and the normalization
chosen is a^o) = 0> hence otT? (aTx
^aLwaJjJ.
one can choose a bivariate probit model. However,
Alternative
in a bivariate
between
status
model
the
female
and the
interrelationship
employment
probit
presence of children is only allowed for through the error terms. The multi

This content downloaded from 136.145.174.50 on Fri, 5 Dec 2014 17:53:40 PM

All use subject to JSTOR Terms and Conditions

232

A.S.

Kalwij

nomial
allows for an interrelationship
logit model
through the observable
as well. For instance, the multinomial
variables
model
allows educational
to affect the presence of children differently when the woman
attainment
is
not
reason
than
a
when
she
is
For
this
multinomial
employed
employed.
logit
is favored in this paper. The multinomial
model
logit model yields a more
flexible empirical specification with respect to the observable
characteristics
but at the costs of imposing a restriction on the relationship
through the error
terms. Basically,
the case that the unobservables
female employ?
determining
ment status and the presence of children are not correlated is not nested within
In contrast to the univariate case (probit versus logit), a compari?
this model.
son between the bivariate probit and multinomial
is complex and
logit model
there is no straightforward
link between
the parameter
estimates obtained
from these two models.
The conditional
expected value of the number of children, once children
are in the household,
is specified as follows:

=
= 1;
E{Yth\Wth 0,/(rrt>o) Wi)

= 1
+exp(z^);

=
= 1;
E{Yth\Wth l,/(r?>o) W2)

= 1
+exp(z^2).

(10)
(11)

This specification ensures that the expected number of children conditional

on
the presence of children is always greater than one.3 Conditional
on the cor?
rect specification of the first moments,
as specified in
the moment
conditions
estimates of <x,?x and ?2.
(3) to (7) yield consistent GMM
The assumptions
this model as outlined
in Sect. 4.1 may con?
underlying
sidered to be strong but the resulting model
is conceptually
less restrictive than
the count data models employed
in the empirical studies mentioned
in Sect. 2.
In the same way that a hurdle specification and left-censored data are closely
related issues, the fact that not all women have reached the end of their fertile
data. The approach
period is closely related to the issue of right-censored
taken above implicitly takes this into account, for instance by conditioning
the
on

moments

first

4.3.

Generalized

age.

of moments

method

A standard GMM
timation procedure
as follows:

(GMM)

estimator is employed. For completeness,

the es?
however,
is briefly described. The moment
conditions can be written

E{J(zih)p(yth,Wth,Zth,0)}
where /(
condition

estimator

) is the matrix
is:

with

= 0
instruments.

The

sample analog

of these moment

Th

= 0.
Yl J(zth)p(yth,wth,zthi9)/N
Y2
A=l
t=Th

N is the total number of observations,

h and Th the last. Given panel data

of household
Th is the first observation
GMM
estimates
{yth^wth^th}^":HTh,

This content downloaded from 136.145.174.50 on Fri, 5 Dec 2014 17:53:40 PM

All use subject to JSTOR Terms and Conditions

Female

and number

employment

are obtained

233

of children

by solving:
H

Ogmm= argmin^^ ^p{yth,Wth,zth\9)'J{zth)'

h=\
t=Th

Th

h
["

rh

]_1

]P ^J{zth)J{zth)'
\_h=\

t=Th

Th

^2
^2j{zth)p{ythiWth,zth',9).
h=\
t=Th

valid under heteroscedasticity

The standard errors reported are asymptotically
and the goodness of fit measure
(the R2) reported is defined as the square of
between
the number of children observed and its estimated
the correlation
conditional
expectation.

5. Empirical
5.1. Empirical

results
specification

In the empirical analysis two different models are estimated. The first model
is
a hurdle model where female employment
status is not explicitly modeled
(model I). In model I, the first step iswhether or not children are present in the
household
and the distribution
function for this is taken to be of a logit type.
In the second stage, the conditional
expected value of the number of children,
once children are present in the household,
is specified as (1 + Qxp(z'th?)). The
is as specified in Sect. 4.2 (Eqs. (9), (10) and (11)) and takes the
second model
status and the presence of chil?
between female employment
interrelationship
seem
dren into account (model II). Although
these two models
conceptually
to be nested, they are not nested from a statistical point of view. Also a stan?
is not nested in the hurdle count data model as for?
dard count data model
are educational
in Sect. 4. The exogenous
mulated
variables in both models
attainment of the both the man and the woman
in the household,
age and age
and year of birth of the woman. Educational
attain?
squared of the woman
ment variables are included as a proxy for the lifetime earnings of the both the
man and woman. A priori one may expect higher educated women
to have
to lower edu?
fewer children because of higher opportunity
costs, compared
cated women. The educational
of the man may have a positive
attainment
effect on the likelihood of having children and the number of children (an in?
come effect). This would be in line with the results of Becker (1960). The year
of birth is included to control for possible birth-cohort
effects. One may argue
that the empirical
allows for little heterogeneity
in household
specification
income as an ex?
earnings and one should, for instance, include household
is not possible since household
income is
planatory variable. This, however,
the result of previous
In other
female employment
and fertility outcomes.
income is bound to be endogenous
and the only way to
words, household
this properly
model
the income process jointly with the female
is to model
and
the scope of this
employment
fertility process. This is clearly beyond
paper, and for this reason time-constant
regressors such as the educational
are included to proxy lifetime earnings. These variables are as?
attainment
sumed to be exogenous.

This content downloaded from 136.145.174.50 on Fri, 5 Dec 2014 17:53:40 PM

All use subject to JSTOR Terms and Conditions

234 A.S.

Table

4. Estimation

results, model

I
Presence

Explanatory

variables

of children

Education
Education
Education

level 2, woman
level 3, woman
levelman
2,
levelman
3,

R2

*
**

moment

test

at the 5% level.
Significant
=
p.e.
estimate,
parameter

5.2. Estimation

-0.55

(0.10)*

-1.37

(0.12)*

0.24

(Year of birth)/10
Conditional

Number

p.e. (s.e.)**
p.e. (s.e.)**
Constant
-11.5(1.98)*

(0.12)*

of children

-8.55(0.79)*

Age/10
8.25(1.05)*
(Age/10)2
-1.02(0.17)*
Education

Kalwij

4.54(0.43)*
-0.61
0.01
-0.05
-0.03

-0.26

(0.13)*

-0.33

0.07(0.04)
(0.15)*

(0.06)*

(0.03)

-0.01

(0.04)
(0.03)
(0.04)

0.47
2.43*

s.e. =

standard

error

results

I are reported in Table 4. Table 4 shows the by

Estimation
results for model
now well-known
are less likely to have
result that higher educated women
to lower educated women,
after controlling
for age. The
children compared
of the man and the presence of
attainment
relationship between educational
The effect of year of birth im?
children appears to have an inverse U-shape.
more
recent
that
the
the
woman's
birth
year the less likely she is to have
plies
children for a given age. An interesting finding is that all of the explanatory
variables, except for age, have no significant effect on the number of children.
test statistic is significant, which may indicate some
The conditional moment
kind of misspecification.
II are reported in Table 5. The condi?
The estimation
results for model
test statistic is insignificant which supports the hypothesis of no
tional moment
In contrast to the results of model
attainment
I, educational
misspecification.
of the man has no significant effect on the presence of children, and appears to
have a negative effect on the number of children in a household where the
woman
is employed. Year of birth has a negative effect on the probability
of
small
this
is
and
effect
children.
relatively
only significant at
Although
having
to cou?
this does suggest that compared
the 10% level for employed women,
are more
likely to
couples of later birth-cohorts
ples of earlier birth-cohorts,
results the conditional
expectations,
stay childless. To clarify the estimation
of the man and year of birth of
attainment
after controlling
for educational
are calculated and plotted in Figs. 5, 6, 7 and 8. Year of birth is
the woman,
set equal to 1960 and educational
attainment of the man is set equal to edu?
cational attainment of the woman. Figure 5 shows the well-known
finding in
the labor supply literature that highly educated women are more likely to be
employed at all ages, relatively to low educated women. Figure 6 shows that
to low edu?
schedule children later in life compared
highly educated women
cated women. Figure 7 shows that highly educated women have fewer chil?
to low educated women. These results are in line with the
dren compared
in Sect. 2. In the
in the fertility literature, as discussed
empirical findings
status has
for these findings, female employment
search for an explanation

This content downloaded from 136.145.174.50 on Fri, 5 Dec 2014 17:53:40 PM

All use subject to JSTOR Terms and Conditions

Female

Table

and number

employment
5. Estimation

II

results, model

Female
=

Wo,
Explanatory
Constant

variables

p.e.

1, Ya,

and

= 0

Education
Education

level 3, man

variables

-0.86

(0.34)

-0.55

Education

level 2, woman
level 3, woman
level 2, man

Education

level 3, man

Education
Education

0.08

Conditional
*
**

-11.9(4.22)*
9.73 (2.08)*
-1.35

(0.22)*

(0.33)*

0.75

(0.20)*

0.61

(0.24)*

(0.21)

-0.06

-0.03

(0.24)

-0.64

(0.32)*

-0.14(0.26)
-0.64
(0.34)

Wih

(0.22)

(s.e.)**

0.06 (0.04)

(0.03)*

-0.05

(0.02)

-0.12(0.05)*
-0.25
(0.06)*

(0.06)

0.03 (0.03)

(0.03)*

0.60

moment

test

at the 5% level.
Significant
=
p.e.
parameter
estimate,

1.94

s.e.

:standard

error

36

Fig. 5. The
tion level

1, Ya, > 0

(s.e.)**

7.64(1.08)*
-1.03
(0.15)*

0.04 (0.03)

(Yearofbirth)/10
R2

-14.2(1.91)*

0.02 (0.02)
-0.01

p.e.

-0.08

p.e.

(0.05)*

0.07

Wth

0.22(0.18)

p.e. (s.e.)**
-7.83
(0.62)*
4.10 (0.34)*

Age/10
(Age/10)2

0, Yth > 0

(s.e.)**

-1.60(0.30)*

Number
of Children
= 0
Wth
Explanatory
Constant

10.8(1.86)*

0.20 (0.22)
0.24 (0.26)
-0.36

Wth

of children

-11.0(3.85)*

1.07 (0.19)*
1.11 (0.24)*

(Yearofbirth)/10

the presence

p.e.

(s.e.)**

2.69(1.87)
-0.61
(0.30)*
level 2, woman
level 3, woman
level 2, man

Education

employment

0.51 (3.89)

Age/10
(Age/10)2
Education

235

of children

conditional

expectation

of female

employment

status, E[Eth\zth]

38

#], by age and educa?

This content downloaded from 136.145.174.50 on Fri, 5 Dec 2014 17:53:40 PM

All use subject to JSTOR Terms and Conditions

236

A. S. Kalwij

Educationlevel2
. Educationlevel3

I ?

-G??
, . .A"

Fig. 6. The conditional

cation level

expectation

of the presence

of children,

E[Yth

, . . .A

'
.Zbf- A

> 0\zth; 6], by age and edu?

>Educationlevel 1
Educationlevel2
i Educationlevel3

Fig. 7. The
level

conditional

expectation

of the number

of children,

E[Yth\zth\

0], by age and education

it possible to
been modeled
jointly with the presence of children. This makes
on
status
the
of
female
and number
the
effects
presence
employment
analyze
results the expected number of children
of children. Based on the estimation
can be
on female employment
status and educational
attainment
conditional
are
in
8
and
that
show
calculated. These conditional
Fig.
plotted
expectations
a large proportion
the
in the number of children between
of the difference
in Fig. 7 is explained by a difference
in female
different levels of education
on female employment
status the difference
status. Conditional
employment

This content downloaded from 136.145.174.50 on Fri, 5 Dec 2014 17:53:40 PM

All use subject to JSTOR Terms and Conditions

Female

and number

employment

Fig. 8. The conditional

woman,
E[Yth\wth,zth;

237

of children

of the number of children

expectation
level
0], by age and education

given

the employment

status of the

in the number of children between the different levels of education is relatively

small and insignificant, given the parameter estimates in Table 5.

6. Conclusions
status on the pres?
This paper has analyzed the effects of female employment
For this pur?
ence and number of children in households
in the Netherlands.
and estimated by the
pose a hurdle count data model has been formulated
The hurdle takes into account the inter?
of moments.
generalized methods
status and the timing of first birth.
female
between
employment
relationship
the number of children, has been
Once children are present in the household,
on female employment
status.
conditional
modeled
to lower edu?
The main results can be summarized as follows. Relatively
schedule children later in life, are less
cated women, highly educated women
likely to have children and have fewer children. These results are in line with
the empirical findings of earlier studies. The empirical results furthermore
effect on the presence
status has a dominant
show that female employment
and number of children: being employed significantly reduces both the likeli?
hood of having children and the number of children. The direct effect of edu?
on the presence and number of children is found to be
cational attainment
In other words,
the effects of educational
relatively small and insignificant.
attainment on the observed fertility pattern runs via the effects of educational
attainment on female employment
status, which in its turn significantly affects
the fertility behavior of households. The observed delay in having children by
shown in Fig. 8, is in line with the empirical findings of
employed women,
Bloemen
and Kalwij
(1996) who stress the importance of state-dependence
status dominates
the effects of educational
and show that female employment

This content downloaded from 136.145.174.50 on Fri, 5 Dec 2014 17:53:40 PM

All use subject to JSTOR Terms and Conditions

238

A.S.

Kalwij

on the timing of first birth. Here, in addition,

attainment
it is shown that fe?
status also has a dominant effect on the number of children
male employment
at all ages. These empirical
results emphasize
the importance
of explicitly
status when analyzing and trying to understand
female employment
modeling
hence life-cycle female employ?
the observed fertility behavior of households,
ment behavior.
as outlined
in this paper only partly solves the
The econometric model
the outcomes
of female
of
simultaneously
analyzing
endogeneity
problems
and fertility decisions.
Intuitively this model may be considered
employment
in this particular empirical analysis. However,
tackling the prob?
appropriate
count
in
data
models
is consid?
issues
lems surrounding
simultaneity
(hurdle)
ered to be a necessary and fruitful route to follow for future research. Fur?
thermore, the model
employed here is a reduced form model used to gain a
and fertility
of observed
better understanding
life-cycle female employment
issues related to, for instance, policy
behavior. To address more fundamental
rate after the birth of the first
schemes to increase the female employment
a
more
to
be
needs
taken. First steps in this direc?
structural
child,
approach
and
tion have been taken in Francesconi
Kalwij
(1999, Chapt. 4).
(1996)

Endnotes
1

a structural model
based on a
(1999) take such an
(1996) and Kalwij
in estimating
such a model
it is fair
involved
difficulties
of the computational
Because
approach.
are not very practical.
is clearly beyond
the
such a model
to say that these models
Estimating
scope of this paper.
2
after the birth of the child. In 1991 a papental
is at most up to 3 months
The maternity
period
leave. The maternity
for a further 6 months
it possible
leave scheme was introduced. This makes
more

A perhaps

model

life-cycle

is to formulate
approach
appealing
Francesconi
behavior.
of household

and estimate

are a bit more

than sketched
leave schemes
here, but they roughly
complex
parental
this
of one year. After
to a maximum
amount
period of leave, and registered as being employed,
has to return to work.
leave the woman
3
one may
and
for the underlying
DGP
Poisson
distribution
favor a truncated
Alternatively
?
as:
The main
the conditional
empirical
expectation
specify
exp(z'th?)/(l
exp(-exp(z^/?))).
is used.
when
this alternative
specification
findings of this paper remain unchanged
and

References
Bureau
National
In: Universities
GS (1960) An economic
analysis of fertility.
in Developed
and Economic
Research
for Economic
Change
(ed.). Demographic

Becker

Princeton
Blau

DM,

University
PK
Robins

Press,
(1989)

Princeton,
Fertility,

N.J.,

Committee
Countries,

pp 209-231
and

Employment,

Child-Care

Costs.

Demography,

26(2):287-299
Bloemen

a Multiple
of Births Decisions:
and Timing
A (1996) Female
Employment
Model.
CentER discussion paper, Tilburg University
and
based on count data: comparisons
PK (1986) Econometric
models
Trivedi

H, Kalwij
State Transition

Cameron

AC,
1:29-53
and tests. Journal of Applied Econometrics
of some estimators
applications
Cameron
Society Mon?
Analysis
of Count Data. Econometric
AC, Trivedi PK (1998) Regression
no. 30, Cambridge
Press, Cambridge
University
ographs,

Cigno A
Davidson

(1991) Economics
R, MacKinnon

1994 ed, Oxford University

Press, Oxford
of thefamily.
in Econometrics.
Oxford
and Inference
JG (1993) Estimation

Press, Oxford
M (1996) Part-Time
and Full-Time
Francesconi
Model.
Working paper, University
Dynamic

Work

and Fertility

of Married

of Essex

This content downloaded from 136.145.174.50 on Fri, 5 Dec 2014 17:53:40 PM

All use subject to JSTOR Terms and Conditions

Women:

University
A

Joint

Female

and number

employment

Groot

W, Pott-buter HA
Economics.
5:155-172

Hansen

LP

ometrica

of children

(1992) The

(1982) Large
50:1029-1054

sample

Heckman

239

properties

of generalized

(1980) A Life-Cycle

TE
JJ, Macurdy
nomic Studies 47-74

in the Netherlands.

timing of maternity

Model

of moments

method

of Female

Journal

Labour

of Population
Econ

estimators.
Review

Supply.

of Eco?

JR (1990) The relationship

between wages
and income and the timing and
JJ, Walker
from Swedish
data. Econometrica
58:1411-1441
spacing of births: evidence
longitudinal
Hotz VJ, Miller KI (1988) An Empirical Analysis
of Life-Cycle
and Female Labor Sup?
Fertility
56:91-118
ply. Econometrica

Heckman

Female Employment
and Fertility Decisions:
Kalwij AS
Consumption,
(1999) Household
econometric
Ph.D. dissertation,
Analysis,
Tilburg University
Moffitt
RA
Labor Supply, and Wages
of Married Women:
(1984) Profiles of Fertility,
Review of Economic
Studies
plete Life-Cycle Model.
TA
Mroz
of an empirical model
(1987) The sensitivity
economic
and statistical assumptions.
Econometrica
J (1986) Specification
Mullahy
Econometrics
33:341-365

and

testing

of

51(2):263-278
of married

women's

55(4):765-799
some modified
count

hours

data models.

J (1998) Much
ado about two: reconsidering
the two-part model
of Wisconsin-Madison
and NBER.
Working paper, University
Newman
CE (1984) A hazard rate approach
to the timing
JL, McCulloch

in health

Mullahy

a Micro
A Com?
of work

to

Journal

of

econometrics.

of births. Econometrica

52(4)
Borg M

O'Malley

(1989) The

Income-Fertility

Demography
26(2):301-310
Pohlmeier W, Ulrich V (1995) An
Demand
Santos

Silva

for Health.
JMC,

Covas

Working

Model

Effect

of the Net

of the Two-Part

of a Child.

Price

Decision

Process

in the

Journal
F

paper, ISEG/Universidade
Santos
Silva JMC, Windmeijer
Demand.

Econometric

Relationship:

paper,

30:339-361
of Human Resources
for Completed
Hurdle Model
(1998) A Modified
T?cnica de Lisboa
FAG

Fertility.

Working

for Health
Care
Spell Models
Multiple
(1998) Two-Part
T?cnica
de Lisboa/The
Institute
for Fiscal

ISEG/Universidade

Studies
en kindertal:
een micro-economische
JJ (1985) Arbeidsaanbod
of Groningen
University
JR (1995) Parental
Walker
Benefits
and Employment
and Fertility
of Wisconsin
University
F (1997) Modeling
Household
Wang W, Famoye
Fertility Decisions
Journal of Population
Economics
10:273-283
Regression.
Siegers

Ph.D.

analyse.
Dynamics.
with

dissertation,

Working

Generalized

paper,
Poisson

R (1995) Duration
and Dispersion
in Count Data Models.
Journal of
Dependence
& Economic
Statistics
13:467-474
Willis RJ (1973) A new approach
to the economic
Journal of political
theory of fertility behavior.
S14-S64.
Economy
Santos Silva JMC (1997) Endogeneity
an application
in count data models:
to
FAG,
Windmeijer
demand
for health care. Journal of Applied Econometrics.
12:281-294
Winkelmann

Business

This content downloaded from 136.145.174.50 on Fri, 5 Dec 2014 17:53:40 PM

All use subject to JSTOR Terms and Conditions