Is there such a thing as too long childcare leave?

Abstract

Purpose
The purpose of this paper is to revisit the question whether women’s employment is negatively affected in countries with very long periods of childcare leave.

Design/methodology/approach
The authors analyzed data on 192,484 individual women, 305 country-years, and 18-countries, combined with country-level data on childcare, unemployment and service sector size.

Findings
The authors found that in countries with short periods of childcare leave the motherhood-employment gap is smaller than in countries with no childcare leave, while in countries with long periods of childcare leave the motherhood-employment gap is bigger than with short periods of leave.

Originality/value
The authors argued that to correctly answer the long-leave question – the relationship between duration of leave and employment of women should be explicitly hypothesized as being curvilinear; and childcare leave should be expected to affect only mothers, not women without children; testing the long-leave hypothesis requires the use of country-comparative data in which countries are observed repeatedly over time; and is best tested against person-level data.

Full text

Is there such a thing as too
long childcare leave?
Rense Nieuwenhuis
Swedish Institute for Social Research (SOFI), Stockholm University,
Stockholm, Sweden, and
Ariana Need and Henk Van der Kolk
Institute for Innovation and Governance Studies (IGS),
University of Twente, Enschede, The Netherlands
Abstract
Purpose –The purpose of this paper is to revisit the question whether women’s employment is negatively
affected in countries with very long periods of childcare leave.
Design/methodology/approach –The authors analyzed data on 192,484 individual women, 305 country-
years, and 18-countries, combined with country-level data on childcare, unemployment and service sector size.
Findings –The authors found that in countries with short periods of childcare leave the motherhood-
employment gap is smaller than in countries with no childcare leave, while in countries with long periods of
childcare leave the motherhood-employment gap is bigger than with short periods of leave.
Originality/value –The authors argued that to correctly answer the long-leave question –the relationship
between duration of leave and employment of women should be explicitly hypothesized as being curvilinear;
and childcare leave should be expected to affect only mothers, not women without children; testing the long-
leave hypothesis requires the use of country-comparative data in which countries are observed repeatedly
over time; and is best tested against person-level data.
Keywords Employment, Public policy, Social policy, Childcare leave, Gender
Paper type Technical paper
1. Background and research question
Trends towards higher female labour force participation rates across OECD countries in
recent decades were identified to have been associated with a combination of demographic
and institutional determinants (Nieuwenhuis, Need and Van der Kolk, 2012). With respect to
demographic determinants, it was found that having children was the most important
determinant of women’s employment (Van der Lippe and Van Dijk, 2002). This means that
the trends towards lower fertility across OECD countries were associated with higher
female labour force participation rates (Kögel, 2004; Engelhardt and Prskawetz, 2004;
also see: Nieuwenhuis, 2015). Higher educated women are more likely to be employed than
lower educated, and women’s rising educational levels –in many countries now exceeding
that of men –contributed to rising female labour force participation rates (Bradley, 2000;
Van der Lippe and Van Dijk, 2002). Finally, declining marriage rates, and the increasing age
of first marriage, have also been associated with higher female employment rates
(Fagan and Norman, 2012).
These demographic changes took place in an institutional context, that further shaped
trends in women’s (opportunities for) employment. This context was characterized by a
changing labour market, with an increasingly large service sector providing employment
opportunities to women (Van der Lippe and Van Dijk, 2002). The expansion of family
policies, and particularly those aimed at the reconciliation of (maternal) employment and
having and raising children, also played an important role. Examples of such work-family
reconciliation policies are maternity leave and childcare leave, continued pay during this
leave, and childcare services and early childhood education (Hegewisch and Gornick, 2011;
Thévenon and Luci, 2012). More recently, paternal leave is becoming available, particularly
in the Nordic countries, but even there the uptake of leave among young fathers still is
International Journal of Sociology
and Social Policy
Vol. 37 No. 1/2, 2017
pp. 2-15
©EmeraldPublishingLimited
0144-333X
DOI 10.1108/IJSSP-07-2015-0074
Received 1 July 2015
Revised 16 October 2015
Accepted 25 October 2015
The current issue and full text archive of this journal is available on Emerald Insight at:
www.emeraldinsight.com/0144-333X.htm
2
IJSSP
37,1/2
Downloaded by Stockholm University At 05:29 07 March 2017 (PT)

substantially lower than among mothers (Eydal et al., 2015), and these gender differences
are persistent (Gornick, 2015).
While the literature on family policy outcomes, with respect to women’s employment, is
fairly well developed, scholars report contradicting findings regarding the impact of the
duration of leave on women’s employment. Childcare leave is argued to provide women the
opportunity to continue their attachment to the labour market after childbirth (Pettit and
Hook, 2005). Consequently, studies hypothesized and found that women’s employment is
positively associated with the availability of childcare leave, because mothers’employment
is facilitated by leave (Akgunduz and Plantenga, 2013; Fagan and Norman, 2012; Gornick
et al., 1999; Maldonado and Nieuwenhuis, 2015; Nieuwenhuis, Need and Van der Kolk, 2012;
Van der Lippe and Van Dijk, 2002). As a result, the “motherhood-employment gap”
(Nieuwenhuis, 2014), that is the difference in employment between mothers and women
without children, was found to be smaller in countries with longer periods of leave. On the
other hand, scholars regard leave as a mechanism of exclusion of women from the labour
market (Pettit and Hook, 2009), reporting a negative association between long periods of
leave and women’s employment (Bruning and Plantenga, 1999; Gornick and Meyers, 2003;
Morgan and Zippel, 2003; Moss and Deven, 1999; Pettit and Hook, 2009; Waldfogel, 2001).
Extensive periods of leave were argued to be associated with the traditional breadwinner
model, with women combining motherhood and employment “sequentially rather than
simultaneously”(Pettit and Hook, 2005, p. 784) and thus leaving the labour market for
extended periods of time after childbirth. In addition, very long periods of leave result in
women becoming detached from the labour market by losing out on experience (human
capital depreciation) (Gorlich and De Grip, 2008; Gornick and Meyers, 2003; OECD, 2001,
2011; Waldfogel, 2001).
Given these seemingly contradicting research findings, we revisit the question what the
impact is of long durations of childcare leave on women’s employment, focussing on the
difference in employment between mothers and women without children (the motherhood-
employment gap). The long-leave question is: to what extent was the motherhood-employment
gap larger between 1975 and 1999 in OECD countries providing long-term childcare leave,
compared to in countries providing short-term leave?
The goal of this paper is to recommend four improvements in answering the long-leave
question, and to empirically demonstrate their importance. Some of these improvements
have been applied before, but not consistently throughout the literature and not all four
simultaneously. The four improvements we address are detailed in the next sections, and
pertain to the (potentially) curvilinear nature between duration of leave and the employment
of women; the argument that childcare leave is expected to affect only mothers, not women
without children; the use of country-comparative data in which countries are observed
repeatedly over time; and the use of person-level data.
The main focus of our analyses is to demonstrate the importance of the four
improvements in estimating the total effect of childcare leave on women’s employment.
In that, we only include a limited number control variables in only one of our regression
models, to account for the demographic and institutional determinants discussed above.
We are interested in the duration effect, and include the sum of both paid and unpaid
periods of leave. It should therefore be emphasized that the goal of this paper is not to
provide an account of all relevant aspects and determinants of women’s employment,
but to provide methodological guidelines for future research on the effect of long durations
of childcare leave.
2. Theory and hypothesis
In this section, we introduce the first two improvements in determining the outcomes of long
durations of childcare leave on women’s employment, based on theoretical considerations.
3
Long childcare
leave
Downloaded by Stockholm University At 05:29 07 March 2017 (PT)

2.1 The relationship between duration of leave and employment of women may be curvilinear
Above, it was already introduced that the effect of childcare leave on women’s employment
is based on the opportunities leave provides for women to combine motherhood
and employment. Regarding childcare leave it was often hypothesized that longer
durations, by providing more opportunities, are associated with more employment among
mothers and therefore with a smaller gap in employment between mothers and women
without children (the motherhood-employment gap). Pettit and Hook (2009), on the other
hand, argued that leave is a mechanism of exclusion for mothers on the labour market.
Based on this argument they hypothesized that longer durations of leave decrease the
employment of mothers.
It is also possible that there is a positive effect on mothers’employment of short-term
leave, and a negative effect on mothers’employment of long-term leave (Akgunduz and
Plantenga, 2013; Del Boca et al., 2009; Jaumotte, 2003; Thévenon and Solaz, 2014). If that is
the case, hypotheses on only the positive effect of leave on women’s employment disregard
possible negative effects, and hypotheses on only the negative effect of leave disregard
possible positive effects. Thus, hypotheses should explicitly differentiate between the effects
of short-term and long-term childcare leave.
2.2 Childcare leave is expected to affect only mothers, not women without children
It is typically hypothesized that childcare leave policies only provide opportunities to mothers,
not to women without children. This differentiates leave policies from institutional factors
such as a large service sector size and low unemployment that were found to stimulate
the employment of both mothers and women without children (Nieuwenhuis, Need and
Van der Kolk, 2012). This seems logical, and it is, but in their empirical strategy not all studies
on the effect of childcare leave explicitly differentiate between the effect of leave on mothers
and on women without children. This was the case in studies that are based on country-level
data on women’s employment using the aggregated female labour force participation rate of a
country as the dependent variable, but also in several studies using person-level data.
A recent OECD (2011) report using country-level data has shown how the employment
of mothers with children below age three was lower in countries with long leave, compared
to countries with short leave. This finding supports the idea that long periods of leave
detach mothers from the labour market, but it would have been more informative to
hypothesize that long periods of leave increase the difference in employment between
mothers and women without children. Akgunduz and Plantenga (2013) found a curvilinear
effect of leave duration on women’s employment using country-level data. Even though in
this study a distinction was made between the effects of short-term and long-term leave,
their analysis of country-level data could not differentiate between the employment of
women with children and women without children. Jaumotte (2003) presented analyses on
both country-level data and on person-level data. The analysis on country-level data
showed a weakly positive (correlation of 0.05) linear association between the duration of
paid leave and the (country-level) female labour force participation rate in a country.
In multivariate regression models based on person-level data (Jaumotte 2003), the
curvilinear effect of leave was explicitly modelled and it was found that short periods
of leave increase the total female labour force participation, and that very long periods of
leave indeed were associated with lower levels of participation. In neither of these
analyses a distinction was made between the effect of leave on mothers and on women
without children. Similarly (Del Boca et al., 2009) found a curvilinear effect of the duration
of leave on the likelihood of a woman’s employment, while controlling for the presence of
children in the household. However, by merely applying a statistical control for the
presence of children in the household, their analysis did not explicitly test whether the
duration of leave only affected the employment of mothers.
4
IJSSP
37,1/2
Downloaded by Stockholm University At 05:29 07 March 2017 (PT)

Based on the arguments that the relationship between duration of leave and employment
of women may be curvilinear and that childcare leave is expected to affect only mothers, not
women without children, the long-leave hypothesis is formulated as follows: in countries
with short periods of childcare leave the motherhood-employment gap is smaller than in
countries with no childcare leave, but in countries with long periods of childcare leave the
motherhood-employment gap is larger than in countries with short periods of leave.
3. Method
In this section, we introduce the third and fourth improvements in determining the outcomes
of long durations of childcare leave on women’s employment, based on methodological
considerations. We introduce our data and statistical method used for the empirical part of
this study.
3.1 Testing the long-leave hypothesis requires the use of country-comparative data in which
countries are observed repeatedly over time
To test our long-leave hypothesis, both the curvilinear effect of leave and the interaction
between this curvilinear effect of leave and motherhood should be accounted for
simultaneously, which increases the complexity of the estimated models. Despite the
complexity of estimating a curvilinear effect of leave simultaneously with an interaction
between leave and motherhood, existing studies of the curvilinear effect of childcare leave
on mothers’employment are based on a limited number of country-level observations. This
is a natural data-limitation common to many country-comparative studies (Van der Meer
et al., 2010). However, the number of available degrees of freedom at the country-level is
small which increases the risk of over-fitting the model (Harrell, 2001). Over-fitting a model
increases the likelihood of influential cases and overly large standard errors. Cases are
regarded as too influential in a regression analysis, if the deletion or addition of a single case
to the data substantially alters the conclusion. This would indicate that the results of the
analysis do not represent a pattern present in all cases in the data, but that the results of the
analysis is solely a result of the inclusion of removal of one specific, single case.
One example is given of how limited the number of countries available for analyses
of the effects of long childcare leave typically are. Budig et al. (2012) presented a
country-comparative study on the curvilinear effect of leave on an aspect of women’s
employment: the motherhood-wage penalty. The analyses were based on 22-countries,
but Budig et al. (2012) specified a complex model including three-way cross-level
interactions (between motherhood, leave and leave squared, and a cultural variable),
which accounted for ten degrees of freedom, in addition to four country-level controls.
With so many country-level parameters to estimate compared to the number of
country-level observations, the model is easily over-specified which increases the risk
of influential cases and overly large standard errors.
3.2 The long-leave hypothesis is best tested against person-level data
It follows from the second suggested improvement, that is to differentiate the effect of
childcare leave between mothers and women without children, that the long-leave
hypothesis is best tested against person-level data. An additional benefit of such data is that
the country-level effects can be controlled for person-level determinants of women’s
employment (other than motherhood).
We test the long-leave hypothesis using person-level observations obtained from the
“Comparative Motherhood-Employment Gap Trend File”, a data set we created by
combining the Mannheim Eurobarometer Trend File (Schmitt and Scholz, 2005), the General
Social Survey (Smith et al., 2010) and the Canadian Election Study (see http://ces-eec.org/).
5
Long childcare
leave
Downloaded by Stockholm University At 05:29 07 March 2017 (PT)

All surveys provided samples that, by using sampling weights, were representative of the
respective country populations. The wording of the questions in the three surveys provided
comparable measures for several demographic background characteristics of individual
women. The Comparative Motherhood-Employment Gap Trend File data set is based on
192,484 person-level observations of women aged 20 to 50, covering 305 country-years and
18 OECD countries. It should be noted that these data are relatively old, but that for
the purpose of our study they provide the unique opportunity of having a total of 305
country-level observations, covering each country repeatedly and annually. The number of
observations and country-years per country, as well as the time-span covered for each
country, are presented in Table I. Even though the time-span covered differs substantially
by country, the statistical techniques applied in this study (detailed below) account for this
to prevent biased results.
Five person-level variables were used:
(1) Employment: a binary dependent variable representing whether or not a woman was
(coded 1) or was not (coded 0) involved in paid employment at the time of the survey.
(2) Motherhood: a binary variable representing whether or not a woman was a mother
(coded 1) or did not have children (coded 0).
(3) Education: an interval level variable representing the number of years of education.
This variable controls for differences in human capital.
(4) Partner: a binary variable differentiating between single women and women living
in a coupled household. This variable controls for the possibility of distributing
tasks within the household.
(5) Cohort: an interval level variable, representing the woman’s year of birth. This is a
control for possible differences in socialization and to account for data sets being
observed in different years.
Country
Earliest observed
year
Latest observed
year
No. of observed
country-years
No. of individual
observations
Austria 1995 1998 4 3,309
Belgium 1975 1997 23 14,729
Canada 1984 1998 4 3,699
Denmark 1975 1998 24 14,872
France 1975 1998 23 16,785
Germany East 1990 1998 9 6,209
Germany West 1975 1998 24 15,060
Greece 1980 1997 18 10,966
Ireland 1975 1998 24 15,183
Italy 1975 1997 23 14,453
Luxembourg 1975 1996 22 4,843
The Netherlands 1975 1998 24 18,058
Norway 1990 1996 6 1,801
Portugal 1985 1998 14 8,514
Spain 1985 1999 15 8,961
Sweden 1995 1998 4 2,351
UK 1975 1999 25 21,733
USA 1975 1998 20 10,958
Total 1975 1999 305 192,484
Note: n¼192,484 individuals from 305 country-years from 18-countries
Table I.
Countries and
country-years:
descriptive statistics
6
IJSSP
37,1/2
Downloaded by Stockholm University At 05:29 07 March 2017 (PT)

The size of the motherhood-employment gap was defined as the effect of motherhood on
employment, and was indicated by estimating the log of the odds ratio.
Country-level data. We combined the person-level data with one country-level variable on
duration of leave. As we are interested here in the total effect of leave on the size of
the motherhood-employment gap, in most of our models we do not control for additional
(country-level) variables. However, in the last model, we control for two variables
representing labour market structures. All our country-level variables were measured
annually, and are therefore time-varying. Our country-level variables are:
Childcare leave: this country-level independent variable indicates the duration of
childcare leave (in weeks) in a country. For easier interpretation, the duration of childcare
leave in weeks was divided by 10 and centred in the full data set on 0. This variable was
measured at the country-level and was obtained from the Comparative Maternity, Parental,
and Childcare Leave and Benefits Database (Gauthier and Bortnik, 2001).
Service sector size: an interval variable representing the size of the service sector,
as a proportion of the total labour force (divided by 10). This variable controls for over time
opportunities for women’s employment created by the growth of the service sector over.
Unemployment: a variable controlling for fluctuating unemployment rates, measured as
the percentage of the labour force that was unemployed (divided by 10).
Descriptive statistics are presented in Table II.
Statistical method. The data were analyzed using multilevel logistic regression, in which
employment was regressed on motherhood, childcare leave duration, and the cross-level
interaction between leave and motherhood. A curvilinear effect of childcare leave duration
was estimated. Person-level observations nested in country-years and countries.
The regression models were estimated using the lme4 package (Bates et al., 2015) in R
(R Core Team, 2016). Above, we argued that testing the long-leave hypothesis requires a
large number of observations. This also reduces the risk of having single influential cases
biasing the conclusions drawn from the regression models. With a small number of
country-level observations it is generally advised to test for influential cases at the country-
level (Van der Meer et al., 2010). The importance of evaluating for the presence of influential
cases is further emphasized by the complexity of the models discussed above, as the “most
common reason (for a case being influential) is having too few observations for the
complexity of the model being fitted”(Harrell, 2001, p. 74). The fact that many multilevel
models are based on large numbers of person-level observations does not help here, as the
accuracy of country-level estimates and cross-level interactions in multilevel modelling are
improved mainly by a large number of countries rather than by a large number of
person-level observations (Hox, 2010).
MSD Min Max
Employment status 0.54 0 1
Mother 0.61 0 1
Education 7.18 2.61 0 15
Partner 0.75 0 1
Cohort 28.25 8.48 0 54
Childcare leave duration 0.00 5.55 −4.44 11.2
Unemployment 0.00 0.37 −0.83 1.53
Service sector 0.00 0.80 −2.16 1.40
No. of countries 18
No. of country-years 305
No. of obs 192,484
Table II.
Descriptive statistics
of motherhood,
employment, and
childcare leave
7
Long childcare
leave
Downloaded by Stockholm University At 05:29 07 March 2017 (PT)

In our analyses we have a large number of country-year level observations, and evaluate
whether this large number of country-year level observations indeed resulted in stable
estimates. We apply software for detecting influential cases in multilevel models, for
which statistical tools have become available (Nieuwenhuis, Te Grotenhuis and Pelzer, 2012;
Van der Meer et al., 2010).
4. Results
To illustrate the possible presence of influential cases, and to demonstrate the sensitivity of
particularly cross-sectional analyses to such influential cases, we first present a graphical
analysis of a cross-sectional subset of our data. In Figure 1 we plotted for 1997 (the year for
which our data contained information on the largest number of countries) the bivariate
relationship between the number of weeks of childcare leave per country and the degree to
which mothers were less likely to be employed than women without children:
the motherhood-employment gap. This gap is represented by the log of the odds ratio
between motherhood and employment, that was calculated separately for each country.
The continuous, curved line represents the LOcal RegrESSion estimate of this association
(Fox, 2002). This line clearly is curved, with durations of childcare leave ranging from
0 to approximately 50-weeks being associated with smaller motherhood-employment gaps,
but with further increases in this duration (up to 156 weeks in countries such as Germany,
Spain and France) being associated with mothers being increasingly less likely to be
employed than women without children.
Closer examination of the bivariate association in Figure 1 shows that several of the
observed countries are outliers to the estimated curve, and therefore may constitute
influential cases. Deletion of a single country thus may result in a differently shaped curve,
and thereby the observed association between the duration of leave and the size of the
motherhood-employment gap. For instance, in 1997 The Netherlands and Austria show a
relatively large motherhood-employment gap relative to other countries at the respective
durations of childcare leave, such as Belgium and Portugal. Particularly Austria may have
Motherhood-Employment Gap
–1.0
–1.5
0 50 100
Weeks Childcare Leave
150
0.0
–0.5 Greece
Denmark
Sweden
Portugal
Germany East
Belgium
Italy
Canada
UK
Ireland
The Netherlands
Austria
Germany West
Spain
France
Notes: Continuous line is based on all observations; dashed line based on observations
excluding Austria
Figure 1.
The curvilinear
association between
childcare leave and
the motherhood-
employment
gap in 1997
8
IJSSP
37,1/2
Downloaded by Stockholm University At 05:29 07 March 2017 (PT)

been an influential case, because the motherhood-employment gap was very large in this
country (indicated by the strongly negative logit). The evaluate the degree to which Austria
has affected the shape of the curve, the curve drawn with the dashed line represents the
association between leave duration and the motherhood-employment gap after Austria was
deleted from the data.
The graphical analysis shown in Figure 1 has two disadvantages for detecting influential
cases. First of all, it only provides an informal test of the influence of Austria on the (shape
of the) association between duration of leave and the motherhood-employment gap. Second,
the selection of Austria was based on this country being an outlier. As influential cases are
not necessarily outliers (Crawley, 2007), the procedure should be performed for all countries
in the data.
Hence, we next estimate a regression model, on the full data. The results are presented in
Table III. We use this regression model to test our long-leave hypothesis, and then evaluate
this model for the presence of influential cases.
In Model Null of Table III, women’s likelihood of being employed was regressed on only
the effect of motherhood, including a random effect of motherhood that represents the
amount to which the association between motherhood and employment varies across
societies. The results show that motherhood is negatively associated with employment with
an estimated effect (logit) of −0.89, and that the estimated variation of this effect is 0.31.
This model was estimated as a baseline, to compare the following models with. In Model I,
the interaction between motherhood and the linear effect of the duration of childcare leave.
It shows that mothers are less likely to be employed than women without children.
The interaction term, estimated at 0.02 shows that this negative effect of motherhood is
reduced in societies with longer durations of childcare leave. The remaining variation of the
effect of motherhood is now 0.30, which only marginally lower than in the baseline model
(Model Null). This means that the linear interaction effect of childcare leave explains little of
how motherhood differently affects employment across societies. In Model II, the
curve-linear interaction between childcare leave duration and motherhood is estimated.
The results now show that in societies with very long periods of leave (indicated by the
squared term of childcare leave, estimated at −0.01), the motherhood-employment gap is
increased. Now, the cross-national variation of the estimated impact of motherhood on
employment is 0.26, which is a 16 per cent reduction from the baseline model. This means
that the curvilinear effect of childcare leave explains cross-national variation in the effect of
motherhood substantially better than the linear effect did. The results, so far, are in line with
the long-leave hypothesis. However, before drawing further conclusions, we evaluated the
extent to which the conclusions drawn based on the longitudinal data was overly influenced
by a single observation at the institutional level, in this case the country-year level.
Specifically, calculated Cook’s distance for estimates of all fixed parameters, to detect the
level of influence each country-year has had on these parameter estimates. In Figure 2 the
25 country-years with the largest value of Cook’s distance are shown (sorted in descending
order). The Cook’s distance of one country-year clearly stands out: East Germany in 1990.
This is thus a different country than in our graphical analysis (Figure 1), which is a result of
using repeated observations of each country in our regression analysis.
In Model III of Table III the results were presented of the same regression model as Model II
of the same table, after the observations of East Germany in 1990 were deleted from the data.
All parameter estimates in Model III of Table III are very similar to those originally reported
in Model II. The conclusions regarding our hypothesis therefore remain the same, despite the
deletion of a single country-year that was much more influential than all others. It should be
noted that the variance components (although mostly the same) and the −2×loglikelihood
cannot be compared across Model III and earlier models, as the number of observations and the
number of groups differs across these models.
9
Long childcare
leave
Downloaded by Stockholm University At 05:29 07 March 2017 (PT)

Model null Model I Model II Model III Model IV
BSE BSE BSE BSE BSE
Fixed effects
Intercept 0.84*** 0.14 0.79*** 0.80*** 0.16 0.77*** 0.16 −0.46*** 0.11
Mother −0.89*** 0.03 −0.89*** 0.03 −0.63*** 0.05 −0.63*** 0.05 −0.48*** 0.05
Education 0.17*** 0.00
Partner −0.61*** 0.01
Cohort 0.01*** 0.00
Unemployment −0.39*** 0.06
Service sector 0.26*** 0.03
Childcare leave 0.06*** 0.01 0.06*** 0.01 0.06*** 0.01 −0.01 0.01
Childcare leave
2
−0.00 0.00 0.00 0.00 0.00*** 0.00
Childcare leave ×mother 0.02*** 0.01 0.07*** 0.01 0.07*** 0.01 0.06*** 0.01
Childcare leave
2
×mother −0.01*** 0.00 −0.01*** 0.00 −0.01*** 0.00
Random effects (SD)
Intercept (country) 0.35 0.44 0.41 0.43 0.17
Intercept (country-year) 0.16 0.09 0.09 0.09 0.09
Mother (country-year) 0.31 0.30 0.26 0.26 0.24
No. of countries 18 18 18 18 18
No. of country-years 305 305 305 304 304
No. of obs. 192,484 192,484 192,484 191,863 191,863
−2×loglikelihood 234,998 234,876 234,827 234,348 224,176
Notes: *po0.05; **po0.01; ***po0.001 (two-tailed)
Table III.
Women’s employment
regressed on the
interaction between
motherhood and the
curvilinear effect of
childcare leave,
1975-1999
10
IJSSP
37,1/2
Downloaded by Stockholm University At 05:29 07 March 2017 (PT)

Finally, in Model IV, we add various controls for demographic and institutional
determinants of women’s employment. The results are in line with findings commonly
reported in the literature: women from more recent cohorts and higher educated women are
more likely to be employed, and women with a partner are less likely to be employed
compared to single women. Higher unemployment rates are naturally associated with lower
employment among women, and women’s employment is found to be higher in societies
with a larger services sector.
As a final sensitivity check, we also estimated a model in which the random intercepts
in countries were replaced by country-level fixed effects (not presented). These fixed effects
are a common econometric technique to account for all unobserved variables at the
country-level, to the extent that they do not vary over time. The results of this model were
substantively identical to those presented in Model IV, suggesting that our results are not
biased by unobserved heterogeneity.
Further calculations (not shown) indicated that the top of the curve that is formed by the
linear and curvilinear effects of childcare leave, lies at about 100 weeks of childcare leave.
This finding holds across the models in Table III, and is very similar to Thévenon and
Solaz (2014), who report that leave is associated with higher rates of women’s employment
“as long as the total period of paid leave does not exceed two years”(p. 2)
5. Discussion
This paper addressed a juxtaposition in the literature on outcomes of childcare leave, with
some studies finding that longer periods of childcare leave improves women’s employment
1990 Germany East
1976 France
1995 Germany West
1975 The Netherlands
1990 Germany West
1994 Denmark
1991 Germany East
1996 Germany West
1985 Belgium
1991 Germany West
1981 France
1998 The Netherlands
1989 Germany West
1995 France
1998 Austria
1996 Austria
1984 Ireland
1995 Germany East
1992 Germany East
1993 The Netherlands
1992 Denmark
1978 France
1978 Ireland
1998 Germany West
1997 Germany West
0.00 0.05 0.10
Cook’s Distance
0.15
Note: Figure presents Cook’s distances of the 25 country-years with the
largest influence on the regression parameters
Source: Based on Model I in Table III
Figure 2.
Influential cases
analysis testing the
long-leave hypothesis
11
Long childcare
leave
Downloaded by Stockholm University At 05:29 07 March 2017 (PT)

(Akgunduz and Plantenga, 2013; Gornick et al., 1999; Nieuwenhuis, Need and Van der Kolk,
2012; Van der Lippe and Van Dijk, 2002), while others regard leave as a mechanism of
exclusion of women from the labour market (Pettit and Hook, 2009) and find that long periods
of leave exclude women from the labour market (Gorlich and De Grip, 2008; Gornick and
Meyers, 2003; OECD, 2001, 2011; Waldfogel, 2001). The main contribution of this paper is
methodological in nature: we recommend and demonstrate four improvements in estimating
the effects of long childcare leave on women’s employment in comparative research.
5.1 The relationship between duration of leave and employment of women is curvilinear
While this improvement has been suggested various times before in the literature (Gorlich
and De Grip, 2008; Gornick and Meyers, 2003; OECD, 2001, 2011; Waldfogel, 2001), it was
not always applied in theorising or analysis (e.g. Nieuwenhuis, Need and Van der Kolk, 2012;
Pettit and Hook, 2009). Our analyses showed that, indeed, a curvilinear effect of the duration
of leave is present, and that applying a curvilinear effect provided a substantially better
explanation of cross-national variation in the size of the motherhood-employment gap than a
linear effect of leave.
5.2 Childcare leave is expected to affect only mothers, not women without children
While seemingly logical, this distinction often was not made. Indeed, we find that childcare
leave most strongly facilitates the employment of mothers. Nevertheless, we also found that
it was positively associated with the employment of women without children. This could be
an anticipation effect, or represent that higher levels of women’s employment are associated
with the implementation of generous leave policies.
5.3 Testing the long-leave hypothesis requires the use of country-comparative data in which
countries are observed repeatedly over time
Country-comparative research is often faced with the limitation of a limited number of
countries available for analysis. A key strategy to overcome that challenge is to observe the
same countries repeatedly over time. Our analyses showed that using a cross-section of just
15-countries is highly susceptible to the presence of influential cases. When analyzing a set
of 305 combined surveys, observing 18-countries repeatedly, we demonstrated that this
model was not sensitive to the presence of influential cases. The only influential case did not
substantively affect our comparative results. The finding that East Germany in 1990 was an
influential case, is not difficult to interpret post hoc: Rosenfeld et al. (2004) have documented
how the regime change in East Germany starting in 1989 affected women’s employment.
Prior to reunification, East Germany had high rates of women’s employment compared to
other countries, particularly with respect to mothers. After the reunification, the duration of
(childcare) leave was on par with that of Western Germany. The degree to which women
combined motherhood and employment declined in East Germany, but not as quickly the
change in leave policy. This explains why in 1990, we observed very long leave and a high
degree to which women combined motherhood and employment.
5.4 The long-leave hypothesis is best tested against person-level data
The use of person-level data is required for differentiating the effect of childcare leave
between mothers and women without children, as mentioned above. It further facilitates
controlling for other determinants of women’s employment. As discussed in our
introduction, such determinants are at the level of the individual in addition to at the
institutional level. We found that women are more likely to be employed when higher
educated, from a more recent cohort and when they are single, as well as when living in a
society with low unemployment and a large service sector. More importantly, for the
12
IJSSP
37,1/2
Downloaded by Stockholm University At 05:29 07 March 2017 (PT)

purpose of this study, we found that after controlling for these determinants of women’s
employment, the duration of childcare leave still had a curvilinear association with the size
of the motherhood-employment gap.
We calculated that the motherhood-employment gap was smallest with about 100-weeks
of childcare leave. While this number is in line with findings reported elsewhere (Thévenon
and Solaz, 2014), we are careful not to draw too strong policy conclusions based on that
finding. This was an estimate based on variation of duration of childcare leave in a total of
18-countries. The exact impact on individual women’s employment later in life, will depend
on a number of other factors not accounted for here. We only included supply-side
explanations of women’s employment, and did not account for possible demand-side
explanations such as the role of employers. We also did not observe whether individual
women actually used the full period of childcare leave they are entitled to, and whether they
will used it all at once or spread it out over a longer period of time.
Nevertheless, our findings do suggest that longer periods of leave can be detrimental to
maternal employment. While short periods of leave can be useful, or even necessary, to
maintain women’s attachment to the labour market after becoming a mother, very long
interruptions of employment indeed seem to be a “mechanism of exclusion”(Pettit and
Hook, 2009). Of findings also suggest that leave should not be abandoned all together,
but that its duration should be limited.
The finding that combining work and family ’sequentially’for a longer period of time,
rather than “simultaneously”(Pettit and Hook, 2009) is harmful for longer term employment
outcomes of women, points towards two relevant alternatives. The first is to stimulate the
availability of affordable and high-quality childcare (Andringa et al., 2015; Van Lancker,
2014). The second is to stimulate availability and uptake of paternity leave. While uptake of
leave among fathers is still substantially lower than among mothers (Eydal et al., 2015),
it forms a way of preventing long interruptions of employment, with both parents taking up
part of the leave. Stimulating the uptake of paternity leave could, in that way, reduce the
negative impact of long durations of leave on maternal employment.
References
Akgunduz, Y.E. and Plantenga, J. (2013), “Labour market effects of parental leave in Europe”,
Cambridge Journal of Economics, Vol. 37 No. 4, pp. 845-862.
Andringa, W., Nieuwenhuis, R. and Van Gerven, M. (2015), “Women’s working hours: the interplay
between gender role attitudes, motherhood, and public childcare support in 23 European
countries”,International Journal of Sociology and Social Policy, Vol. 35 Nos 9/10, pp. 582-599.
Bates, D., Maechler, M., Bolker, B. and Walker, S. (2015), “Fitting linear mixed-effects models using
lme4”,Journal of Statistical Software, Vol. 67 No. 1 pp. 1-48, doi: 10.18637/jss.v067.i01.
Bradley, K. (2000), “The incorporation of women into higher education: paradoxical outcomes?”,
Sociology of Education, Vol. 73 No. 1, pp. 1-18.
Crawley, M.J. (2000), The R Book, Wiley.
Bruning, G. and Plantenga, J. (1999), “Parental leave and equal opportunities: experiences in eight
European countries”,Journal of European Social Policy, Vol. 9 No. 3, pp. 195-209.
Budig, M.J., Misra, J. and Boeckmann, I. (2012), “The motherhood penalty in cross-national perspective:
the importance of work-family policies and cultural attitudes”,Social Politics: International
Studies in Gender, State & Society, Vol. 19 No. 2, pp. 163-193.
Del Boca, D., Pasqua, S. and Pronzato, C. (2009), “Motherhood and market work decisions in institutional
context: a European perspective”,Oxford Economic Papers, Vol. 61 No. S1, pp. i147-i171.
Engelhardt, H. and Prskawetz, A. (2004), “On the changing correlation between fertility and female
employment over space and time”,European Journal of Population/Revue Européenne de
Démographie, Vol. 20 No. 1, pp. 35-62, doi: 10.1023/B:EUJP.0000014543.95571.3b.
13
Long childcare
leave
Downloaded by Stockholm University At 05:29 07 March 2017 (PT)

Eydal, G.B., Gíslason, I., Rostgaard, T., Brandth, B., Duvander, A.-Z. and Johanna, L.-T. (2015), “Trends
in parental leave in the Nordic countries: has the forward march of gender equality halted?”,
Community, Work & Family, Vol. 18 No. 2, pp. 167-181, doi: 10.1080/13668803.2014.1002754.
Fagan, C. and Norman, H. (2012), “Trends and social divisions in maternal employment patterns
following maternity leave in the UK”,International Journal of Sociology and Social Policy, Vol. 32
Nos 9/10, pp. 544-560.
Fox, J. (2002), An R and S-Plus Companion to Applied Regression, Sage, Thousand Oaks, CA.
Gauthier, A.H. and Bortnik, A. (2001), “Comparative maternity, parental, and childcare leave and
benefits database”, Version 2, available at: www.demogr.mpg.de/cgi-bin/databases/fampoldb/
index.plx (accessed 10 February 2017).
Gorlich, D. and De Grip, A. (2008), “Human capital depreciation during hometime”,Oxford Economic
Papers, Vol. 61 No. S1, pp. i98-i121.
Gornick, J. (2015), “Leave policies in challenging times: what have we learned? What lies ahead?”,
Community, Work & Family, Vol. 18 No. 2, pp. 236-243, doi: 10.1080/13668803.2015.1023062.
Gornick, J.C. and Meyers, M.K. (2003), Families that Work. Policies for Reconciling Parenthood and
Employment, Russel Sage Foundation, New York, NY.
Gornick, J.C., Meyers, M.K. and Ross, K.E. (1999), “Public childcare, parental leave, and employment”,
in Sainsbury, D. (Ed.), Gender and Welfare State Regimes, Oxford University Press, Oxford,
pp. 117-145.
Harrell, F.E. Jr (2001), Regression Modeling Strategies. With Applications to Linear Models, Logistics
Regression, and Survival Analysis, Springer, New York, NY.
Hegewisch, A. and Gornick, J.C. (2011), “The impact of work-family policies on women’s employment:
a review of research from OECD countries”,Community, Work & Family, Vol. 14 No. 2,
pp. 119-138.
Hox, J.J. (2010), Multilevel Analysis. Techniques and Applications, 2nd ed., Routledge, New York, NY.
Jaumotte, F. (2003), “Labour force participation of women: empirical evidence on the role
of policy and other determinants in OECD countries”,OECD Economic Studies, Nos 37/2,
pp. 51-108, available at: https://search.oecd.org/eco/growth/34562935.pdf (accessed 10
February 2017).
Kögel, T. (2004), “Did the association between fertility and female employment within OECD countries
really change its sign?”,Journal of Population Economics, Vol. 17 No. 1, pp. 45-65.
Maldonado, L.C. and Nieuwenhuis, R. (2015), “Family policies and single parent poverty in 18 OECD
countries, 1978-2008”,Community, Work & Family, Vol. 18 No. 4, pp. 395-415.
Morgan, K.J. and Zippel, K. (2003), “Paid to care: the origins and effects of care leave policies in Western
Europe”,Social Politics: International Studies in Gender, State & Society, Vol. 10 No. 1, pp. 49-85.
Moss, P. and Deven, F. (1999), Parental Leave: Progress or Pitfall? NIDI CBGS Publications, Brussels.
Nieuwenhuis, R. (2014), “Family Policy outcomes: combining institutional and demographic
explanations of women’s employment and earnings inequality in OECD countries,
1975-2005”, University of Twente, Enschede.
Nieuwenhuis, R. (2015), “Association, aggregation, and paradoxes: on the positive correlation between
fertility and women’s employment”,Demographic Research, Vol. 32, pp. 1-5.
Nieuwenhuis, R., Need, A. and Van der Kolk, H. (2012), “Institutional and demographic explanations of
women’s employment in 18 OECD countries, 1975-1999”,Journal of Marriage and Family,
Vol. 74, No. 3, pp. 614-630.
Nieuwenhuis, R., Te Grotenhuis, M. and Pelzer, B. (2012), “Influence.ME: tools for detecting influential
data in mixed effects models”,R Journal, Vol. 4 No. 2, pp. 38-47.
OECD (2001), available at: www.oecd-ilibrary.org/social-issues-migration-health/oecd-employment-
outlook-2001_empl_outlook-2001-en;jsessionid=el7tx7sgpj60.x-oecd-live-03 (accessed 10
February 2017).
14
IJSSP
37,1/2
Downloaded by Stockholm University At 05:29 07 March 2017 (PT)

OECD (2011), Doing Better for Families, OECD Publishing, Paris.
Pettit, B. and Hook, J. (2005), “The structure of women’s employment in comparative perspective”,
Social Forces, Vol. 84 No. 2, pp. 779-801.
Pettit, B. and Hook, J.L. (2009), Gendered Tradeoffs. Family, Social Policy, and Economic Inequality in
Twenty-One Countries, Russel Sage Foundation, New York, NY.
RCoreTeam (2016),“R: a language and environment for statistical computing”,RFoundationfor
Statistical Computing, Vienna, available at: www.R-project.org/ (accessed 10 February 2017).
Rosenfeld, R.A., Trappe, H. and Gornick, J.C. (2004), “Gender and work in Germany: before and after
reunification”,Annual Review of Sociology, Vol. 30 No. 1, pp. 103-124.
Schmitt, H. and Scholz, E. (2005), “The Mannheim Eurobarometer trend file, 1970-2002”,
Inter-university Consortium for Political and Social Research, Ann Arbor, MI.
Smith, T., Van Marsden, P., Hout, M. and Kim, J. (2010), General Social Surveys, 1972-2010,
National Opinion Research Center, Chicago, IL.
Thévenon, O. and Luci, A. (2012), “Reconciling work, family and child outcomes: what implications for
family support policies?”,Population Research and Policy Review, Vol. 31 No. 6, pp. 855-882.
Thévenon, O. and Solaz, A. (2014), “Parental leave and labour market outcomes: lessons from 40 years
of policies in OECD countries”, available at: http://ideas.repec.org/p/idg/wpaper/199.html
(accessed 10 February 2017).
Van der Lippe, T. and Van Dijk, L. (2002), “Comparative research on women’s employment”,
Annual Review of Sociology, Vol. 28 No. 1, pp. 221-241.
Van der Meer, T., Te Grotenhuis, M. and Pelzer, B. (2010), “Influential cases in multilevel modeling:
a methodological comment”,American Sociological Review, Vol. 75 No. 1, pp. 173-178.
Van Lancker, W. (2014), “To whose benefit? An empirical and comparative investigation into the (un)
intended consequences of family policy in the social investment state”, Universiteit Antwerpen,
Antwerp.
Waldfogel, J. (2001), “International policies toward parental leave and child care”,The Future of
Children, Vol. 11 No. 1, pp. 99-111.
Corresponding author
Rense Nieuwenhuis can be contacted at: rense.nieuwenhuis@sofi.su.se
For instructions on how to order reprints of this article, please visit our website:
www.emeraldgrouppublishing.com/licensing/reprints.htm
Or contact us for further details: permissions@emeraldinsight.com
15
Long childcare
leave
Downloaded by Stockholm University At 05:29 07 March 2017 (PT)