With Strings Attached: Grandparent-Provided Child Care and Female Labor Market Outcomes∗ Eva Garc´ıa-Mor´ an† Zo¨ e Kuehn‡ December 2015 Abstract Grandparents are regular providers of free child care. Similar to other forms of child care, availability of grandparent-provided child care has positive effects on fertility and mothers’ employment. However, grandparent-provided child care requires residing close to parents or in-laws which may imply spatial restrictions for labor markets. We find that West German mothers who reside close to parents or in-laws have lower wages. We build a model of residence choice, fertility decisions, and female labor force participation that can account for the observed relationships. We simulate our model to analyze how women’s decisions would change if the availability of grandparent-provided child care or family policies were altered. If child care subsidies were raised to the Swedish level, fertility, mothers’ labor force participation, and geographical mobility would increase. Increased availability of grandparents on the other hand, while also increasing fertility and mothers’ labor force participation would reduce mobility.

JEL classification: J13, J61, H42, R23 Keywords: grandparent-provided child care, fertility, employment, spatial restrictions, regional labor markets

∗

Previous versions of this paper have been awarded UWIN Unicredit’s Prize for Best Paper in Gender Economics 2012 and the Etta Chiuri Prize 2012. We would like to thank all participants of the Workshop on Gender Equality at Bocconi University for their interesting comments and observations, especially Barbara Petrongolo and Alberto Alesina. We are grateful to Vincenzo Galasso for his helpful comments. We would also like to thank two anonymous referees for their very helpful comments. Zo¨e Kuehn gratefully acknowledges financial aid by FEDEA (Fundaci´on de Estudios de Econom´ıa Aplicada) in the context of the project “Evaluaci´ on de Pol´ıticas P´ ublicas (ECO2011-30323-C03-01) of the Spanish Ministry of Science and Research. † [email protected] · Universit¨at W¨ urzburg· Sanderring 2 · 97070 W¨ urzburg · Germany. ‡ [email protected] · Universidad Aut´ onoma de Madrid · Departamento de An´alisis Econ´omico: Teor´ıa Econ´ omica e Historia Econ´ omica · Campus de Cantoblanco · 28049 Madrid · Spain.

1

Introduction

The availability of child care and especially cheap or even costless child care has important effects on fertility and mothers’ labor force participation. While female labor force participation has increased tremendously over the past decades, mothers are still participating significantly less. Across OECD countries, the average difference in labor force participation rates between women and mothers (of children age 3 or younger) is around 10 (20) percentage points. For Italy, Del Boca [2002] shows that both the availability of child care and the possibility of part time work increase labor force participation and fertility. Blau and Robins [1989] establish a similar pattern for the US. Within the context of already high female labor force participation rates in Sweden, M¨orck et al [2013] is one of the few papers that focuses exclusively on the positive effect of lower child care costs on fertility. In a literature summary, Del Boca and Vuri [2007] point out that most studies find that high child care costs deter female labor supply, while availability of child care is found to have positive effects on mothers’ labor force participation. These findings suggest that the main barrier that mothers face at the time of working is to obtain affordable child care. Whereas the aforementioned studies analyze how availability of formal child care affects mothers’ labor force participation, the current paper focuses on the potentially different effects of grandparents as providers of child care. Grandparents are an important source of child care. According to data from the 2nd wave of the Survey of Health, Ageing and Retirement in Europe (SHARE), 16% (Denmark), 32% (Germany), and 48% (Italy) of grandparents take care of their grandchildren age six or younger on a daily or weekly basis (see Figure A-1 of the Appendix). In the US, 28.3% of children under 5 with employed mothers are regularly cared for by their grandparents (Overturf Johnson [2005]). According to the same source, these children spend on average 24 hours per week in grandparent-provided child care.1 In order to enjoy grandparentprovided child care on a regular basis, residence choices of adult children and elderly parents have to coincide. Figure 1.1 displays data from the 2nd wave of SHARE on the average fraction across several European countries, the fraction in Germany, and the crosscountry standard deviation of grandparents who provide daily care by distance between grandparents and their grandchildren age six or younger. The high standard deviation implies large cross-country differences, but on average grandparents who live closer than five kilometers from their grandchildren are much more likely to provide daily care. In 1

This is similar to the average 22 hours per week that children of employed mothers spend in nursery or preschool arrangements. In 46% of cases grandparent-provided child care is not the only child care arrangement, but given that average time in any child care arrangement is 28 hours (excluding time in parental care), on average child care by grandparents when provided amounts to over 85% of average child care time.

Figure 1.1: Grandparents providing daily care by distance to grandchild age 0-6 Mean Std. Deviation Mean Germany

Percentage (%)

40 30 20 10 0

< 5 km

5-25km

25-100km

> 100 km

Data: Survey of Health, Ageing and Retirement in Europe (SHARE), 2nd wave. Mean and standard deviation are calculated across all fourteen European countries included in the 2nd wave of SHARE, see Figure A-1 of the Appendix for the list of countries.

Germany, – similar to the average across countries – approximately 18% of grandparents who live close by provide daily child care. Within the next distance category – 5 to 25 kilometers – the share of daily care providers among grandparents drops to an average 6% (1% in Germany).2 Grandparent-provided child care – similar to other types of child care – can have positive effects on fertility and mothers’ labor force participation. However, different from other types of child care, it imposes spatial restrictions on labor markets which may affect female labor market outcomes. Looking at data for West Germany we find that mothers residing close to parents or in-laws have 4-5% lower hourly and monthly wages. Similar to other studies we also find that women residing close to parents or in-laws are more likely to have children, and that as mothers they are more likely to be employed. Decisions regarding residence, fertility, child care arrangements, and female employment, are highly interdependent, and hence our empirical analysis faces problems of endogeneity and reverse causality. While our findings thus cannot be interpreted as causal, the observed relationships provide the motivation for our theoretical model. 2

This pattern becomes less pronounced for child care at weekly or even lower frequencies, see Figures A2 and A-3 of the Appendix.

2

We build a model of married women’s residence choice, fertility and employment decisions. In our model there are two regions: a home region close to parents or in-laws and another region further away. Conditionally on their earnings capacity in the home region, women receive wage offers for the region further away– both for themselves as well as for their spouses. In our model, working mothers need child care. In the home region, close to parents or in-laws, there is a certain chance that free grandparent-provided child care is available. A woman decides where to reside, taking into account differences in expected child care costs and her potential wage-were she to work- and her spouse’s income in each region. In order to be able to enjoy free grandparent-provided child care, women might reject better offers in the region further away. This mechanism amplifies the relationship between lower wages and living close to parents or in-laws. In addition, given that women are more likely to have children and to work if child care is free, our model also replicates the positive correlation found in the data of fertility and employment with living close to parents or in-laws. In our model, the simultaneous interaction of fertility decisions and child care costs with two important determinants of family migration decisions – women’s wages and their spouses’ income– generates the observed relationships.3 While there are other reasons why individuals - including those without children - decide to stay close to parents or in-laws (care of the elderly, affection, etc.), we abstain from modeling these motives here. We focus on young women and mothers with young children (0-6 years) for whom fertility decisions and child care arrangements are particularly important. We then simulate the model to analyze how women’s decisions would change if the availability of grandparentprovided child care or family policies were altered. We find that if child care subsidies were raised to the Swedish level, fertility, mothers’ labor force participation, and geographical mobility would increase. Increased availability of grandparents – something which a recent proposal for shared parental leaves between adult children and their parents (“granny leaves”) in the United Kingdom tries to achieve – would also increase fertility and mothers’ labor force participation, but it would reduce mobility. To the best of our knowledge the current paper is the first one that explicitly incorporates spatial restrictions imposed by grandparent-provided child care into a model of fertility and labor force participation and that analyzes the effect of those restrictions on participation and wages. There is an important strand of literature analyzing fertility and labor force participation decisions in general equilibrium models; see for instance Attana3

Mincer [1978] finds spouses’ wage offers to be determinant for family migration decisions, but to a lesser extent if the woman is working and if her work is permanent and well paid. Hence, husbands’ and wives’ wages tend to jointly determine migration decisions.

3

sio et al [2010], Erosa et al [2010], Fern´andez and Wong [2014], Garc´ıa-Mor´an [2010], Greenwood et al [2000], or Guner and Knowles [2009]. A closely related paper is Cardia and Ng [2003] who – different from the current paper – explicitly incorporate grandparents’ decisions into a general equilibrium model of grandparent-provided child care. The authors suggest subsidizing grandparents’ time to be the most effective child care policy in terms of output and capital accumulation. However, the authors do not consider the spatial restrictions and potential costs in terms of labor market outcomes implied by grandparent-provided child care. Also closely related are the works by Bick [forthcoming] and Mendez [forthcoming]. Within a life cycle model, the former analyzes data for Germany and concludes that informal child care (by relatives) plays an important role given that mothers’ labor force participation exceeds child care enrollment for children up to 2 years. However, different from the current paper the author does not model relativeprovided child care, nor the spatial restrictions that it imposes. Mendez [forthcoming] provides a model of residence choice, fertility, and female labor force participation and attempts to account for differences in geographical mobility and female participation across European countries. Different from the current paper, the author does not address the effect on wages but focuses exclusively on the positive effect on female participation associated with living close to parents or in-laws. Recent work by Gemici [2010] provides a dynamic model of couple’s location decisions and finds that family ties hinder mobility and wage growth. Similarly, in our model, specific family ties related to grandparentprovided child care reduce mobility and lead to mothers accepting lower wages. Thus, our paper is also - to the best of our knowledge - the first one to document lower wages of mothers related to the geographical proximity between parents and adult children. The existing literature, on the contrary, has highlighted positive aspects associated with living close to parents or in-laws. Studying fertility intentions rather than outcomes, Raymo et al [2010] find that these intentions are higher for Italian and Japanese women who live close to their parents. Holdswoth and Dale [1998] estimate that the probability of being in employment is 1.24 times higher for Spanish women whose parents live in the same town. For the US, Compton and Pollak [2014] find that married women with small children living close to mothers and mothers-in-law have a 10 percentage point higher probability to be in employment. The studies by Dimova and Wolff [2011] and Zamarro [2011] use SHARE data and estimate simultaneous equation models of labor supply and grandparent-provided child care. Zamarro [2011] only finds a positive effect of grandparent-provided child care on mothers’ labor force participation for Greece and the Netherlands. For ten European countries, Dimova and Wolff [2011], who also take into account financial transfers between parents and adult children, find a positive effect of grandparent-provided child care on the extensive margin of female labor force participation but no effect along the intensive margin. Arpino et al [2014] and Posadas and 4

Vidal-Fern´andez [2013] for Italy and the US respectively, find that grandparent-provided child care – instrumented for by grandparents being alive – increases in particular labor force participation of low educated mothers with small children. Our paper is also related to the broader literature on intergenerational time transfers which tends to focus on time transfers from children to elderly parents. One interesting paper that also incorporates residence choices is Konrad et al [2002]. The authors develop a game theoretical model of strategic choice of residence among siblings who try to avoid having to take care of elderly parents. Looking at German data, they find support for their model’s predictions of older siblings locating further away from their parents than younger siblings. With a similar approach in mind, Stern [1995] estimates care choices of elderly parents together with location decisions of children. His work is closely related to the current paper as he also takes into account how the child’s location decision affects his or her work decision. Alesina and Giuliano [2010] argue that the extent of intergenerational time transfers within the family – care for children or the elderly – is strongly determined by the value of family ties in a society. The authors find that across countries a higher value of family ties is associated with lower geographical mobility, higher fertility, but also with more traditional gender roles and thus lower female labor force participation. Also in our model, increased availability of grandparent-provided child care within a country raises fertility and reduces mobility, but – on the contrary – leads to higher female participation. The remainder of this paper is organized as follows: the next section presents our empirical analysis. Section 3 presents the model and Section 4 describes our calibration strategy. In Section 5 we present the results of the model, and in Section 6 we discuss the model’s mechanisms in detail. In Section 7 we perform two counterfactual experiments, and Section 8 concludes.

2 2.1

Empirical Analysis Data

For our empirical analysis, we consider data from the German Socio-Economic Panel (SOEP), an annual household survey. The SOEP provides extensive information on individuals’ labor force participation, marital and family status, wages, education, etc. (for more details see SOEP [2005] or Wagner et al [2007]). It also includes variables that are of particular interest for our analysis: child care provided by relatives and geographical 5

distance to parents. In five waves (1991, 1996, 2001, 2006, and 2011) survey participants were asked to indicate where family members, including parents, siblings, and grandchildren live. Answers are categorized in the following way: i) in the same building, ii) the same neighborhood, iii) the same town, iv) another town but within one hour by car, v) further away, or vi) in a foreign country. We also have information if parents are part of the household. We construct a dummy variable “parents or in-laws live close” that takes on value one for individuals whose mother, father, or in-law lives in the same building, neighborhood or town. We also construct a dummy variable indicating if parents or in-laws live in the same household as their adult children. We construct similar distance dummies for siblings including brothers and sisters in-laws and for grandchildren. Finally, we also construct a dummy variable indicating if no parent or in-law is alive. Regarding relative-provided child care, we create a dummy variable “child care by relatives” that takes on value one for mothers with children up to the school-entry age of six who indicate that relatives take care of the child. To construct this variable we have to rely on slightly different questions for the first two and the last three waves. For 2001, 2006, and 2011 we consider the question “Are there additionally other persons outside of the household who regularly watch or take care of your children?” and we use the answer “Yes, relatives”. In waves 1991 and 1996 the survey asked “Who usually takes care of the children during the workday? (first, second, third person).” We use answers that refer to “ Relatives who are not members of the household (separated spouse, grandmother, grandfather, older siblings, other relatives) for any of the three persons mentioned. For these two earlier waves we also have information on which relatives provide child care and we observe that in 98.5% of cases these are the child’s grandparents. Hence, we feel fairly comfortable to use our variable “child care by relatives” as an effective measure of grandparent-provided child care. Given pronounced differences between East and West Germany regarding mothers’ labor force participation and child care provision, we focus in our analysis on women in and from West Germany.4 We exclude individuals born in a foreign country, because for them both key variables of our analysis, (i) availability of child care by grandparents and (ii) residence relative to parents, might be determined by very different aspects. 4

In 1990 (1991 for East Germany), there were 542 and 18 slots for every 1000 children under 3 in East and West Germany respectively (see Statistisches Bundesamt [2015]). Labor force participation rates of East German mothers of small children (0-3 years) have traditionally been very high and continue to be around 15 percentage points higher than rates for West German mothers (Bundesministerium f¨ ur Familie, Senioren, Frauen und Jugend [2005].)

6

Samples For our analysis, we pool the data from the five available waves. Given that we are interested in the relationships between labor market outcomes, child care, and residence choice, our main sample considers women age 20 to 49, potential mothers of children, age 0-6. The sample is restricted to individuals who provide information on the residence of parents or in-laws or who indicate that all are deceased. We define women on maternity leave and those working less than 20 hours per month as not working given that they have a reduced need for child care. We also exclude self employed from our sample and those defined as employed but who do not provide any information on wages or firm tenure. Our main sample consists of 10,909 observations: 7,061 for mothers and 3,484 for childless women. We also construct a smaller sample of 2,326 observations for mothers with children age 0-6 who provide information on child care arrangements and distance to parents or inlaws and distance to siblings or brothers- and sisters in-law. We use this sample to test whether geographical proximity is a good indirect measure of grandparent-provided child care. Finally, a third sample consists of 5,224 observations of West German grandmothers age 40 to 92 who provide information on distance to grandchildren as well as their own health status. We excluded wave 1991 from this sample because it did not include a comparable question for self-reported health status. We use this sample to check that geographical proximity between adult children and their parents is not mainly driven by grandparents’ bad health status and their need for care.

Descriptive Statistics Table A.1 of the Appendix provides summary statistics for our main sample. Childless women are on average 31 years old, while mothers are on average 39 years old. Around 80% of mothers and 25% of childless women are married. Approximately 28% of mothers have children age three or younger. Town sizes are grouped into small (up to 20,000 inhabitants), medium-sized (20,000-100,000 inhabitants), and large (more than 100,000 inhabitants). The majority of mothers lives in small towns (45%) compared to large or medium-sized towns, whereas childless women are almost equally distributed across the three types of towns. Childless women are somewhat more educated than mothers. Around 26% have tertiary education (higher vocational or university degree) compared to 20% of mothers. Approximately 90% of childless women are employed, while this is only the case for 58% of mothers. Real hourly wages of childless women and mothers are around 14A C (in 2010 A C).5 Whereas 25% of childless women live in the same household as their parents, this 5

We use data from the Statistische Bundesamt on the German consumer price index to adjust for

7

is only the case for 2.4% of mothers. Approximately 32% of childless women and 54% of mothers live close to parents or in-laws; i.e. they live in the same building, neighborhood, or town while 38% and 39% respectively live at least one hour away. Summary statistics for our two additional samples are displayed in Table A.2 of the Appendix. The first column refers to the sample of mothers with children age 0-6. Compared to mothers in our main sample, these women are somewhat younger, more likely to be married, more educated, and less likely to be employed. They are similar to mothers in our main sample in most other aspects. Regarding child care, around 1/3 uses relativeprovided child care, whereas 2/3 uses formal child care (nursery, babysitter, governess). Around 41% live close to at least one sibling or brother- or sister in-law, while 36% live close to both parents and siblings, including in-laws. Statistics for our sample of grandmothers are displayed in the second column of Table A.2. Around 47% of grandmothers live close to at least one grandchild. Most grandmothers report satisfactory health conditions (41%), followed by poor or very poor health conditions (32%), and good or very good health conditions (27%).

2.2

Proximity to Grandparents and Child Care

The availability of grandparents as child care providers is not only likely to influence mothers’ labor force participation, but at the same time its use might be determined by mothers’ decisions to work. This reverse causality introduces a potential bias into a direct measure of grandparent-provided child care when trying to estimate its effect on labor market outcomes. Using geographical proximity to grandparents as an indirect measure could solve this problem. In order to test how well proximity to parents or in-laws predicts use of grandparent-provided child care, we use our sample of mothers with children age 0-6, and we run a probit regression of relative-provided child care on individual controls including distance to parents or in-laws. Table A.3 of the Appendix displays marginal effects for married women from this regression. Results show that living close to parents or in-laws increases the likelihood that a married mother uses relative-provided child care by 68%. The second column of Table A.3 presents results for disaggregated distance categories. Our definition of living “close” seems to capture well the required proximity for grandparent-provided child care. Given the reference category “in same town”, we estimate positive and significant marginal effects for categories “in same building” and “in same neighborhood”. Marginal effects are negative for distance categories corresponding to“one hour away” or “further away.” In the third column we display results when also wage inflation.

8

controlling for proximity to siblings or brothers- or sisters-in-law and proximity to both, siblings and parents. If rivalry among siblings for grandparent-provided child care played an important role, the inclusion of these controls should affect our estimated marginal effects. Our results remain basically unchanged. However, there might also exist some concern that geographical proximity between adult children and elderly parents could be an indicator of parents’ bad health and their own need for care, rather than the possibility of free child care. We investigate this matter using our sample of grandmothers. We run a probit regression of self-reported health status on a variety of control variables and an indicator if a grandchild lives close by. Table A.4 of the Appendix displays average marginal effects from this regression. Results show that living close to grandchildren seems to be positively related to reporting a satisfactory health status but negatively to reporting a good or very good health status. Marginal effects for the proximity to grandchildren for those reporting a bad or poor health status are not significant. Hence, we do not find any evidence that proximity is driven by grandparents’ bad health status and their need for care. On the other hand, we established a positive relationship between proximity to parents or in-laws and relative-provided child care. We thus feel confident to use geographical proximity of mothers to parents or in-laws as an indirect measure of grandparent-provided child care. Furthermore, geographical proximity reflects more than just currently provided child care by grandparents. It might also reflect child care provided by “potential” grandparents, and thus geographical proximity proves particularly useful to test effects on fertility. However, as residence decisions might not be independent of women’s fertility choices and mothers’ labor force participation, a caveat remains.

2.3

Proximity to Grandparents, Fertility, and Employment

In line with findings in the literature discussed before, we find that living close to parents or in-laws is associated with higher fertility, and that mothers living close are more likely to be employed. Table 2.1 summarizes these empirical findings; for detailed results see Tables A.5 and A.6 of the Appendix. Married women living in the same building, neighborhood, or town as their parents or in-laws have a 22% higher probability to have children. This effect is even stronger (40.3%) for women with tertiary education. Regarding employment, married mothers living close to parents or in-laws are 13% more likely to be employed. Here, the effect is stronger for women with lower educational attainment. Compton and Pollak [2014] argue that geographical proximity is a good instrument for child care arrangements because its positive effect on labor force participation does not 9

extend to groups for which grandparent-provided child care is not a determinant for labor supply. Similarly, in our data the effect does not extend neither to single men nor to single childless women. Table 2.1: Marginal effects of proximity to parents or in-laws for fertility and employment decisions+ fertility Sample: women women with tertiary education women with less than tertiary education

employment

0.223*** (0.03) 0.403*** (0.07) 0.168*** (0.04)

mothers mothers with tertiary education mothers with less than tertiary education

0.136*** (0.04) 0.103 (0.08) 0.142*** (0.04)

single childless women single men

-0.033 (0.10) -0.016 (0.08)

Marginal effects are marked with * if the level of significance is between 5% and 10%, ** if the level of significance is between 1% and 5% and *** if the level of significance is less than 1%. + For women and mothers marginal effects are evaluated for married individuals. In the case of single childless women and single men we report average marginal effects.

2.4

Proximity to Grandparents and Wages

Controlling for selection effects into employment, we find that mothers living close to parents or in-laws earn lower hourly wages. The first column of Table 2.2 displays coefficients from a Heckman selection model for log hourly wages for mothers from our main sample. Living close to parents or in-laws is associated with 4.2% lower hourly wages.6 Other control variables show the expected signs. Wages increase with firm tenure and tertiary education. On the other hand, having more children, being married, having neither parents nor in-laws, and living in smaller towns is associated with lower hourly wages.7 6

Using log monthly wages, controlled for by hours worked, leads to slightly more negative coefficients for living close, as does not controlling for selection effects (see Table A.7 of of the Appendix). 7 Results are robust to the inclusion of a polynomial for age instead of age group dummies. We also check the robustness of our results to the exclusion of the variables marital status and spouse’s income (see Table A.7 of the Appendix). Due to the limited number of individuals in the panel who change their labor force participation over the observed period, individual fixed effect estimations do not provide any significant results.

10

Table 2.2: Heckman Selection Model for Mothers’ Log Hourly Wages log hourly wage

is employed

has tertiary education no parent or in-law alive in education/training number of children married nationality not German parents or in-laws close parents or in-laws in household in small town in large town has child 0-3 log (monthly wage of spouse) firm tenure constant

0.354*** -0.076** -0.003 -0.080*** -0.042** -0.044 -0.042*** -0.041 -0.044** 0.032

(0.017) (0.030) (0.049) (0.009) (0.017) (0.054) (0.015) (0.049) (0.018) (0.020)

0.017*** 2.361***

(0.001) (0.035)

observations

7,061

0.218*** 0.153* 0.284** -0.248*** -0.376*** -0.125 0.134*** -0.001 0.005 -0.017 -1.211*** 0.018***

(0.042) (0.079) (0.131) (0.019) (0.049) (0.110) (0.035) (0.109) (0.042) (0.047) (0.046) (0.005)

0.905***

(0.080)

7,061

Data: SOEP unbalanced panel 91,96, 01, 06, 11. The dependent variable in the selection equation is employment. Mothers on maternity leaves and those working less than 20 hours a month are defined as not employed. The coefficients are marked with * if the level of significance is between 5% and 10%, ** if the level of significance is between 1% and 5% and *** if the level of significance is less than 1%. All regressions include year, age group, and state dummies. Reference group: unmarried mothers of age 30-34, with children older than 3, without tertiary education, who live in a medium-sized town in North Rhine-Westphalia, at least one hour away from parents and in-laws, in 1991.

We also run the model separately for individuals with and those without tertiary education to check if results are driven by individuals with low education. On the contrary, the penalty in hourly wages for staying close to parents or in-laws turns out to be higher – 6% – for the group of tertiary educated individuals. However, results might also be driven by individuals with very low-paying jobs, independently of education. Using a sample that excludes those making less than the bottom 10% of wage earners shows that this is not the case. While, the wage penalty is somewhat smaller (3.3%) results remain significant (see Table A.8 of the Appendix.) The two exclusion restrictions in our Heckmann selection model are: (i) having a child age three or younger and (ii) spouse’s income. Both variables significantly affect mothers’ labor force participation (see column one of Table A.6 of the Appendix). However, none is directly related to mothers’ hourly wages. Given that we only consider mothers, differences in a child’s age are mostly explained for by the mother’s own age and her years of education. We control for both variables. The second exclusion restriction might be invalid if assortative matching leads to similar incomes of wife and spouse. As long as these similarities are mostly explained for by a common level of education or the size of the community they live in – both variables that we include as controls – the use of this 11

exclusion restriction is justified. Given strong interdependencies of decisions regarding residence, fertility, child care arrangements, and female labor force participation, our empirical analysis faces problems of endogeneity and reverse causality. By using geographical proximity between adult children and their parents as an indirect measure of grandparent-provided child care, we address the interdependency of child care arrangements and female labor force participation. However, we are not able to control for the selection effect due to choice of residence. Hence, certain caveats remain because we cannot dismiss a reverse causality between geographical proximity and labor market outcomes or fertility. Nevertheless, from the observed relationships between labor force participation, fertility, and wages with grandparent-provided child care a set of interesting questions arise: How valuable is grandparent-provided child care in terms of fertility and employment? How does this type of free location-dependent child care affect mobility and wages? How do family policies compare, regarding aggregate employment, mobility, and fertility? In order to answer these questions and to better disentangle women’s decisions, we build a model economy that incorporates the spatial restrictions imposed by grandparent-provided child care.

3

The Model

In our model economy there are two regions (r) where individuals can reside, ’Home’ denoted by H and ’Far’, denoted by F , r = H, F . The economy is populated by a continuum of married women of mass one.8 Women differ across two dimensions: (i) their own type (i) and (ii) their spouse’s type (j). Types determine wage rates for women and their spouses in the labor market in H, wiH and w˜j H respectively. Women in our economy live for two periods, each of three years, corresponding to the first six years of a child’s life. Essentially we want to capture mothers’ decisions during the time when child care is most important, i.e. during early childhood. We choose two periods because labor force participation by mothers experiences important changes when their children reach the pre-school age of three. Women have to decide whether to become mothers, how much to work, and where to live. Each period a woman is endowed with one unit of productive time. Spouses are assumed to always work one unit of time. 8

Around 87% of mothers of children age 0-6 in our SOEP sample are married (see Table A.2). Hence, we only model married women’s decisions. Even though marriage and residence decisions might be related, in order to keep the analysis tractable we abstain from modeling a marriage market, and we simply assign an exogenous income to each woman to represent her husband’s income.

12

At the beginning of the first period and conditional on her type i, women are matched with a spouse of type j. Women then draw two independent elements, w and m from a distribution D(µ , σ ) such that wage offers in F are given by: wiF = wiH w w˜j F = w˜j H m ,

with Cov(w , m ) = 0, and where w , m ∼ D(µ , σ ) indicate change in wages that individuals would experience if moving to F . Upon observing the offer and comparing it to wages in H, women have to decide where to reside. We assume that residence choices are only made once during a woman’s life time and cannot be reconsidered. In the first period, women also have to decide whether or not to have children, k = 1 or k = 0. Every period women also have to decide how much to work. Working mothers need child care. The price of child care per unit of time, pt depends on the age of the child, with t ∈ 1, 2. Mothers with small children – from age 0 to 3 – pay p1 . Mothers with older children – ages 3 to 6 – pay p2 . Living in H potentially provides access to free child care by grandparents. However, with a certain probability grandparents fall sick, die, or are otherwise unable or unwilling to take care of their grandchildren. In the first period, only a share g1 of women who live in H have access to free child care. The remaining (1 − g1 ) have to purchase child care at price p1 . In the second period, among those who obtained grandparent-provided child care in the first period only a share g2 continue to have to free child care. The remaining (1 − g1 g2 ) face child care costs equal to p2 . Women might receive a subsidy for child care ω from the government. In this case they pay (1 − ω)pt , for t ∈ 1, 2 . Women might also receive family benefits T conditional on having children. Women with children (k = 1) care about the quality of their children, e and about consumption, c. Hence, they enjoy the following utility U (c, e, k) =


c1−σ + σ e eα − σ k k, 1−σ

where σ k are fixed utility costs per child. Childless women (k = 0) only care about consumption and they thus work all their disposable time (l = 1). Mothers spend time working (l) or taking care of their children (tm ), with l + tm = 1. Time spent in child care is assumed to be equal to the time the mother is at work (l = tc ). The quality of 13

children, e is a weighted sum of the time a mother spends with her children, tm and the time her children spend in child care: e = φm tm + φc l, which is equivalent to: e = φm − l(φm − φc ). How a mother divides her time between working and taking care of her children, crucially depends on how decisive her time is for her children’s quality. This is captured by the two weights, φm for time spent with the mother and φc for time spent in child care or with grandparents. We assume that time in formal child care and time spent with grandparents are of equal importance for children’s quality.9 The government in this economy provides family benefits conditional on having children (T ), it might subsidize child care costs (ω), and it imposes an income tax τ .

3.1

Value functions

We solve our model backwards. Thus we first present the value functions for women in the second period. Value functions in the second period For childless women the value of residing in H is given by c1−σ 2 H H , H (wi , w˜j ) = 1−σ subject to the budget constraint c = (1 − τ )(wiH + w˜j H ), 9

Hansen and Hawkes [2009] find that for the first nine months formal child care is associated with higher school readiness scores, while grandparent-provided child care is associated with a higher vocabulary test score, both scores measured at the age of three. Findings in Bernal and Keane [2011] show that informal child care, including grandparent-provided child care, has negative effects on children’s test scores while center-based care does not. However, the authors only study single mothers which introduces a bias towards more disadvantaged backgrounds of mothers and towards informal care providers of lower quality. In another study, Bernal [2008] finds that non-maternal child care is detrimental for children’s test scores. Hence if grandparent-provided child care is similar to the type of care a mother provides, then grandparents might be the second best option. Moreover, children in informal care might receive more individual attention, see Clarke-Stewart et al. [1994] and grandparents tend to guarantee a stable providerchild relationship, something found to be determinant for the quality of child care (see Walker [1991]). Hence, while formal child care and grandparent-provided child might be very different it is not clear whether one of them is of higher quality than the other.

14

where τ denotes the income tax. Childless women only care about consumption, they work all their disposable time l = 1, and they consume all disposable income. When deciding how much to work, mothers on the other hand take into account that their children’s quality depends on how much time they spend taking care of them. Thus the value function for a mother living in H who has access to free grandparent-provided child care (g) during the second period is given by   1−σ c e α k 2 H H + (σ e − σ ) , Hg (wi , w˜j , k) = max l 1−σ subject to the budget constraint c = (1 − τ )(wiH l + w˜j H ) + T,

(3.1)

and given the children’s quality production function e = φm − l(φm − φc ). A mother in F has to purchase child care at price p2 per unit of time worked. Her value function is as follows:   1−σ c e α k 2 F F + (σ e − σ ) , F (wi , w˜j , k) = max l 1−σ subject to the budget constraint c = (1 − τ )(wi F l + w˜j F ) + T − (1 − ω)p2 l, and given the children’s quality production function. Mothers in H without access to grandparent-provided child care (ng) have a similar value function given by  1−σ  c H 2 H e α k Hng (wi , w˜j , k) = max + (σ e − σ ) , l 1−σ subject to the budget constraint c = (1 − τ )(wiH l + w˜j H ) + T − (1 − ω)p2 l, and given the children’s quality production function.

15

Value functions in the first period In the first period, a woman who resides in H has to decide whether to have children or not and how much to work. Only a share g1 of working women has access to free child care provided by grandparents. Their value function is given by   1−σ c e α k 2 2 1 H H + (σ e − σ )k + β(g2 Hg (; ) + (1 − g2 )Hng (; ) , Hg (wi , w˜j ) = max l,k 1−σ subject to the budget constraint c = (1 − τ )(wiH l + w˜j H ) + T k, and given the children’s quality production function. Given that residence choices are only made at the beginning of the first period, the continuation value for a woman living in H is equal to the discounted value of living in H in the second period. In the second period, with probability g2 , she will continue to have access to grandparent-provided child care. On the other hand, a share (1 − g2 ) of those who had access in the first period will have to pay for child care in the second period. Living in H but without access to free grandparent-provided child care (ng), a woman’s value function is as follows   1−σ c e α k 2 1 H H + (σ e − σ )k + β(Hng (; )) , Hng (wi , w˜j ) = max l,k 1−σ subject to the budget constraint c = (1 − τ )(wiH l + w˜j H ) + T k − (1 − ω)p1 lk and given the children’s quality production function. A woman who did not have access to free child care in the first period will neither have access in the second period. A woman residing in F has to decide whether or not to have children and how much to work, taking into account that if she has children and works, she has to purchase child care at price p1 . Her first period value function is given by F 1 (wiF , w˜j F ) = max( l,k

c1−σ + (σ e eα − σ k )k + βF 2 (wiF , w˜j F , k)) 1−σ

subject to the budget constraint c = (1 − τ )(wi F l + w˜j F ) + T k − (1 − ω)p1 lk and given the children’s quality production function.

16

Residence Choice When deciding where to reside, women do not know for sure if they will have access to grandparent-provided child care in H. They thus consider the expected value of living in H, given by 1∗ (wiH , w˜j H ), EH 1∗ = g1 Hg1∗ (wiH , w˜j H ) + (1 − g1 )Hng 1∗ (wiH , w˜j H ) denote the value functions evaluated at the opwhere Hg1∗ (wiH , w˜j H ) and Hng timal decisions for labor supply and number of children. Women decide where to reside by comparing the expected value of staying in H to the value of living in F . They will decide to stay if EH 1∗ > F 1∗ ; i.e. if the expected value of living in H is strictly higher than the value of living in F .10

Without considering the decision to have children or not, there are only two reasons why women would decide to reside in F : (i) a higher wage rate and/or (ii) a higher spouse’s income. However, once fertility decisions and differences in child care costs are introduced a woman’s residence choice depends on the wage offers, the relative importance of children’s quality for her utility, and the differences in child care costs across regions. A woman who receives a wage offer for her and her spouse in F equivalent to the wages both can earn in H will always decide to stay, because in H she may obtain free child care. But wage offers could include a higher wage rate in F , (wiF > wiH ) , a lower wage rate in F , (wiF < wiH ) or wages could be the same, (wiF = wiH ). Moreover, spouses’ incomes might also be different. In this case the residence choice becomes non-trivial. With the possibility of free grandparent-provided child care in H, a higher wage offer in F might not be enough to make up for the higher cost of child care. Furthermore, higher wages in F imply higher opportunity costs of staying at home, and thus there could exist wage offers such that in F the optimal decision for the mother could be to work while in H faced with lower wages and lower opportunity costs, the mother could optimally decide to stay at home, thus raising her children’s quality. Depending on the relative weight of children’s quality in utility, the situation in H with lower wages might provide for a higher utility. Once children and child care costs are taken into account, higher wages do not always determine residence choices because with different wages mothers’ optimal labor supply changes and so does their time with children. In the following subsection, we study a special case of our model where utility in children’s quality is linear. This allows us to derive an analytical solution for optimal labor supply, and we can discuss how changes in the price of child care and wages affect women’s employment decisions. 10

Individuals who are indifferent are distributed equally across the two regions.

17

3.2

Optimal Labor Supply

With children’s quality entering utility linearly (α = 1), we are able to obtain a closedform solution for the optimal labor supply. Childless women work all their disposable time, l∗ = 1. For mothers who do not have access to free grandparent-provided child care on the other hand, the optimal labor supply is given by ∗

l =



1 σ e (φm − φc )

 σ1

(wir (1 − τ ) − (1 − ω)pt )

1−σ σ

−

(1 − τ )w˜j r + T , (1 − τ )wir − (1 − ω)pt

subject to 0 ≤ l∗ ≤ 1, for t = 1, 2.and r = H, F. By setting pt = 0 for t = 1, 2 we obtain the optimal labor supply for mothers who have access to free grandparent-provided child care in both periods. Mothers’ optimal labor supply depends crucially on how important time spent with children is for their quality, compared to time spent in child care or with grandparents. We assume that time with mothers is at least as important as time in child care (φm ≥ φc ). Mothers’ labor supply also depends on the relationship between mothers’ wages and the cost of child care, as well as on the relative value of the spouse’s income compared to mothers’ potential income. Mothers whose marginal benefit from working – their wage rate wir – is lower than the marginal cost of working – the cost of child care, pt – will decide to stay at home. In case mothers’ time for children’s quality is of equal importance as time spent in child care, φm = φc , women whose wage rate is high enough to cover child care costs, work all their disposable time  0 for (1 − τ )wir < (1 − ω)pt and φm ≥ φc ∗ l = 1 for (1 − τ )wir > (1 − ω)pt and φm = φc . In case φm > φc , the optimal labor supply of women whose hourly wage is sufficient to cover child care costs will depend on the other parameters as well as on spouse’s income and family benefits. The effect of the other parameters on the optimal labor supply is as expected. A higher value of σ which affects the curvature of utility in consumption, increases labor supply. An increase in the importance of time spent in child care for children’s quality and an increase in women’s wages have similar effects. On the other hand, an increase in the weight of a mother’s time for children’s quality reduces labor supply. Similarly an increase in the cost of child care and an increase in spouse’s income decrease labor supply. In particular, mothers supply no labor at all if their spouses’ income is large enough,  r 1/σ wi (1−τ )−(1−ω)pt 1 ∗ r l = 0 if w˜j > 1−τ [ − T ]. σ e (φm −φc ) 18

4

Calibration Strategy

In order to be able to quantify the importance of grandparent-provided child care for women’s decisions on residence, labor supply, and fertility, we calibrate our model. A model period corresponds to three years and hence the discount factor β is set to 0.889 to match a yearly interest rate of 4%. Policy parameters are taken directly from data for Germany. All remaining parameters are calibrated to match model moments to several labor market and fertility statistics for West Germany. Unless otherwise stated, the statistics used for calibration come from a subsample of our previously described main sample of West German women age 20-49. In particular, we consider weighted statistics for married women and their husbands who either have children 0-6 or who have no children at all. We define women in H as those who live in the same household, building, neighborhood, or town as their parents or in-laws while women in F are those who live in another town but within one hour by car or further away. We first describe in detail how we construct our matching matrix that assigns spouses to women and how both are assigned wages in H. We then detail how wage offers for F are drawn. Finally, we discuss how we calibrate the parameters of the model, and how we set policy parameters to represent German family policies. In order to determine how a spouse of type j is allocated to a woman of type i, we approximate types by education or schooling levels s such that i, j ∈ s. Following the International Standard Classification of Education (ISCED 1997) the SOEP defines the following ISCED levels: (1) primary schooling, (2) lower secondary, (3) upper secondary or vocational, (4) upper secondary and vocational, (5) higher vocational, and (6) university. To pair up women and men we use a matching matrix obtained from our sample Φ(s, s), with Φ(i, j) being a particular element of this matrix and where i ∈ s and j ∈ s denote women’s and men’s education levels respectively (see Table A.9 of the Appendix). Following Guner et al [2012], we use average hourly wages for each education level from our sample to assign wage rates to spouses in H. To estimate women’s wages which are more likely to be affected by selection into employment we normalize the observed wage rates by education level by the average wage rate of women in our sample, µw . This way we obtain a vector of educational wage factors as a fraction of the average wage rate, denoted by γw . Our potential wages for women in H – before participation decisions are made – are then given by γw w, ¯ where we choose w¯ such as to match the observed average wage rate of women in our model to µw .11 Secondly, we specify the distribution D(µ , σ ) from which individuals draw wage offers 11

See Table A.10 for average hourly wages by education for women and spouses.

19

in F . Ideally, we would like to compare women’s observed wage changes when moving in our model to those in the data. However, to calculate wage changes in the data we have to observe individuals’ wages in the labor market before and after the move. For childless women and women with children age 0-6 we do not have sufficient observations that fulfill these criteria. This is why we follow the convention in the literature and base our calculations on wage data for men. In particular, we consider wage changes of married men in West Germany who moved to another federal state or another city. We calculate the ratio of their monthly wages after the move to wages before the move. To maximize the number of observations we consider a different sample that includes all years from 1991 to 2012. This empirical distribution of the ratio of wages after and before moving can be approximated by a log-normal distribution with mean, (¯) 1.08, median 1.03, and standard deviation, (sd ) 0.3. However, in our model D(µ , σ ) represents an underlying distribution which includes accepted and rejected offers, whereas in the data we only observe accepted offers of those who participate. In order to estimate this underlying distribution we thus choose its mean (µ ) and its standard deviation (σ ) in order to target the observed mean (¯) and standard deviation (sd ) of wage changes. In particular, to calculate wage changes in our model we consider wage offers in F that women have accepted by moving to F but which are independent of their participation decisions. Hence we choose parameters of the underlying distribution such as to map data on men’s realized wage changes to women’s potential wage changes in our model. This is somewhat different from what has been done in the literature. However, as mentioned before, given data limitations we are unable to construct the ideal mapping that would require data on women’s realized wage changes. Alternatively, matching the data on men’s wage changes to women’s realized wage changes in the model seems more problematic given that the latter are affected by a stronger selection bias. We could also compare the data to spouses’ wage changes in the model. But men make no decisions in our model and their wage changes are simply an outcome of women’s choices. In our model economy individuals differ along women’s potential wage rates and their spouses’ income in H and F . This distribution is denoted by Π[(w, w) ˜ H , (wH , w˜ H )F ], chosen to be consistent with the matching matrix Φ(s, s) and with the initial distribution of women over education levels, Ω(s) and the initial distribution of men over education levels, Θ(s). These last two distributions correspond to the fraction of individuals by education in our sample (see Table A.11 of the Appendix). Each element of the distribution Π[(wi , w˜j )H , (wiH w , w˜j H m, )F ] is determined as follows. The number of women who receive a particular offer ((w1 , w˜3 )H , (w1H 2 , w˜3 H 5 )F ) depends on four elements: (i) the number of women with wage rate w1H – given by Ω(1) – (ii) the number of men with wage 20

rate w˜3 H , – given by Θ(3) –, (iii) the probability of this match happening, Φ(1, 3) and (iv) the probability of receiving the offer w1H 2 and w˜3 H 5 . These last probabilities are equal to the probability of drawing 2 and 5 from the distribution D(µ , σ ), which correspond to the densities d(2) and d(5). Given that there is a mass one of women, the number of women receiving a particular offer is given by the probability of receiving the offer which is equal to the product of the probabilities discussed above, normalized by the mass of spouses of education type 3, Θ(3).12 We calibrate the parameters of the utility function, (σ, σe , σk , α), the cost of child care in both periods, p1 and p2 , the probabilities of having access to free child care, g1 and g2 , women’s average potential wage rate w, ¯ mean and standard deviation of the distribution of wages offers, µ and σ , and the parameters of the children’s quality production function, (φm , φc ). Note that we impose φm + φc = 1, and hence we only need to calibrate one of these last two parameters. We now relate the calibrated parameters to the data moments which they are most likely to affect. We set women’s underlying average wage rate w¯ to 13.50 in order to match working women’s observed mean hourly wage, µw . The curvature of consumption in the utility function, σ is set to 0.8, to match the percentage of working women in H. Given that on average these women face lower child care costs their participation decision is very much determined by the value of consumption. We use three moments related to fertility and participation to match the fixed utility cost of children, σk , the curvature of children’s quality in the utility function, α and the weight of children’s quality, σe . This last parameter crucially affects mothers’ decision to participate in the labor market and we use the aggregate labor force participation rate of mothers as the related target. The curvature of children’s quality in the utility function is set to match the participation rate of mothers with older children [3-6) in H. Whether a woman decides to have children or not is related to the fixed utility cost of children. We thus use the percentage of women who are mothers in F to match σk . We set σk , α, and σe to 0.58, 0.73, and 1.72 respectively. The weight of mothers’ time for children’s quality, φm strongly determines how much a mother works depending on her child’s age. We calibrate this parameter to a value of 0.75 to match the percentage of working mothers with small children [0-3) in H. To determine child care cost for small children (p1 ) we consider SOEP data for 2011.13 According to 12

Π[(wi , w˜j )H , (wi w , w˜j m )F ]= Ω(1)Θ(3)Φ(1, 3)d(2)d(5)/Θ(3). This is the only wave that includes a question related to overall child care costs - including payments to babysitters, etc. Questions in earlier waves, e.g. 1996, exclusively referred to costs of child care institutions and are thus likely to underestimate child care costs, especially for small children. 13

21

Table 4.1: Parameters Parameter

Explanation

Value

Parameters set a priori β discount factor

0.889

Calibrated Parameters σ utility curvature and weight of consumption α utility curvature of children’s quality σe weight of children’s quality in utility σk fixed utility cost of children φm weight of mother’s time for children’s quality p1 cost of child care, period 1, [0-3) years p2 cost of child care, period 2 [3-6) years g1 probability of free care in H, period 1, [0-3) g2 conditional probability of free care in H, period 2 [3-6) w¯ underlying mean hourly wage of women µ mean underlying offer function for wage in F σ standard deviation underlying offer function

0.8 0.73 1.72 0.58 0.75 3.3 2.4 0.07 0.3 13.50 1.00 0.013

Policy Parameters τ T ω

income tax rate family benefits child care subsidy

0.37 0.9625 0

this data married mothers with small children face child care costs equal to 12.1% of household income. Thus we set p1 to 3.30 to match this statistic. The price of child care when children are older (p2 ) is calibrated to match the participation rate of mothers with older children in F . This parameter takes on value 2.40. We set the probability of women having access to grandparent-provided child care in the first period, g1 to 0.07 to match the percentage of working mothers who exclusively rely on grandparent-provided child care for small children in the data. The probability that those women continue to have access in the second period, g2 , is set to 0.3 to match the percentage of working mothers using this type of care for older children. Finally, we calibrate the mean and standard deviation of the distribution of wage offers. We set µ to 1 in order to match the observed mean of wage changes, ¯. We set σ to 0.013 to match the share of the population living in H. The fraction of women in each region is closely related to wages in F which are to 22

a large extent determined by the variation in wage offers. Finally, the model’s policy parameters are the income tax rate, τ , child care subsidies, ω and family benefits, T . We set τ to 37% which is equivalent to the income tax revenue collected by the German government as a fraction of GDP (OECD [2010]). According to the OECD [2002], all German families receive benefits for each child up to the age of eighteen (Kindergeld ). In particular, they receive 154A C per month for the first child, for the second, and for the third, and 174A C for the fourth, fifth child etc. We set the amount of family benefits T to 0.9625 such as to match the Kindergeld received for one child. In Germany, public child care slots are highly subsidized (OECD [2008]). However, availability of these slots for small children is very restricted, especially in West Germany. In 1990, there were 18 slots for every 1000 children under 3 in West Germany (see Statistisches Bundesamt [2015]). By 2011, 85% of small children who attended child care were in some form of private day care while only 15% attended public or publicly subsidized nurseries ¨ (Statistische Amter des Bundes und der L¨ander [2011]). The large majority of mother thus does not have access to highly subsidized child care for their small children. Hence, effective child care costs are high and our calibrated parameter p1 reflects this. We set child care subsidies, ω equal to zero in our benchmark economy and use this parameter as a policy instrument to lower average child care costs in our counterfactual experiments. Such a reduction in effective child care costs can then be interpreted as an increase in the provision of public slots or as a subsidy to private child care. All parameters are displayed in Table 4.1.

5

Results - Benchmark Economy

Table 5.1 presents targeted moments from our benchmark model together with the corresponding data moments. Our model matches labor force participation rates of women who live close to parents or in-laws, particularly well. In the data, 22.76% and 56.75% of those women with small and older children respectively work. The model estimates these numbers to be 22.83% and 57.63%. Considering mothers as well as childless women, labor force participation rates close are 49.11% in the data and 50.33% in our model. Without the possibility of free grandparent-provided child care, mothers who live far from parents or in-laws participate less. In the data, the participation rate of those mothers with older children is 50.6%, six percentage points lower compared to those living close by. Our model estimate of 46.13% somewhat overestimates the difference in labor force participation rates across regions for mothers with older children.

23

Table 5.1: Data and model moments: targeted

LFP of mothers with children [0-3) in H LFP of mothers with children [3-6) in H LFP of women LFP of mothers with children [3-6) in F % of women being mothers in F Aggregate LFP rate of mothers Child care costs as % of average household income, children [0-3) % of working mothers using free care in H when children are [0-3) % of working mothers using free care in H when children are [3-6) Mean hourly wage rate of women Mean of offer function Share of population living in H

Data

Model

22.76 56.75 49.11 50.62 67.53 33.29 12.1 34.50 4.20 15.61 1.08 57.21

22.83 57.63 50.33 46.13 65.41 36.42 12.52 36.50 4.42 15.22 1.05 58.09

The model also performs fairly well in terms of matching data on fertility. We closely match the percentage of married women who are mothers in F , 65.41% in the model compared to 67.53% in the data. However, the model overestimates the fraction of women who are mothers in H (a non-targeted moment). Combined with the fact that mothers in H have higher participation rates leads to an overestimation of the participation of mothers by three percentage points. Regarding child care costs for small children, the model predicts these to be equivalent to 12.52% of household income, closely matching the 12.1% in the data. The model also matches data on the use of grandparent-provided child care rather well. In the data, 34.5% of working mothers with small children exclusively rely on grandparent-provided child care, compared to 36.5% in the model. When children are older, this fraction drops to 4.2% in the data and 4.42% in the model. Women’s mean hourly wage in the data is 15.61A C compared to 15.22A C in our model. Regarding wage changes, we underestimate the mean of the distribution. In the data, individuals who move obtain average wage gains of 8%. In our model the average increase in the potential wage rate – before participation decisions are made but after the move – is 5%. In terms of mobility, with 58.09% of women in our model living in H the model produces a close estimate of the fact that 57.21% of women live close to parents or in-laws. Model moments in Table 5.1 were targeted to calibrate certain parameters. In order to assess the model’s validity for carrying out policy analysis, we need to consider the model’s 24

performance in matching moments that have not been used for calibration. Table 5.2 displays these non-targeted moments together with the corresponding data moments. As mentioned before, mothers who live far from parents or in-laws have lower participation rates. In the data, 20.5% of those mothers with small children participate. The model produces a fairly good estimate of this rate, 17.02%. In our empirical analysis we showed that married women who live close to parents or in-laws are more likely to have children. The model is able to capture this fact qualitatively, but it overestimates the percentage of mothers in H by around 9 percentage points. Moreover, our model can only account for about a third of the observed variation in accepted wage offers. While the standard deviation of wage changes in the data is equal to 0.30, in our model this number is 0.13. The main reason why our model underestimates this number is due to the fact that there are other reasons for individuals to move (e.g house prices, amenities, etc.) which are not present in our model but which lead to a larger variation in observed wage changes of movers in the data. Table 5.2: Data and model moments: non-targeted

LFP rate of mothers with children [0-3), F % of women being mothers, H Standard deviation of offer function Ratio time spent with child, all mothers ’vs’ working mothers Average hourly wage rate of women, H Average hourly wage rate of women, F Own wage elasticity of married women Cross-wage elasticity of married women

Data

Model

20.52 73.91 0.30 (1.17-1.43) 14.64 16.90 (2.8-5) (-0.2,-2.4)

17.02 83.10 0.13 1.16 14.57 16.04 3.27 -1.67

In our model, individuals do not value leisure and hence when a mother is not working, she is taking care of her children. While this limits the comparability of time use in model and data somewhat, we check how the model performs in relative terms. In particular, we consider data by the OECD [2011] on time mothers who live in a couple and who have at least one child 0-6, devote to child care both as primary or secondary activity. In Germany, mothers spend 11.9% of their time taking care of children. Mothers working part time spend 10.1% and those working full time 8.3%. Hence, compared to working mothers, all mothers spend 17-43% more time taking care of children. We take these values to be lower and upper bounds for time spent taking care of children by working 25

mothers in our model. In our model, all mothers spend around 16% more time taking care of children. This number is thus close to the lower bound determined by part-time working mothers in the data. Despite the lack of leisure, the model produces sensible estimates of the relative time working mothers spend taking care of children. Women’s mean hourly wage is one of our calibration targets but we did not target average wages in F and H. Our model does quite a good job in matching a higher average wage rate in F . We also report women’s own wage elasticity and the elasticity of their wages with respect to their spouses’ income. Our measure of own wage elasticity is the percentage change response in aggregate labor force participation to a 10% increase in women’s wages. Similarly, the measure of cross wage elasticity is the response to a 10% increase in spouses’ income. In our model, women can decide whether to participate in the labor market, how much to work, and whether to have children or not. Empirical elasticity estimates have to be such as to reflect these options. The survey by Keane and Rogerson [2011] reports that estimates from structural models that allow for adjustments along these margins range from 2.8 to 5. Women’s labor force participation is thus found to be very responsive to changes in their own wages. Our model produces an elasticity of 3.27. Most of the effect comes from women who respond to the increase in income by not having children, and as childless women they work all their disposable time. For the response of women’s labor force participation to changes in their spouses’ income, estimates range from −0.2 to −2.4, see Blau and Kahn [2007] and Hausman and Ruud [2009]. Our model predicts a cross-wage elasticity of −1.67. This is partly driven by childless women in our model who decide to become mothers and to stay at home as their spouses’ income increases.

6 6.1

Discussion Residence, fertility and labor force participation by type

In our empirical analysis we saw that living close to parents or in-laws is associated with positive effects for fertility and participation that differ in magnitude by women’s education. In particular, the fertility effect is stronger for women with tertiary education, while proximity to parents matters more for labor force participation of women with lower levels of education. Considering disaggregated results by women’s types, we are able to analyze the model’s predictions along similar lines. Table 6.1 display the model’s distribution of women by type in H and F . In H, 1.43% of women have the lowest level of education, while in F the share of these women is 0.88%. On the other hand, women with 26

university education represent 14.06% of women in H but 16.06% of women in F . Disproportionately more high type women reside in F . We observe a similar pattern looking at disaggregate statistics of residence choice by education in our data sample, see Table A.12 of the Appendix. In our model, high type women are able to obtain high enough wages in F such that lower expected child care costs in H do not affect their decisions. For women of low education who earn low wages on the other hand, the possibility of free grandparent-provided child care makes a difference, and thus more of them decide to stay. Table 6.1: Model Distribution of women by type, in ’Home’ and ’Far’ ISCED level

’Home’

’Far’

1 2 3 4 5 6

0.0143 0.1460 0.5065 0.0901 0.1025 0.1406

0.0088 0.1007 0.5220 0.0970 0.1108 0.1606

Considering fertility, compared to women of the three lowest education levels, high type women in our model in both regions are less likely to be mothers. Table 6.2 displays the share of mothers among women by type and region in our model. In F only 37% of high-type women are mothers compared to 74% in H. Table 6.2: Share of mothers among women in model by type, in ’Home’ and ’Far’ ISCED level

’Home’

’Far’

1 1 0.92 0.49 0.56 0.74

1 0.98 0.71 0.57 0.57 0.37

1 2 3 4 5 6

27

Women who reside in F have received higher wage offers and hence they face higher opportunity costs of taking care of children. This is why these women prefer not to have children and to work all their disposable time. In H, this effect is less pronounced because these women have lower wages and thus face lower opportunity cost of having children. Secondly, there are also more low type mothers in H because access to free grandparentprovided child care lowers the costs for having children. Women of medium education levels move mainly because their spouses receive better wage offers even if her potential wage in F is lower. In that case the opportunity cost of having children in F is lower and more women with medium education have children in F . In line with our empirical findings regarding a stronger fertility effect for women with tertiary education, fertility rates in our model by region differ most strongly for these women. Looking at disaggregate statistics on fertility by education from our data sample, we observe that the model is also able to generate the observed higher fertility rates of women with lower education, see Table A.13 of the Appendix. We then analyze our model’s predictions regarding the effect of living close to parents or in-laws on mothers’ labor force participation by education, see Table 6.3.14 In H around 5% of mothers with the lowest level of education work. These women have access to free grandparent-provided child care. In F , low type women do not work because they have to pay for child care and the utility they could obtain from additional consumption is lower than the one they obtain by staying at home and taking care of children. Table 6.3: Labor force participation rate of mothers in model by type, in ’Home’ and ’Far’ ISCED level

’Home’

’Far’

0.05 0.04 0.40 0.27 0.45 1

0 0.09 0.32 0.40 0.30 0.66

1 2 3 4 5 6

14

See Table A.12 of the Appendix for labor force participation rates by education and region in our data sample.

28

For mothers with the second lowest level of education, wages and thus the opportunity cost of not working are higher and they participate more in F than in H. In ’Home’ all mothers with university education work while this is not the case in ’Far’. This is due to the fact that some university educated mothers reside in ’Far’ because of a higher wage offer for their spouses while their own wages are lower. Hence, for these women the opportunity costs of staying at home and taking care of children in F are lower than in H.

6.2

Grandparent-provided child care and wages

Our main result in the empirical analysis of the paper established a negative relationship for mothers between living close to parents or in-laws and hourly wages. In our model some women decide to live in H to have the possibility of grandparent-provided child care, rejecting a better wage offer in F . Hence, the same negative relationship arises. In order to separate the influence of higher wage offers in F from lower child care costs in H, we shut down the availability of grandparent-provided child care. The first row of Table 6.4 reports the share of women who live in H in our benchmark economy, and when there is no grandparent-provided child care available. We observe that under the latter scenario more women decide to live in F . Even though this effect is small, it shows how grandparent-provided child care reduces geographical mobility. Table 6.4: Share of population and working mothers’ mean wage rates by region: Benchmark economy versus no grandparents Benchmark No economy grandparents Share of population living in H Working mothers’ mean wage rate, in ’H’ Working mothers’ mean wage rate, in ’F’

58.09 14.47 14.81

57.70 14.66 14.77

A larger difference in mothers’ wages across regions also reflect this. Rows two and three of Table 6.4 display the average wage rate for working mothers in F and H in the benchmark economy and when there are no grandparents. Working mothers in F have higher wages. For these women the opportunity costs of not working are high, even when taking into account the costs of child care. In H, given the possibility of free child care provided 29

by grandparents even mothers of the lowest type work. Their low wages contribute to the lower average wage in H. When there is no grandparent-provided child care available, higher wage offers still imply higher average wages in F , but the difference with wages in H is smaller. In our benchmark economy there are certain low type women who receive a higher wage offer for F but decide to stay in H in order to enjoy free grandparent-provided child care. Without grandparents, these women move to F and work. Their lower wages reduce the average wage rate in F . On the other hand, those women of low education who do not receive better wage offers stay in H, but without grandparent-provided child care they do not work. This leads to an increase in the average wage in H. Aggregate differences in mothers’ wages across regions are hence driven by changes in labor force participation of mothers of different types as well as the availability of grandparentprovided child care. In our model, we can observe accepted wage offers after residence choices are realized but before participation decisions are made, and we can thus isolate the effect of access to grandparent-provided child care from labor force participation decisions. Controlling for mothers’ education we regress mothers’ accepted log wages (before participation decisions are made) on a dummy variable that takes on value one if the mother lives in H. Table 6.5 displays the results from this regression. Table 6.5: Regression of model generated mothers’ accepted log wages (before participation decisions are made) on type and residence Benchmark No economy grandparents (1) (2) log wages log wages lives in H

-0.1165

-0.1030

ISCED 1 ISCED 2 ISCED 3 ISCED 5 ISCED 6 constant

-0.4513 -0.2935 -0.1326 -0.0118 0.1723 2.6633

-0.4491 -0.2882 -0.1326 -0.0118 0.1723 2.6633

Reference Group: Mothers with ISCED 4 in “Far”. Note that the coefficient for ISCED 5 is negative because averages wages by education group are lower than those of ISCED 4 (see Table A.10 of the Appendix). Observations are weighted by the share of mothers of each type in each region.

30

The coefficient for the residence dummy indicates that mothers who live in H have lower potential wages. In particular, living in H implies a wage penalty of 11.65%. When we remove access to grandparent-provided child care, the wage penalty drops to 10%. The possibility of free grandparent-provided child care thus comes with “strings attached” accounting for almost 12% of the difference in a mother’s potential wage in “Home” and “Far.”

6.3

Disentangling the effect of wage offers for women and spouses

Women in our model decide where to reside based on wage offers and taking into account the higher costs of child care in F . A wage offer in F consists of a wage rate for her and a spouse’s income. In order to provide a better understanding of the model’s mechanisms, we shut down one dimension of these offers at a time. First, we consider offers that consist of different wages for her but that keep spouse’s income constant across regions. In a second step we maintain her wage rate constant across the two regions and only vary spouses’ incomes. Column 2 of Table 6.6 reports model moments when we shut down changes in spouses’ income. Shutting down this particular channel for living in F leads to lower mobility. Without a change in spouse income only women who receive a high enough wage offer for themselves move. In particular, 73.78% of women stay in H compared to 58.09% in our benchmark economy. This reduction in mobility increases the gap between women’s average wages in H and F . Incentives to have children in F are also affected and only 18.24% of women in F are mothers. In our benchmark economy some women move to take advantage of higher spouse’s income, even if their own potential wage in F is lower. These women have children and work less or do not work at all. When women only move because of a higher wage rate for her this scenario becomes impossible. A higher wage rate in F thus makes working all the time and to not have children more attractive. In particular, when shutting down changes in spouse’ income across regions we observe that all high type women in F stay childless. Among women in F who are mothers, participation rates almost double – for those with older children – and increase more than threefold for those with small children. Variation in spouses’ income across regions therefore increases mobility and fertility while leading to lower female labor force participation. On the other hand, if we shut down different wage offers for her, the only reason for moving to F lies with a higher spouse’s income. Again as an additional channel for mobility is shut down fewer women move. Approximately 69% stay in H, compared to 58% 31

Table 6.6: Shutting down each element of wage offers in F - changes in women wages and spouses’ income - one at a time

Share of population living in H % of women being mothers, H % of women being mothers in F LFP of mothers with children [0-3) in H LFP of mothers with children [3-6) in H LFP rate of mothers with children [0-3), F LFP of mothers with children [3-6) in F Average hourly wage rate of women, H Average hourly wage rate of women, F

Benchmark economy

No change in spouses’ income in F (m = 1)

No change in in women’s wages in F (w = 1)

58.09 83.10 65.41 22.83 57.63 17.02 46.13 14.57 16.04

73.78 83.27 18.24 22.49 56.21 60.25 90.70 14.58 16.13

69.15 82.15 89.43 23.74 60.62 10.81 51.41 14.60 15.04

in our benchmark economy (see column 3 of Table 6.6). Given the lack of better wage opportunities for her in F the mean wage rate for women in F is lower than in the benchmark economy -15.04 instead of 16.04. Hence, more women in F have children, 89.43% compared to 65.41% in the benchmark economy. Given that her wage is irrelevant for the residence choice, we observe that selection by educational types is less pronounced. However, the average level of a woman’s education in F remains high due to assortative matching and the fact that low educational couples still prefer to stay in H given the possibility of grandparent-provided child care. Fewer mothers of small children in F work because child care costs for these children are high and women’s wages are lower than in the benchmark economy. However, when children are older child care costs are lower and participation of mothers is higher compared to the benchmark economy. This is due to the fact that now more high type women are mothers and their opportunity cost of not working is high, and when child care costs are lower they work.

7

Counterfactual Experiments

In order to evaluate and compare the importance of grandparent-provided child care for mother’s labor force participation decisions and fertility choices to alternative policies we carry out two counterfactual experiments. In our first experiment we analyze a situation in which availability of grandparents increases while our second experiment considers sub-

32

sidies for paid child care. In 2012, a legislative proposal by the German government suggested to extent “grandparental” leaves to all families, making them equivalent to parental leaves of up to 3 years. While this particular legislative proposal was turned down in 2013, recently the British government has brought forward a proposal for “granny” leaves that would allow for parental leaves to be shared with grandparents. Such policies are likely to increase grandparent-provided child care. In our experiment we arbitrarily double the probability of having access to grandparent-provided child care from 7% to 14%. Our second experiment that considers a subsidy for child care is motivated by the positive relationship found in the literature between low cost of child care and female labor force participation. We expect such a policy to lead to an increase in mothers’ labor force participation. We set the subsidy to 62% of child care costs. This is the amount of subsidy needed to reduce the cost of child care for small children in our benchmark economy to be equal to the cost that Swedish families face as a fraction of average household income. In Sweden, child care costs – for children age 2 – are one of the lowest among OECD countries, and they amount to 4.4% of average household income (OECD [2008]).

More grandparent-provided child care Column two of Table 7.1 displays results from our first experiment, next to the corresponding moments from our benchmark economy. The increase in the probability of grandparent-provided child care in the first period (g1 ) from 7% to 14% implies that also in the second period the fraction of those who continue to have access to free child care (g1 g2 ) increases from 2.1% to 4.2%. As a result, the percentage of working mothers with small children who use grandparent-provided child care increases by around 19 percentage points. Labor force participation of mothers with small children increases by 6 percentage points. Moreover, the percentage of women who decide to have children in H increases by around 1.5 percentage points, compared to the benchmark economy. In F , fewer women decide to have children. Mobility drops by 1.65%, as more women decide to stay in H. In particular, women of lower education stay in H. They participate in the labor market which lowers the average wage in H and increases the gap in wages between the two regions.

Child Care Subsidies In column three we display moments from our second experiment that considers an increase in child care subsidies from 0 to 62%. Note that we do

33

Table 7.1: Counterfactual Experiments Benchmark More economy grandparent care (g1 = 0.14) % of women being mothers, H % of women being mothers, F LFP rate of mothers children [0-3), H LFP rate of mothers children [0-3), F LFP rate of mothers children [3-6), H LFP rate of mothers children [3-6), F % using free care in H, children [0-3) % using free care in H, children [3-6) Share of population living in H Average hourly wage rate of women, H Average hourly wage rate of women, F

83.10 65.41 22.83 17.02 57.63 46.13 36.50 4.42 58.09 14.57 16.04

84.55 64.71 29.35 15.58 58.18 44.28 55.83 8.76 59.05 14.45 16.08

Higher child care subsidy (ω = 0.62) 100 100 90.21 83.57 97.89 87.96 7.68 2.12 56.25 13.60 14.79

not adjust income taxes and thus while this policy is not neutral in terms of government savings, we make sure that expenses for these subsidies are covered by income tax. Introducing a subsidy that effectively lowers costs for child care leads to almost full labor force participation of mothers. Moreover, the subsidy raises incentives for women to have children. In our model economy, all women decide to have children. Increased participation rates lead to lower average wage rates in both regions. Low-wage earners can afford paid child care and decide to work. With lower prices of child care, free grandparent-provided child care makes less of a differences and hence the two regions become more similar. This is why we observe an increase in mobility of around 3%, and a smaller gap in wages between F and H. Subsidizing child care costs thus allows women to accept wage offers in F that they reject in our benchmark economy.

8

Conclusion

Looking at West German data we document that women residing close to parents or in-laws are more likely to have children and that as mothers they are more likely to be employed. However, we also find that hourly wages of mothers living close to parents or 34

in-laws are lower. We build a model of residence choice, fertility decisions, and female labor force participation to account for these relationships. We calibrate our model to match key statistics on fertility and female labor force participation for Germany. Our model performs reasonably well in matching non-targeted moments and in particular different patterns in residence and fertility choice and labor force participation for women with different earnings capacities. We perform two counterfactual experiments to analyze how women’s decisions on residence, fertility, and labor force participation change under distinct scenarios regarding availability of grandparent-provided child care and different family policies. While both, child care subsidies and increased availability of grandparents lead to higher fertility and female labor force participation, they have opposite effects for geographical mobility and mothers’ wages. Child care subsidies lower the cost of child care independently of a mothers’ location and hence wages of mothers who live close to parents or in-laws become more similar to those of mothers living further away. Increased availability of grandparents on the other hand, restricts mothers’ location choices. In order to take advantage of this type of free child care women have to remain close to parents or in-laws which leads to lower mobility and intensifies wage differences. In absence of readily available subsidized child care slots, grandparent-provided child care plays an important role for mothers’ labor force participation. However, as long as mobility of individuals decreases with age, this type of child care implies spatial restrictions for mothers, likely to affect their labor market outcomes. When designing policies aimed at increasing labor force participation of mothers policy makers should take into account that grandparent-provided child care is only an imperfect substitute for location-independent affordable child care.

References Alesina, Alberto and Paola Giuliano (2010): “The Power of the Family,” Journal of Economic Growth, 15, pp 93-125. Attanasio, Orazio; Low, Hamish and Virginia Sanchez-Marcos (2010): “Explaining Changes in Female Labor Supply in a Life-Cycle Model,” American Economic Review, 98(4), pp 1517-52. Arpino, Bruno; Pronzato, Chiara D. and Lara P. Tavares (2014): “The effect of grandparental support on mothers’ labour market participation: an instrumental variable approach,” European Journal of Population, 30(4), pp 369-390. 35

Bernal Raquel (2008): “The Effect of Maternal Employment and Child Care on Children’s Cognitive Development,” International Economic Review, 49(4), pp 11731209. Bernal Raquel and Michael P. Keane (2011): “Child Care Choices and Children’s Cognitive Achievement: The Case of Single Mothers,” Journal of Labor Economics, 29(3), pp 459-512. Bick, Alexander (forthcoming): “The Quantitative Role of Child Care for Female Labor Force Participation and Fertility,” Journal of the European Economic Association., forthcoming. Blau, Francine D. and Lawrence M. Kahn (2007): “Changes in the labor supply behavior of married women: 1980-2000,” Journal of Labor Economics, 25(3), pp 393-438. Blau, David M. and Philip K. Robins (1989): “Fertility, Employment, and Child-Care Costs,” Demography, 26, (2), pp 287-299. Bundesministerium f¨ ur Familie, Senioren, Frauen und Jugend (2005): Gender-Datenreport, 1. Datenreport zur Gleichstellung von Frauen und M¨annern in der Bundesrepublik Deutschland, M¨ unchen. Cardia, Emanuela and Serena Ng (2003): “Intergenerational time transfers and childcare,” Review of Economic Dynamics, 6, pp 431-454. Clarke-Stewart, K. A.; Gruber, C. P., and L. M. Fitzgerald (1994): “Children at home and in day care,” Hillsdale, NJ: Erlbaum. Compton, Janice and Robert A. Pollak (2014):“Family Proximity, Childcare, and Women’s Labor Force Attachment,”Journal of Urban Economics, 79, pp 72-90. Dimova, Ralitza and Fran¸cois-Charles Wolff (2011): “Do downward private transfers enhance maternal labor supply? Evidence from around Europe,” Journal of Population Economics, 24(3), pp 911-933. del Boca, Daniela (2002): “The effect of child care and part time opportunities on participation and fertility decisions in Italy,” Journal of Population Economics, 15, pp 549-573. del Boca, Daniela and Daniela Vuri (2007): “The mismatch between employment and child care in Italy: the impact of rationing,” Journal of Population Economics, 20, pp 805-832. 36

Erosa, Andr´ es; Fuster, Luisa and Diego Restuccia (2010): “A general equilibrium analysis of parental leave policies,” Review of Economic Dynamics, 13(4), pp 742-758. Fern´ andez, Raquel and Joyce Wong (2014): “Divorce Risk, Wages, and Working Wives: A Quantitative Life-Cycle Analysis of Female Labour Force Participation,” Economic Journal, 124(576), pp 319-358. Garc´ıa-Mor´ an, Eva (2010): “Childcare Costs, Female Employment, and Public Policy,” mimeo Universidad Carlos III de Madrid. Gemici, Ahu (2011): “Family Migration and Labor Market Outcomes,” Ph.D. dissertation, University of Pennsylvania. Greenwood, Jeremy; Guner, Nezih and John A. Knowles (2000): “Women on Welfare: A Macroeconomic Analysis,” American Economic Review, 90(2), P&P, pp 383-388. Guner, Nezih; Kaygusuz, Remzi and Gustavo Ventura (2012): “ Taxation and Household Labour Supply,” Review of Economic Studies, 79(3), pp 1113-1149. Guner, Nezih and John A. Knowles (2009): “ Why is the Rate of Single Parenthood Lower in Canada than in the U.S.? A Dynamic Equilibrium Analysis of Welfare Policies,” Canadian Journal of Economics, 42(1), pp 56-89. Hansen, Kristine and Denise Hawkes (2009): “Early Childcare and Child Development,” Journal of Social Policy, 38(2), pp 211-239. Hausman, Jerry and Paul Ruud (2009): “Family Labor Supply with Taxes,” American Economic Review, 74 (2), pp 242-248. Holdswoth, Clare and Angela Dale (1998): “Working mothers in Great Britain and Spain: A Preliminary Analysis,” CCSR Occasional Paper No. 14. Keane, Michael P. and Richard Rogerson (2011): “Reconciling micro and macro labor supply elasticities: a structural perspective,” NBER Working Paper No. 17430. Konrad, Kai A.; Kunemund, Harald; Lommerud, Kjell Erik and Julio R. Robledo (2002): “Geography of the Family,” American Economic Review, 92(4), pp 981-998. Mendez, Ildefonso (forthcoming): “Child care and geographical mobility,” International Labour Review, forthcoming. 37

Mincer, Jacob (1978): “Family Migration Decision” Journal of Political Economy, 86(5), pp 749-773. M¨ orck, Eva; Sj¨ ogren, Anna and Helena Svaleryd (2013): “Childcare costs and the demand for children - evidence from a nationwide reform,” Journal of Population Economics, 26(1), pp 33-65. OECD (2011):“LMF2.5: Time use for work, care and other day-to-day activities,” OECD Family database, Directorate for Employment, Labour and Social Affairs, Organisation for Economic Co-Operation and Development, Paris. OECD (2010):“Country statistical profile: Germany 2010,” OECDiLibrary, Organisation for Economic Co-Operation and Development, Paris. OECD (2008a): “OECD Economic Surveys: Germany ”, Organisation for Economic Co-Operation and Development, Paris. OECD (2008b): “Family Database”, Directorate for Employment, Labour and Social Affairs, Organisation for Economic Co-Operation and Development, Paris. OECD (2002):“Germany Country chapter,” Benefits and Wages, OECD series, Organisation for Economic Co-Operation and Development, Paris. Overturf Johnson, Julia (2005): “Who’s Minding the Kids? Child Care Arrangements: Winter 2002,” Current Population Reports, U.S Census Bureau, pp 70-101. Posadas, Josefina and Marian Vidal-Fern´ andez (2013): “Grandparents’ Childcare and Female Labor Force Participation,”IZA Journal of Labor Policy, 2(14), pp 1-20. Raymo, James M.; Mencarini, Letizia; Iwasawa, Miho and Rie Moriizumi (2010): “Intergenerational Proximity and the Fertility Intentions of Married Women,” Asian Population Studies, 6(2), pp 193-214. SOEP (2005):“DTC- Desktop Companion to the German Socio-Economic Panel (SOEP),” edited by John P. Haisken-DeNew (RWI Essen) & Joachim R. Frick (DIW Berlin), Version 8.0 - Dec 2005 Updated to Wave 21 (U). Stern, Steven (1995): “Measuring Child Work and Residence Adjustments to Parents’ Long-Term Care Needs,” The Gerontologist, 36(1), pp 76-87. ¨ Statistische Amter des Bundes und der L¨ ander (2011): “Kindertagesbetreuung regional 2011, Vergleich aller 412 Kreise in Deutschland,” Statistisches Bundesamt, Wiesbaden. 38

Statistisches Bundesamt (2015): “Tageseinrichtungen f¨ ur Kinder nach Art der Pl¨atze und Platz-Kind Relation nach L¨andern,” Statististik der Kinder-und Jugendhilfe, available upon request for years 1990 (West), 1991 (East), 1998 (Total). Wagner, Gert G.; Frick, Joachim R. and J¨ urgen Schupp (2007): “The German Socio-Economic Panel Study (SOEP) - Scope, Evolution and Enhancements,” Schmollers Jahrbuch 127(1), pp.139-169. Walker, James R. (1991): “Public Policy and the Supply of Child Care Services,” in The Economics of Child Care edited by David M. Blau, Russell Sage Foundation, New York. Zamarro, Gemma (2011): “Family Labor Participation and Child Care Decisions. The Role of Grannies,” RAND Working Paper WR-833.

A

Appendix Figure A-1: Grandparents providing care for grandchildren age 0-6, by frequency Denmark

Daily

Weekly

Daily or Weekly

Sweden France Czech Republic Spain Germany Austria Netherlands Ireland Switzerland Poland Belgium Greece Italy 0

5

10

15

20

25

30

35

40

Percentage (%) Data: Survey of Health, Ageing and Retirement in Europe (SHARE), 2nd wave

39

45

50

Percentage (%)

Figure A-2: Grandparents providing weekly care by distance to grandchild age 0-6 Mean Std. Deviation Mean Germany

40

20

0

< 5 km

5-25km

25-100km

> 100 km

Data: Survey of Health, Ageing and Retirement in Europe (SHARE), 2nd wave. Mean and standard deviation are calculated across all fourteen European countries included in SHARE, see Figure A-1 for the list of countries.

Percentage (%)

Figure A-3: Grandparents providing monthly or less frequent care by distance to grandchild age 0-6

40

20 Mean Std. Deviation Mean Germany 0

< 5 km

5-25km

25-100km

> 100 km

Data: Survey of Health, Ageing and Retirement in Europe (SHARE), 2nd wave. Mean and standard deviation are calculated across all fourteen European countries included in SHARE, see Figure A-1 for the list of countries.

40

Table A.1: Summary statistics: Main sample age age 20-24 age 25-29 age 30-34 age 35-39 age 40-44 age 45-49 married nationality not German number of children has child 0-3 has child 4-6 in small town in large town has tertiary education is employed full time part time weekly hours worked* in education/training firm tenure* hourly wage (in 2010 A C)* monthly wage of spouse (in 2010 A C)* no parent or in-law alive parents or in-laws in household parents or in-laws close - parents or in-laws in same building - parents or in-laws in same neighborhood - parents or in-laws in same town parents or in-laws far - parents or in-laws one hour away - parents or in-laws further away - parents or in-laws in foreign country 1991 1996 2001 2006 2011 Observations

Mothers Mean(Std. Dev.) 38.63 (6.83) 0.03 (0.16) 0.08 (0.28) 0.18 (0.38) 0.22 (0.42) 0.26 (0.44) 0.24 (0.43) 0.80 (0.40) 0.02 (0.15) 1.92 (0.89) 0.28 (0.45) 0.22 (0.42) 0.45 (0.50) 0.26 (0.44) 0.20 (0.40) 0.58 (0.49) 0.17 (0.37) 0.36 (0.48) 25.77 (12.21) 0.02 (0.13) 8.26 (7.44) 14.34 (8.09) 3790.17 (2246.77) 0.05 (0.22) 0.02 (0.15) 0.54 (0.50) 0.12 (0.33) 0.21 (0.40) 0.22 (0.41) 0.38 (0.49) 0.28 (0.45) 0.10 (0.29) 0.01 (0.08) 0.14 (0.35) 0.15 (0.35) 0.27 (0.44) 0.25 (0.43) 0.20 (0.40)

Childless women Mean(Std. Dev.) 31.03 (8.12) 0.26 (0.44) 0.26 (0.44) 0.18 (0.38) 0.12 (0.32) 0.11 (0.31) 0.09 (0.28) 0.25 (0.44) 0.06 (0.23) 0.36 (0.48) 0.36 (0.48) 0.25 (0.44) 0.90 (0.29) 0.69 (0.46) 0.09 (0.29) 38.62 (9.77) 0.17 (0.38) 6.20 (6.55) 13.77 (7.23) 3346.29 (3302.21) 0.04 (0.19) 0.25 (0.43) 0.32 (0.47) 0.06 (0.23) 0.10 (0.30) 0.16 (0.37) 0.39 (0.49) 0.26 (0.44) 0.13 (0.34) 0.01 (0.08) 0.14 (0.35) 0.15 (0.36) 0.24 (0.43) 0.24 (0.43) 0.23 (0.42)

7,061

3,848

Pooled data from SOEP unbalanced panel 91,96,01,06,11; West German women 20-49; *mean(std.dev) are calculated only including strictly positive observations.

41

Table A.2: Summary statistics for additional samples: (1) Mothers of children age 0-6 and (2) Grandmothers age 40-92 (1) Mean(Std. Dev.) 33.53 (5.41) 0.87 (0.34) 0.04 (0.21) 1.98 (0.93) 0.62 (0.49) 0.46 (0.50) 0.26 (0.44) 0.24 (0.43) 0.34 (0.47) 0.06 (0.24) 0.27 (0.44) 0.02 (0.12) 3653.68 (2102.82) 0.01 (0.11) 0.01 (0.11) 0.58 (0.49) 0.13 (0.34) 0.22 (0.41) 0.23 (0.42) 0.39 (0.49) 0.30 (0.46) 0.09 (0.29) 0.01 (0.09) 0.41 (0.49) 0.36 (0.48)

age married nationality not German number of children has child 0-3 in small town in large town has tertiary education is employed full time part time in education/training monthly wage of spouse (in 2010 A C)* no parent or in-law alive parents or in-laws in household parents or in-laws close - parents or in-laws in same building - parents or in-laws in same neighborhood - parents or in-laws in same town parents or in-laws far - parents or in-laws one hour away - parents or in-laws further away - parents or in-laws in foreign country siblings or brothers-or sisters-in law close siblings and parents (incl. in-laws) close grandchild close child age 0-6 cared for by relatives child age 0-6 in formal care health good or very good health satisfactory health poor or very poor 1991 1996 2001 2006 2011

(2) Mean(Std. Dev.) 67.08 (10.11) 0.63 (0.48) 2.43 (1.22) 0.44 0.50) 0.27 (0.44) 0.10 (0.30) 0.25 (0.43) 0.08 (0.26) 0.10 (0.30) 3458.23 (4001.38)

0.47 (0.50) 0.31 (0.46) 0.63 (0.48)

0.17 0.17 0.31 0.21 0.14

Observations

(0.37) (0.38) (0.46) (0.41) (0.34)

0.27 (0.45) 0.41 (0.49) 0.32 (0.47) 0 0.12(0.32) 0.25 (0.43) 0.30 (0.46) 0.34 (0.47)

2,326

5,224

Column (1): Pooled data from SOEP unbalanced panel 91,96,01,06,11; West German mothers 20-49 with children age 0-6; Column (2): Pooled data from SOEP unbalanced panel 96 01,06,11; West German grandmothers age 40-92; for this sample we do not consider data for 1991 because in this wave the question regarding self-reported health status was different and is not directly comparable. *mean(std.dev) are calculated only including strictly positive observations.

42

Table A.3: Marginal effects for married mothers of children age 0-6 from probit regression of relative-provided child care on distance to parents or in-laws has tertiary education in education/training in small town in large town no parent or in-law alive number of children age age2 parents or in-laws close parents or in-laws in household parents or in-laws in same building parents or in-laws in same neighborhood parents or in-laws one hour away parents or in-laws further away parents or in-laws in foreign country married is employed has child (0-3) part time nationality not German log (monthly wage of spouse) siblings or brothers-or sisters-in law close siblings and parents ( inlc. in-laws) close Observations

(1) 0.106 0.294 0.214** 0.005 0.273 -0.047 -0.008 -0.000 0.687*** 0.373

(0.07) (0.24) (0.07) (0.08) (0.28) (0.04) (0.05) (0.00) (0.06) (0.26)

-0.046 0.377*** 0.147* 0.366*** -0.036 0.006

(0.10) (0.09) (0.07) (0.08) (0.14) (0.01)

2,326

Child age 0-6 cared for by relatives (2) (3) 0.151* (0.07) 0.103 0.280 (0.24) 0.302 0.170* (0.08) 0.215** 0.040 (0.08) 0.009 -0.234 (0.28) 0.251 -0.052 (0.04) -0.046 -0.011 (0.06) -0.005 -0.000 (0.00) -0.000 0.659*** -0.135 (0.27) 0.444 0.344*** (0.10) 0.245** (0.08) -0.373*** (0.08) -1.221*** (0.16) -1.102* (0.52) -0.052 (0.10) -0.041 0.360*** (0.09) 0.378*** 0.147* (0.07) 0.147* 0.385*** (0.09) 0.366*** -0.053 (0.14) -0.034 0.007 (0.01) 0.006 -0.217 0.224 2,326

(0.07) (0.24) (0.07) (0.08) (0.28) (0.04) (0.06) (0.00) (0.08) (0.27)

(0.10) (0.09) (0.07) (0.08) (0.14) (0.01) (0.16) (0.18)

2,326

Standard errors in parentheses: *** p<0.01, ** p<0.05, * p<0.1 Pooled data from SOEP unbalanced panel 91,96,01,06,11; West German mothers 20-49 with child 0-6. Reference group: mothers of child 3-6, of German nationality, without tertiary education, currently not in education/training, not married, in 1991, in a medium-sized town in North Rhine-Westphalia, with at least one parent or in-law alive, living far from parents or in-laws and siblings and who do not use relative-provided child care. in column (2): living in same town as parents or in-laws. All regression include year and state dummies.

Table A.4: Average marginal effects for grandmothers from probit regression of self-reported health status on distance to grandchild has tertiary education no parent or in-law alive number of children in small town in large town age age2 grandchild close grandchild in household married is employed part time log (monthly wage of spouse)

Observations

Health bad or poor

Health satisfactory

Health good or very good

-0.341*** (0.07) 0.110 (0.06) 0.010 (0.02) 0.000 (0.05) 0.051 (0.05) -0.062** (0.02) 0.001*** (0.00) -0.025 (0.04) 0.007 (0.12) -0.101* (0.04) -0.189** (0.06) -0.150 (0.08) 0.009 (0.01)

0.055 (0.06) 0.053 (0.05) -0.007 (0.01) 0.094* (0.05) -0.015 (0.05) 0.013 (0.02) -0.000 (0.00) 0.094** (0.04) -0.022 (0.11) 0.044 (0.04) -0.017 (0.06) 0.141 0.141 -0.010 (0.01)

0.240*** (0.06) -0.147** (0.05) -0.002 (0.02) -0.117* (0.05) -0.034 (0.05) 0.074** (0.02) -0.001*** (0.00) -0.099* (0.04) 0.027 (0.12) 0.056 (0.04) 0.186** (0.06) -0.043 (0.07) 0.003 (0.01)

5,224

5,224

5,224

Standard errors in parentheses: *** p<0.01, ** p<0.05, * p<0.1 Pooled data from SOEP unbalanced 96,01,06,11 West German grandmothers 40-92. Reference group: grandmothers without tertiary education and not married, in 1996, in a medium-sized town in North Rhine-Westphalia, far from grandchild. All regression include year and state dummies.

43

Table A.5: Marginal effects evaluated for married women from probit regression for having children for: (1) all women (2) tertiary educated women and (3) women with less than tertiary education

married in education/training in small town in large town has tertiary education nationality not German log (monthly wage of spouse) parents or in-laws close parents or in-laws in household no parent or in-law alive

Observations

(1) has children

(2) has children

(3) has children

1.057*** (0.04) -0.523*** (0.07) 0.126** (0.04) -0.181*** (0.04) -0.342*** -0.059 (0.08) -0.001 (0.00) 0.223*** (0.03) -0.588*** (0.06) 0.097 (0.07)

1.217*** (0.08) -0.318* (0.15) 0.149 (0.08) -0.129 (0.08) (0.04) 0.323 (0.22) 0.008 (0.01) 0.403*** (0.07) -0.450** (0.17) 0.181 (0.16)

1.006*** (0.04) -0.610*** (0.08) 0.115** (0.04) -0.196*** (0.05)

10,909

2,417

8,492

-0.128 (0.09) -0.005 (0.01) 0.168*** (0.04) -0.647*** (0.07) 0.058 (0.08)

Standard errors in parentheses: *** p<0.01, ** p<0.05, * p<0.1 Pooled data from SOEP unbalanced panel 91,96, 01,06, 11; West German women 20-49. Reference group: women of age 30-34, of German nationality, in 1991, in a medium-sized town in North Rhine-Westphalia, currently not in education/training, with at least one parent or in-law alive, far from parents or in-laws, not married and in column (1) without tertiary education. All regression include year, age group and state dummies. All results are robust to the inclusion of a polynomial for age instead of age group dummies. Results are also robust to the exclusion of the variables marital status and spouse’s income.

44

Table A.6: Marginal effects evaluated for married mothers from probit regression for employment for: (1) all mothers (2) tertiary educated mother (3) mothers with less than tertiary education and average marginal effects for: (4) single childless women and (5) single men.

has child 0-3 number of children in education/training in small town in large town married has tertiary education nationality not German log (monthly wage of spouse) parents or in-laws close parents or in-laws in household no parent or in-law alive

Observations

(1) is employed

(2) is employed

(3) is employed

-1.196*** (0.05) -0.250*** (0.02) 0.264* (0.13) 0.005 (0.04) -0.013 (0.05) -0.383*** (0.05) 0.213*** (0.04) -0.125 (0.11) 0.019*** (0.01) 0.136*** (0.04) 0.001 (0.11) 0.156* (0.08)

-1.335*** (0.10) -0.231*** (0.05) 0.102 (0.24) -0.085 (0.10) -0.041 (0.10) -0.410*** (0.12)

-1.170*** (0.05) -0.249*** (0.02) 0.330* (0.16) 0.011 (0.05) -0.021 (0.05) -0.392*** (0.05)

-0.087 (0.28) -0.003 (0.01) 0.103 (0.08) 0.183 (0.28) 0.403 (0.23)

-0.161 (0.12) 0.024*** (0.01) 0.142*** (0.04) -0.019 (0.12) 0.113 (0.09)

7,061

1,436

5,623

(4) is employed

(5) is employed

0.232* (0.10) 0.132 (0.10) -0.156 (0.09)

0.523*** (0.09) 0.127 (0.07) -0.031 (0.07)

0.267** (0.09) -0.478*** (0.12)

0.509*** (0.08) -0.252* (0.11)

-0.033 (0.10) -0.565*** (0.10) -0.203 (0.17)

-0.016 (0.08) -0.391*** (0.08) -0.011 (0.13)

2,872

4,251

Individuals on maternity leaves and those working less than 20 hours a month are defined as not employed. Standard errors in parentheses: *** p<0.01, ** p<0.05, * p<0.1 Pooled data from SOEP unbalanced panel 91,96, 01,06, 11; Columns (1)-(3): West German mothers 20-49. Reference group: mothers of age 30-34 with children > 3, of German nationality, in 1991, in a medium-sized town in North Rhine-Westphalia, currently not in education/training, with at least one parent or in-law alive, far from parents or in-laws, not married and in column (1) without tertiary education. For column (4): Single childless women age 20-49. Reference group: women of age 30-34, of German nationality, in 1991, in a medium-sized town in North Rhine-Westphalia, currently not in education/training, with at least one parent or in-law alive, far from parents or in-laws. For column (5): single men age 20-49. Reference group: men of age 30-34, of German nationality, in 1991, in a medium-sized town in North Rhine-Westphalia, currently not in education/training, with at least one parent or in-law alive, far from parents or in-laws. All regression include year age group, and state dummies. All results are robust to the inclusion of a polynomial for age instead of age group dummies. Results for mothers are also robust to the exclusion of the variables marital status and spouse’s income.

45

Table A.7: Mothers’ wages and distance to parents or in-laws: (1) Heckman selection model for monthly wages (2) Heckman selection model for hourly wages excluding variables posing a possible endogeneity problem (marital status and spouse’s income) (3) OLS regression of hourly wages (1) log(monthly wage) has tertiary education no parent or in-law alive number of children in education/training married nationality not German hours worked firm tenure parents or in-laws close parents or in-laws in household in small town in large town

0.345*** (0.019) -0.098*** (0.033) -0.081*** (0.010) -0.053 (0.053) -0.016 (0.019) -0.056 (0.059) 0.011*** (0.000) 0.018*** (0.001) -0.048*** (0.016) -0.010 (0.053) -0.048** (0.019) 0.037* (0.021)

5.739*** (0.044) 7,061

7,061

log (monthly wage of spouse)

Observations R-squared

0.216*** (0.042) 0.154* (0.079) -0.248*** (0.019) 0.277** (0.131) -0.376*** (0.049) -0.125 (0.110)

0.135*** (0.035) -0.001 (0.109) 0.005 (0.042) -0.016 (0.047) -1.202*** (0.046) 0.018*** (0.005) 0.901*** (0.080)

has child 0-3

constant

is employed

(2) log(hourly wage)

is employed

0.352*** (0.017) -0.069** (0.030) -0.083*** (0.009) 0.002 (0.049)

0.206*** (0.042) 0.181** (0.078) -0.258*** (0.019) 0.331** (0.131)

-0.045 (0.054)

-0.152 (0.110)

0.017*** (0.001) -0.045*** (0.015) -0.035 (0.049) -0.045** (0.018) 0.035* (0.020)

0.109*** (0.035) 0.018 (0.108) -0.001 (0.042) 0.001 (0.046) -1.220*** (0.046)

(3) log(hourly wage) 0.349*** (0.017) -0.085*** (0.030) -0.065*** (0.009) -0.020 (0.048) -0.024 (0.017) -0.037 (0.054)

0.017*** (0.001) -0.052*** (0.015) -0.037 (0.049) -0.046*** (0.018) 0.033* (0.020)

2.333*** (0.035)

0.744*** (0.074)

2.421*** 2.421***

7,061

7,061

4,068 0.218

The dependent variable in the selection equation is employment. Mothers on maternity leaves and those working less than 20 hours a month are defined as not employed. Standard errors in parentheses: *** p<0.01, ** p<0.05, * p<0.1 Pooled data from SOEP unbalanced panel 91,96, 01,06, 11; West German mothers 20-49. Reference group: mothers of age 30-34 with children > 3, of German nationality, in 1991, in a medium-sized town in North Rhine-Westphalia, currently not in education/training, without tertiary education, with at least one parent or in-law alive, far from parents or in-laws, not married. All regressions include yea, age group, and state dummies.

46

Table A.8: Mothers’ wages and distance to parents or in-laws: Heckman Selection Model for mothers’ hourly wages: (1) tertiary educated mothers (2) mothers with less than tertiary education (3) mothers earning more than the lowest 10% (1) log(hourly wage)

is employed

(2) log(hourly wage)

is employed

has tertiary education no parent or in-law alive in education/training number of children married nationality not German parents or in-laws close parents or in-laws in household in small town in large town

0.064 (0.071) -0.096 (0.085) -0.078*** (0.021) -0.057 (0.040) 0.038 (0.119) -0.060* (0.032) 0.113 (0.116) -0.028 (0.041) 0.037 (0.040)

has child 0-3 log (monthly wage of spouse) firm tenure constant

Observations

0.012*** (0.002) 2.793*** (0.076) 1,438

0.391* (0.226) 0.120 (0.240) -0.229*** (0.047) -0.408*** (0.119) -0.100 (0.278) 0.098 (0.077) 0.152 (0.279) -0.083 (0.100) -0.043 (0.101) -1.348*** (0.099) -0.002 (0.011)

-0.095*** (0.033) 0.024 (0.059) -0.078*** (0.010) -0.039** (0.019) -0.051 (0.061) -0.034** (0.017) -0.068 (0.054) -0.051*** (0.020) 0.029 (0.023)

1.606*** (0.199)

0.018*** (0.001) 2.336*** (0.039)

1,438

5,623

0.110 (0.085) 0.352** (0.159) -0.247*** (0.022) -0.385*** (0.055) -0.160 (0.122) 0.141*** (0.040) -0.018 (0.120) 0.012 (0.047) -0.024 (0.053) -1.186*** (0.054) 0.023*** (0.006)

(3) log(hourly wage) 0.296*** (0.015) -0.081*** (0.026) -0.046 (0.042) -0.064*** (0.009) 0.009 (0.015) -0.085* (0.048) -0.033** (0.013) -0.097** (0.041) -0.027* (0.016) 0.028 (0.017)

is employed 0.335*** (0.045) 0.131 (0.084) 0.301** (0.140) -0.309*** (0.021) -0.451*** (0.052) -0.119 (0.119) 0.128*** (0.038) -0.011 (0.115) -0.020 (0.046) 0.020 (0.050) -1.153*** (0.050) 0.015*** (0.005)

0.814*** (0.089)

0.011*** (0.001) 2.481*** (0.032)

0.874*** (0.087)

5,623

6,069

6,069

The dependent variable in the selection equation is employment. Mothers on maternity leaves and those working less than 20 hours a month are defined as not employed. Standard errors in parentheses: *** p<0.01, ** p<0.05, * p<0.1 Pooled data from SOEP unbalanced panel 91,96, 01,06, 11; West German mothers 20-49. Reference group: mothers of age 30-34 with children > 3, of German nationality, in 1991, in a medium-sized town in North Rhine-Westphalia, currently not in education/training, with at least one parent or in-law alive, far from parents or in-laws, not married. in column (3) without tertiary education and earning more than 588A C per month. All regressions include year, age group, and state dummies.

47

Table A.9: Matching matrix ISCED level W o m e n

1 2 3 4 5 6

1 0.234 0.077 0.016 0.012 0.021 0.01

Spouses P 2 3 4 5 6 0.122 0.473 0.134 0.037 0 1 0.192 0.504 0.039 0.118 0.07 1 0.092 0.605 0.047 0.113 0.127 1 0.05 0.337 0.154 0.08 0.367 1 0.077 0.302 0.073 0.206 0.321 1 0.02 0.15 0.03 0.08 0.71 1

Pooled data from SOEP unbalanced panel 91,96,01,06,11; weighted statistics for West German married women age 20-49 with children age 0-6 or no children at all.

Table A.10: Average hourly wages (in 2010 A C) by education level: women and spouses ISCED level 1 2 3 4 5 6

Women

Spouses

10.25 11.87 14.20 16.36 16.11 20.16

18.90 15.71 17.77 19.18 20.78 27.51

Pooled data from SOEP unbalanced panel 91,96,01,06,11; weighted statistics for West German married women age 20-49 with children age 0-6 or no children at all.

48

Table A.11: Distribution of individuals by education ISCED level 1 2 3 4 5 6

Women

Spouses

0.012 0.127 0.513 0.093 0.106 0.149

0.0255 0.0893 0.4666 0.0568 0.1146 0.2472

Pooled data from SOEP unbalanced panel 91,96,01,06,11; weighted statistics for West German married women age 20-49 with children age 0-6 or no children at all.

Table A.12: Distribution of women by education and proximity to parents or in-laws ISCED level 1 2 3 4 5 6

close to parents or in-laws

far from parents or in-laws

0.0103 0.1406 0.5602 0.0964 0.0997 0.0930

0.0146 0.1086 0.4493 0.0884 0.1145 0.2245

Pooled data from SOEP unbalanced panel 91,96,01,06,11; weighted statistics for West German married women age 20-49 with children age 0-6 or no children at all.

49

Table A.13: Share of mothers among women by education and proximity to parents or in-laws ISCED level 1 2 3 4 5 6

All

close to parents or in-laws

far from parents or in-laws

0.7339 0.7646 0.7060 0.7037 0.7139 0.6700

0.8259 0.7871 0.7288 0.7207 0.7134 0.7616

0.6883 0.7477 0.6799 0.6728 0.7164 0.6191

Pooled data from SOEP unbalanced panel 91,96,01,06,11; weighted statistics for West German married women age 20-49 with children age 0-6 or no children at all.

Table A.14: Labor force participation rate of mothers by education and proximity to parents or in-laws ISCED level 1 2 3 4 5 6

All

close to parents or in-laws

far from parents or in-laws

0.2032 0.2309 0.3292 0.3228 0.3911 0.4101

0.1264 0.2464 0.3447 0.3615 0.3737 0.4664

0.2905 0.1889 0.2992 0.2520 0.4083 0.3450

Pooled data from SOEP unbalanced panel 91,96,01,06,11; weighted statistics for West German married women age 20-49 with children age 0-6 or no children at all.

50