Paternity Leave and the Gender Wage Gap: New Causal Evidence

PATERNITY LEAVE AND THE GENDER WAGE GAP: NEW CAUSAL EVIDENCE

signe hald andersen

The Research Unit Sølvgade 10, 2nd floor 1307 Copenhagen, Denmark

Danmarks Statistik Sejrøgade 11 ISSN 0908-3979 2100 Copenhagen, Denmark study paper 118 march 2017 The ROCKWOOL Foundation Research Unit Study Paper No. 118

PATERNITY LEAVE AND THE GENDER WAGE GAP: NEW CAUSAL EVIDENCE

Signe Hald Andersen

Copenhagen 2017 Paternity Leave And The Gender Wage Gap: New Causal Evidence

Study Paper No. 118

Published by: © The ROCKWOOL Foundation Research Unit

Address: The ROCKWOOL Foundation Research Unit Soelvgade 10, 2.tv. DK-1307 Copenhagen K

Telephone +45 33 34 48 00 E-mail [email protected] web site: www.en.rff.dk

March 2017 PATERNITY LEAVE AND THE GENDER WAGE GAP: NEW CAUSAL EVIDENCE

ABSTRACT This paper tests how father’s uptake of paternity leave affects the intra-household gender wage gap. For this purpose, the paper uses full sample, administrative data from Statistics Denmark and facilitate causal inference by exploiting five reforms of the parental leave. However rather than focusing on the actual number of days of leave the father takes as is done in existing studies, the paper tests the effect of the extent of father’s leave, relative to the extent of mother’s leave. Results show that father’s involvement in the childcare benefits mother’s wage income in both the short and longer run, but that it also affects his own wage income to an extent where the overall effect of his leave on the gender wage gap becomes less pronounced. Results do however show that families where fathers take more leave have higher total household wage incomes.

Keywords: IV-models, the gender wage gap, paternity leave, reforms.

INTRODUCTION Despite decreasing differences between men and women’s wages, women in most countries still earn significantly less than men. In the US and UK the gap is at 17 percent, and even in countries like Denmark and Sweden, where both the labor market and the welfare state is predominantly organized such as to promote equal rights and opportunities for men and women, women still earn 7 respectively 15 percent less than men (source: OECD.stat). The leave mothers take in the period immediately before and after childbirth is among the many explanations for the persistent gender wage gap in most western countries. This has spurred an interest in the effect of leave taken by the father, and the extent to which it affects mother’s labor market outcomes, but also his own – and through this – the gender wage gap (Cools et al. 2011; Hart, Andersen & Drange, 2016; Johansson, 2010; Rege & Solli, 2013). To answer this question, studies have so far focused on the effects of the absolute length of the paternity leave. However to the extent that the negative effect of mother’s leave reflects how the leave increases her specialization, relative to the father, in taking care of children and related domestic chores, we need a more refined measurement of the way families organize the leave. Such measure should capture the leave taken by the father, relative to the mother, as this will be a more direct indicator of how the parents share the responsibilities and provide us with a clearer reflection of how the leave affects the degree of intra-household specialization. For this purpose, my study tests how exogenous changes in the leave taken by fathers relative to mothers affects mother’s and father’s wage income – and essentially, the gender wage gap.

BACKGROUND The literature has provided a range of explanations for the persistent gender wage gap. These include differences in human capital – such as schooling and experience -, gender differences in choice of occupations, industries and firms, and discrimination. Also, scholars have more recently begun to scrutinize the influence from gender differences in norms, psychological attributes and non-cognitive skills. And while women have largely caught up with men when it comes to human capital, many of the other explanations still account for substantial parts of the variation in the gender wage gap (Blau & Kahn, 2016). However, an important additional and non-negligible explanation of the gender wage gap pertains to the actual and expected division of labor in the household by gender, in particular with regards to the responsibilities related to child birth and subsequent child care. This is the motherhood penalty, which, like the gender wage gap, is documented empirically by numerous studies, across both socioeconomic groups (e.g. Budig and Hodges 2010; Evertsson and Duvander 2010) and institutional contexts (as

3 documented in e.g. Aisenbrey et al. 2009; Gangl and Ziefle 2009). According to the literature, the motherhood penalty works through both the selection of specific women into motherhood (see e.g. Datta Gupta and Smith 2002), statistical discrimination of mothers at the labor market (as found in Datta Gupta et al. 2008) and the human capital depreciation caused by time spent away from the labor market due to leave and the need for reduced or flexible work hours to accommodate responsibilities related to child care (Staff and Mortimer 2012; for empirical support, see Kahn et al. 2014; Lundberg and Rose 2000; Pertold-Gebick et al. 2016). Similarly – or maybe even in contrast – there is talk of a fatherhood premium (in line with men’s marriage premium), as empirical evidence suggest that becoming a father results in a wage gain of between 3 and 10 percent (e.g. Killewald, 2012; Lund- berg & Rose, 2002). Possible explanations of this effect include the expected increased gender specific specialization that (may) result from the childbirth, and which will reduce the time the father spend on household chores and increase his labor market specialization – and hereby also increase his productivity and wage gains. According to another explanation, men associate fatherhood with responsible and pro-social behavior, which means that becoming a father inclines men to desist from delinquency, and instead increase their pro-social integration and institutional involvement (Augus- tine, Nelson & Edin, 2009; Knoester & Eggebeen, 2006). This is then likely to increase father’s labor market participation as well as his dedication to this work, and through this, increase his wages. Given the motherhood penalty and the fatherhood premium, having children then seem to be an important independent driver of the gender wage gap. However aside from the labor market choices that men and women make when becoming parents, the type of support that society provides for parents may also further reinforce this effect of parenthood on the gender wage gap. More specifically, by granting parents the right to take leave before and after childbirth, but also by earmarking parts of the leave to mothers and other parts to fathers, and – what is most often the case – only little or no leave to fathers but a significant amount to mothers, the system then promotes both mothers’ stronger involvement in child care and fathers’ dedication to (and even importance at) the labor market. The question then remains if a reorganization of the parental leave system can be used as a lever to reduce the gender wage gap? At the speculative level, we may think of least two routes through which increases in father’s leave may matter. First, to the extent that the duration of father’s and mother’s leave is negatively correlated, e.g. because policies allow parents to share a given number of weeks, father’s longer leave will reduce the amount time that mother spend away from the labor market and reduce the depreciation of her labor market capital. In addition, parallel to the effect of mother’s leave on her labor market capital, we may also expect father’s time away from work due to leave to affect his labor market capital negatively (a matter that is still disputed, though, see Cools et al. 2011; Johannsson 2010; Rege and Solli 2013). Furthermore,

4 since father’s leave increases his household capital, it also increases the probability that parents share domestic responsibilities in the longer run, by simultaneously giv- ing him more say in household matters and her more legitimacy in counting on his participation in taking care of chores. This will increase the time and energy she has for pursuing a labor market career. In combination, both the expected immediate and longer run processes will serve to reduce the within-household wage gap, and through this, the macro level gender wage gap. Second, father’s leave may also affect the degree to which the mother becomes the victim of statistical discrimination due to her motherhood. His leave will send a strong signal to the mother’s employer about her husband’s involvement in child care and other domestic chores, and may reduce the employer’s expectations that the motherhood will lower the woman’s productivity (Schober 2014). In sum, we have valid reasons to expect father’s leave to affect the motherhood penalty – and possible also reduce the fatherhood premium, and through this reduce the gender wage gap. Thus, if the system somehow incentivizes families to increase the leave taken by fathers and reduce leave taken by mothers, it may reduce the effect of childbirth on the gender specific specialization in the household, and improve women’s possibilities of investing at the labor market post-birth. My study is an attempt to test this hypothesis. It does so by testing the effect on men and women’s wage income of five reforms of the parental leave system that changed men and women’s incentives for taking leave when becoming parents. As I explain later, some reforms change both parents’ incentives and others only fathers’, and in sum, insights from these reforms will improve our understanding of the reasons for the gender wage gap.

Literature review

Mine is not the first study to investigate how changes in the parental leave system affects mother’s and father’s labor market outcomes, and through this, the gender wage gap. In particular, scholars from the Scandinavian countries – where the parental leave system often reflects the broader gender egalitarian ideologies of those countries – have been active in producing scientific evidence for use in this debate. Here e.g. Jo- hansson (2010) uses two reforms of the Swedish system for parental leave and Swedish data to test how own and partner’s leave affect labor market outcomes. She finds that while own leave reduces labor market outcomes for both fathers and mothers, there is a positive effect of father’s leave on mother’s earnings (even if effects vary by model specification). This effect is larger than the negative effect of mother’s leave on mother’s earnings, which then means that father’s leave serves to reduce the gender wage gap. In contrast, Cools et al. (2011) find evidence of a negative effect on Norwegian data. They exploit a reform of the Norwegian parental leave system that earmarked 4 of 42 weeks of leave to fathers, and show that while paternity leave does not affect father’s outcomes, it reduces mother’s labor market outcomes, hereby increasing the

5 gender wage gap. Exploiting the same reform, but a different empirical setup, Rege & Solli (2013) present evidence that paternity leave reduces father’s future earnings from childbirth till the child is 5 years old (which is their last point of observation). Last, a study exploiting a later reform in Norway, which increased the number of weeks earmarked for fathers from 6 to 10, find no effect of the daddy quota for neither father’s nor mother’s labor market outcomes (Hart, Andersen & Drange, 2016). In sum, existing evidence is mixed an provide no clear insights as to if and to what extent father’s leave benefits mother’s labor market outcomes and through this, the gender wage gap. My study contributes to this literature, not only by presenting evidence from reforms implemented in a different Scandinavian country (which may be of minor interest to most readers), but also by using another design than existing studies. The studies presented above focus on the effects of the absolute length of father’s leave, however, my interest lies on the effects of the relative length of father’s leave - or put differently, the effect on the gender wage gap of how much leave the father takes relative to the mother. My line of reasoning is that if one of the important drivers of the motherhood penalty is the effect of maternity leave on the gender specific specialization in the household, we should not study variation in father’s absolute leave, because in principle this variation could be paralleled with similar variation in mother’s leave (if e.g. she also responds to the reform in her uptake of leave), and cause only parallel offsetting of the gender specific specialization in the household. Instead, we need to study variation in how much leave the father takes, relative to the mother, because it is this variation that we may plausibly think will affect how labor is divided in the home and how much room the mother will have for investing at the labor market. And only this way will we improve our understanding of how the organization of the parental leave system affects the gender wage gap.

METHOD When considering household dynamics, we cannot, however, disregard issues pertaining to the selection of specific women into motherhood, or for that matter, the fact that it is likely to be specific types of fathers that appreciate and take advantage of the opportunities for paternity leave. This means that if we observe higher wages for mothers whose partners take long leave it may simply be an indication that such couples are already committed to egalitarian principles and that the duration of father’s leave – and his relative share of the total leave - is a symptom of this commitment, rather than the actual cause of the mother’s subsequent wage level. Existing literature widely acknowledges such problems of selection and solves them through various routes, of which the predominant strategies involve the implementation of fixed effects methods (Johannsson, 2010; Gangl and Ziefle 2009) and the use of policy reforms in regression discontinuity designs (Lalive et al, 2011), and difference- in-differences models (Cools et al. 2011; Johannsson 2010; Rege and Solli 2013). My

6 study follows this tradition and use five reforms of the parental leave system as instruments in standard IV-models (estimated as two stage least squares models). Equation 1 and 2 show the model

Father’s share of total leaveh = αh + δ2xh + θreformh + rh

Mother’s labor market outcomesh = α2 + δ2xh + β father’s share of total leaveh + uh

In both equations, h is the household (h=1,…N ). Equation 1 is the first stage equation, in which the endogenous variable ( father’s share of total leave) is regressed on the vector of exogenous household controls xh and the instrument. The second stage uses the predicted (rather than the actual) value of the endogenous variable to predict the outcome variable, mother’s and father’s wage income, along with the vector of controls, xh. The standard errors in the second stage are adjusted to account for the fact that we use the predicted value of the endogenous variable. Random error terms are rh and uh. If the instrument is valid, there is no correlation between the predicted value of the endogenous variable and the error term in the second stage equation, and the model will produce consistent, plausibly causal, results.

Parental leave in Denmark

As in most countries, the parental leave system in Denmark has changed and become more generous over time. The Danish Factory Act from 1901 was the first to stipulate mother’s rights at childbirth, and it granted female factory workers the right to 4 weeks of leave (starting from the time of birth). Women with few economic resources could get a small financial compensation during the four weeks. Since then, parental leave in Denmark has undergone very significant changes, and today, Danish parents are entitled to 52 weeks of leave in total, of which 4 respectively 14 are earmarked for mothers immediately before respectively after the child birth, 2 are earmarked for fathers and parents may share the remaining 32 weeks as they wish. However, between 1980 and now, there have been 4 substantial changes – April 1989, July 1997, March 1998, March 2002 - that I will use as instruments in my IV model. These four significant changes concern two important aspects of the leave. First, the 1998 and 2002 reforms affected the number of weeks of leave earmarked for fathers. The first reform increased the number from 2 to 4, and the last reform reduced the number from 4 to 2, while at the same time increasing the total number of weeks of leave (by combining the parental leave system with the child care leave system that I describe below). Both reforms are directly relevant for the my study as they aim at changing father’s leave relative to mothers – the 1998 reform is likely to increase father’s share relative to mother’s and the 2002 reform will most likely reduce father’s share. Second, the 1989 and the 1997-reforms changed the financial incentives for taking leave, by introducing wage compensation during the leave (for both mothers and

7 fathers) for public respectively private sector employees. These reforms address one of the major concerns voiced by a lot of families, when deciding on the organization of the parental leave; fathers are often the main breadwinner in most families, which makes it difficult for families to afford his prolonged leave. However the introduction of wage compensation during the leave reduces this obstacle. Importantly though, the wage compensation introduced with the 1989 and 1997 reforms also changes mothers’ financial incentives of taking leave. Thus, using these reforms, a change in father’s share of the total leave may arise from a change in both the number of days or weeks that the father is on leave and a change in the number of days and weeks that the mother is on leave. This is a potential threat to the causal interpretation of my results, as there might be a correlation between father’s relative share of the total leave and the total leave taken in the family, such that e.g. father’s share may be larger when the total leave is shorter and vice versa. In such cases father’s share of the total leave also captures the length of mother’s actual leave, something which may affect her wage income directly. The coefficient I estimate in the IV-model would then not only represent the causal effect of his share, but also of her absolute leave. While both elements represent the reform effect, the interpretation is less clear, and in particular, we will not be able to answer the research question of the importance of father’s share of the total leave. Thus, to separate out these two reform effects, my model include an indicator of the total number of days of leave taken by the parents.

The 1994 labor market reform

When studying parental leave in a Danish context we need to also consider the labor market reform that was announced June 30th 1993 and implemented January 1st 1994. One of the purposes of this reform was to introduce a new type of leave, the so- called child care leave. The leave targeted not only parents of new born babies, but all parents with children below the age of 9, and granted them the right to a minimum of 13 (for unemployed) and a maximum of 26 (for parents with jobs) weeks of paid leave (the compensation equaled 80 percent of unemployment insurance benefits). At this point in time the duration of the traditional leave was 32 weeks (8 weeks before the birth and 24 weeks after) and with this new addition parents now had the right to a total of 58 weeks of leave (of which 10 of the 24 weeks and all the additional 13-36 weeks could be shared freely between the parents). Politicians soon realized that the scheme did not have the intended effect of reduc- ing unemployment, and implemented several subsequent reforms throughout the 1990s to make the leave less attractive (these reforms both concerned the level of the compensation given during the leave and the duration of the leave). Despite these absent labor market effects, studies show that the Danish families welcomed the new scheme, but also that women were more likely than men to use this new option (Olsen, 2000). The reform is then likely to have affected the ratio between parents’ child related

8 leave. In principle, this means that I have an additional reform that I may use as a fifth instrument in my analysis. However, because the whole Danish labor market was being reorganized in this period, it is possible that the date in time that this instrument represents does not live up to the exclusion restriction (i.e. because it may be correlated with other changes at the labor market, and hereby affect wage income through other channels than the parental leave). For this reason, I include the 1994-reform as the fifth instrument in the analyses, but encourage readers interpret the results relying on this reform with caution. Table 1 below lays out the content of the five reforms.

Table 1: The five reforms

Reform date Who did it Content Post birth leave after Policy implication (announcement affect? the reform for fathers date) April 1989 Public sector Full wage compensation during the 14 w for mother, 10 w Increased economic employees 32 weeks of leave. 22 weeks ear- sharable incentive for fathers marked for mothers and 10 weeks to take leave sharable between parents

January 1994 All parents of 13-36 additional weeks of paid 14 w for mother, 10 w Increased economic (June 1993) children below leave (80 percent of unemployment sharable + 36 w of child incentives for parents age 9 insurance benefits) care leave to take leave

March 1997 Private sector Full wage compensation during 14 w for mother, 10 w Increased economic (March 1995) employees leave. 14 weeks earmarked for sharable + 36 w of child incentives for parents mothers and 2 weeks earmarked care leave to take leave for fathers

April1998 All fathers, 2 more weeks earmarked for fathers 14 w for mother, 4 w for Increased non-mon- (April 3rd, 1998) children born – a total of 4 weeks earmarked for father, 8 w sharable + etary incentive for after October fathers 36 w of child care leave father to take leave 15th, 1997

March 2002 All families, No. of weeks earmarked for fathers 14 w for mother, 2 w for Reduced non-mon- (March 25th, 2002) children born reduced to 2, but no. of sharable father, 32 w sharable etary incentive for after January weeks increased to 32 father to take leave 1st 2002

Figure 1 shows the average days of leave taken by first time mothers respectively fathers, by the child’s birth month from January 1985 to December 2003. As shown, especially the duration of mother’s leave increases over time – from an average of a little more than 160 days to a little more than 250 days – and we see some discrete jumps in the duration at January 1994, which marks the implementation of the child care reform, and in 2002 at the time of the implementation of the new, prolonged leave scheme. Interestingly – and unexpectedly – we see a slight response among mothers to the 1998-reform, even if this reform only targets men. This could be an indication that mothers did not wish to reduce their total leave as a result of “losing” 2 of the shar-

9 able 10 weeks to the fathers, and that they then used 2 more weeks of child care leave as compensation. There also seem to be a decrease in leave taken by mothers at the beginning of 1996. This does not coincide with any particular reform, but may result from the changes in the child care leave scheme mentioned earlier. The duration of father’s leave also increases during this period, however far less. At the beginning of the period fathers take approximately 7 days of leave and at the end of the period, fathers’ leave has increased to a little more than 14 days on average. For fathers, the biggest increase seems to happen with the implementation of the child care leave scheme in 1994, just as there is a small increase from 1998 to 1999. The effect of the additional reforms seems less evident.

Figure 1: Leave by birth month. First time mothers and fathers

Importantly though, we cannot expect the reform in 1989 and 1997 which introduces wage compensation during leave to affect all men equally; the first reform concerns only public sector employees and the second reform concerns only private sector employees. Hence, figure 2 shows the average duration of leave taken by men employed in the private respectively the public sector at the time of their child’s birth from 1985 to 2003. The figure not only demonstrates an expected higher take-up rate of men employed in the public sector, it also shows how the reform effects differ by sector. Whereas men employed in the public sector respond to the 1989 reform by increasing the average duration of their leave from April 1989 onwards, there is no action at this point in time for privately employed men. We also see how the implementation of parental leave in 1994 increases the duration of leave taken by publicly employed men, and also, but less so among the privately employed men. And while there is no clear response to the 1997 reform which introduces full wage compensation

10 in the private sector (for neither group of men), we see a non-negligible increase in the duration of leave taken by men employed in both sectors with the introduction of earmarked leave in 1998. The last reform considered here – the 2002 reform which increases the total number of weeks of leave, but reduces the number of weeks earmarked for fathers – appears to increase leave taken by men employed in both sectors.

Figure 2: Leave by birth month. Fathers, by sector

In sum, the combined information gained from figures 1 and 2, is that we may use the five reforms as sources of exogenous variation in the duration of father’s leave and hence as exogenous variation in the ratio between parents leave.

Problems pertaining to time trends

A key assumption when using IV-models is that the instrument only affects the outcome through the endogenous regressor – this is the untestable exclusion restriction. However because my instrument is a date in time, I face the risk that (estimated) differences on outcome measures between families affected by the reforms (who give birth after the reform) and families unaffected by the reform (who give birth before the reform) reflect more general time trends - rather than the reform effect – simply because my date-in-time instrument is correlated with time trends in my outcome measures. In fact, during the time period covered by my 5 reforms, we see a) increasing wages (as illustrated in figure 3a below using men and women’s average wage incomes) and b) increased age at family formation which implies that over time, families are likely to have a stronger foothold at the labor market at the time that they start producing children (as illustrated in figure 3b).

11 Figure 3a: Men and women’ s average wage income Figure 3b:Age first childbirth, mothers and fathers

All this implies that before-/after-reform samples may differ on various characteristics, which may be indicators that the instrument does not fulfill the exclusion restriction. I address these possible threats to the exclusion restriction through two routes. First, rather than using raw wage income, I use a deflated version, to ensure that changes over years – and thus differences between the before and after groups - do not just reflect general time trends in favor of one of the two groups. In addition, rather than focusing on wage income in a given year after childbirth, I define my income measure for men and women as before and after childbirth differences in wages to hereby account for differences in wage levels between my before- and after-reform samples that persists even after deflating the measure. Second, I control for a number of background characteristics in my models, to adjust for differences caused by time trends, and I test the robustness of my results on re-specified samples based on matched before-after samples. With this test, I ensure that these two samples are directly comparable on all relevant observed characteristics, and that differences in age at parenthood – and the possibly resulting differences in life situations - between them do not drive my results.

The monotonicity assumption

An additional concern in the IV model is that the design – or more specifically, the instrument - lives up to the monotonicity assumption. This assumption specifies that the while the instrument may have no effect on some families, all the affected families are affected in the same way. If this assumption cannot be fulfilled in a specific set up, it is not guaranteed that the IV estimator produces a causal effect. In my setup, the assumption implies that none of the reforms I expect will increase (or, in case of the 2002 reform, decrease) fathers share of the leave, may result in less (more) leave taken among fathers in a subsample of families. Since the monotoniticy assumption is – by defini-

12 tion – untestable, we can only speculate if a specific setup fulfills the assumption. I my case, we may easily imagine that e.g. the 1998 reform fulfills the assumption, as it only targets fathers, however both the 1989 and the 1997 reform concern employees in the public respectively private sector, regardless of gender, which means that they changes both father’s and mother’s economics incentive for taking leave. As a result, there may be families, who are committed to traditional gender role ideologies and who have a preference for long leave, where these two reforms increase mother’s share of the leave. Similarly, some families may respond to the 1994 reform by increasing mother’s leave, and the 2002 by increasing father’s leave (this is probably less likely given all existing knowledge on gender specific preferences for leave). But while the monotoniticy assumption is untestable, we may design empirical setups where it is more or less likely to be violated. For instance, for the analyses of the 1989 and the 1997 reforms we may increase the credibility of the assumption by restricting the analytical samples to include families where only the father can be affected, simply because the mother is not employed in the sector affected by the reform. While it is less obvious how such samples could be constructed for the 1994 and 2002 reforms, I conduct a robustness check with these re-specified 1989- and 1997 samples.

DATA To study how paternity leave affects the gender wage gap I rely on administrative data from Statistics Denmark. All residents in Denmark have a unique personal identification number (similar to the social security number in the US) that is used to identify the residents in all major transactions with public authorities as well as in transactions with a range of private institutions as e.g. banks. Statistics Denmark collects an exten- sive part of the information registered by this unique personal identification number, and makes these data available for statistical and research purposes. These registers are available as a yearly panel dating back from 1980 and they contain information on dealings with the welfare system, including parental leave, childbirth, marital status and partner ID. The Danish administrative registers are well suited for the study that I undertake in this paper, because the registers allow me to link parents of a child, and because we observe the labor markets histories of both parts, as well as a range of other characteristics. As described above, my identification strategy relies on the implementation of five different reforms, which I use in five different IV-regression models based on samples including households with child births happening from one year before till one year after the reform. I use this 2 year window to prevent seasonal variation in who gives birth when from influencing my results. Given the target groups of these reforms models relying on the 1989 and 1997 reforms for identification of causal effect, only rests on data from households in which the male part is publicly respectively privately employed.

13 I use only first births to avoid interference from previous parental leave, and to hereby simplify the model, and I exclude single headed households. I choose not to use another obvious strategy where I include all five instruments in one model which would then cover the long time span from before 1989 till after 2002; with such a long time span, quite a few other macro level changes and time trends will have affected both labor market and family policies, and the causal interpretation results based on such a model will be less straightforward.

The endogenous regressor

To calculate my endogenous regressor, I divide the number of days of leave (both parental leave and child care leave) taken by the father by the total number of days taken by both parents. For simplicity, I only consider leave taken within the first year from the childbirth – which is also where families take most leave. I measure leave as the number of days the mother respectively the father has received parental/child care leave benefits. This measurement is useful, even for parents who receive full wage compensation during their leave, since the government reimburses employers’ wage expenses (up to a certain level) through these parental leave benefits (which are registered by employee and not by employer). The resulting indicator varies between 0 and 1 with 0 indicating no days taken by the father and 1 indicating that father takes all leave in the family. Table 2 below shows that this share varies between 8 to 14 percent in my five samples, and that all reforms except for the one implemented in 2002 significantly increases father’s share of parental leave. According to the descriptive statistics – and in line with what we would expect given the content of the reform - this last reform significantly reduced father’s share of the total leave. Note that the presented shares are higher than what is found in previous studies on paternity leave in Denmark (e.g. Olsen, 2007), but this reflect the delimitation of my samples where I only focus on first born children.

Outcome variables

My study focuses on the effect of the leave father takes, relative to mother, on the gender wage gap. To operationalize this question, I use three sets of outcome variables. The first set measures differences in women’s wage the year before child birth (t-1) and 2-4 years after the child birth (t+2, t+3, t+4), and consists in three separate measures (1. Woman’s waget+2-woman’s waget-1, 2. Woman’s waget+3- woman’s waget-1, 3. Woman’s waget+4-woman’s waget-1). The second set measures differences between men’s wage the year before child birth (t-1) and 2-4 years after the child birth (t+2, t+3, t+4). Also this measure consists in three separate measures (4. Man’s waget+2-man’s waget-1, 5. Man’s waget+3-man’s waget-1, 6. Man’s waget+4- man’s waget-1). The last set measures the intra-household differences in men’s and women’s wages in the same

14 period – i.e. differences between the before-after birth change in women’s respectively men’s wages (7. (Man’s waget+2-man’s waget-1)-( Woman’s wage t+2-woman’s waget-1), 8.

(Man’s wage t+3-man’s waget-1)-( Woman’s wage t+3-woman’s waget-1), 9. (Man’s waget+4- man’s wage t-1)-( Woman’s wage t+4-woman’s waget-1) ). With the last set of outcome variables, I directly address the relationship between father’s leave and the gender wage gap (only intra household, that is), whereas with the first two sets of variables provides a clarification of what drives the (potential) effects (i.e. if it is changes in mother’s or father’s wage that reduces/increase the wage gap in the family). But in addition to these three sets of outcomes variables I furthermore test how father’s share of the total leave affects the household total in the same time period; this outcome does not follow strictly from this study’s focus on the gender wage gap, yet we cannot disregard the fact that the prospect of income losses – or gains for that matter – is an important decisive factor for most families when they make labor market choices, and that this may be a particularly strong focus in the period that surrounds the birth of the first child. Hence, even if families and the broader society may appreciate a reduction the gender wage gap at the ideological level, we have to acknowledge that whether or not we can expect each family to respond to new knowledge on whether or not a specific policy (in this case policies related to parental leave) reduces this gap, will depend on the components contributing to this reduction, and how the policy and its individual level consequences affects the everyday lives and economies of families. On these grounds, I include an additional set of variables that capture the effect of father’s leave on household wage income (10. (Man’s waget+2-man’s waget-1)+( Woman’s waget+2-woman’s waget-1), 11. (Man’s waget+3-man’s waget-1)+( Woman’s wage t+3-woman’s wage t-1), 12. (Man’s waget+4-man’s waget-1)+( Woman’s wage t+4-woman’s waget-1) ). Table 2 show the before-after reform descriptive statistics of these variables by sample. The table provides a range of interesting information. For instance, we see that most means for women – at least for years t+2 and t+3 – are negative, indicating that on average, women experience a wage drop after child birth. But we also see that the drop seems smaller in most of the after-reform samples. In contrast all means for men are positive, indicating that becoming a father interferes little – or maybe not at all – with their wage progression. Also there is no clear pattern in differences in wage developments before and after the reform or in total household wage income before and after the reform. And while there is no clear pattern with regards to the gender wage gap before and after the reforms, most means measuring the household total are larger after the reforms, compared to before. Table 2 also shows statistics for the total leave taken by parents during the first year after their child is born (which I include in the models for reasons specified above). As seen, total leave increases over the years, from approximately 176 days to 250 days. The table also includes statistics for mother’s respectively father’s leave. While I do not include these indicators in the models, I show them to demonstrate exactly how each reform affects the use of parental leave in families. As shown, the 1989- and 1997-

15 0.14 0.23 0.23 0.33 -0.12 -0.10 (1.71) (1.93) (1.99) (1.79) (0.95) (1.03) (1.58) (1.09) (1.57) (1.55) (1.84) (1.35) (0.28) (0.00) 0.13** 0.46** 0.46** 0.34** (125.10) 0.47*** Treated Treated 0.49*** 1.00*** 0.02*** (125.07) mean/sd 248.96*** 233.86*** 2002 0.19 0.12 0.14 0.42 0.42 0.31 0.39 0.32 0.40 0.22 0.00 -0.11 -0.10 -0.02 (1.92) (1.79) (1.52) (1.02) (0.92) (1.62) (1.28) (1.53) (1.08) (1.40) (1.68) (1.80) (0.00) (0.27) 217.56 202.86 (117.16) (114.93) Controls mean/sd 0.45 0.37 0.45 0.39 0.46 0.44 -0.01 (1.76) (1.75) (1.49) (0.93) (1.59) (1.23) (0.99) (1.89) (1.04) (1.39) (1.48) (1.64) (0.00) (0.27) 196.18 0.35** 210.86 -0.02** 0.14*** 0.51*** (111.57) Treated Treated (114.42) 1.00*** 0.03*** 0.48*** mean/sd 1998 0.13 0.47 0.59 0.45 0.32 0.06 0.39 0.53 0.35 0.46 0.00 0.44 -0.01 -0.03 (1.15) (1.61) (1.74) (1.72) (0.96) (1.49) (1.49) (1.02) (1.86) (1.36) (1.39) (0.26) (0.86) (0.00) 196.13 209.65 (109.17) (112.38) Controls mean/sd 0.47 0.47 0.49 0.51 0.04 0.54 0.44 -0.03 -0.04 (1.15) (1.73) (1.78) (1.49) (1.52) (1.02) (0.97) (1.87) (1.40) (0.26) (1.38) (1.64) (0.00) (0.87) 0.35** 0.44** 0.40** 0.13*** Treated Treated 1.00*** (103.62) (102.30) mean/sd 219.43** 206.66*** 1997 0.12 0.49 0.49 0.51 0.31 0.37 0.02 0.39 0.43 0.00 0.54 0.44 -0.05 -0.06 (1.19) (1.61) (1.71) (1.51) (1.51) (1.75) (0.95) (1.36) (1.87) (1.44) (1.00) (0.86) (0.25) (0.00) 210.67 222.55 (100.14) (101.25) Controls mean/sd 0.44 (1.19) (1.70) (0.94) (1.42) (1.47) (1.59) (1.56) (0.98) (1.32) (1.07) (1.34) (0.23) (0.86) (0.00) 0.11** 0.51** (98.74) 0.50** Treated Treated 0.37*** 0.04*** 0.54*** 1.00*** 0.44*** 0.23*** 0.58*** 0.33*** (101.40) -0.11*** mean/sd -0.07*** 188.07*** 200.36*** 1994 0.16 0.10 0.41 0.47 0.47 0.25 0.32 0.06 0.00 0.36 0.44 -0.15 -0.19 -0.06 (1.16) (0.96) (1.42) (1.26) (1.46) (1.66) (0.90) (1.05) (0.99) (1.55) (1.55) (1.35) (0.23) (0.00) 161.86 171.06 (82.28) (80.60) Controls mean/sd 0.31 0.27 0.27 0.23 0.34 -0.10 -0.08 (1.18) (1.10) (0.94) (0.99) (0.90) (1.28) (1.45) (1.33) (1.20) (0.22) (0.89) (0.83) (0.00) (0.80) 0.15** 0.08** (74.00) (73.24) 0.11*** 0.18*** 0.21*** Treated Treated 1.00*** mean/sd -0.07*** 171.42** 159.99*** 1989 0.14 0.31 0.31 0.21 0.27 0.32 0.32 0.09 0.22 0.00 0.30 -0.05 -0.08 -0.00 (1.18) (1.18) (0.91) (0.96) (1.32) (1.23) (1.39) (1.08) (0.86) (0.20) (0.89) (0.00) (0.80) (0.80) 167.12 176.14 (69.06) (69.89) Controls mean/sd

1 Wage total (100,000 (t+4-t-1) DKK) Wage (100,000 total DKK) (t+3-t-1) Wage (100,000 diff DKK) (t+3-t-1) Wage diff (100,000 (t+4-t-1) DKK) Wage diff (100,000 (t+4-t-1) DKK) Wage (100,000 diff DKK) (t+3-t-1) Wage diff (100,000 (t+4-t-1) DKK) Wage diff. (t+2-t-1) (100,000Wage diff. DKK) (t+2-t-1) Wage (100,000 diff DKK) (t+3-t-1) Wage diff. (t+2-t-1) (100,000Wage diff. DKK) (t+2-t-1) (100,000Wage diff. DKK) (t+2-t-1) (100,000Wage total DKK) (t+2-t-1) Fathers share of leave Child born after the reform Total leave Mother’s leave Mother’s leave Outcome variables Mother Father Gap Total variable Dependent Instrument Covariates Table 2: Descriptive 2: Table statistics by reform

16 0.14 0.23 0.23 0.33 -0.12 -0.10 (1.71) (1.93) (1.99) (1.79) (0.95) (1.03) (1.58) (1.09) (1.57) (1.55) (1.84) (1.35) (0.28) (0.00) 0.13** 0.46** 0.46** 0.34** (125.10) 0.47*** Treated Treated 0.49*** 1.00*** 0.02*** (125.07) mean/sd 248.96*** 233.86*** 2002 0.19 0.12 0.14 0.42 0.42 0.31 0.39 0.32 0.40 0.22 0.00 -0.11 -0.10 -0.02 (1.92) (1.79) (1.52) (1.02) (0.92) (1.62) (1.28) (1.53) (1.08) (1.40) (1.68) (1.80) (0.00) (0.27) 217.56 202.86 (117.16) (114.93) Controls mean/sd 0.45 0.37 0.45 0.39 0.46 0.44 -0.01 (1.76) (1.75) (1.49) (0.93) (1.59) (1.23) (0.99) (1.89) (1.04) (1.39) (1.48) (1.64) (0.00) (0.27) 196.18 0.35** 210.86 -0.02** 0.14*** 0.51*** (111.57) Treated Treated (114.42) 1.00*** 0.03*** 0.48*** mean/sd 1998 0.13 0.47 0.59 0.45 0.32 0.06 0.39 0.53 0.35 0.46 0.00 0.44 -0.01 -0.03 (1.15) (1.61) (1.74) (1.72) (0.96) (1.49) (1.49) (1.02) (1.86) (1.36) (1.39) (0.26) (0.86) (0.00) 196.13 209.65 (109.17) (112.38) Controls mean/sd 0.47 0.47 0.49 0.51 0.04 0.54 0.44 -0.03 -0.04 (1.15) (1.73) (1.78) (1.49) (1.52) (1.02) (0.97) (1.87) (1.40) (0.26) (1.38) (1.64) (0.00) (0.87) 0.35** 0.44** 0.40** 0.13*** Treated Treated 1.00*** (103.62) (102.30) mean/sd 219.43** 206.66*** 1997 0.12 0.49 0.49 0.51 0.31 0.37 0.02 0.39 0.43 0.00 0.54 0.44 -0.05 -0.06 (1.19) (1.61) (1.71) (1.51) (1.51) (1.75) (0.95) (1.36) (1.87) (1.44) (1.00) (0.86) (0.25) (0.00) 210.67 222.55 (100.14) (101.25) Controls mean/sd 0.44 (1.19) (1.70) (0.94) (1.42) (1.47) (1.59) (1.56) (0.98) (1.32) (1.07) (1.34) (0.23) (0.86) (0.00) 0.11** 0.51** (98.74) 0.50** Treated Treated 0.37*** 0.04*** 0.54*** 1.00*** 0.44*** 0.23*** 0.58*** 0.33*** (101.40) -0.11*** mean/sd -0.07*** 188.07*** 200.36*** 1994 0.16 0.10 0.41 0.47 0.47 0.25 0.32 0.06 0.00 0.36 0.44 -0.15 -0.19 -0.06 (1.16) (0.96) (1.42) (1.26) (1.46) (1.66) (0.90) (1.05) (0.99) (1.55) (1.55) (1.35) (0.23) (0.00) 161.86 171.06 (82.28) (80.60) Controls mean/sd 0.31 0.27 0.27 0.23 0.34 -0.10 -0.08 (1.18) (1.10) (0.94) (0.99) (0.90) (1.28) (1.45) (1.33) (1.20) (0.22) (0.89) (0.83) (0.00) (0.80) 0.15** 0.08** (74.00) (73.24) 0.11*** 0.18*** 0.21*** Treated Treated 1.00*** mean/sd -0.07*** 171.42** 159.99*** 1989 0.14 0.31 0.31 0.21 0.27 0.32 0.32 0.09 0.22 0.00 0.30 -0.05 -0.08 -0.00 (1.18) (1.18) (0.91) (0.96) (1.32) (1.23) (1.39) (1.08) (0.86) (0.20) (0.89) (0.00) (0.80) (0.80) 167.12 176.14 (69.06) (69.89) Controls mean/sd

17 - 0.15 0.10 0.31 0.46 0.34 (1.16) 15.10 (0.41) (4.49) (1.22) (0.49) (1.63) (1.40) (0.47) (0.46) (0.43) (0.36) (0.39) (0.39) (0.30) 18594 0.24** 0.21** 0.35** 0.39** (26.33) 0.18*** 0.19*** 1.73*** Treated Treated 2.36*** mean/sd 28.71*** 2002 0.17 0.15 0.10 1.67 2.31 0.45 0.23 0.32 0.39 0.40 0.20 0.22 0.35 (1.12) 14.70 (0.41) (1.31) (1.53) (0.49) (0.42) (1.27) (0.47) (4.46) 28.52 (0.36) (0.48) (0.38) (0.30) (0.40) 20384 (23.35) Controls mean/sd 0.15 0.16 0.41 0.24 0.09 0.22 0.35 (1.18) (1.05) (0.49) (0.42) (1.68) (1.37) (0.43) (0.36) (4.45) (0.39) (0.48) (0.48) (0.29) (0.37) 20364 0.19** (25.10) 0.35** 2.17*** 1.57*** 0.47*** Treated Treated 0.63*** 28.02** mean/sd 14.68*** 1998 2.11 0.15 0.16 0.18 0.74 0.41 1.52 0.25 0.09 0.23 0.53 0.35 0.36 27.90 (1.02) 13.52 (1.23) (0.49) (0.42) (1.88) (1.34) (0.43) (0.36) (4.45) (0.48) (0.48) (0.29) (0.38) (0.37) 21964 (24.96) Controls mean/sd 0.17 0.15 0.07 1.63 0.23 2.46 0.39 0.50 (1.18) (0.41) (1.21) (1.59) (0.99) (4.26) (0.49) (0.42) (0.26) (0.47) (0.36) (0.50) (0.38) (0.35) 15079 0.16** 0.34** 0.22** (16.53) 27.86** Treated Treated 0.46*** 0.58*** mean/sd 12.77*** 1997 0.15 0.16 0.14 0.71 0.51 1.62 0.07 2.45 0.23 0.23 0.32 0.48 0.40 27.74 (1.13) (1.19) (0.96) 11.88 (1.79) (0.49) (0.42) (0.26) (0.42) (0.47) (4.22) (0.50) (0.37) (0.35) (0.35) 16675 (16.38) Controls mean/sd 0.17 0.98 1.43 0.09 0.43 0.34 (2.16) (0.98) (4.24) (1.28) (0.99) (0.47) (0.36) (0.43) (0.50) (0.48) (0.29) (0.44) (0.38) (0.35) 22493 0.15** 0.15** 0.37** 0.50** (22.46) 1.94*** 0.26*** 0.25*** Treated Treated mean/sd 27.36*** 12.29*** 1994 0.17 0.14 0.14 1.98 0.96 9.20 1.45 0.27 0.27 0.09 0.43 0.53 0.33 0.36 27.11 (2.12) (0.97) (1.05) (1.29) (0.47) (0.28) (4.25) (0.50) (0.45) (0.48) (0.44) (0.38) (0.35) (0.35) 23190 (13.14) Controls mean/sd 0.14 0.10 1.68 0.26 0.25 0.25 2.20 0.23 0.56 0.29 0.36 3446 (4.19) (1.01) (1.02) (0.90) (0.42) (1.60) (0.43) (0.45) (0.45) (0.48) (0.44) (0.30) (0.44) (0.35) (15.10) 0.44** 0.28** 27.67** Treated Treated mean/sd 11.43*** 1989 0.11 2.16 0.15 1.67 9.02 0.51 0.24 0.24 0.27 0.25 0.62 0.25 0.38 0.30 3668 (4.11) (1.13) 27.45 (0.98) (0.31) (0.49) (1.64) (0.46) (0.43) (0.43) (0.43) (0.43) (0.44) (0.87) (0.35) (12.72) Controls mean/sd Intermediate training Higher education N Higher education Vocational training Father’s leave Age at t Married to child’s father Wage income (100,000 t-1 DKK) Intermediate training No higher education Wage income (100,000 t-1 DKK) Unemployment t-1 Unemployment Unemployment t-1 Unemployment Vocational training No higher education Covariates used in the matching characteristicsMother’s Father’s characteristicsFather’s Table 2: Descriptive 2: Table statistics by reform, continued †p<0.1; *p<0.05; **p<0.01; ***p<0.001 **p<0.01; *p<0.05; †p<0.1; fiedleave using the combined information from two registers(the DREAM statistics]), register whereas the current and version only reliesthe the on latter Sammenhængende register. Second, all Socialstatistik analyses the of previous version relied matched on samples,[The whereas coherent socialthe current version only use matched samples in a robustness test. 1. Notethat 1. these ratios differ from theones presented in a previous version the of paper athttp://www.rockwoolfonden.dk/app/uploads/2016/11/ (found Study-paper-114-Final_WEB-1.pdf). The differences between the two versions of the paper reflect two changes. First, in previousthe version, I speci

18 - 0.15 0.10 0.31 0.46 0.34 (1.16) 15.10 (0.41) (4.49) (1.22) (0.49) (1.63) (1.40) (0.47) (0.46) (0.43) (0.36) (0.39) (0.39) (0.30) 18594 0.24** 0.21** 0.35** 0.39** (26.33) 0.18*** 0.19*** 1.73*** Treated Treated 2.36*** mean/sd 28.71*** 2002 0.17 0.15 0.10 1.67 2.31 0.45 0.23 0.32 0.39 0.40 0.20 0.22 0.35 (1.12) 14.70 (0.41) (1.31) (1.53) (0.49) (0.42) (1.27) (0.47) (4.46) 28.52 (0.36) (0.48) (0.38) (0.30) (0.40) 20384 (23.35) Controls mean/sd 0.15 0.16 0.41 0.24 0.09 0.22 0.35 (1.18) (1.05) (0.49) (0.42) (1.68) (1.37) (0.43) (0.36) (4.45) (0.39) (0.48) (0.48) (0.29) (0.37) 20364 0.19** (25.10) 0.35** 2.17*** 1.57*** 0.47*** Treated Treated 0.63*** 28.02** mean/sd 14.68*** 1998 2.11 0.15 0.16 0.18 0.74 0.41 1.52 0.25 0.09 0.23 0.53 0.35 0.36 27.90 (1.02) 13.52 (1.23) (0.49) (0.42) (1.88) (1.34) (0.43) (0.36) (4.45) (0.48) (0.48) (0.29) (0.38) (0.37) 21964 (24.96) Controls mean/sd 0.17 0.15 0.07 1.63 0.23 2.46 0.39 0.50 (1.18) (0.41) (1.21) (1.59) (0.99) (4.26) (0.49) (0.42) (0.26) (0.47) (0.36) (0.50) (0.38) (0.35) 15079 0.16** 0.34** 0.22** (16.53) 27.86** Treated Treated 0.46*** 0.58*** mean/sd 12.77*** 1997 0.15 0.16 0.14 0.71 0.51 1.62 0.07 2.45 0.23 0.23 0.32 0.48 0.40 27.74 (1.13) (1.19) (0.96) 11.88 (1.79) (0.49) (0.42) (0.26) (0.42) (0.47) (4.22) (0.50) (0.37) (0.35) (0.35) 16675 (16.38) Controls mean/sd 0.17 0.98 1.43 0.09 0.43 0.34 (2.16) (0.98) (4.24) (1.28) (0.99) (0.47) (0.36) (0.43) (0.50) (0.48) (0.29) (0.44) (0.38) (0.35) 22493 0.15** 0.15** 0.37** 0.50** (22.46) 1.94*** 0.26*** 0.25*** Treated Treated mean/sd 27.36*** 12.29*** 1994 0.17 0.14 0.14 1.98 0.96 9.20 1.45 0.27 0.27 0.09 0.43 0.53 0.33 0.36 27.11 (2.12) (0.97) (1.05) (1.29) (0.47) (0.28) (4.25) (0.50) (0.45) (0.48) (0.44) (0.38) (0.35) (0.35) 23190 (13.14) Controls mean/sd 0.14 0.10 1.68 0.26 0.25 0.25 2.20 0.23 0.56 0.29 0.36 3446 (4.19) (1.01) (1.02) (0.90) (0.42) (1.60) (0.43) (0.45) (0.45) (0.48) (0.44) (0.30) (0.44) (0.35) (15.10) 0.44** 0.28** 27.67** Treated Treated mean/sd 11.43*** 1989 0.11 2.16 0.15 1.67 9.02 0.51 0.24 0.24 0.27 0.25 0.62 0.25 0.38 0.30 3668 (4.11) (1.13) 27.45 (0.98) (0.31) (0.49) (1.64) (0.46) (0.43) (0.43) (0.43) (0.43) (0.44) (0.87) (0.35) (12.72) Controls mean/sd Intermediate training Higher education N Higher education Vocational training Father’s leave Age at t Married to child’s father Wage income (100,000 t-1 DKK) Intermediate training No higher education Wage income (100,000 t-1 DKK) Unemployment t-1 Unemployment Unemployment t-1 Unemployment Vocational training No higher education Covariates used in the matching characteristicsMother’s Father’s characteristicsFather’s †p<0.1; *p<0.05; **p<0.01; ***p<0.001 **p<0.01; *p<0.05; †p<0.1; fiedleave using the combined information from two registers(the DREAM statistics]), register whereas the current and version only reliesthe the on latter Sammenhængende register. Second, all Socialstatistik analyses the of previous version relied matched on samples,[The whereas coherent socialthe current version only use matched samples in a robustness test. 1. Notethat 1. these ratios differ from theones presented in a previous version the of paper athttp://www.rockwoolfonden.dk/app/uploads/2016/11/ (found Study-paper-114-Final_WEB-1.pdf). The differences between the two versions of the paper reflect two changes. First, in previousthe version, I speci

19 reforms actually reduce the numbers of days of leave taken by mothers, while at the same time increasing the days taken by fathers. In contrast, the 1994-, 1998- and 2002 reforms increase both mother’s and father’s leave.

Covariates

To adjust for differences in my before and after samples, I rely on a range of standard control variables, which are all key indicators of the families’ life situation and labor market affiliation. The group of control variables include mother’s age and marital status at child birth, mother’s and father’s unemployment and income the year before the childbirth (varying between 0 and 10, with 0 indicating no unemployment during the year and 10 indicating full unemployment during the year), and last, mother’s and father’s educational level. The lower part of Table 2 shows the descriptive statistics for before and after samples (including t-values for differences in means). As shown, the before- and after-samples at each reform differ on a range of characteristics. Control- ling for these variables in the models will adjust for the differences, and the robustness check that I conduct on matched before and after samples will demonstrate the extent to which

FINDINGS Table 3 shows the results of my analyses. I run a total of 45 models, reflecting my use of five different reforms, my focus on wage income in both the short and longer run (year t+2 to t+4 from the year of the childbirth) for mothers and fathers, and on the resulting change in the wage gap between partners. Due to the high number of models, I only present the coefficient of interest from the second stage model – i.e. the instrumented effect of father’s share of the total leave on wages. In addition, the lower part of the table shows the effect of the instrument (the reforms) on the endogenous regressor (father’s share of the total leave) as well as the F-test for the excluded instrument. However, both the first and second stage models include all controls presented in table 2. Note that the coefficients presented in table 3 might appear rather large, but this reflects the nature of the endogenous regressor: The regressor varies between 0 and 1 with 1 indicating that father takes all leave. The coefficients then indicate the difference in mother’s labor market outcomes between household in which father take 0 percent of the leave and in which fathers take 100 percent of the leave. Starting from the bottom, we see that the instrument performs well in all models. All reforms significantly affects fathers share of the total leave, and all F-tests have values above 10. Observe that all reforms including the one implemented in 2002 positively affects father’s share of the total leave by between 1 and 6 percent. While the first four results comply with expectations, the fifth might seem surprising, given the expected reform effect and the descriptive evidence presented in table 2. Importantly,

20 though, the coefficient becomes positive, when I include the indicator of total leave taken by the parents, in the model. This means that holding constant the total leave taken in the family, the reform actually increases the share taken by the father.

Table 3: Results 1989-reform 1994-reform 1997-reform 1998-reform 2002-reform Coef. Std. Coef. Std. Coef. Std. Coef. Std. Coef. Std.

Mother Wage t+2-t-1 -2.58† 1.46 1.94*** 0.15 1.69 1.18 1.78*** 0.62 0.44*** 0.15 Wage t+3-t-1 -3.81* 1.69 1.93*** 0.16 1.32 1.23 0.44 0.59 0.45** 0.16 Wage t+4-t-1 -7.18*** 2.41 1.96*** 0.16 0.77 1.25 -1.76* 0.65 1.32*** 0.18 Father Wage t+2-t-1 -2.15 1.32 1.04*** 0.17 1.85 1.51 1.27† 0.75 0.12 0.21 Wage t+3-t-1 -1.73 1.42 1.75*** 0.19 1.13 1.71 0.25 0.85 0.53* 0.24 Wage t+4-t-1 -2.47 1.58 1.89*** 0.21 -2.64 1.96 -3.29*** 1.00 1.07*** 0.25 Gap (Father-Mother) Gap t+2-t-1 0.43 1.75 -0.90** 0.21 0.17 1.76 -0.51 0.90 -0.32 0.26 Gap t+3-t-1 2.08 1.97 -0.18 0.23 -0.18 2.01 -0.19 1.01 0.08 0.28 Gap t+4-t-1 4.71* 2.40 -0.08 0.24 -3.41 2.34 -1.52 1.10 -0.25 0.30 Household total (Father+Mother) Total t+2-t-1 -4.74* 2.16 2.98*** 0.24 3.54† 2.07 3.05*** 1.04 0.56* 0.27 Total t+3-t-1 -5.54* 2.42 3.68*** 0.27 2.45 2.20 0.70 1.07 0.98*** 0.29 Total t+4-t-1 -9.65*** 3.29 3.85*** 0.28 -1.87 2.31 -5.05*** 1.28 2.39*** 0.32 First stage Ex. Res. 0.01*** 0.00 0.06*** 0.00 0.01*** 0.00 0.01*** 0.00 0.06*** 0.00 F-test 12.84*** 1171.5*** 18.95*** 63.03*** 966.10*** N 7114 45683 31754 42328 38978

†p<0.1; *p<0.05; **p<0.01; ***p<0.001

Note: All models include variables for total leave taken by parents (in months), mother’s age a childbirth, mother’s and father’s labor market experience, unemployment experience and wage income (all measured the year before childbirth), and whether mother and father are married at childbirth (as opposed to just cohabiting). Endogenous variable (the exclusion restriction) measures father’s leave as a share of total leave (father’s leave/(mother’s leave + father’s leave)). I restrict the samples to include mothers and fathers of children born from one year before the reform till one year after the reform (for the assessment of each reform).

21 Next, focusing on the coefficients from the second stage models, results from the specifications relying on the 1994- and 2002-reforms suggest that father’s leave benefits mother’s wages both short, medium and long term, and results from the 1998-reform suggest a positive short run effect but a negative long run effect. While there is no effect of the 1997 reform, results based on the 1989-reform show negative wage effects for mothers on father’s increased up-take of paternity leave. Results for fathers largely show the same effects, even if the coefficients from the 1989-specification are not significant. The results presented in the next rows of the table combine the effects for mothers and fathers by showing how the exogenously induced change in father’s relative share of the total leave taken in the family affects the wage gap between the parents. Recall that I calculate this outcome as the father’s wage at a given point in time, mi- nus mother’s wage at the same point in time. This means that a negative value is an indication of a reduction in the wage gap between the two, and a positive value is an indication of an increased wage gap. From the table we get the depressing result that father’s increase uptake of parental leave has limited effects on the wage gap between the parents. From the 1994- and 2002-reforms we get short run reductions, but in contrast, the effects we saw for fathers and mothers based on the 1989- and the 1998 specification nets out and reflect only parallel offsetting changes, that does not affect the wage balance between the parents. The last rows of the table shows the degree to which an increase in father’s share of the total leave affects total household wage income. For obvious reasons, the pattern here is relative consistent with the results that focused on mother and father separately – the 1989 reform reduces total household wage income, both in the short, intermediate and long run, and the 1994- and 2002-reforms increases it. While there is no effect of the 1997-reform the 1998-reform results in positive short run, but negative long run outcomes.

Eligibility criteria and announcement effects

Importantly, reforms are not only consequential from the day they are installed. Their mere announcement may change people’s behavior and, in case of the specific reforms exploited in this paper, also eligibility criteria relating to the child’s birthday are likely

Table 4: Re-specified instruments

Reform New before group New after group New reform date reflects (birth month of child) (birth month of child) 1994 July 1992-June 1993 July 1993-June 1994 Announcement 1997 March 1994-February-1995 March 1995-February-1996 Announcement 1998 November 1996-October 1997 November 1997-October 1998 Eligibility criteria 2002 January 2001-December 2001 January 2002-December 2002 Eligibility criteria

22 to matter. As specified in Table 1, politicians announced the 1994-reform at the end of June 1993 (expanded parental leave), and the 1997-reform (private sector, full wage compensation) was announced, but also initiated as early as March 1995. In addition, parents of children born from October 15th 1997 onwards benefitted from the rules

Table 5: Eligibility criteria and announcement effects

1994-reform 1997-reform 1998-reform 2002-reform Coef. Std. Coef. Std. Coef. Std. Coef. Std. Mother Wage t+2-t-1 2.01*** 0.29 1.96*** 0.49 1.58*** 0.55 0.30* 0.14 Wage t+3-t-1 3.08*** 0.33 3.08*** 0.57 1.42* 0.56 0.25† 0.14 Wage t+4-t-1 3.01*** 0.33 1.32* 0.51 -0.23 0.55 0.95*** 0.15 Father Wage t+2-t-1 2.46*** 0.34 0.19 0.53 0.51 0.66 -0.17 0.19 Wage t+3-t-1 3.01*** 0.38 -0.12 0.62 0.08 0.77 0.26 0.21 Wage t+4-t-1 3.88*** 0.44 0.30 0.73 -2.96*** 0.90 0.60* 0.22 Gap Gap t+2-t-1 0.46 0.41 -1.77** 0.70 -1.07 0.82 -0.47* 0.23 Gap t+3-t-1 -0.06 0.43 -3.19*** 0.83 -1.34 0.92 0.01 0.25 Gap t+4-t-1 0.87† 0.47 -1.02 0.87 -2.73** 1.03 -0.35 0.26 Total Total t+2-t-1 4.47*** 0.49 2.15*** 0.74 2.09* 0.89 0.13 0.24 Total t+3-t-1 6.10*** 0.56 2.96*** 0.85 1.50 0.97 0.51* 0.26 Total t+4-t-1 6.89*** 0.61 1.63† 0.91 -3.20*** 1.08 1.55*** 0.27

Ex. Res. 0.03*** 0.00 0.02*** 0.00 0.02*** 0.00 0.07*** 0.00 F-test 356.08*** 110.41*** 76.95*** 1212.9*** N 45,853 32,778 42,884 39,303

†p<0.1; *p<0.05; **p<0.01; ***p<0.001

Note: All models include variables for total leave taken by parents (in months), mothers age a childbirth, mother’s and father’s labor market experience, unemployment experience and wage income (all measured the year before childbirth), and whether mother and father are married at childbirth (as opposed to just cohabiting). Endogenous variable (the exclusion restriction) measures father’s leave as a share of total leave (father’s leave/(mother’s leave + father’s leave)). I restrict the samples to include mothers and fathers of children born from one year before the reform till one year after the reform (for the assessment of each reform).

23 specified in the 1998-reform (four weeks earmarked for fathers), and parents of children born after January 1st 2002 could use the new parental leave scheme specified in the 2002-reform (less weeks earmarked for fathers, but more leave in total). Hence specifying the before/after samples according to the reform date may not provide the sharpest distinction of who is and who is not affected by the reform. This is particularly true with regards to the 1998- and the 2002-reforms, where parents of children born both before the announcement and the effectuation date are eligible for the new rules. But is may also, to a certain extent, be true for the announcement of the 1994-reform, since the period between the announcement and the effectuation of the reform is suffi- ciently short to allow parents to start planning according to the new rules at the time of the announcement (especially since father’s tend to take the last part of the leave). Note that I have not been able to find evidence of the announcement of the 1989-reform. To test how these alternative dates – reflecting either the eligibility criteria or the announcement of the reform – matter for my effects, I rerun my models with re-specified instruments, as described in table 4 above. Table 5 shows the results from these new analyses. Looking first at the output from the first stage regression we see that for all reforms, the instruments remain positive and significant, and that their power increases compared to what we saw in Table 3, as illustrated by the size of the F-test. The results from the second stage models support – or even reinforces - previous findings. First, the exogenous variation in father’s paternity leave that we get from all reforms now results in positive effect on mother’s wage in the short, intermediate and long run (except for the long run effect in the 1998-reform specification, which is insignificant). Second, for fathers we now only get positive effects of increased leave in the 1994-reform specification and in the long run with the 2002-reform specification. Again the exogenous variation that we get from the 1998-reform in father’s leave points to a negative long run effect on his wage. Third, the re-specification provides stronger evidence that an increase in father’s leave reduces the household wage gap, even if the time period at which we can detect the effects vary by specification. Last, there is strong evidence across the reforms that an increase in father’s share of the total leave increases household wage levels. Given that these results rely on more precisely specified before- and after- groups, they are probably more valid than the ones presented in table 3, and should be the preferred specification.

The monotonicity assumption

As described previously, I test the robustness of my results in a setup that is likely to fulfill the monotonicity assumption. While it is difficult to come up with setups that allow for a test of the results based on the 1994 and the 2002 reforms, there is a plau- sible setup for the test of the results based on the 1989 and the 1997 reforms, just as I expect the monotonicity assumption to hold in my original 1998 specification. For this

24 purpose, table 6 show results based on this re-specification for the years 1989, 1997 and 1998 (which is not an actual re-specification). Note that both the 1997 and 1998 results are based on the preferred specification, as described in the previous section.

Table 6: Reduced samples to fulfill the monotonicity assumption

1989-reform 1997-reform 1998-reform Coef. Std. Coef. Std. Coef. Std. Mother Wage t+2-t-1 -1.18 1.45 1.13*** 0.31 1.58*** 0.55 Wage t+3-t-1 -2.18 1.57 1.89*** 0.35 1.42* 0.56 Wage t+4-t-1 -4.74** 1.82 0.91** 0.33 -0.23 0.55 Father Wage t+2-t-1 -0.33 1.10 -0.20 0.37 0.51 0.66 Wage t+3-t-1 -0.64 1.25 -0.47 0.42 0.08 0.77 Wage t+4-t-1 -0.62 1.37 -0.51 0.50 -2.96*** 0.90 Gap Gap t+2-t-1 0.84 1.81 -1.33** 0.48 -1.07 0.82 Gap t+3-t-1 1.55 1.98 -2.35*** 0.55 -1.34 0.92 Gap t+4-t-1 4.12† 2.23 -1.43** 0.61 -2.73** 1.03 Total Total t+2-t-1 -1.51 1.83* 0.94* 0.48 2.09* 0.89 Total t+3-t-1 -2.82 2.03** 1.42** 0.54 1.50 0.97 Total t+4-t-1 -5.36* 2.33 0.40 0.60 -3.20*** 1.08

Ex. Res. 0.02*** 0.01 0.04*** 0.00 0.02*** 0.00 F-test 21.40*** 187.4*** 76.95*** N 2304 18147 42,884

†p<0.1; *p<0.05; **p<0.01; ***p<0.001

25 As shown, results do not differ much from the results presented in tables 3 and 5. As we could expect, given the smaller sample, significance levels drop in the 1989-specification, and we do, to some extent get smaller coefficients. However conclusions based on this specification are largely similar to what the remaining results would point to; when father takes a higher share of the leave, it reduces the wage gap between mother and father, and increases household income.

Placebo tests

An important concern when using reforms for making causal inference is whether the estimated effects of the reforms in the first stage regressions, reflect the actual effect rather than time trends that would have affected parental leave also in the absence of the reform. A related concern is the possibility that politicians implement the reforms as responses to changes in social conditions and preferences, rather than as means to change these parameters. This then makes the reforms endogenous to the first stage outcome variable (the so-called policy endogeneity, see Besley and Case 2000). There are multiple ways of testing this concern, through placebo tests – one is to assess the effect of the reform on a sample of parents not targeted by the reform and another is to artificially move the reform date. The two reforms implemented in 1989 and 1997 each have a comparison group that we may expect is unaffected by the reform; since the 1989 reform only targets the publicly employed and the 1997 reform only targets those employed in the private sector, I may conduct placebo tests relying on employees of the other, unaffected sector. This will also test the concern related to policy endogeneity mentioned above (however only for these two reforms). For the three other specifications I test whether artificially moving the reform date affects the results, and for the 1994 and 2002 reforms I test the “reform effect” a year before the actual reform. However using this strategy with the 1998-reform would collide with the impact of the 1997-reform, and my robustness check for the 1998 reform therefore consists in moving the reform data a year a forward in time. Table 6 shows the results, and as can be seen, for all specifications except for the 2002-reform, we get very poor first stage results, with insignificant effects of the instrument, and F-test values for the exclusion of the instrument way below the acceptable threshold value of 10. This then also means that the results from the second stage are unreliable (notice also their large values which is a classic indication of a weak instrument). These are solid indications that my reform effects do not merely reflect time trends that are also present during other periods. With regards to the effects we see for the 2002-reform, this may be a signal that the reform is in fact endogenous to time trends in parental leave, even if the strength of the instrument in this specification is significantly lower than what we saw in tables 3 and 4.

26 Table 7: Placebo test

1989-reform 1994-reform 1997-reform 1998-reform 2002-reform Coef. Std. Coef. Std. Coef. Std. Coef. Std. Coef. Std. Mother

Wage t+2-t-1 -31.7 22.8 38.2 37.3 9.6 14.6 -1.0 2.6 -2.8 1.3 Wage t+3-t-1 -29.0 21.0 -8.4 9.7 2.7 7.4 -19.2 11.4 1.9 1.3 Wage t+4-t-1 -42.7 30.4 -41.4 40.1 -0.3 6.4 -19.2 11.5 2.6 1.4 Father Wage t+2-t-1 -2.5 6.4 -41.8 40.6 -3.4 7.8 -11.2 7.3 -9.2 2.8 Wage t+3-t-1 0.5 6.9 -19.7 20.1 3.4 8.5 -27.5 16.4 -1.1 1.7 Wage t+4-t-1 -12.7 11.5 -25.5 25.4 13.4 20.2 -28.5 17.0 0.1 1.9 Gap Gap t+2-t-1 29.2 21.9 -80.0 77.5 -13.0 19.9 -10.2 7.4 -6.4 2.5 Gap t+3-t-1 29.5 22.4 -11.4 13.9 0.7 9.2 -8.3 6.7 -3.0 2.2 Gap t+4-t-1 30.0 23.0 16.0 17.8 13.8 21.3 -9.4 7.3 -2.6 2.3 Total Total t+2-t-1 -34.2 25.3 -3.6 8.9 6.1 12.4 -12.2 8.2 -12.0 3.5 Total t+3-t-1 -28.6 21.8 -28.1 28.2 6.0 13.0 -46.7 27.5 0.9 2.2 Total t+4-t-1 -55.3 39.8 -66.9 64.7 13.1 21.0 -47.7 28.1 2.7 2.4

Ex. Res. 0.0 0.00 0.0 0.0 0.01 0.01 0.0 0.00 0.01*** 0.00 F-test 2.0 2.2 0.2 2.2 16.46*** N 14,883 43,691 1,513 40,381 38,604 Content Couples with A year before Couples with no A year after A year before no public sector announcement private sector announcement announcement affiliation affiliation

†p<0.1; *p<0.05; **p<0.01; ***p<0.001

27 Matched samples As described previously, I wish to further test the robustness of my results but re-run- ning my models on matched before- and after-samples. For this purpose I conduct a 1-to-1 matching based on an estimated propensity score, of families who have children before the reform with families who have children after the reform. In the matching, I use the control variables specified in table 2, which are all key indicators of the families’ life situation and labor market affiliation. The last rows of the table shows the balancing properties of the matching procedure, and we see that the matching has been successful in all specifications, except for the one relying on the 1994-reform. Here, the after-reform mothers are significantly older than the before-reform mothers, but after-reform fathers earn significantly less that before reform fathers. I meet the concerns pertaining to such less than perfectly balanced samples through controlling for these background characteristics in the models. As shown, results are very similar to what I presented in table 3, which is a good indication that my results are not driven by compositional differences between the before and after samples.

CONCLUSION This paper departs from the growing literature on how father’s paternity leave affects mother’s labor market outcomes - and through this, the gender wage gap - but offers a new approach to understanding when father’s take up of paternity leave may matter. By claiming that father’s leave only matters if it changes the relative distribution of labor market respectively domestic capital in the household, I refrain from looking at the effect the actual days, weeks or months that fathers spend on paternity leave, but focus on how many days he spends relative to how many days the mother spends. For this purpose, I exploit a series of reforms of the Danish parental leave system that change the total months of leave available for Danish parents, and introduces or removes earmarked leave for fathers. These reforms facilitate exogenous variation in both fathers’ and mothers’ take up of leave and enables causal inference. Relying the preferred specification from table 5, we learn that father’s leave does indeed seem to positively influence mother’s wage income. However, the effect is stronger on this outcome, than on the overall gender wage gap, because the father also benefits from his own leave, hereby offsetting the positive effect for mothers. This does however mean that households where fathers take more leave experience increased total household wage income, with contributions from both mother and father.

28 Table 8: Results based on matched samples 1989-reform 1994-reform 1997-reform 1998-reform 2002-reform Coef. Std. Coef. Std. Coef. Std. Coef. Std. Coef. Std. Mother Wage t+2-t-1 -2.23 1.39 1.91*** 0.15 1.60 1.26 1.87*** 0.65 0.35* 0.15 Wage t+3-t-1 -3.68* 1.64 1.91*** 0.16 1.74 1.36 0.47 0.62 0.30† 0.16 Wage t+4-t-1 -7.04*** 2.32 1.95*** 0.16 1.03 1.36 -1.95*** 0.69 1.05*** 0.17 Father Wage t+2-t-1 -2.22† 1.30 1.06*** 0.17 1.65 1.63 1.50† 0.80 -0.15 0.21 Wage t+3-t-1 -1.83 1.40 1.77*** 0.19 0.80 1.84 0.58 0.90 0.30 0.24 Wage t+4-t-1 -2.58 1.55 1.90*** 0.21 -3.38 2.20 -2.93*** 1.03 0.83*** 0.25 Gap Gap t+2-t-1 0.01 1.71 -0.85*** 0.21 0.05 1.91 -0.37*** 0.94 -0.50* 0.26 Gap t+3-t-1 1.85 1.91 -0.14 0.23 -0.94 2.19 0.11 1.06 0.00 0.28 Gap t+4-t-1 4.46* 2.30 -0.05 0.25 -4.40 2.64 -0.98*** 1.14 -0.22 0.30 Total Total t+2-t-1 -4.44* 2.07 2.97*** 0.24 3.24 2.20 3.38*** 1.11 0.20 0.27 Total t+3-t-1 -5.51* 2.37 3.69*** 0.27 2.54 2.38 1.05 1.13 0.60* 0.29 Total t+4-t-1 -9.62*** 3.21 3.86*** 0.29 -2.35 2.53 -4.87*** 1.34 1.88*** 0.31

Ex. Res. 0.01*** 0.00 0.06*** 0.00 0.01*** 0.00 0.01*** 0.00 0.06*** 0.00 F-test 13.32*** 1124.6*** 16.63*** 57.30*** 1000.4*** N 6898 44608 30021 40666 36953 Balancing statistics from the matching procedure R2 0.001 0.000 0.000 0.000 0.000 2 7.85 25.63* 0.50 1.67 0.98

χContent Couples with no A year before Couples with no A year after A year before public sector announcemen private sector announcement announcement affiliation affiliation

†p<0.1; *p<0.05; **p<0.01; ***p<0.001

Note: All models include variables for total leave taken by parents (in months), mother’s age a childbirth, mother’s and father’s labor market experience, unemployment experience and wage income (all measured the year before childbirth), and whether mother and father are married at childbirth (as opposed to just cohabiting). Endogenous variable (the exclusion restriction) measures father’s leave as a share of total leave (father’s leave/(mother’s leave + father’s leave)). I restrict the samples to include mothers and fathers of children born from one year before the reform till one year after the reform (for the assessment of each reform).

REFERENCES Aisenbrey, Silke, Evertsson, Marie & Grunow, Daniela (2009): Is There a Career Pen- alty for Mothers’ Time Out? A Comparison of Germany, Sweden and the United States. Social Forces, 88(2): 573-605 Augustine, Jennifer March, Nelson, Timothy & Edin, Kathryn (2009): Why Do Poor Men Have Children? Fertility Intentions among Low-Income Unmarried U. S. Fathers. ANNALS of the American Academy of Political and Social Sciences, 624: 99-117 Becker, Gary S. (1981): A Treatise on the Family. Harvard University Press Besley, Timothy & Case, Anne (2000): Unnatural experiments? Estimating the inci- dence of endogenous policies. The Economic Journal, 110: 672-694 Blau, Francine & Kahn, Lawrence (2016): The Gender Wage Gap: Extent, Trends, and Explanations. IZA DP, no. 9656 Bonke, Jens, Datta Gupta, Nabanita & Smith, Nina (2004): Timing and Flexibility of Housework and Men and Women’s Wages. In D.S. Hamermesh and G. A. Pfann (eds.). Contributions to Economic Analysis, vol. 204. Elsevier Press. Budig, Michelle J. & Hodges, Melissa J. (2010): Differences in Disadvantage: Variation in the Motherhood Penalty across White Women’s Earnings Distribution. American Sociological Review, 75(5): 705-728 Cools, Sara, Fiva, Jon H. & Kirkebøen, Lars Johannessen (2011): Causal effects of paternity leave on children and parents. Statistics Norway, Research Department Discussion Papers, no. 657 Datta Gupta, Nabanita, Smith, Nina & Verner, Mette (2008): The impact of Nordic countries’ family friendly policies on employment, wages, and children. Review of Economics of the Household, 6: 65-89 Datta Gupta, Nabanita Datta & Smith, Nina (2002): Children and Career Interrup- tions: the Family Gap in Denmark. Economica, 69(276): 609-629 Ekberg, John, Eriksson, Rickard & Friebel, Guido (2005): Parental Leave – A Policy Evaluation of the Swedish “Daddy-Month” Reform. IZA Discussion Paper, 1617 Evertsson, Marie & Duvander, Ann-Zofie (2010): Parental Leave – Possibility or Trap? Does Family Leave Length Effect Swedish Women’s Labour Market Opportuni- ties? European Sociological Review, 27(4): 435-450 Gangl, Markus & Ziefle, Andrea (2009): Motherhood, Labor Force Behavior, and Women’s Careers: An Empirical Assessment of the Wage Penalty for Motherhood in Britain, Germany and the United States. Demography, 46(2): 341-369 Hansen, Anette Eklund (2003): Barselsorlovens historie i 100 år. Found at www.aba.dk Hart, Rannveig K., Andersen, Synøve N., & Drange, Nina (2016): The Father’s Quo- ta: An Earnings Equalizer? Paper presented at PAA, Washington, 2016. Jacobsen, Sara Jul, Klynge, Alice Heegaard & Holt, Helle (2013): Øremærkning af barsel til fædre. Et litteraturstudie. København: SFI, 13:22 Johansson, Elly-Ann (2010): The effect of own and spousal parental leave on earnings. IFAU Working Paper, 2010:4

31 Kahn, Joan R., García-Mangland, Javier & Bianchi, Suzanne M. (2014): The Moth- erhood penalty at Midlife: Long-Term Effects of Children on Women’s Careers. Journal of Marriage and Family, 76: 56-72 Karu, Marre (2012): Parental Leave in Estonia: Does Familization of Fathers Lead to Defamilization of Mothers? NORA – Nordic Journal of Feminist and Gender Research, 20(2): 94-108. Killewald, Alexandra (2012): A Reconsideration of the Fatherhood Premiun: Mar- riage, Coresidence, Biology, and Fathers’ Wages. American Journal of Sociology, 78(1): 96 -116 Knoester, Chris & Eggebeen, David J. (2006): The Effects of the Transition to Par- enthood and Subsequent Children on Men’s Well-being and Social Participation. Journal of Family Issues, 27: 1532-60 Kotsadam, Andreas & Finseraas, Henning (2011): The State Intervenes in the Battle of the Sexes. Causal Effects of Paternity Leave. Social Science Research, 40(6): 1611- 1622. Lalive, Rafael, Schlosser, Analía, Steinhauer, Andreas & Zweimüller, Josef (2011): Pa- rental Leave and Mothers’ Careers: the relative Importance of Job Protection and Cash Benefits. IZA Discussion Paper, 5792 Lundberg, Shelly & Rose, Elaina (2000): Parenthood and the earnings of married men and women. Labour Economics, 7: 689-710 Lundberg, Shelly & Rose, Elaina (2002): The Effects og sons and saughters on men’s labor supply and wages. The review of Economics and Statistics, 84(2): 251-268 Nielsen, Helena Skyt, Simonsen, Marianne & Verner, Mette (2004): Does the Gap in Family-friendly Policies Derive the Family Gap. Scandinavian Journal of Economics, 106(4): 721-744 Olsen, Bente Marianne (2000): Nye fædre på orlov. Sociologisk Institut, Københavns Universitet Olsen, Bente Marianne (2007): Evaluering af den fleksible barselsorlov. SFI 07:29 Patnaik, Ankita (2015): Reserving time for daddy: The short and long-run consequences of fathers’ quotas. Available at SSRN: http://ssrn.com/abstract=2475970 Pertold-Gebicka, Barbara, Pertold, Filip & Datta Gupta, Nabanita (2016): Employ- ment Adjustments around Childbirth. IZA Discussion Paper, no. 9685 Pylkkänen, Elina & Smith, Nina (2003): Career Interruptions due to Parental leave: A Comparative Study of Denmark and Sweden. OECD SOCIAL, EMPLOYMENT AND MIGRATION WORKING PAPERS, 2003(1) Rasmussen, Astrid Würtz (2010): Increasing the length of parents’ birth related leave: The effect on children’s longterm educational outcomes. Labour Economics, 17(1): 91-100 Rege, Mari & Soli, Ingeborg F., (2013): The Impact of Paternity Leave on Fathers’ Future Earnings. Demography, 50: 2255-2277

32 Schober, Pia S. (2014): Daddy leave: does it change the gender division of domestic work? DIW Roundup, 46 Staff, Jeremy & Mortimer, Jeylan T. (2012): Explaining the Motherhood Wage Penalty During the Early Occupational Career. Demography, 49(1): 1-21 Ugreninov, Elisabeth (2013): Can Family Policy reduce Mothers’ Sick Leave Absence? A Causal Analysis of the Norwegian Paternity Leave Reform. Journal of Family and Economic Issues, 34(4): 435-446