Master of Public Administration – Economics & Governance
Leiden University
Faculty of Governance & Global Affairs
Master Thesis
Pontus Korsgren
s2547996
The Effects of Individual Parental Leave Quotas on Parents’
Self-Reported Health and Wellbeing
Supervisor: Professor Max Van Lent
2
Keywords: Parental Leave, Parental Leave Quotas, Gender Equality and Labour Market Incentives, Subjective Health and Wellbeing, Regression Discontinuity Design
Table of Contents
Abstract ... 2
1. Introduction to the Research ... 3
2. Theory and Hypotheses ... 4
3. Institutional Setting: The Nordic Parental Leave Model ... 9
3.1. The Nordic Parental Leave Quotas ... 10
4. Methodology and Data ... 15
4.1. Estimation Strategy ... 16
4.1.1. Key RD Assumptions ... 16
4.1.2. Regression Discontinuity Design... 17
4.1.3. Equations and Descriptive Statistics ... 19
5. Results ... 25 5.1. Robustness Checks ... 28 6. Discussion... 41 7. Conclusion ... 44 Bibliography ... 48 Appendix ... 53
Abstract
Many governments are currently attempting to find ways of increasing fathers' share of parental leave, in order to enhance gender equality on the labour market and within the household. One policy approach that has been found to be successful, is earmarking parental leave exclusively for the father. In fact, following the recent EU Directive on The Work-Life Balance for Parents and Carers, all EU member states are required to implement such policies within the next three years. In light of the directive, this paper examines the effects of earmarked parental leave on parents’ self-reported happiness, health, life-satisfaction, job-satisfaction and work-life balance, by estimating the causal effects of the Nordic parental leave quotas. Using European Social Survey data, I apply a regression discontinuity design to compare Nordic parents’ who had a child just before the policy off, with those who had a child immediately after the policy cut-off. I find positive and statistically significant effects on high-income parents, older parents, and on parents with a high educational attainment. I also find that fathers in general experience a large and significant positive effect on how they manage to balance between their job and other aspects of life. Meanwhile, low-income parents, young parents and parents with a low educational attainment experience negative and significant effects on their subjective health and wellbeing.
3
1. Introduction to the Research
Family benefit systems are firmly established in virtually all modern welfare states. The goals are typically to encourage fertility and/or improve the wellbeing and opportunities of families and children (Gonzáles, 2013). In recent years, however, there has been a rising concern that many of the “old” family benefit systems are contributing to cementing gender biases on the labour market. Ruhm (1998), for example, finds that the gender wage-gap often increases due to parental leave. As a response, a growing literature is concerned with how to improve various areas of social policy, and most OECD countries are looking over their existing family benefit programs in light of that gender mainstreaming is emerging into the forefront of the policy debate (Ekberg, et al. 2013). At the European level, the European Council has adopted the Commission’s proposal for a Directive on The Work-Life Balance for Parents and Carers which “seeks to promote a good balance between family and professional commitments and to provide more equal opportunities for women and men in the workplace and at home” (The European Council, 2019). The proposal was adopted on the 14th of June 2019 and is also
included in the new European Gender Equality Strategy (2020-2025), which was launched on the 5th of March 2020. Two elements of the directive stand out:
i. paternity leave – fathers (or second parents) have been granted the right to 10 working days of paid leave around the time of birth of a child.
ii. Earmarked parental leave – a paid, individual, parental leave quota where 2 months are non-transferable between the parents has been introduced.
All EU Member States now have three years to adopt laws, regulations and administrative provisions necessary to comply with the directive. Therefore, we can expect that many European countries will be implementing new family benefit systems within the near future. One region that has been vocal in the debate leading up to the new EU directive, and where several relatively generous family benefit systems already have been introduced, is the Nordic countries. As described by Ellingsæter & Leira (2006), the Nordic welfare states have been forerunners in transforming “parenthood” into “political issues”, through multileveled policy-interventions aimed at gender and family arrangements. Already in 1993, the Swedish government, for instance, announced that;
“It is important that fathers take parental leave. An increased use of parental leave by fathers should contribute to a change in attitudes among managers; they will view parental leave as something natural to consider when planning and organizing the work. This change in attitudes is necessary for both men and women to dare to take parental leave without a feeling of jeopardizing their career or development opportunities at work. Another reason for increasing fathers' use of parental leave is that women's prospects of achieving equal opportunities to men in the labour market will be limited, as long as women are responsible for practical housework and children. A shared responsibility for the practical care of children would mean a more even distribution of interruptions in work between women and men, and women would thereby gain better opportunities of development and making a career in their profession.” From Government Bill, 1993/94:147 to the Swedish Parliament, translation by Ekberg, et al. (2013)
4 In other words, according to the Swedish Government it is essential to make parental leave more equally divided between the parents (which in practice means increasing the father’s share) if we are to effectively reduce gender inequalities. Specifically, by increasing the share of leave taken by fathers we can enhance gender equality within the household, reduce some barriers to parenthood and strengthen women’s position on the labour market (see Orloff, 2009; Ruhm, 1998; Thévenon & Solaz, 2013). Importantly, while it has been found that general parental leave benefits tied to pre-birth earnings often widen the wage-gap within the household (Ginja et al., 2018; Ruhm, 1998), individual earmarked parental leave for the father can be expected to increase the women’s intra-household share of the labour income (Duredahl et al., 2019) and an increase in leave-sharing is associated with better career outcomes for the mother (Nielsen, 2009). Furthermore, providing fathers’ with access to flexible parental leave in the immediate postpartum period has been found to improve maternal health – especially childbirth-related complications reduce (Persson & Rossin-Slater, 2019). Yet, relatively little is known about the effects of a more equally divided parental leave on the parents’ health and wellbeing measured over a longer time-period.
In light of the new EU directive, this paper aims to contribute towards filling this gap, by studying what effects earmarked parental leave is likely to have on parents’ self-reported happiness, health, life-satisfaction, job-satisfaction and work-life balance. This is done by applying a regression discontinuity design to estimate the causal effects of the Nordic parental leave quotas. Specifically, Denmark’s 1997, Finland’s 2013, Norway’s 1993, and Sweden’s 1995 and 2002 parental leave reforms are studied.
2. Theory and Hypotheses
Gender equality is a fundamental principle of the EU. In fact, Article 3 of the Treaty on the European Union stipulates that the EU shall actively promote equality between men and women. Further, Article 23 of the Charter of Fundamental Rights of the European Union requires “equality between men and women to be ensured in all areas, including employment, work and pay”. In line with these goals, the new EU directive on work-life balance for parents and carers aims to “increase the participation of women in the labour market and the take-up of family-related leave and flexible working arrangements […meaning] that parents and carers will be better able to reconcile their professional and private lives, and companies will benefit from more motivated workers” (The European Council, 2019). Importantly, the EU hopes that the quotas will achieve more “equal sharing of caring responsibilities between men and women, and the closing of the gender gaps in earnings and pay” (Ibid). In other words, the EU hypothesises that earmarked parental leave will increase fathers take-up of household responsibilities, meaning that the mother is able to resume her career earlier following child-birth. In the long run, this is expected to reduce underlying gender biases and discrimination on the labour market, as employers will change their belief that women are more likely to drop out from the labour force.
5 Indeed, this paper is not primarily concerned with exploring how the introduction of parental leave quotas impact on the labour supply of men and women or on the gender pay-gap. It is interested in the policy effects on health and wellbeing, meaning that the overarching goal is to investigate the EU’s hypothesis that earmarked parental leave will improve parents “motivation”, ability to “reconcile their professional and private lives” and overall wellbeing. This can be done by either measuring subjective (i.e. self-reported) or objective (involving indicators such as life expectancy, per capita income, etc.) outcomes (see Deeming & Jones, 2015; Sudhir & Sen, 1994). Most existing research papers have studied objective measures, by using longitudinal administrative data. However, subjective wellbeing has become an increasingly important topic in public policy debates, and improving subjective health and wellbeing is emerging as a key societal aspiration (Deaton et al., 2015, p640). For example, in
The Report on the Measurement of Economic Performance and Social Progress, economists
Joseph Stiglitz, Amartya Sen, and Jean Paul Fitoussi conclude (based on an extensive literature review) that most economic research “have assumed that it is sufficient to look at people’s choices to derive information about their wellbeing, and that these choices would conform to a standard set of assumptions. In recent years, however, much research has focused on what people value and how they act in real life, and this has highlighted large discrepancies between standard assumptions of economic theory and real-world phenomena” (Fitoussi, Sen & Stiglitz, 2009, p43). Further, they argue that objective measures are insufficient to capture social progress and misses key information about people’s quality of life. Therefore, “statistical offices and researchers should focus more on incorporating questions to capture people’s [subjective] life evaluations, hedonic experiences and priorities” (Ibid, p16). Hence, to contribute to this relatively understudied side of health and wellbeing, this paper focuses on subjective measures. Naturally, only the individual respondent can give information about his or her subjective states and values. Yet, an extensive literature on health and wellbeing1 shows that such indicators can relatively accurately predict individuals’ behaviour, and self-reported values are found to correlate with electrical readings of the brain (Ibid, p43). Typically, research papers that use subjective indicators try to cover how individuals evaluate their life as a whole or specific areas, such as family, work and financial circumstances. For example, Wassell (2015) has studied how different maternity leave policies impact on self-reported measures of happiness and self-reported life-satisfaction. She finds a significant negative correlation between weeks of full pay before the maternity leave and life-satisfaction outcomes (p157). Wassell (2014) has also examined how different family care leave policies impact on long-term wellbeing outcomes of children. Her results confirm that paid maternity leave increases the immediate health of infants, and increases the self-reported physical and mental health of mothers (p85).
Following the school of research that focuses on subjective health and wellbeing, this thesis estimates the effects of the Nordic parental leave quotas on self-reported measures of general happiness, health, life-satisfaction, job-satisfaction and work-life balance. The reason for specifically studying these outcomes are because several recent research papers have showed
1See for exampleDeaton et al., 2015; Deeming & Jones, 2015; Diener & Tov, 2012; Engster & Stensota, 2011; Clark &
6 that happiness and satisfaction indicators are vital for understanding how institutional conditions and policies influence individuals’ health and wellbeing (Frey & Stutzer, 2002; Wassell, 2014). And arguably, self-reported measures are most important, since it is impossible for a third person to decide whether another person is happy or satisfied with his/her job, life, free-time, etc. (Wassell, 2014, p36). Furthermore, previous studies have found health to be a particularly important variable for explaining wellbeing (Clark & Oswald, 2002). For this purpose, subjective health is important to measure, “because it captures more than what one health test can measure” (Wassell, 2014, p33). Therefore, I test if these indicators were improved by the Nordic parental leave quotas, and define hypothesis 1 as follows:
H1) Parental leave quotas improve parents’ self-reported health and wellbeing.
As explained above, the key theory underpinning H1 is that parental leave quotas alter the take-up of maternal and paternal leave, and by doing so, they improve the parents’ health and wellbeing. However, the connection between leave take-up, health and wellbeing, is not intuitive. In the remainder of this chapter, I therefore elaborate on the causal mechanisms.
First of all, to understand the mechanisms, it is important to understand how parents’ can be expected to allocate parental leave between themselves. This can be done through a theoretical model developed by Ekberg, et al. (2013). For simplification purposes, we can understand the framework by considering a household consisting of one child (c), one mother (m) and one father (f). In this setting, the household’s total utility (UHH) is dependent on the total days of
parental leave (d) taken out by each parent and the households total net income (y):
UHH (d, y) = Umf (y) + Uc (d)
However, the total utility is also subject to a budget constraint (BC) and a parental leave constraint (PLC). In economic models, the BC represents all the combinations of goods and services that a person can purchase given current prices and his/her given income. In this applied case, the BC refers to all consumption and saving necessary for the household, and is dependent on the wage of the respective parent, the replacement rate (the percentage of the parent's annual employment income that is replaced by the parental leave income) and the number of days that each parent takes in parental leave. For the sake of simplification, we can assume that the replacement rate is 80 percent. The PLC simply refers to that the parents cannot take out more leave than the total number of leave days available to them:
BC(y) = Wagem (1 – 0.2 * dm) + Wagef (1 – 0.2 * df)
(PLC) Maximum number of parental leave days ≥ dm + df
Notably, this model makes a number of additional assumptions. Particularly that parents do not have any preference concerning working or spending time at home – they only care about how much they can consume with their disposable income. Additionally, the child is assumed to
7 have zero consumption by him/herself and is indifferent about spending time with the mother or the father. Under these assumptions, it is rational for parents to allocate most (or all) of the available parental leave days to the parent with the lowest opportunity cost (that is the person with the lower wage). Due to underlying gender biases and discrimination, this tends to be the mother – an assumption that is applied in the formulas and equations defined below. In fact, the highest earner (usually the father) can be expected to only take parental leave days if some UHH(df) ≥ UHH(Wagef). From this we can derive the following expectation: all else being equal,
the amount of the parental leave taken by fathers should be higher, the higher the mother’s income is – since replacing the fathers wage with parental leave will generate less of a loss for the household.
I now apply this framework on the situation where a one month parental leave quota has been implemented. In this setting, an additional individual PLC is introduced, meaning that each parents´ total days of parental leave is capped (𝑑̅). For example, before the reform parents might have had 365 parental leave days to allocate between themselves. Now, the reform stipulates that 30 days are earmarked for each parent. This implies that the lowest earner can maximum take 335 of the available leave days. Since the 30 days reserved for the highest earner are provided on a “use it or lose it” basis, the reform increases the cost of not using them, since the days cannot be reallocated to the lower earner. Therefore, we can formulate the following expectation:
U(𝑑̅f) ≥ U(df).
As such, we can assume that the reform increases the probability that the highest earner (the father) takes more parental leave days, and that the lower earner (the mother) takes fewer parental leave days. However, the mother is still likely to take the majority of the total parental leave period (at least in a short to mid-term perspective). Thus, the father can be expected to increase his proportion of parental leave relatively more than the mother decreases her proportion (since an increase from 0 days to 30 days is a large increase in percent, while a decrease from 365 days to 335 days is relatively small). This has an important implication for modelling the policy effects on health and wellbeing: parental leave quotas are predicted to have a larger impact on fathers than on mothers. Consequently, I also expect that the policy has a larger impact on fathers’ health and wellbeing outcomes. Using classic economic theory, this can be explained by the law of diminishing marginal utility. Precisely, that the first unit of consumption of parental leave generates more utility for the individual than the second and subsequent units. Therefore, when fathers go from taking almost no parental leave to taking at least some leave, this has a larger impact on utility than the effect of mothers going from taking a large amount of parental leave to taking marginally less parental leave. The same applies for the effect of mothers increase in labour supply, as it is a very marginal increase in relation to her total working life. Accordingly, hypothesis 2 is formulated as followed:
H2) Parental leave quotas have a larger effect on fathers’ than on mothers’ self-reported health and wellbeing.
8 The theoretical considerations above also imply that different subgroups are likely to react differently to the reform. Income has stood out as an especially important covariate, since it is expected to have a strong direct impact on the parents’ allocative decision. Intuitively, income change has a positive effect on consumption and utility, which should translate into a positive effect on subjective wellbeing. Also, the existing literature has found strong evidence of a significant positive effect of income changes on health, both across and within countries (see Adler et al., 1994; Case, 2001; Chetty et al., 2017; Van Doorslaer et al., 1997; Frijters et al., 2005). In this regard, we should expect that parents’ with higher incomes have higher self-reported health and wellbeing. However, this is true regardless of the existence of parental leave quotas. Therefore, what is more important for this study is how income changes impact on the probability that the mother takes less parental leave and that the father take more parental leave. Although relatively little is still known about the associations between individual characteristics and parental leave, Andersson, Duvander and Evertsson (2020) have studied how take-up of paternal leave is influenced by socioeconomic and demographic characteristics. Specifically, the authors study the differences in relative changes in fathers’ take-up of parental leave over the period 1993-2010 in Sweden, by using Swedish register data from the Social Insurance Agency (individual-level data and longitudinal information on socioeconomic and demographic features of all Swedish residents). Notably, they find that “as the parental leave benefit is related to previous earnings, the take-up of parental leave is quite different for fathers with high and low incomes” – fathers with lower incomes have substantially higher probability of taking no leave than fathers with higher incomes (Ibid, p372). The authors also find that “younger fathers had a depressed probability of taking a leave of more than two months” (Ibid, p378). The most feasible explanation for these trends is that low-income parents and young parents in the beginning of their careers, are more restricted to take leave due to unstable labour market situations. A second important finding is that if the mother has a low income, the father is especially likely to take no leave at all, while he is more likely to take a long leave duration if the mother has a high income (Ibid). Indeed, this finding is in line with the theoretical considerations above: parents will allocate most of the available parental leave days to the parent with the lowest opportunity cost. Therefore, if the mother has a high income, the household has stronger incentives to allocate parental leave to the father. A final central finding is that “fathers with the lowest educational attainment (below secondary education) have a 28% higher risk of taking no leave at all than fathers with a secondary education, while the highest educated fathers (with more than two years of tertiary education) have 67% higher odds of taking a leave of more than two months, as compared to those with a secondary education” (Ibid, p370). Therefore, educational attainment also stands out as an important determinator. One explanation for this is that higher education promotes positive attitudes towards gender equality and sharing household responsibilities more equally between the mother and the father. A second explanation is that parents with higher education also have stronger labour market positions, meaning that they have fewer labour market constraints to take leave and stronger bargaining positions in relation to their employers.
Based on Andersson’s, Duvander’s and Evertsson’s (2020) research, I expect that income, age and educational attainment are positively correlated with the probability that the father takes more parental leave, and that the mother takes less leave. Therefore, I expect that younger, less
9 educated and poorer parents are less affected by the parental leave reforms, since they are still relatively likely to allocate most of the leave to the mother. This means that the hypothesized positive effects on health and wellbeing will be more pronounced among older parents, high-income parents, and parents with a high educational attainment. As such, hypothesis 3 states that:
H3) The parental leave reforms especially benefit high-income, older and highly educated parents.
Lastly, as having several children logically increases the number of parental leave days that parents need to take out, it is a reasonable expectation that the number of children is an important determinator. Explicitly, several children require more time off work than having only one child. But at the same time, having several children can be assumed to negatively impact on the parents’ disposable income, due to increased child-related costs. In this sense, parents are faced with a dilemma – they are required to take more leave, while they are also in need of a higher household income. In this situation, it is increasingly rational for the parents to allocate most of the available parental leave days to the parent with the lowest opportunity cost/the lower wage. This is because the highest earners’ wage should generate more utility for the household than his/her time off work → UHH(df) ≤ UHH(Wagef). This implies that parents
with many children are expected to be less likely to divide the parental leave more equally between themselves, than parents with fewer children. Therefore, parents’ with several children are expected to benefit less from the parental leave reforms. Accordingly, hypothesis 4 is formulated as followed:
H4) Parental leave quotas especially benefit parents with fewer children.
To investigate the four hypotheses, I first run regressions where I estimate the overall treatment effect on the full sample of parents. Next, I run five separate robustness checks to see if the results are sensitive to variation in the key variables defined in H2, H3 and H4.
3. Institutional Setting: The Nordic Parental Leave Model
The Nordic welfare states have been categorized by Esping-Andersen as a social democratic regime, distinct from the liberal and conservative ones (Esping-Andersen, 1990). Their parental leave model stands out due to the strong emphasis on creating conducive conditions for mothers’ to combine family and work (Esping-Andersen, 2002). In existing literature, this model is often described as a “best practise” when it comes to gender-mainstreaming, and empirical studies have found that the large negative impact of having children on the mothers’ labour supply is significantly smaller in the Nordic countries than in most other OECD countries (Smith et al., 2003). Researchers have also found that Nordic mothers’ have a particularly high probability (well over 90 percent) to return to work after taking parental leave (Pylkkänen & Smith, p2, 2004). Ray et al. (2009) have conducted an extensive review of the
10 national parental leave policies of 21 high-income countries, focusing on “the level of support provided to parents” and the“degree to which leave policies promote an egalitarian distribution between mothers and fathers of the time devoted to child care”. They find that the Nordic countries stand out as the countries with the highest generosity and most gender equal policies (Denmark scores high, although slightly lower than its Nordic neighbours). Specifically, they stand out due to the generosity of the paid leave, universal coverage combined with modest eligibility restrictions, financing structures that pool risk among many employers, scheduling flexibility and paid parental leave quotas (Ray et al., 2009, p19-21). While many countries have introduced one or two weeks of paternity leave in connection with the birth of a child (several also offer paid leave), surprisingly few countries provide fathers with longer earmarked leave after the two first weeks following child birth. The Nordic countries have, however, established this form of paid individual leave for fathers.
In 1974, Sweden became the first country in the world to provide fathers with the right to parental leave. Fathers were given the right to take out three of the six paid months of parental leave, but this right could be (and in practise almost always was) transferred to the mother. Norway became the second country to provide fathers with the right to take parental leave: in 1978 they granted parents the right to share 12 of the 18 weeks of the maternity leave between themselves. Finland followed its Nordic neighbours in 1980, by giving fathers the right to use the last four weeks of the parental leave, and in 1984 Denmark granted fathers the right to share up to 10 weeks with the mothers (Haas & Rosgaard, 2011, p179-180). Table 1 illustrates when fathers were granted the right to parental leave and when parental leave quotas were first established in each of the Nordic countries.
Table 1: Paternal Leave and Parental Leave Quotas in the Nordic Countries
Denmark Finland Norway Sweden Paternal Leave 1984 1980 1978 1974
Parental Leave Quota 1997* 2003** 1993 1995
*abolished in 2002.
**conditional parental leave quota.
3.1. The Nordic Parental Leave Quotas
What arguably makes the Nordic parental leave model stand out are the parental leave quotas (see Haas & Rosgaard, 2011; Ray et al., 2009; Smith et al., 2003). These are reforms that earmark parental leave exclusively for each of the parents on a ‘use it or lose it’ principle. So, if the father does not take his earmarked leave, the leave-period is lost to the household meaning that the households total parental leave period will be shortened. In this sense, these quotas are designed to give fathers a strong incentive to take out parental leave.
11 Parental leave quotas were first introduced in Norway, when the Government introduced a reform that stipulated that fathers who had a child on after the 1st of April 1993 would be
allocated four weeks of the parental leave and that this period could not be transferred to the mother. It immediately turned out to be successful in dividing the leave more equally between the parents, illustrated by the large increase in paternal leave that followed (Branth & Kvande, 2011, p3). Cools et al. (2015) have studied how the 1993 reform impacted on a broad range of outcomes, using Norwegian register data. They find that the reform drastically increased the number of fathers taking out parental leave. In their sample of families, the fraction taking paternity leave immediately increased from 4% in March 1993 to 39% in April 1993. They also find that the fraction kept increasing over time, to above 80% in 2005.
In 1995 Sweden became the second country to introduce a parental leave quota by also reserving one month of the parental leave exclusively for the father, and in 2002 the Swedish Parliament passed a second reform which extended the quota by one additional month. Ekberg, et al. (2013) and Eriksson (2006) have investigated how these reforms impacted the take-up of parental leave by fathers, whether fathers changed their behaviour in terms of household work and child care and whether there were any long-term effects on the labour-market outcomes of mothers and fathers. They find a strong effect on fathers’ take-up of parental leave. Following the 1995 reform, the share of fathers taking no days of parental leave decreased from 54% to 18% and the average amount of parental leave taken by fathers increased by roughly 50% (similar in size as the effects of the 2002 reform). The reforms were thus highly successful in increasing fathers’ take-up of parental leave. However, Ekberg et al. (2013) find no evidence that fathers who take more parental leave take more leave for care of sick children measured over an eight-year period. They also studied effects on long-term wages and employment within the household, without finding evidence for substantial effects of the reform. Hence, although the reforms increased fathers use of parental leave, it can be questioned if they succeeded with increasing gender equality. Nonetheless, the reforms can be said to have made the parental leave more equally divided between the parents.
Inspired by Norway and Sweden, Denmark gave fathers of children born on or after the 15th October 1997 two weeks of earmarked parental leave. However, the policy turned out to be controversial in Denmark as there was no political consensus on what the policy should target (gender equality vs work flexibility). As a consequence, the quota was abolished in the beginning of 2002 (Haas & Rosgaard, 2011, p180). Yet, many trade unions continued to advocate for parental leave quotas, and as a result, all public employees have since year 2008 the right to six months of earmarked paid leave. In this sense, the parental leave quota was re-installed, but only for certain sectors (Eydal et al. 2015, p174). Duredahl et al. (2019) have studied the effects of Denmark’s 1997 reform using longitudinal administrative Danish register data. They find a sharp increase in fathers’ take-up of parental leave at the policy-cut-off. They also find a statistically significant positive effect of the reform on mothers’ share of the household’s net labour income, which increased by approximately 1.2 percentage-points (Drudahl et al., 2019, p86-88).
12 The abolished parental leave quota in Denmark is interesting, as it provides an additional natural experiment. Specifically, it allows the researcher to compare parents who had a child just before the abolishment of the quota (the control group), with parents’ who had a child just after the parental leave reform (the treatment group), and then compare these results with the estimated treatment effect of implementing parental leave quotas. Supposedly, abolishing the quota should reduce the incentive for parents to allocate parental leave to the father, since U(𝑑̅f)
≥ U(df). Therefore, the Danish 2002 reform can be expected to reduce the equal sharing of
parental leave, which should have the opposite effect on treated parents’ health and wellbeing compared to introducing parental leave quotas. Nielsen (2009) has used administrative register data covering the entire Danish population, to study “the causal effect of economic incentives by estimating how price sensitive the leave-sharing decision is” (p2). She finds that there was no reduction in fathers’ take-up of parental leave at the 2002 policy cut-off. In fact, there was an increase in the share of fathers taking out more than four weeks of leave by 9 percentage-points and the number of fathers taking no leave decreased by 4 percentage-percentage-points (Ibid, p8). However, this can be explained by the fact that the policy simultaneously extended the period with full family benefit compensation from 28 weeks to 50 weeks. As such, while there was a small increase in fathers’ take-up, there was a large increase in mothers’ take-up – the fraction of mothers using more than 24 weeks increased by about 30 percentage-points (Ibid). Therefore, even though the policy slightly increased fathers’ take-up of parental leave, the policy can be argued to have contributed to making the parental leave less equally divided between the parents, since the increase in mothers’ take-up was so much larger. In the Appendix (table A.12), I estimate how Denmark’s 2002 policy-reform affected the self-reported health and wellbeing outcomes of treated parents. Although the sample of Danish parents exposed to the specific policy-treatment are too few to support strong conclusions, the results give an indication that abolishing earmarked parental leave had the opposite effects, compared to introducing such quotas. Particularly, while I find evidence that earmarked parental leave improved parents’ self-reported work-life balance and job-satisfaction (as will be described in detail in the discussion), it appears as if Denmark’s 2002 reform may have had a negative impact on these outcome variables.
Unlike its Nordic neighbours, Finland opted for a conditional parental leave quota. Specifically, in 2003 they introduced the so called ‘fathers month’ that stipulated that father would receive two additional weeks of leave if they take a minimum of two weeks of the shared parental leave. This was extended to four “bonus” weeks in 2010 (Haas & Rosgaard, 2011, p181). However, the conditional parental leave quota was less successful than the policy-makers had hoped, as there only was a modest increase in fathers’ take-up of parental leave at the cut-off – in 2007 only 11.9% of fathers took out parental leave, compared to 2.6% in 2002 (Kamerman & Moss, 2009, p96). Further, although the father's month gradually increased in popularity, its actual contribution towards making the parental leave more equally divided between the parents is questioned by the fact that most Finish mothers still stayed at home during the father's month (Salmi & Lammi-Taskula, 2013, p115). As a consequence, Finland abolished the conditional quota on the 1st of January 2013, and replaced it by a 9 week earmarked (individual)
paternity leave of which only 3 weeks can be taken simultaneously with the mother (Lammi-Taskula, 2017, p92). In this sense, the policy practically introduced a parental leave quota.
13 Salmi & Lammi-Taskula (2019) have found that the proportion of fathers taking out parental leave significantly increased by the reform. Just before the policy was introduced, about 32% of all fathers took out his earmarked parental leave; by 2015 the proportion had increased to 50% (p12-13).
One question that comes to mind is if we can infer results from the Nordic countries and their parental leave model onto a broader population of countries, or at least onto all EU member states which now are required to introduce parental leave quota systems. Afterall, it is conceivable that gender norms are so much more egalitarian in the Nordic countries than elsewhere so the effect of more equal leave has a substantially different effect on Nordic parents’. Although the effects are likely to vary somewhat between countries, Kleven et al. (2019) find that the difference in attitudes may be exaggerated. Even in the Nordic countries attitudes are still relatively traditional, with a surprisingly large proportion responding that women should not be working full time when they have children living at home. Inspired by Kleven’s et al. (2019) analysis of gender roles in Sweden, Denmark, the US and the UK, I use data provided by the International Social Survey Programme (ISSP) to compare attitudes towards gender equality and equal sharing of child-care responsibilities across Europe. The ISSP provides an especially appropriate source for this analysis, as the 2002 survey round covered the topic “family and changing gender roles”. Specifically, I select one country from each of Esping-Andersen’s (1990) three welfare state regime types, to get an idea of how attitudes differ between the type of countries: Sweden is selected from the social democratic (Nordic) model, the United Kingdome from the Anglo-Saxon model and Germany from the continental model. Additionally, I include one Mediterranean country into the analysis (Spain), as Esping-Andersen has been critiqued for not covering these countries in his typology (see for example Arts & Gelissen, 2002). I also include one of the “Eastern accession” EU member state (Latvia) to create a more inclusive spread. The four questions below are used to study attitudes towards gender equality and sharing of household responsibilities, and Figure 1 illustrates how the attitudes differ between the countries.
1. “do you think that women should work outside the home full-time, part-time or not at all when they are married but with no children?” (panel A)
2. “do you think that women should work outside the home full-time, part-time or not at all when there is a child under school age?” (panel B)
3. “do you think that women should work outside the home full-time, part-time or not at all when the youngest child is still in school?” (panel C)
4. “do you think that women should work outside the home full-time, part-time or not at all if the child has left the home?” (panel D)
14
A) Married but with no children B) There is a child under school age
C) The youngest child is still in school D) the youngest child has left home
Figure 1: Attitudes Towards Gender Equality across Europe
Figure 1 illustrates that there indeed are differences in attitudes towards gender equality across Europe. For example, the Swedish population is more open to the idea that women with young children can work part-time, while a relatively large share of the UK population is of the opinion that the mother should stay at home completely to take care of young children. However, the similarities in gender attitudes appear to stand out much more than the differences. This raises doubts about the degree to which Nordic countries are strong outliers in terms of attitudes towards gender equality within the household and on the labour market, and thus makes the results increasingly inferable onto other EU member states2.
2 In the Appendix (Fig A.1), I provide an additional illustration of how gender norms vary across countries. These graphs
recreate Kleven’s et al (2019) analysis of gender roles and attitudes in families with children in Sweden, Denmark, UK and the US.
15
4. Methodology and Data
In this paper, a multivariate analysis is performed on parents living withing the four Nordic countries Denmark, Finland, Norway and Sweden. For this purpose, the paper relies on European Social Survey (ESS) Data, collected as near as practically possible to each of the countries’ respective parental leave reforms. The ESS is a pan-European cross-sectional time-series survey running every two years – all of the Nordic countries have participated in every round. The ESS asks respondents many questions about self-reported health and wellbeing and links answers at the individual level via personal identification numbers. Out of particular relevance for this paper are the five interview questions summarized in table 2 below. In the quantitative analysis, these function as the outcome variables of interest.
Table 2: ESS Outcome Variables
Outcome Variable ESS Question Scale
1. Happiness “How happy are you?” 1 – 10
2. Health “How is your health in
general?”
1 – 5
3. Life-satisfaction “How satisfied are you
with life as a whole?”
1 – 10
4. Job-satisfaction “How satisfied are you
with your job?”
1 – 10
5. Work-life balance “How satisfied are you
with how you balance between time on your job and time on other aspects?”
1 – 10
Besides subjective health and wellbeing, the ESS contains rich information on children, labour supply, occupation and many other variables. For example, we are able to identify the gender and age of the respondent and other household member’s relationship to the responded. We are also able to identify and measure potentially important covariates, such as educational attainment, labour status, if the person is an immigrant, etc. Importantly, the 2006 survey-round (ESS3) included the module titled “the timing of life”, which was later repeated in the 2018 round (ESS9). The module aims to understand the views of European citizens on the organisation of the life course and of their strategies to influence and plan their own lives. It also includes measures on youngest age and oldest age of life events, planning for retirement and the timing of key life events. Of particular importance for this study is the variable “year (first) child was born”. Therefore, the ESS3 and the ESS9 are appropriate dataset for this paper, as we are able to identify Nordic parents who had their first child around the respective policy reform and their self-reported health and wellbeing.
16
4.1. Estimation Strategy
The estimation strategy used in this paper exploits the sharp eligibility cut-offs established by the reforms. Furthermore, it exploits the assumption that birth can be seen as a random event where “nature” determines if a child is born into the control group or into the treatment group (González, 2013, p161). For this setting, the Regression Discontinuity Design (RD) is particularly applicable for estimating the effect of the policy, as the cut-off determines the group assignment. More specifically, I apply a ‘Multiple Cut-Off RD Design’, as we have five different policy thresholds. The approach is based on pooling the data and normalizing the score so that the cut-off is zero for all units. This allows us to apply the standard RD design based on the single cut-off approach to make the estimates.
4.1.1. Key RD Assumptions
Lee and Lemieux (2010) have written an extensive review of RD designs in econometrics and a guide for researchers interested in applying the method. To avoid important identification, interpretation, and estimation issues, this paper relies on their recommendations concerning RD assumptions.
Firstly, it is crucial that individuals cannot manipulate the treatment. For this paper, the key
question is if the ‘birth of the first child’ can be manipulated by the parents. Although this is a valid and important concern for the identification strategy, the timing of conception cannot be perfectly controlled by the parents as the duration of pregnancy is normally distributed with a mean of 40 weeks and a standard deviation of 2 weeks (Ekberg, et al. 2013, p135). It is impossible for parents to postpone child-birth. Parents can hypothetically trigger the birth, but this is not a concern as “triggering birth (except for health reasons) is considered highly unethical and against professional standards” (Ibid) in the Nordic countries. Still, to test for manipulation I run initial regressions to investigate if there is a discontinuity in number of first born children at the policy cut-off. I also investigate if the other variables that were determined before the realizationof the treatment variable have the same distribution just before and just after the policy cut-off. “If there is a discontinuity in these baseline covariates, then at a minimum, the underlying identifying assumption of individuals’ inability to precisely manipulate the assignment variable is unwarranted” (Lee & Lemieux, 2010, p283). More concretely, I run regressions where the dependent variables are different household characteristics that the existing literature identifies as potentially important for fertility and health/wellbeing outcomes (see for example Ang, 2015; Baker & Milligan, 2014; Ginja et al., 2018; Gonzáles, 2013; Persson & Rossin-Slater, 2019; Schellekens, 2009; Whittington et al., 1990). As a final validity check, I also perform placebo tests by measuring the effects of artificial parental leave reforms implemented two years (t – 2) before the real implementation dates. If we find significant effects on the health and wellbeing outcomes, this would indicate that there is no real effect of the studied parental leave quotas.
Secondly, when the researcher chooses a parametric functional form (e.g. a low-order
17 nonparametric approach (such as local linear regression) may also generate bias (Lee & Lemieux, 2010, p284). In this paper, it is difficult to assess which functional form will generate the least bias, due to the finite sample size. To get around this, Lee and Lemieux (2010) recommends to not rely on one particular method or specification. Rather, the researcher should run separate regressions using different functional forms, as “results that are stable across alternative and equally plausible specifications are generally viewed as more reliable than those that are sensitive to minor changes in specification” (p285). Therefore, I run both linear and polynomial regressions to estimate the effect of the policy reforms. An equally important consideration concerns what bandwidth to choose – it should be wide enough to reduce the amount of noise, but narrow enough to compare observations “close enough” on both sides of the threshold (Ibid, p309). The optimal bandwidth for this paper is to look at the year just before and just after the reform. Yet, I run additional regressions with wider bandwidths to test how the coefficients and their statistical significance level change depending on the specification.
An important challenge for the analysis is that the Norwegian reform was implemented on the 1st of April 1993 and the Danish reform was implemented in mid-October 1997. This is problematic, as the ESS does not report the exact moth of birth – it only reports year of birth, which limits our ability to detect a discontinuity right at the policy implementation date. To approximate, I assume that all Norwegian parents who had a child in 1993 or later are treated, even though three of these months are before the actual cut-off date. Similarly, I approximate that all Danish parents who had a child in 1997 are in the control group, and only those who had a child in 1998 or later are treated, even though there are two and a half months in 1997 that are in fact in the post-reform period. Although I recognize that this impacts on the validity of the estimations, these are necessary assumptions to make due to the data limitations. For Sweden and Finland, however, we are able to use the exact policy cut-offs as these reforms were implemented on the 1st of January.
4.1.2. Regression Discontinuity Design
Inspired by Cattaneo’s et al. (2016) paper on Interpreting Regression Discontinuity Designs
with Multiple Cut-offs, this paper normalizes the threshold to zero for each of the policy
cut-offs, then indexes all years relative to that year. By doing so, we are able to study all of the policy reforms simultaneously, regardless of when the policy was introduced, using the standard single cut-off design. In this approach, Y0i denotes the outcome when a parent is not
exposed to the treatment, and Y1i denotes the outcomes when they are exposed. Year of birth of
the first child is the running variable, and Di is the treatment indicator. The reason for using
year of birth of the first child as the running variable rather than the exact date of birth, is that
the ESS does not report on the exact date of birth. Nonetheless, even if we were able to run the regressions using the exact dates of birth, the limited number of observations would not allow us to look much closer to the policy cut-off. Similarly, the reason for basing the running variable on the birth of the first child, rather than basing it on the birth of any child, is because the ESS module on ‘the timing of life’ only explicitly studies when parents had their first child.
18 that any parent who had a child on or after Ci will be treated. Hence, the assignment is
dependent on both the running variable and the cut-off variable. As already described, however, I normalize the running variable, denoted as Ybi = ‘Year of birth of first child’ – Ci.Next, I pool
the data as if there only is one single cut-off (Ci = 0). So, ybi is the year of birth of the parent’s
first child relative to the policy cut-off which equals zero. This entails that if ybi ≥ 0, parent i is
assigned to the treatment group; if not, he/she is assigned to the control group. We can summarize these assumptions in the expectation function below, illustrating how the expected outcome of interest is dependent on the normalized running variable and the treatment indicator:
E[Yi ⃒ ybi, Di]
Having defined the expectation function, I now move on to elaborate on the strategy used to estimate the treatment effect of the policy reforms. Formally, the treatment effect (T) captures the outcome of treated parent i minus the contrafactual outcome of untreated parent i → Ti(Ci)
= Y1i(Ci) – Y0i (Ci). In any RD design, the contrafactual situation for parents just above the
cut-off refers to the parents that are right below the cut-cut-off. However, as I have normalized the running variable and pooled the data, we can assume that there only is one single cut-off, meaning that all units will face the same treatment effect denoted T(C). Under this assumption, the approach is directly comparable to the single cut-off RD design, “where the assumption of homogeneous treatment effects leads to the identification of the overall constant effect of treatment” (Cattaneo et al. 2016, p1237). In this setting, the “point estimation amounts to fitting a weighted least squares regression of the outcome (Yi) on a polynomial basis of the running
variable for observations within a small region around the cut-off” (Ibid, p1241). Since I pool and normalize the data, the cut-off = C = 0, which means that the constant treatment effect is equal to the difference between the intercept of the polynomial (or linear) fits at the cut-off (since there is one fit for the pre-reform period, and one for the post-reform period).
A potential objection to this approach is that there is heterogeneity between the Nordic parental leave quotas. Most notably, they differ in geographical location and time of implementation. Therefore, the assumption that the treatment effect can be kept constant can be questioned. However, as I am interested in the effect of introducing a parental leave quota, the reforms are assumed to be similar enough to allow for this comparison. Still, to further limit the risk of omitted variable bias, I control for unobserved country fixed effects, by including distinct country intercepts into my regression model. As will be further explained below, these are estimated as the coefficients on country binary indicator variables.
19
4.1.3. Equations and Descriptive Statistics
(1) Yiyb = α + γ1 yb + γ2 (yb ⋅ Di) + β ⋅ Di + Π X iyb′ + FEc + ε iyb .
Y is an outcome variable for any of the self-reported health and wellbeing indicators for parent i who had his/her first child in the year yb. Yb is the running variable which is normalized to
zero for the year that the policy was introduced and thus takes values −1 for one year before the reform, 1 for one year after the reform, etc. α is the intercept and Di is a dummy variable
that takes the value 1 in all years after the reform and 0 otherwise. The main parameter of interest in the equation is β, which captures any potential discontinuity effect in Y at the cut-off. Notably, I use a linear term in yb that accounts for any smooth trends in Y, and it is allowed to change after the cut-off. However, as alluded to above, it is difficult to assess (due to the finite sample size) if parametric linear regressions capture the true functional form. Indeed, this is a well-studied problem in econometrics and statistics, and several nonparametric methods have been suggested to provide flexible estimates of the regression (Lee & Lemieux, 2010, p316). In applied studies, an especially common method used to relax the linearity assumption, is to include polynomials in the regression model (Ibid). In this regard, by experimenting with more flexible specifications by adding polynomial terms, we can better assess the robustness of the RD estimates of the treatment effect. Therefore, despite using the linear trend in my main specification, I also explore local polynomial RD point estimators with robust confidence intervals (an approach described in detail by Cattaneo and Farrell 2018; Calonico, Cattaneo, Farrell and Titiunik, 2019; and Calonico, Cattaneo and Farrell, 2020), and report the local polynomial estimations in the appendix (table A.5 – table A.11). Reassuringly, the results remain stable regardless of the specification, which makes them increasingly reliable.
The vector X includes a battery of controls. Based on a thorough literature review, I control for variables that are found to have an impact on fertility, the intra-household division of parental leave, health and wellbeing. Firstly, I control for key respondent characteristics, including age of respondent, age when respondent had the first child, two educational attainment dummies, a dummy for if the respondent is in any form of paid work and a dummy for if the respondent is an immigrant. Secondly, I control for a number of additional household characteristics, including household net income, number of children and two educational attainment dummies for the respondent’s partner. Thirdly, as already alluded to I control for fixed country effects, denoted as FE in country C. As such, the model includes distinct country intercepts which are estimated by generating separate binary indicators for each country, and including these as control variables in the regressions. By doing so, we control (to a certain extent) for any heterogeneity between the Nordic parental leave quotas, such as the time-invariant factors and other institutional factors that may differ, meaning that we reduce the risk of omitted variable bias. Lastly, ε is the residual.
In the main specification, Y is observed for parents who had their first child within one year on either side of the policy reform. In additional specifications, I experiment with bandwidths of two years before and after the reform and three years before and after the reform. Initially,
20 however, I run a validity check of the selected RD approach by investigating if the assignment is “as good as randomized” (Lee and Lemieux, 2010, p282), by using equation 2 below to check if the observables in the years before and after the treatment are balanced. The results are reported in table 4. As shown, equation 2 is almost identical to the main equation, except that the different key observable covariates now are the outcome variables of interest.
(2) Xiyb = α + γ1 yb + γ2 (ybi ⋅ Di) + β ⋅ Di + FEc + ε iyb
Table 3 reports important descriptive statistics for the sample of parents by their first child’s year of birth. The full sample includes all Nordic parents who had their first child between three years before to three years after the respective reform. I estimate specifications with the full sample, then restrict the bandwidth closer to the cut-off. As illustrated, 46.6% of the observed parents had a child in the post-reform period, meaning that 53.4% had a child in the pre-reform period. Out of these, 51% are male and 49% are female. On average, the parents were about 38 years old at the time of the interview and had two children. When they had the first child, they were on average 29 years. We also see that 73.3% were in some sort of paid work and 10.9% were born outside the country. Furthermore, 39.7% of the respondents had at least some university level education while 6.7% only had basic education. Concerning the respondents’ partners, 28.7% had completed university level education, while 5.6% only had basic education. We also see that 27.6% had a household income that was higher than the median household income in the country of residence, and 37.2% had a household income that was lower than the median household income.
Table 3: Descriptive Statistics
Mean SD Median
Respondent’s age3 38.0331 6.2075 38
Age when respondent had first child4 28.9612 5.2420 29
Respondent is male 0.5101 0.5003 1
Respondent is female 0.4899 0.5003 0
Respondent’s number of children 2.0043 0.7746 2
Respondent has at least some university level education
0.3966 0.4895 0
3
Calculated by subtracting year of birth from year of the interview
21
Respondent only has basic education5 0.0661 0.2486 0
Partner has completed university level education6
0.2874 0.4529 0
Partner only has basic level education7
0.0560 0.2302 0
Respondent is born outside the country 0.1092 0.3121 0
Respondent is currently in paid work of any kind
0.7342 0.4421 1
Income8 < median income in country of
residence
0.3721 0.4837 0
Income8 > median income in country of residence
0.2759 0.4473 0
Post-reform dummy 0.4655 0.4991 0
Year of first child’s birth -0.1695 2.2118 -1
Notes: The sample includes all Danish, Norwegian and Swedish parents that were interviewed in 2006 (ESS3) and all Finish
parents that were interviewed in 2018 (ESS9), who had a child any time between three years before to three year after each reform. The number of observations is 696. All numbers are rounded up to four decimals.
Figure 2 shows the number of first born children by yb over the full sample. As illustrated, both a linear function and a polynomial function is fitted separately for the years before and after the reform. Visually, it appears as if there is no discontinuity in number of births right at the policy cut-off. This is also confirmed by the regression results. Specifically, by using equation 1, but replacing the health and wellbeing outcome variable with the (natural log) number of children born, we find that there is no statistically significant discontinuity at the cut-off. However, after the reform it seems like there may be a positive trend in the number of first born children (made visible by fig. 2, panel A). Hence, although it is outside the scope of this study, the post-reform trend gives some indication that the reform may have had a positive effect on fertility within the following three years of the reform. This is an interesting hypothesis, as existing literature have found positive effects of increasing various family benefits on fertility (see for example Ang, 2015; Garganta et al., 2015; Gonzáles, 2013; Schellekens, 2009; Whittington et al., 1990), and Andersson and Duvander (2006) specifically find that increased paternal involvement in child care is positively related to continued
50 – 9 years
6
International Standard Classification of Education (ISCED) level 5-6 7
International Standard Classification of Education (ISCED) level 1-2
22 childbearing. Meanwhile, Farré and Gonzáles (2019) find that parents who were just entitled to a new paternity leave in Spain were less likely to have a second child within the following six years. Hence, for future research, it would be interesting to specifically study the fertility-effects of increasing and or implementing earmarked parental leave.
A) Linear fit showing number of first born children by yb B) Polynomial fit showing (natural log) number of first born children by yb
Figure 2: Number of First Born Children by Yb
Table 4 shows the balance in covariates estimated by using equation 2. Specifically, I report the coefficients for the treatment indicator taking value 1 for years after the reform and 0 otherwise. This means that the regressions compare parents who had their first child after the reform with parents who had their first child before the reform. Column 1 shows the main specification, where the sample is limited to those parents who had the first child within one year on either side of the reform. Column two shows parents who had their first child two years before to two years after the reform and column three shows the full sample. Additionally, figure 3 visually displays the key characteristics over the full sample.
Table 4 – Balance in Covariates
Column 1 +/- 1 year Column 2 +/- 2 years Column 3 +/- 3 years Respondent’s age -0.7430 (0.8308) 0.6693 (1.3521) 0.0211 (0.9808)
Age when respondent had first child 0.0243
(0.7400)
0.2261 (1.1892)
-0.1035 (0.8677)
Respondent has at least some university level education -0.0182 (0.0657) -0.0796 (0.1127) 0.0603 (0.0814)
23 Respondent only has basic education (0-9
years) 0.0032 (0.0347) 0.01200 (0.0599) -0.0210 (0.0419)
Gender X university level education (respondent) -0.0089 (0.1087) -0.0616 (0.1866) 0.0839 (0.1340)
Gender X basic education (respondent) -0.0043
(0.0530)
-0.0043 (0.0920)
-0.0338 (0.0645)
Partner has completed university level education -0.0601 (0.0588) -0.0689 (0.1035) -0.0280 (0.0742)
Partner only has basic level education -0.0147
(0.0346)
-0.0235 (0.0609)
-0.0347 (0.0421)
Gender X university level education (partner) -0.1354 (0.0922)
-0.1705 (0.1671)
-0.1040 (0.1180)
Gender X basic level education (partner) -0.0215 (0.0608)
-0.0478 (0.1048)
-0.04741 .0738239
Respondent is born outside the country -0.0039
(0.0424)
-0.0127 (0.0729)
-0.0072 (0.0520)
Respondent is currently in paid work of any kind -0.0575 (0.0605) -0.0312 (0.1013) -0.0578 (0.0733)
Household net income -0.0440
(0.1035)
0.1578 (0.1845)
0.0007 (0.1324)
Respondent has several children -0.0021
(0.0580) 0.0155 (0.0992) 0.0779 (0.0702) Observations 223 435 696
Notes: The coefficients reported are for the post reform binary indicator taking value 1 for years after the reform. Every
coefficient is from a different regression. Robust standard errors are shown in parentheses and the dependent variable is indicated in each row header.
*** Significant at the 2 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level.
24 The results indicate that the parents’ characteristics are relatively balanced around the policy cut-off, as there is no significant discontinuity in age, age when the respondent had the first child, number of children, educational attainment, employment, immigrant status or household net income. I also check the balance of the parents’ educational attainment interacted with a binary gender indicator, without finding any significant discontinuity. Therefore, we can conclude that controlled and treated parents are similar in these key observable covariates, meaning that the assignment is as good as randomized. Nonetheless, to conduct one final validity test of the RD design, I run placebo tests by assigning a placebo treatment to parents who had a child before the cut-off. If the Nordic parental leave quotas are the driving force behind the health and wellbeing results, we would expect to find no significant discontinuity effect on parents who were not exposed to the policy treatment. Therefore, since the parental leave reforms were implemented at t – 0, we should not find any statistically significant effect if we change the cut-off to t – 2. Setting the placebo treatment at t – 2 allows us to study parents who had a child within one year on either side of the placebo reform, and parents who had a child within two years on either side of the placebo reform. Henceforth, placebo estimates are included in all results tables below.
A) Age when respondent had first child | B) Respondent has university education | C) Respondent only has basic education
D) Respondent is born outside the country | E) Respondent is in paid work | F) Respondent has several children
G) Partner has completed university | H) Partner only has basic education | I) Household net income
25
5. Results
In the following chapter, I empirically investigate if the parental leave quotas had an effect on the five different health and wellbeing outcomes. Table 5 reports the regression results from estimating equation 1 with the three different specifications, using the (natural log) self-reported happiness, health, life-satisfaction, job-satisfaction and work-life balance scores. Specifically, in column 1 I compare parents who had their first child during the year before the reform with those who had the first child during the year following the reform. In column two, I compare parents who had their first child during the two years prior to the reform, with parents who had the first child during the two years following the reform; and in column 3, I compare parents who had their first child during the tree years before, with parents who had their first child during the three years following the reform. The two last columns report the placebo (t – 2) estimates. As I find that the parents’ characteristics are balanced over the full sample (see table 4 above), all specifications can be assumed to show how the reforms impact on parents health and wellbeing. Still, as parents who had a child within a narrow window around the cut-off are assumed to be most similar, the results in column 1 (and to an extent in column 2) are most reliable for drawing causal inference. The coefficients reported are for the binary treatment indicator taking value 1 for years after the reform and 0 otherwise. As I use the natural log of Yi while maintaining the linear trend in ybi, the estimated coefficients should be
interpreted as followed: if the treatment indicator changes from 0 to 1, this corresponds to a change in Yi equivalent to the particular coefficient, which can be expressed as a change in
percent.
Table 5: Health and Wellbeing Results
Column 1 +/- 1 year Column 2 +/- 2 years Column 3 +/- 3 years Placebo +/- 1 year Placebo +/- 2 years Panel A: Happiness -0.0256 (0.0275) -0.0716* (0.0438) -0.0340 (0.0331) 0.0058 (0.0297) 0.0255 (0.0480) Number of observations 223 434 694 256 472 Panel B: Health -0.0148 (0.0263) -0.0371 (0.0427) -0.0027 (0.0312) -0.0132 (0.0276) 0.0161 (0.0196) Number of observations 223 435 696 258 472 Panel C: Life-satisfaction -0.0143 (0.0282) -0.0856* (0.0489) -0.0320 (0.0349) -0.0049 (0.0318) 0.0246 (0.0547) Number of observations 220 429 689 256 468
