A financial incentive: How the introduction of the income-dependent combination tax credit affected hours worked and employment levels of Dutch mothers

(1)

1

A financial incentive:

How the introduction of the income-dependent combination tax credit

affected hours worked and employment levels of Dutch mothers

Istvan Bruggeman S1529943

Thesis MA Public Administration; Economic and Governance Thesis coordinator: Eduard Suari-Andreu

Final Version / 10-06-2019

(2)

2 Table of contents

1. Introduction…………..……….3

2. Theoretical Background and Hypotheses………...………..4

2.1 Female labor force participation trends in OECD countries………...4

2.2 Female labor force participation trends in the Netherlands……….6

2.3 Overview of the hypotheses………..………...9

3. Institutional Context and retrenchment of childcare benefits……..……..…..……9

4. Methodology………...……10

4.1 Difference-in-difference method & regression model..………..………….…10

4.2 Placebo test…………..………..………...…12

4.3 Data and sample selection………...……….………13

4.4 Common trends assumption………..……14

4.5 Descriptive statistics………...………...………...….15

5. Analysis and results…...………….………....……21

5.1 Regression analyses…...………..……….…….21

5.2 Placebo test results…..………...………..……….…………23

5.3 Childcare benefits results………...………..………...….…24

5.4 Policy effect on income levels…...……….……….…….26

6. Discussion & Conclusion...……….…27

6.1 Findings related to other literature concerning in-work benefits….………..……..27

6.2 Dual effect of childcare benefits and in-work benefits ……….………..28

6.3 Implications for women labor force participation trends in the Netherlands…...29

6.4 Limitations of the research……….………...30

6.5 Concluding remarks………..…….31

7. References………...…….32

(3)

3 1. Introduction

Over the last decades, female labor force participation has increased drastically in OECD states (Jaumotte, 2003, p.5). There are many explanations for this trend, such as economic changes, medical advancements, institutional changes and a better accessibility of childcare. Another explanation is a change is social norms about women’s position in the labor force (Fernandez, 2007, p.1). There are authors that explain trends in women’s labor force participation via the u-shaped hypothesis, which states that economic development and women’s labor force participation are connected. Namely, in states with low levels of economic development, women’s participation is high. When the economy grows, women’s participation first drops and then increases again (Goldin, 1994, p.2). However, there is also a lot of variation in trends of female labor force participation between countries. Increasing the labor force participation of women became an important topic recently, because of the aging population in many countries. More female labor participation could be a part of the solution for that problem (Jaumotte, 2003, p.6).

The Netherlands has a relatively high level of women that work part-time, while men work mostly full-time (Keck & Saraceno, 2013, p.300). In 2018, women in the Netherlands worked 28 hours per week on average, while men worked 39 hours per week (Jong, de, 2019, online). Female labor force participation rose the last decades, but this was mostly the case for part-time employment in the Netherlands. This is unique, as the percentage of women that work part-time is lower in other EU countries (Portegijs & Keuzenkamp, 2008, p.12). A possible explanation for the unique situation in the Netherlands is that part-time employment facilitates the combination of work-family responsibilities. Besides that, part-time employment is institutionalized and protected by law. Employers cannot easily increase hours worked of part-time employees for that reason. The issue has come on the political agenda, because working part-time has negative consequences on the use of female talent, career opportunities and being economically dependent. Besides that, more female employment could help to maintain financing the welfare state (Portegijs & Keuzenkamp, 2008, p.9).

The Dutch government implemented different policy changes to promote female labor participation in the period between 2005 and 2009. The aim of these changes was to increase both the number of women that work and the amount of hours they work per week. (Boer, de, Jongen & Kabatek, 2014, p.6). This series of change began with the Law on Childcare, introduced in 2005, which lowered the parental costs for formal and informal daycare. In-work benefits also changed in that period with the aim to lower costs of working. The combination tax credit was introduced in 2001 and lowered taxes to be paid for all working parents. The income-independent combination credit gradually replaced it and it was only eligible for second-earners in dual-parent households and for single parents, instead for all working parents. This credit became income-dependent in 2009 (Boer, de, Jongen & Kabatek, 2015, pp.7-11). The reason to change the tax credit in 2009 was to increase hours worked per week and employment levels of second-earners in dual-parent households and single parents (Tweede Kamer, 2008, online). The policy change was particularly aimed at women. The policy change showed that increasing female labor participation was indeed an issue on the political agenda in the Netherlands. The aim of this research is to examine if the policy change of 2009 indeed influenced hours worked per week and employment levels of second earners in dual-parent households and single parents. The research question is “was the effect of the 2009 introduction of the income-dependent combination tax credit on hours worked per week and employment levels for second earners in dual-parent households and single parents positive?” To answer

(4)

4 that question, a difference-in-difference method is used, using data from the Dutch Household Survey. Based on the literature, the expectation is that the policy change had a positive effect on the employment levels and hours worked of second earners and single parents. The policy change in 2009 was about income, as second earners and single parents now needed to earn more than a threshold to receive the tax credit (Rijksoverheid, 2019, online). It is therefore also interesting to know if the policy change also lead to more income for these parents.

This research contributes to the existing literature on the effect of social policies on hours worked and employment levels. Namely, existing literature does not provide consistent findings of that effect. Besides the relation with other literature, it is important to know if the policy change affected the treatment group or not. The treatment group in this research are second-earners in dual parent households and single parents with a child younger than 12 years old. Otherwise, the governmental policy failed. Finally, high levels of part-time employment have negative effects for the economic development of women and for the welfare state. It is therefore socially desirable that the policy was effective.

The research is structured as follows. First, existing literature and similar research on female labor force trends are examined. Based thereon, three hypotheses are derived. Then, the institutional context in the Netherlands in the years before the policy change and thereafter is explained. Thirdly, the methodology section explains the reason to use a difference-in-difference method and the dataset. The analysis section thereafter shows the results of the regression analyses, which are used to test the hypotheses. Only the first hypothesis can be tested due to too little observations to properly test the second and third hypotheses. The results show positive, although not significant effects on hours worked. There are significant effects on employment levels. However, following the placebo test, it is likely that another policy change was the cause. Namely, in 2005, childcare benefits were increased in the Netherlands. As this policy change affects the same treatment group, that policy change is also included and examined through a difference-in-difference regression. The effect on income levels is also not significant. The discussion section relates the findings to other literature. Overall, the results show that the 2009 policy change had no significant effect on the treatment group though. It is likely that the 2005 childcare benefit change did influence the treatment group.

2. Theoretical background and hypotheses

2.1 Female labor force participation trends in OECD countries

Female labor force participation has grown in all high-income economies during the last decades (Hegewisch & Gornick, 2011, p. 120). Data from the OECD show that the average participation rate of women in OECD countries grew from 42% in 1960 to 64% in 2017 (OECD. Stat, 2019, online). Hegewisch and Gornick (2011) state that a changing attitude towards women with family responsibilities can explain this growing participation rate. Examples are work-family supports, like leave policies for mothers, more flexible work arrangements and more public funding for childcare (p. 120). Besides that, the amount of policies that facilitates childcare of fathers is also growing. These policies help to create an equal balance in the time of unpaid care of children between fathers and mothers. It is relevant to know what the effect of different policies is on the labor force participation of women, as these results help to derive hypotheses from.

(5)

5 Hegewisch and Gornick (2011) review existing literature to assess the effect of leave policies and childcare benefits on labor outcomes for fathers and mothers (p.120). Both policies are good examples on how these policies differ between countries, as the generosity, duration and focus on mothers or fathers varies (Hegewisch & Gornick, 2011, pp.122-128). An important point that the authors make, is that social policies can have negative or unintended outcomes (2011, p.130). The question they try to answer is if policies that are more generous lead to a worsening gender wage gap. They provide a supply side and a demand side argument. The supply argument states that part-time employment and longer leave entitlements lead to a reduction in the focus of women on their future careers. Women are then likely to self-select for less ambitious and less paid jobs. Besides that, repeated use of leave policies reduces female human capital (Hegewisch & Gornick, 2011, p.130). The demand argument states that employers are likely to discriminate against women, if they are entitled to generous policies, like leave policies. The reason is that women are more likely than men are to make use of these entitlements (2011, p.130).

An important contribution of Hegewisch and Gornick is that they point out that policies have different effects on the labor market outcome of women in other countries. Besides that, they point out that some policies may even have negative effects. They conclude that knowledge is missing in some areas concerning the effect of work-family policies (2011, p. 132). It is therefore relevant to examine the effect of a policy change on labor force participation of women. Namely, it is possible that the effect of a social policy can have negative consequences

for female labor participation.

Thévenon (2013) focusses on economic development as an important explanation for rising female labor force participation (p.8). An important determinant for that participation is the economic change from the manufacturing sector to a service-based economy. Besides economic development, governmental policies that encourage parents to combine work and family are important determinants of the growing female labor participation rate (Thévenon, 2013, p.12). There are three main types of such policies: leave policies, childcare services for working parents for pre-school aged children and tax transfers and benefits that lead to financial advantages for women to be employed (p.12). Thévenon states that, just like Hegewisch and Gornick, these policies vary between OECD countries. However, Thévenon provides an econometric analysis, instead of a literature review (2013, p.19). The analysis includes aggregated labor force participation rates of women (between 25 and 54 years old) of 18 OECD countries of the period 1980-2007 (Thévenon, 2013, p.19).

Thévenon (2013) runs a regression to examine the effect of labor market characteristics and policy characteristics on full time female labor force participation (p.19). Both labor market characteristics, as policy characteristics, influence the rate of female labor force participation. An increasing service sector in the economy positively affects full-time female labor force participation. Policy characteristics have mixed effects on the employment rate. The presence of leave policies has a positive effect on full-time employment, while a longer duration negatively affects full-time employment. Childcare policies positively affect full-time employment of women (Thévenon, 2013, pp.26-27). The findings of Thévenon resemble those of Hegewisch and Gornick, as they also stated that leave policies and childcare services affect labor market outcomes for women. Thévenon (2013) concludes that three points influence female labor force participation, namely the structure of the labor market, the institutional setting supporting work-life balances and the improvement in the average level of educational attainment of women (p.39). The results therefore show that not only social policies affect women’s labor force participation, but also labor market characteristics. The article focusses on

(6)

6 aggregated data from 18 OECD countries, which is useful to create a broad picture of the determinants of female labor force participation.

It is also relevant to know if social policies have other effects on different households’ types. Single-parent households have a higher chance of being poor than dual-parent households (Maldonado & Nieuwenhuis, 2015, p.396). Maldonado and Nieuwenhuis (2015) examine the effect of reconciliation policies and financial support policies on differences in poverty between single-mother, single-father and dual-parent households (p.397). Reconciliation policies are policies that facilitate combining work and family responsibilities. The authors point out that there are differences in the number of resources and the capacity to use these resources to combat poverty between single-parent households and dual-parent households (Maldonado & Nieuwenhuis, 2015, p.398). Dual-parent households have a higher chance of being employed, as these households have two potential earners, instead of one. Besides that, dual-earners can more easily combine work-family responsibilities, because they can share the responsibility for childcare (2015, p.398). The authors state that households that have more resources, which usually are dual-parent households, benefit the most from social policies (2015, p.400). Namely, single-parent households need all their employment earnings to fulfill their needs, which means that they will not benefit from longer unpaid leave policies. The authors analyze differences in poverty rates by looking at the situation before a policy and thereafter, using data from the Luxembourg Income Study. They use datasets of 18 OECD countries for the period 1978-2008 (Maldonado & Nieuwenhuis, 2015, p.400). The analyses show that for all time and in all countries single-parent households are poorer than dual-parent households are. Single-mothers are also more likely to be poor than single-fathers. A reason is that employment reduces the risk of being poor and single-mother households work less than parents in dual-parent households do (Maldonado & Nieuwenhuis, 2015, p.404). The analyses also show that poverty rates of single-parent households lower relatively more than dual-parent households due to family allowances. The authors conclude that both leave policies, as family allowances, lowered poverty rates of single-parent households relatively more than the poverty rates of parent households. This indicates that their expectation that dual-parent households benefit more from social policies than single-dual-parent households do not hold (Maldonado & Nieuwenhuis, 2015, pp.410-411).

Maldonado and Nieuwenhuis point out that social policies can affect household types differently. Therefore, it would be useful to know if the introduction of the income-dependent combination tax credit had different effects for single-parent households and dual-parent households in the Netherlands. Maldonado and Nieuwenhuis (2015) point out is that single-parent households are more likely to need all their earnings from employment to fulfill their basic needs (p.400). The income-dependent combination tax credit is an in-work benefit, which is therefore likely to affect second-earners in dual parent households more than single parents are. These findings lead to the hypothesis that the 2009 policy change affected second earners in dual-parent households more than single parents.

2.2 Female labor force participation trends in the Netherlands

This section focusses on the trends and determinants of women’s labor force participation in the Netherlands. Bosch, Deelen and Euwals (2010) state that societal models that facilitate women’s employment differ between countries. They state that the Netherlands have chosen a model in which women’s employment rate is high, but most of the women work part-time. This makes combining work and family responsibilities possible (p.35). However, the authors point out that some scholars argue that this model is not socially desirable. There are two reasons for

(7)

7 this discussion. First, part-time employment may lead to a reduction of the use of female human capital. This can harm women’s participation. Secondly, more female full-time employment can be a part of the solution for the sustainability of the welfare state. Namely, there is a growing problem due to the aging population in the Netherlands (Bosch, Deelen & Euwals, 2010, p.35). The Dutch government has implemented different policies to protect part-time employment since the 1980s, which can explain this high part-time employment rate of women (Bosch, Deelen & Euwals, 2010, p.36). The authors use a regression analysis to estimate the effect of age of women on the development of working part-time or not. They use data from the Dutch Labor Survey (Bosch, Deelen & Euwals, 2010, p.38). The dependent variable is working hours of women in the Netherlands. The authors divide women in different age cohorts and include personal and family characteristics in the regression model. Having young children has a large effect on the number of working hours of women. (Bosch, Deelen & Euwals, 2010, p.38). The authors conclude that it is not likely that the situation in the Netherlands will change. Over the years, the number of women working full-time remains stable, and the authors find no reason to believe that this trend will change. The authors argue that higher childcare subsidies did not affect the propensity of women working full-time in the period between 2006 and 2008. More full-time employment of women is not likely to occur in the Netherlands when there will be no change in social norms and implementation of effective policies (Bosch, Deelen & Euwals, 2010, p.52). These findings are in line with the findings of Statistics Netherlands, which found that the full-time employment rate of women was stable for the years 2006 until 2016 (CBS, 2017, online).

An interesting argument is that higher childcare subsidies did not change the full-time employment rate of women (Bosch, Deelen & Euwals, 2010, p.38). Namely, this rate has been stable over the last years. However, the authors did not include social policies in their regression, so they do not provide statistical findings on the effect of social policies on the number of hours worked of women. It could be that higher childcare subsidies did led to more hours worked for women. However, it did not lead to more full-time employment. This is possible, because someone needs to work more than 35 hours per week to work full-time.

Bettendorf, Jongen & Muller (2015) examine the effect of large changes in Dutch childcare policies on labor supply. In 2005, the Law on Childcare was introduced. Besides changes in childcare subsidies, the Dutch government also increased earned income tax credits (EITCs) in the period from 2004 until 2009. This policy change affects the same group, namely parents with children under the age of 12 (Bettendorf, Jongen & Muller, 2015, pp. 112-113). The authors use a difference-in-difference method to estimate the combined effect of both policy changes on the hours worked per week (Bettendorf, Jongen & Muller, 2015, p. 113). The treatment group are parents between 20 and 50 years old, with their youngest child being under the age of 12. The control group are similar parents, only with a youngest child above the age of 12. They retrieve the data from the Dutch Labor Force Survey. The authors estimate an intention-to-treat effect, as it is not possible to link individual data on labor supply to the use of childcare.

The treatment effect could be difficult to estimate when parents anticipated these policy reforms (Bettendorf, Jongen & Muller, 2015, p.116). However, the authors find no changes in trends in the years before the childcare reform. They run two regressions, one for the period 2005-2007, for the short run effect, and one for the period 2008-2009, for medium run effect. Besides that, they compare different groups, namely parents with a youngest child between 0 and 3 years old, between 4 and 7 years old and 8 and 11 years old (Bettendorf, Jongen & Muller, 2015, p.114).

(8)

8 The estimated results of the analysis show that the participation rate of women has grown 2.7 percentage points in the period 2005-2007 and 4.2 percentage points in the period 2008-2009 (Bettendorf, Jongen & Muller, 2015, p.118). The effect on hours worked for the period 2005-2007 is positive with an increase of 1.0 hours worked per week and an increase of 1.6 hours for the 2008-2009 period. The authors also run a placebo test, which results are not significant, and robustness checks, which lead to no effect in the 2005-2007 period and a positive effect of 3.3 percentage points in the 2008-2009 period (Bettendorf, Jongen & Muller, 2015, p.118). Both effects are larger for single mothers than for coupled mothers. Overall, the results show a positive effect of these reforms for the treatment group. However, the public expenditures are quite high for a relatively small change in participation and hours worked (Bettendorf, Jongen & Muller, 2015, p.120). The effect of on labor supply and hours worked of men is lower than of women.

The results of Bettendorf et al. lead to two hypotheses. The first hypothesis states that the introduction of the income-dependent combination tax credit has a positive effect on hours worked and employment levels of women. The second hypothesis states that this introduction has a larger effect on women than on men.

The findings of Bettendorf et al. are in line with the argument of Bosch et al. that social policies can positively affect the participation rate of women, but that effect is small. This strengthens the argument of Bosch et al. that childcare benefits that are more generous do not lead to more full-time employment of women. The article of Bettendorf et al. is useful, as its methodology and analyses provide good results that show the effect of policy changes. The article also shows that the policy changes in childcare policies should be included, as these changes affect the same treatment group.

The article of Bettendorf et al. poses the question about the relation between the height of public expenditure on social policies and their actual effect. De Boer, Jongen and Kabatek (2015) examine this question in their research. They look at the public expenditure on each policy and compare the costs of each policy with the actual effect on employment (Boer, de, Jongen & Kabatek, 2015, p.2). The authors examine the household utility levels, which consists of disposable income, hours worked of both the men and women and the hours of formal childcare (Boer, de, Jongen & Kabatek, 2015, p.11). The authors derive data from the Dutch Labor Market Panel. They compare two groups with each other, namely parents with children between 0-3 years of age and children between 4-11 years of age (Boer, de, Jongen & Kabatek, 2015, p.14). The reason to compare these groups is that the childcare needs and labor supply incentives may differ for non-trivial reasons.

The authors firstly use descriptive statistics to compare the effect of different policies on men and women. Thereafter, they calculate the wage elasticity of women and men and the elasticity of childcare (Boer, de, Jongen & Kabatek, 2015, p.20). The wage elasticity of women is higher, meaning that a wage increase should lead to relatively larger increase of women in labor participation than of men. The authors then compare their systematic results with the difference-in-difference results of Bettendorf et al. (2015). Finally, the authors examine the effectiveness of these policies regarding their costs. Both childcare benefits and the income-dependent combination tax credit positively affect hours worked of women (Boer, de, Jongen & Kabatek, 2015, p. 20). The most effective policy is in-work benefits for second earners, which increases with income. Fiscal stimuli do not affect primary earners much. There is also a trade-off between equity and efficiency, as public expenditure on middle and high incomes have larger effects (Boer, de, Jongen & Kabatek, 2015, p.30).

(9)

9

2.3 Overview of the hypotheses

There are thee hypotheses that help to answer the research question. The discussed literature provides the theoretical background of these hypotheses:

1) The introduction of the income-dependent combination tax credit (IDCTC) has a positive effect on hours worked and employment levels of the treatment group.

2) The IDCTC has a larger effect in the change of hours worked and employment levels of mothers than of fathers.

3) The IDCTC has a larger effect in the change of hours worked and employment levels of coupled parents than of single parents.

3. Institutional context and retrenchment of childcare benefits

The Dutch government implemented different policies to promote female labor participation in the period between 2005 and 2009 (Boer, de, Jongen, Kabatek, 2015, p.6). Two important policy adjustments were changes in childcare benefits and in-work benefits. The childcare benefits changed due to the Law on Childcare of 2005. The law unified the subsidy for formal daycare places. Besides that, the subsidy rate for parents rose in 2006 and 2007. This led to lower parental costs for childcare services (Boer, de, Jongen, Kabatek, 2015, pp.6-7). Following the reform, public spending on childcare rose from 1 billion euros in 2005 to 3 billion euros in 2009 (0.5% of the GDP). However, the Dutch government implemented cuts on childcare benefits in 2012. (Algemene Rekenkamer, 2014, p.2). These retrenchments could affect the labor participation rate of parents with younger children (Algemene Rekenkamer, 2014, p.3). If that is the case, the policy change of 2012 interferes with the 2009 policy change, as both policies affect the same group. It is not possible to focus on that policy change in this research though. Therefore, the years that are examined are up until 2012. There were no cuts in the income-dependent combination tax credit in that period.

Another important policy adjustment in that period was the change of in-work benefits. An important in-work benefit was the combination discount. This discount was introduced in 2001 and was aimed at all working parents with a child younger than 12 years old to facilitate the combination of work and family responsibilities (Boer, de, Jongen & Kabatek, 2015, p.9). The combination discount lowered taxes to be paid for these parents. However, this combination discount faded in the period 2005-2009. The income-independent combination discount gradually replaced it. This discount was only eligible for second-earners and single parents with a child younger than 12 years old, instead of all working parents (Boer, de, Jongen & Kabatek, 2015, p.11).

The income-dependent combination tax credit replaced the income-independent combination credit in 2009. The policy change was that earned income determined if second earners and single parents with a child younger than 12 years old were eligible for the tax credit or not. Namely, someone had to earn more than a threshold to receive the tax credit. Therefore, not every single parent or second earner would receive the tax credit anymore. When both parents earn the same, then the tax credit will be transferred to the oldest parent (Belastingdienst, 2019, online). For the past few years, the minimal amount of yearly earnings was around €4800. The height of the tax credit depends on what someone earns. For 2019, the tax credit was 11,450% of the earned income from labor minus the threshold (€4993 for 2009) when earned income

(10)

10 was between €4993 and €29752. Thereafter, the tax credit remains the same. The height of the tax credit is also different for people that reach the age to receive their old age pension (AOW), as the percentage lowers from 11,450% to 5,86%. The maximum earning also remains the same, however the standard tax credit lowers from €2835 to €1452 (Belastingdienst, 2019, online). Public expenditure on the in-work benefits rose from 724 million euros in 2008 to 1290 million euros in 2009 (Boer, de, Jongen & Kabatek, 2015, p.11). Following the governmental tax plan of 2008, the income-dependent combination tax credit has the purpose to stimulate working among single parents and the second-earners (Tweede Kamer, 2008, online). It is argued that this group reacts strongly to changes in spendable income. Therefore, the tax credit should stimulate them to work more hours.

Timeline of the policy changes

2001 : Introduction of the Combination Discount for all working parents with a child below the age of 12

2004: Introduction of the Income-Independent Combination Discount for all working second earners and single parents with a child below the age of 12

2005: The introduction of the Law on Childcare, which unified the subsidies for all formal daycare places

2006/2007: Increase of the subsidy rates for childcare, which lowered the parental fees 2005-2009: Fading of the Combination Discount

2009: Introduction of the Income-Dependent Combination Discount for all working second earners and single parents, who earned more than a threshold per year

2009: Both the Combination Discount and the Income-Independent Combination Discount are abolished

2012 onwards: Retrenchments of childcare benefits

4. Methodology

4.1 Difference-in-difference method & regression models

The aim of this research is to examine the effect of the introduction of the income-dependent combination tax credit (IDCTC) on hours worked and employment level of the treatment group. Besides these outcome variables, changes in income levels are also examined. The treatment group consists of second-earners in dual parent households and single parents with a child younger than 12 years old. A randomized trial is useful, as it can filter out the selection bias. This refers to the differences in the outcome variables between the treatment and control group, which is not caused by the treatment effect of the 2009 policy change. This means that the differences in outcomes for these groups is accounted for the treatment (Angrist & Pischke, 2014, p.30). However, it is not possible to examine a policy change via a true experiment, as the treatment group and control group are not randomly sampled and randomly assigned to a treatment. Therefore, examining the effect of a policy change is a natural experiment. That method uses a phenomenon that causes randomization for groups if they are eligible for a treatment or not (Blundell & Costa Dias, 2009, p.578).

Therefore, a difference-in-difference method is useful to examine the effect of the policy change on the treatment group. Namely, the policy states that the IDCTC only is applicable for parents with a child that is younger than 12 years old. This means that similar groups of parents, which do have children older than 11 years, are not eligible. The difference-in-difference

(11)

11 method firstly analyzes the common differences in outcome variables between both groups of parents before the policy change. Secondly, the method analyzes the differences in outcome variables between both groups after the policy change, regarding the initial common difference in outcome (Blundell & Costa Dias, 2009, p.578). This method needs longitudinal data from the same observations to analyze differences in the outcome variables. An important feature of this approach is the common trends assumption. This means that both groups should have evolved in the same way, if the treatment had been absent (Angrist & Pischke, 2014, p.184). It is very difficult to test this assumption. However, data from before the reform could show if both groups faced similar trends in hours worked, employment levels and income (Bettendorf, Jongen & Muller, 2015, p.116). The common trends assumption can also be violated if the government was anticipating on changing behavior of women or that the parents anticipated on the reform.

The data can show if anticipation is likely to happen or not. If there were no large changes in the years before the reform, then it is not likely that the government anticipated on changing behavior (Bettendorf, Jongen & Muller, 2015, p.116). The policy reform that lead to the introduction of the IDCTC was part of the 2009 tax plan, which started on January 1, 2009. The cabinet published this plan on October 7, 2008 (Tweede Kamer, 2008, online). It is therefore not likely that parents anticipated on the reform, as the publication date is close to 2009. Besides that, the tax plan had to be approved by parliament thereafter. A final threat to the common trends assumption is the change in characteristics of both groups. It is therefore important to check changes in fertility rates between the groups, as having a young child would change the composition of the treatment group (Bettendorf, Jongen & Muller, 2015, p.116). Descriptive data of the years before the reform can show if the common trends assumption holds or not. Besides that, descriptive data can show if other characteristics of the treatment and control group changed after the policy change.

The treatment effect on the outcome variables of the treatment group is estimated through a regression analysis. A basic difference-in-difference regression analysis has three components. A treatment dummy variable, which in this case would refer to the treatment group of parents with children below the age of 12. A time dummy, which refers to the situation before and after the implementation in 2009. Finally, the regression contains an interaction term of the time and treatment effect. The coefficient of the interaction term is the difference-in-difference estimator, which refers to the treatment effect (Angrist & Pischke, 2014, p.187).

The treatment group in this research consists of two groups. The first group consists of parents with a youngest child below the age of 12 and above the age of 3. The second group consists of parents with a youngest child below the age of 4. The research design is like the one of Boer et al. (2014). They include two treatment groups into their regression analysis The reason is that the type of childcare differs between those groups (p.18). Children between 4 and 11 years old go to school and need out-of-school care. Younger children need other type of daycare. Although the main purpose of this research is to examine the effect of the introduction of the income-dependent combination tax credit (IDCTC), it is important to include the effect of childcare benefits. The control group are parents with a youngest child above the age of 11

and below the age of 18.

One difference-in-difference model is used for testing all three hypotheses and the research question. The same model is used for both treatment groups:

(12)

12 Yit = α + β * CHILDi + γ * POST2009t + δ * (CHILDi * POST2009t) + λit * CONTROLit +εit

The regression equation consists of different variables. The outcome variable Y stands for 1: hours worked per week, 2: employment levels and 3: income for both treatment groups, depending of the regression. CHILD is a dummy variable, which has value 1 if parents have a youngest child below the age of 12 and above the age of 3 in the model for the first treatment group. It has value 0 if parents have a youngest child below the age of 4 or above the age of 11. For the second treatment group, CHILD takes value 1 if parents have a youngest child below the age of 4 and value 0 if parents have a youngest child above the age of 3 in the model.

The beta-coefficient captures the group effect, which is the difference between the control group and the treatment group before the policy change. POST2009 is a time dummy variable, which has value 1 when observations are from 2009 onwards and value 0 if observations are from before 2009. The gamma-coefficient captures the time effect, which is the common difference in outcomes for the treatment groups and the control group before and after the policy change. The CHILD*POST2009 dummy variable has value 1 for observations after the reform and 0 for observations before the reform (2009). The delta-coefficient captures the treatment effect, which compares the differences between the control and treatment from before and from after the policy change in 2009.

CONTROL stands for variables that control for changes in both groups over time. The control variables are age, sex, education, number of children, marital status and having a partner. The reason to include these variables is to control if the composition of the treatment and control groups do not change significantly, as this could harm the difference-in-difference model. The control variables are mostly like the variables Bettendorf et al. use (2015, pp.13-14). First, the regressions are run for the treatment groups to test the first hypothesis. The regressions are also run for women and men to test the second hypothesis. To test the third hypothesis, the regressions are run for single parents and coupled parents. The only variable that changes in these regressions is the CHILD variable, as it then stands for mothers, fathers, single parents and coupled parents. εit stand for the error term. In both models, robust standard errors are used

to account for the problem of heteroscedasticity (Angrist & Pischke, 2014, p.97).

A possible problem regarding inferences is the effect of the changes in childcare benefits from 2005 onwards. It is not possible to separate the effect of the childcare benefits and the tax credit on the outcome variables, if the effect still holds in 2009. However, it is possible to run a similar regression, but before 2009. The regression model is therefore the same, except for the time dummy. The regression is run multiple times for the years 2005 until 2009. The only variable that changes is the time dummy, which will take value 1 for the year 2005 and 0 for the years before 2005 for the first regression. The time dummy will take value 1 for 2006, 2007 and 2008 for the other regressions. The outcome of the regression per year can show if the treatment effect of the childcare benefit change increased, decreased or stabilized over the years. This information is useful to interpret the results of the regression of the tax credit, as it shows if the effect of the childcare benefits is still present in 2009.

4.2 Placebo test

It is useful to conduct a placebo test, as it helps to assess the significance of difference-in-difference estimate (Fowler, 2013, p.177). The idea behind such a test is that it examines if the

(13)

13 results of the difference-in-difference regression differ when the “treatment” was introduced in another year than it did. Therefore, the results of the placebo test should not be significant, as the placebo should not affect outcome variables. If the placebo is indeed significant, this could mean that the treatment effect is not accounted for the actual treatment, but due to something else (like another policy change or other external phenomenon). The placebo time dummy stands for the year 2008.

Yit = α + β * CHILD<12i + γ * Placebot + γ * POST2009t + δ * (CHILDi * Placebot)

+ δ * (CHILDi * POST2009t) + λit * CONTROLit + εit

4.3 Data and sample selection

The data are derived from the Dutch Household Survey (DHS). Approximately 2000 Dutch households take part in this yearly survey. The panel data contains information about employment, pensions, income, economic and personal characteristics (Centerdata, 2019, online). The data of the DHS is personal, so the analyses are on micro level. Therefore, it becomes possible to analyze differences in hours worked and employment levels per year. The years that are included in the regressions are 2003 up and until 2011, as the difference-in-difference estimator needs to be singled out to capture the treatment effect. Therefore, some years before and some years after the reform are examined. Some years before the law on Childcare are included, to examine the treatment effect of that policy change as well. The years from 2012 onwards are excluded, as childcare benefit retrenchments may interfere with treatment effect of the income-dependent combination tax credit.

The DHS does not include information about receiving the tax credit. This makes the model less precise than when this information was included. However, the effect of the reform is still possible to estimate. All observations that are not eligible for the tax credit are excluded from both the treatment groups and the control group. The reason therefore is to make the treatment groups and the control group more alike to make a comparison between these groups possible. A problem occurs when a child ages and becomes 12. Namely, the observations that first were the treatment group then become the control group. Only the households that are constantly in the same group are examined for that reason. Table 1 shows the number of observations that are excluded, because of the requirements to receive the tax credit. The total number of observations are 400 for the below 4 treatment group, 1057 for the below 12 & above 3 treatment group and 701 for the control group.

Table 1; Sample selection Number of observations (2003 -2011) Household with youngest child <4 Household with youngest child >3 & <12 Control group with youngest child >11 & <18

From the sample 2003-2011 1800 5416 3368

Exclusion of non-parents 968 3156 1950

Exclusion of prime earners in dual-parent households

(14)

14 Exclusion of observation

that underwent changes in household type

7 70 51

Total N = 2158 400 1057 701

Source; the Dutch Household Survey (2003-2011)

Following the Dutch central government, persons should request the tax credit themselves (Rijksoverheid, 2019, online). The problem that may occur is that eligible persons do not request the tax credit. This problem with non-compliance of the treatment cannot be controlled for exactly, as the data is missing from the DHS. Therefore, the estimated effect of the treatment is an intention-to-treat effect (Angrist & Pischke, 2014, p.119). This means that the effect of the introduction of the tax credit captures the causal effect of being assigned to a treatment, but without exactly knowing who complied and who did not.

4.4 Common trends assumption

As discussed earlier, the common trends assumption is an important feature of the difference-in-difference method (Angrist & Pischke, 2014, pp 184-185). All groups should follow the same trend before the policy change. The difference-in-difference method compares the outcomes of the treatment group with a counterfactual, which is the outcome if the treatment had not occurred (Toshkov, 2016, p.233). Figure 1 and figure 2 graphically show the trends of both treatment groups and the control group for hours worked per week and employment levels to show if the groups follow the same trend before 2009.

0 5 10 15 20 25 30 35 2003 2004 2005 2006 2007 2008 2009 2010 2011 H o u rs work ed Year

Figure 1; mean hours worked per week

(15)

15 Figure 1 shows the mean of hours worked per week of the three different groups. The children below 12 & above 3 and the children above 12 group roughly follow the same line, with a somewhat different trend in 2006. The lines converge again in 2008. The children below 4 group has another trend before 2005 but does seem to follow the same trend as the below 12 & above 3 group from 2006 onwards. The only difference is that this group works more hours per week from 2007 onwards. The 2009 cutoff point is not very sharp and does not have to seem a positive effect on hours worked for the treatment groups and the control group. All groups’ employment level lower after 2009, except for the control group, which increases after 2010.

Figure 2 shows the percentage of being employed per group. An observant is employed when he/she works on a contractual basis, works in their own business or is a freelance/ self-employed (Dutch Household Survey, 2003-2011). All the groups follow the same trend for being employed before the policy change in 2009. A cutoff is visible for 2005, as the percentage of being employed rose for all groups then. A possible reason for this increase is the introduction of the Law on Childcare (Boer, de, Jongen & Kabatek, 2015, p.20). The treatment group with a youngest child below the age of 4 has the overall highest percentage of being employed. The percentage of being employed for that group rose until 2008 and thereafter lowered again. The other two groups are more alike, but the treatment group with the youngest child below 12 & above 3 has the lowest percentage of being employed for all years. Both treatment groups also follow the same trend, except for 2003 and the treatment group below four’s employment level rose more in 2008. There is not clear cutoff visible at 2009, which may suggest that the effect of the policy change on the treatment groups is not very high.

4.5 Descriptive Statistics

It is important to include variables into the analysis that affect the outcome variables. A problem may occur when these variables change much over time and therefore harm the common trends assumption. It then becomes more difficult to examine the treatment effect of the policy change,

0 10 20 30 40 50 60 70 80 2003 2004 2005 2006 2007 2008 2009 2010 2011 Perc en ta ge b ein g em p loy ed Year

Figure 2; percentage of being employed

(16)

16 as these variables may also be a cause for changes in the outcome variables. When these confounding variables are adjusted for, they become control variables (Toshkov, 2016, p.207). Table 2 shows the descriptive statistics of all groups before the implementation of the tax credit in 2009. For the hours worked variable, outliers are excluded, as some respondents claimed to work more than 100 hours a week. Therefore, only respondents that worked 60 hours or less are included, to provide more suitable results for the mean of hours worked per week. Therefore, 8 observation are excluded for the hours worked regression. It is important to compare the variables per group to check if the group’s characteristics are not too different. The majority of the observed parents are women. Therefore, the results of this research mostly provides information on Dutch mothers.

Table 2; descriptive statistics for the years 2003-2008 for all groups

Mean values for the outcome and control variables

Parents with youngest child <4 Parents with youngest child >3 & <12 Control group with youngest child >11 & <18

Hours worked per week 25.82

(1.0351)a 108 observationsb 23.90 (0.599) 302 observations 27.25 (0.875) 182 observations Employed 56.27% 48.76% 52.99% Age 31.25 (0.250) 36.41 (0.186) 44.33 (0.211) Male/Female 5.42%/94.58% 4.67%/95.33% 8.97%/91.03%

Highest level of education

Special Education - 0.96% 0.21%

Primary Education - 0.69% -

High School 26.11% 37.52% 48.51%

Senior Vocational (MBO) 22.37% 30.49% 28.42%

College (HBO) 38.64% 23.76% 18.59% University 12.54% 5.36% 3.42% Other 0.34% 1.51% 0.85% Number of children 1 Child 33.56% 5.08% 10.04% 2 Children 42.37% 47.12% 47.86% 3 Children 14.24% 35.44% 31.20% 4 Children or more 9.84% 12.36% 10.90% Marital status Married 44.02% 44.41% 44.02% Divorced 2.99% 6.10% 2.99% Living together 1.36% 1.36% 0.21% Widowed - - 0.21% Never Married 0.34% 0.34% 2.14% Do not know 53.90% 53.90% 50.43% Has a partner 98.98% 97.94% 89.74% Total N = 1491 295 728 468

Source; Dutch Household Survey (2003-2008) a) _{Standard deviation}

(17)

17 Table 2 shows that the mean value of hours worked before the 2009 policy change was around 24-27 hours per week. The mean value for hours worked is higher for the control group (around 27 hours per week) than for the treatment groups. Employment levels show an opposite direction. The employment level of the treatment group with the youngest child below 4 is the highest (56%). The control group relatively works more hours per week than the below 4 group, but they are less employed. This could suggest that the below 4 group must combine work-family responsibilities more than the other groups and can therefore work less hours per week. The other treatment group (below 12 & above 3) has the lowest

employment rate and works less hours per week than the other groups.

The table also includes the control variables. All groups differ in age, which makes sense. Namely, the below 4 group is more likely to be young than the other groups. Level of education and number of children also differ between the groups. Having two children is the most common for all three groups. The treatment group below 12 & above 3 has relatively more children than the other two groups, while the below 4 group has the lowest number of children. The level of education is also different per group. The groups do differ in

characteristics. Therefore, the control variables should be included into the analysis to avoid an omitted variables bias.

Table 3; descriptive statistics for the years 2009-2011 for all groups

Mean values for the outcome and control variables

Hours worked per week 31.59

(1.613)a 44 observationsb 25.24 (0.905) 114 observations 27.57 (1.072) 83 observations Employed 70.48% 65.05% 67.81% Age 34.04 (0.408) 40.40 (0.283) 47.24 (0.304) Male/Female 16.19%/83.31% 6.69%/93.31% 11.16%/88.64%

Highest level of education

Special Education - 0.91% -

Primary Education - 0.61% -

High School 16.99% 31.92% 36.91%

Senior Vocational (MBO) 13.33% 31.00% 33.91%

College (HBO) 40.95% 23.40% 19.31% University 26.67% 11.55% 8.15% Other 2.86% 0.61% 1.72% Number of children 1 Child 25.71% 8.81% 14.16% 2 Children 35.24% 47.42% 45.49% 3 Children 30.48% 30.70% 30.04% 4 Children 8.57% 13.07% 10.30% Marital status Married 45.71% 42.25% 34.76% Divorced 0.95% 1.52% 7.73% Living together 1.90% 0.61% 0.43% Widowed - - 0.43%

(18)

18

Never Married 2.86% 1.82% 1.72%

Do not know 48.57% 53.80% 54.94%

Has a partner 94.29% 94.83% 81.97%

Total N = 667 105 329 233

b)_{Differs from the total number of observations due to non-response}

Table 3 shows the mean values for the period after the implementation of the tax credit. The outcome variables hours worked per week and employment levels have changed. For the treatment group with a youngest child below 4, hours worked per week increased from 25.82 to 31.59 hours a week. The employment level also increased from 56.27% to 70.48%. This is a large increase for both outcome variables. Hours worked also increased for the other treatment group, the effect seems to be smaller though. The employment level for that group increased largely, from 48.76% to 65.05%. For the control group, hours worked per week increased a little, and the employment level rose largely (from 52.99 % to 67.81%). Employment levels rose for all three groups with around 12%. Therefore, the tax credit does not seem to have affected the groups differently concerning employment levels. Hours worked, however, did increase for the treatment groups, but not for the control group. This could be due to the tax credit.

The results of table 2 and 3 show the fertility rates before and after 2009. A large variation in fertility rates changes the composition of the control and treatment groups. The fertility rate of the below 12 & above 3 group and the control group remains somewhat the same. The fertility rate of the below 4 group increased, but not that much. Therefore, there are no large changes in the composition of the groups regarding fertility. This research does not only examine the difference in outcomes for the treatment groups and the control group, as it is also interested in the differences in outcomes for fathers and mothers and for single mothers and coupled mothers. Table 4 shows the descriptive statistics for hours worked per week and employment level for fathers and mothers before the introduction of the tax credit in 2009. Table 5 shows the same, but for the period after the introduction.

Fathers work relatively more hours and have a higher employment rate than mothers in all three groups. The information of table 4 and table 5 shows that the mothers and fathers in the below 4 group worked relatively more hours after 2009 than before 2009. For the other treatment group, fathers worked relatively more after 2009 than before, while this increase is smaller for mothers in that group. Employment levels rose for all mothers, but not for all fathers. This is in line with the second hypothesis that mothers are more affected by the policy change than fathers are. However, this does not seem to hold for hours worked per week. The number of observed fathers poses a problem for a regression analysis. Namely, the number of fathers is very low and therefore not suitable for a regression analysis that provides good difference-in-difference estimators. It is therefore not possible to test the second hypotheses with a regression analysis.

(19)

19 Table 4; descriptive statistics for fathers and mothers for the years 2003-2008

Mean values for hours worked and employment

Hours worked per week for mothers 25.26 (0.592)a 285 observationsb 23.26 (0.592) 285 observations 26.22 (0.902) 166 observations

Hours worked per week for fathers 37.60 (1.721) 5 observations 34.71 (2.806) 17 observations 38.00 (1.945) 16 observations Employed for mothers 54.48% 46.83% 50.00% Employed for fathers 87.50% 88.24% 83.33% Total N = 1712 279 mothers 16 fathers 694 mothers 34 fathers 426 mothers 42 fathers

Table 5; descriptive statistics for fathers and mothers for the years 2009-2011

Mean values for hours worked and employment

Hours worked per week for mothers 29.64 (1.728)a 36 observationsb 24.50 (0.991) 107 observations 26.96 (1.055) 77 observations

Hours worked per week for fathers 40.38 (2.686) 8 observations 36.57 (2.091) 7 observations 35.50 (5.464) 6 observations Employed for mothers 68.18% 63.19% 66.18% Employed for fathers 82.35% 90.91% 80.77% Total N = 667 88 mothers 17 fathers 307 mothers 22 fathers 207 mothers 26 fathers

Table 6 and 7 show similar data as table 4 and 5, but the groups are now divided for coupled parents and single parents. Unfortunately, there are too little observations to run a proper regression analysis. Therefore, these descriptive statistics do provide information that gives a direction of the treatment effect. However, there are simply not enough observations

(20)

20 that are single to execute a proper analysis. Therefore, the third hypothesis cannot be tested with a regression analysis.

Table 6; descriptive statistics for coupled and single parents 2003-2008

Mean values for 2003-2008 Parents with youngest child <4 Parents with youngest child >3 & <12 Control group with youngest child >11 & <18

Hours worked per week for coupled parents

25.82 (1.035)a 108 observationsb 23.92 (0.611) 293 observations 25.79 (0.896) 158 observations

Hours worked per week for single parents

- - 0 observations 23.33 (3.069) 9 observations 36.88 (2.047) 24 observations Employed

for coupled parents

56.16% 48.81% 51.19%

Employed

for single parents

66.67% 46.67% 68.75% Total N = 1712 292 coupled 3 singles 713 coupled 15 singles 420 coupled 48 singles

Table 7; descriptive statistics for coupled and single parents 2009-2011

Mean values for 2009-2011 Parents with youngest child <4 Parents with youngest child >3 & <12 Control group with youngest child >11 & <18

Hours worked per week for coupled parents

31.46 (1.646)a 41 observationsb 24.94 (0.934) 108 observations 26.56 (1.255) 64 observations

Hours worked per week for single parents

33.33 (8.819) 3 observations 27.00 (4.434) 6 observations 31.00 (1.854) 19 observations Employed

for coupled parents

70.71% 64.42% 61.78%

Employed

for single parents

66.67% 76.47% 95.24% Total N = 669 99 coupled 6 singles 312 coupled 19 singles 191 coupled 42 singles

(21)

21 5. Analysis and results

5.1 Regression analyses

Table 8 and table 9 show the regression results for the outcome variables. Only the first hypothesis is tested. The regressions are run for both treatment groups, which are split as the below 12 & above 3 group receives formal childcare benefits and the below 4 group does not.

Table 8; regression results for hours worked per week

Difference-in-Difference coefficients for hours worked

Parents with youngest child <4 Parents with youngest child >3 & <12 Group dummy -2.580* (1.444)a -4.661*** (1.000) Policy dummy -0.058 (0.198) -0.043 (0.199) Interaction dummy 3.301* (1.923) 0.915 (1.122) Total N = 1457 R-squared = 400 0.2503 1057 0.2535

Notes: * = p<0.10, ** = p<0.05 & *** = p<0.01. A complete list of the dif-in-dif coefficients of the control variables is found in the appendix.

The interaction dummy’s coefficient is the most relevant, as it provides information about the overall treatment effect of the policy change. The interaction effect is only significant for the below 4 group at the 0.10 level. The coefficient is 3.3, meaning that the policy change lead to an increase of the hours worked of that group with more than 3 hours a week. The coefficient for the other treatment group is 0.915, meaning that the policy change lead to an increase of not even 1 hour worked per week.

Table 9 show similar results as table 8, but now for being employed instead of hours worked per week. The interpretation of the results differs from the other table, as the outcome variable employed is a binary variable. The coefficients of the dummies provide other information than in table 8, as hours worked is a continuous variable. The interaction dummy is positive and significant at the 0.01 level for the below 12 group, and at the 0.05 level for the below 4 group. The coefficient is 0.163 for the below 12 group. That means that the policy change increased the chance of being employed with 16.3% for that group. That is 14.1% for the below 4 group.

(22)

22 Table 9; regression results for employment levels

Difference-in-Difference coefficients for employment

Parents with youngest child <4 Parents with youngest child >3 & <12 Group dummy -0.042 (0.041)a -0.035 (0.035) Policy dummy -0.021*** (0.198) -0.024*** (0.006) Interaction dummy 0.141** (0.059) 0.163*** (0.035) Total N = 1457 R-squared = 400 0.2769 1057 0.2774

Notes: * = p<0.10, ** = p<0.05 & *** = p<0.01. A complete list of the dif-in-dif coefficients of the control variables is found in the appendix.

The first hypothesis states that the policy change has a positive effect on hours worked and employment levels of the treatment group. Although there is a positive effect visible for both treatment groups, the effect is not significant at 0.05 level. This suggests that the effect of the introduction of the income-independent tax credit did have a positive effect, but the effect was not statistically significant. Therefore, the analysis does not support the first hypothesis. However, the treatment lead to an increase of more than three hours for the below 4 group. That is a large increase, although not statistically significant. Therefore, the policy does seem be effective. The effect is positive, but much smaller for the below 12 & above 3 group, meaning that the policy change does not affect both groups equally. The treatment effect on employment levels is statistically significant for both treatment groups. These findings support the first hypothesis for employment levels.

As discussed in the methodology section, the number of observations for fathers and single parents is very low. This makes it difficult to test the second and third hypotheses with a regression analysis. However, the results of the regression analysis used to test the first hypothesis can provide some information about the differences between fathers and mothers, and the difference between having a partner or not. Being a mother has a negative effect on hours worked per week for both treatment groups, as mothers work 9 hours per week less than men (the coefficients are found in the appendix). The descriptive statistics of tables 4 and 5 show that hours worked of mothers increased largely for the below 4 treatment group (from 25 hours to 29 hours per week). However, this large increase is not visible for the mothers in the other treatment group (an increase of not even 1 hour per week). The fathers in both groups worked more hours after the policy change than before. This may suggest that the treatment affected fathers more than mothers, which contradicts the second hypothesis.

The third hypothesis states that coupled parents benefit more from the policy change than single parents do. Having a partner negatively affects hours worked and employment levels for both groups (see appendix for the exact coefficients). This suggests that having a partner does lowers the necessity to work. This could also mean that single parents are therefore more employed than coupled parents are in the first place. The data from table 6 is in line with that argument. Only the single parents in the below 12 & above 3 group are less employed than the coupled parents of that group. The single parents of the other groups are relatively more

(23)

23 employed than the coupled parents before the reform. There are too little observations too really make a strong argument about a difference in the effect of the policy change for coupled and single parents. The data from table 6 and 7 do not provide a clear direction of differences in the treatment effect for coupled and single parents. This could mean that there is not a large variation in the treatment effect on coupled and single parents, which would reject the third hypothesis.

5.2 Placebo test results

Table 10 shows the results of the placebo regression analysis for hours worked per week. Table 10; regression results of the placebo test for hours worked.

Parents with youngest child <4 Parents with youngest child >3 & <12 Group dummy -2.777** (1.411)a -4.805*** (1.005) Policy dummy -0.014 (0.213) -0.003 (0.215) Interaction dummy 3.491* (2.077) 1.058 (1.144)

Placebo time dummy 0.273

(0.231)

0.247 (0.233) Placebo interaction dummy 1.396

(2.691) 1.114 (1.379) Total N = 1457 R-squared 400 0.2504 1057 0.2535

Notes: * = p<0.10, ** = p<0.05 & *** = p<0.01.

Both placebo dummies are not significant. Therefore, the common trends assumption is not violated. Table 11 shows the results of the placebo regression analysis for employment levels. Contrary to hours worked per week, the placebo interaction dummies are significant for employment levels for both groups. This is a problem for interpreting the results of the regression analyses for employment levels. Namely, the treatment effect observed is probably not due to the policy change of 2009. The placebo test is significant, meaning that there is a treatment effect visible for 2008.

The results of the placebo test for employment levels are in line with figure 2, which graphically displays the mean values of employment levels per year for all three groups. This figure shows that employment levels for all three groups began to rise from 2005 onwards, with a very sharp increase in employment for the below 4 group in 2008. There also was no sharp cutoff visible for 2009. The results of the regression for estimating the treatment effect on employment levels are significant, as the placebo tests are as well. This means that it is likely that something else caused the rising employment levels. Consequently, there is no support that the 2009 policy change positively affect employment levels for the treatment groups.

(24)

24 Table 11; regression results of the placebo test for employment levels.

Difference-in-Difference coefficients for employment

Group with children below 4 Group with children below 12 Group dummy -0.061 (0.434)a -0.047 (0.035) Policy dummy -0.022*** (0.006) -0.025*** (0.006) Interaction dummy 0.160 (0.063) 0.174*** (0.037)

Placebo time dummy -0.007

(0.006)

-0.008 (0.006) Placebo interaction dummy 0.161**

(0.073) 0.088** (0.042) Total N = 1457 R-squared 400 0.2770 1057 0.2774

Notes: * = p<0.10, ** = p<0.05 & *** = p<0.01.

5.3 Childcare benefits results

It is not possible to separate the treatment effect of the changes in childcare benefits (2005) and the change in the tax credit (2009). Table 12 and 13 show the results of the regressions for the outcome variables per year. Each regression includes another year with 2005 as the treatment year for all regressions. The results show the evolvement of treatment effect of the 2005 childcare benefit change.

Table 12; regression results for different years for hours worked

Group with children below 4 Group with children below 12 Interaction dummy 2005 -1.732 (2.852)a [0.2680]b 1.741 (1.507) [0.2717] Interaction dummy 2006 -1.314 (2.550) [0.2660] 0.599 (1.225) [0.2704] Interaction dummy 2007 -1.634 (2.539) [0.2616] 1.009 (1.187) [0.2653] Interaction dummy 2008 -1.246 (2.546) [0.2615] 1.141 (1.150) [0.2651] Total N = 1023 295 728

b)_{R-squared value}