Effect of Parental Leave Benefit Reform
on Mother
's and Child's Health
Empirical Evidence from Germany
Sharfaa S.N. Azeez
Faculty of Economics and Business Economics,University of Groningen
Supervisor: Dr. M. Groneck
June 8, 2018
Abstract
This paper analyses whether the new parental leave law in 2007 resulted in better health of both mother and child. To reform is meant to have better financial support for families and encourages more father involvement in child rearing. Using the German Social Economics Panel (SOEP), a difference-in-difference method is used, which compares (mothers of) chil-dren born before 2007 with (mothers of) chilchil-dren born after 2007. Using this dataset, no evidence of better health after the reform is found for mothers. Moreover, mixed evidence is found for children of these mothers.
JEL Classifications: I1
Keywords: Health, Parental leave, mother's health, child's health
1
Introduction
An important reform of the German parental leave law included a parental leave allowance based on earnings rather than a flat rate. Germany has implemented a parent allowance of 67% of previous income (if that person has worked in the 12 months before their baby is born). The amount that is received ranges from 300 to 1800 euros, in contrast to the 300 euro before the reform. Parents are eligible for the parental leave for 12 (only one parent take the leave) to 14 (both parents take the leave) months. This encourages fathers also to take a leave for two months minimum. To reform is meant to have better financial support for families and encourages more father involvement in child rearing. The underlying aim of such a reform is to achieve higher fertility rate and labor participation rate of women, as these rates are very low in Germany compared to other European countries. In 2017, the fertility rate in Germany was 1.45 children per woman. In their neighboring country France and the Netherlands, this was 2.07 and 1.76 children per woman, respectively1.
Although the aim of the reform is not the improve the health of mothers nor the children, this paper examines whether the parental leave benefit reform in 2007 resulted in an increase in health. In other words, whether the reform has resulted in positive externalities in the form of better health for those who were affected by the reform. For example, better mental health for mother is achieved through having a more secure income in the period that the mother is taking care of her child (Hewitt et al., 2017). Moreover, due to parental leave mothers are breastfeeding their child for a longer period, which results in better child's health (Baker and Milligan, 2008).
To measure the effect of the reform two methods will be used; a paired t-test and Ordi-nary Least Square (OLS). The latter will be done through a difference-in-difference (DiD) approach, which will indicate the effect of the reform. To show the impact of the parental leave benefit reform in Germany on child's and mother's health, the German Socio-Economic Panel (SOEP) will be used. Interviews have started from 1984 and are still ongoing. By using the DiD approach, (mothers of) children born before the 2007 reform are compared with (mothers of) children born after the 2007 reform (Baker and Milligan (2008a); Dust-mann and Sch¨onberg (2012); Rossin (2011); Hewitt et al. (2017)). Moreover, this method will also be used to eliminate the endogeneity problem (Avendano et al. (2015)). That is that mothers decide to return to work later if they experience low health. Moreover, no information about parental leave (i.e. the number months and amount received) is necessary
1See Central Intelligence Agency, The World Factbook:https://www.cia.gov/library/publications/
to show the effect of the reform.
The DiD is the interaction between a treatment dummy and reform dummy, which will show the health differentials between a group that did experience change after the reform and a group that was not affected by the reform. Since the latter group includes individuals with low income, it will also show whether the reform resulted in more health inequality. When the coefficient of this variable is positive, it indicates that there are indeed health benefits for mother and/or child after the reform. In this paper, I aim to expand previous research by analyzing both mother's and child's health.
The hypothesis that I would like to verify in this research is that changes in the parental leave law result in better health of both the mother and child. In other words, improved financial support and more father involvement in child rearing result in better health. Several studies already have found a positive relationship between maternity leave and child's health. There is also evidence that the length of maternity leave has an impact on the mental health of the mother. This will be discussed in the literature review of this paper.
Hence, two question will be answered in this paper:
1. Does change in the parental leave law have an impact on mother's health? 2. Does change in the parental leave law have an impact on early childhood health?
This paper finds no evidence of increase in the health of mothers, for which several health measures are used. Moreover, for children mixed evidence of the reform on health is found. The positive effect that is found is only a short-term effect.
The remainder of this paper is outlined as follows. Section 2 analyses the theory and main findings of parental and maternity leave on health. Section 3 discusses the estimation model that is used to analyze the effect of parental leave benefit reform on health. Section 4 discusses the data, sample selection, and descriptive statistics. Section 5 presents the empirical results of mother's and child's health. The final section discusses several concerns of the model and draws a conclusion.
2
Related Literature
2.1
Mother
's health
That mother's health improved with the length of the maternity leave was found by
Chatterji and Markowitz (2012). Short maternity leave consisting of less than 12 weeks was associated with having more depressive symptoms. Moreover, they found that an even shorter leave of less than 8 weeks was associated with having worst general health. They concluded that longer maternity leaves result in better health. This was also found by Avendano et al. (2015), who did similar research but instead looked at the long-run impact. They found evidence that depression among women above 50, is caused by the maternity leave policies that are implement at the time when the first child is born. This relationship between maternity leave policies and mothers at older age having less depressive symptoms is explained through the short-term effect of these policies.
Above described research focussed mainly on the length of maternity leave. Hewitt et al. (2017) analyzed whether having a near full paid parental system positively affects health. Firstly, they found that there was a significant mental and physical health improvement for mothers who were applicable to this system. Better physical health is explained by the hold up of mothers making use of formal child care, which in turn reduces the chance of obtaining infectious diseases. Moreover, improvement in mental health is explained by having a more secure income and less immediate need to return to work. Health improvement was most present in the most advantaged and disadvantaged groups. The advantaged groups were able to return later to work without affecting their income too much. For the disadvantaged group, improvements in the mental health were found as they obtained a more secure income.
Similar research was done by B¨utikofer et al. (2018). A new policy in Norway was
that when a reform is introduced, where the government's aim is to stimulate mothers to take longer leaves, overall the policy is effective. However, the degree of this effect is much stronger for the disadvantaged group.
Poor health has negative effects on labor market outcomes. Currie and Madrian (1999) find that “poor health reduces the capacity to work and has substantive effects on wages, labor force participation, and job choice”. Lalive and Zweimm¨uller (2009) did research on the effectiveness of parental leave in Austria, and found that it has a significant effect on fertility in both the short and long run in two ways. Firstly, parental leave benefits result in more births in a shorter period of time. Secondly, the cost per child reduces, which
increases the number of births. Moreover, Ruhm (1998) found that women's employment to
population (EP) ratio increased due to the implementation of parental leave. This is a result of women deciding to enter the market, prior to having children, who otherwise would not have done this. In addition to this, mothers are more likely to return to work after having children. The latter was also found by Baker and Milligan (2008b); they found evidence that this holds for all lengths of maternity leave. However, longer maternity leaves are associated with mothers not having the decision to never return to work (e.g. to care for their children) or mothers making the switch from a full-time job to part-time job (e.g. when their infants are still very young).
2.2
Child
's health
Berger et al. (2005) found a positive relationship between maternity leave and child's health. They found that short maternity leave (i.e. mother returning to her job in less than 12 weeks) results in mothers having less ability to give the care that the child needs, such as breastfeeding their baby for a longer period and/or to vaccinate their child for certain immunizations. As was found by Hewitt et al. (2017) and, Baker and Milligan (2008a), that subgroup analysis show mixed results of the impact of parental leave policies on mother's health; also evidence is found by Rossin (2011) on the effect on child's health. For the advantaged group (e.g. highly-educated and married mothers) are more likely to be eligible for the maternity leave, which means that the children of these mothers have higher birth weight, less chance to be born prematurely and decrease in the mortality rate.
However, for the mothers of the disadvantaged group (e.g. low-educated and single
implemented.
Also, Ruhm (2000) finds evidence that suggests that parental leave leads to poorer health in early childhood, which is measured by lower child mortality rates. These lower rates are explained by the prevention of certain health risks due to increased child care by parents. He also finds that when the leave is longer, the mortality rate becomes even smaller. Both these papers suggest a positive relationship between time investment in early childhood care and child's health. This is in line with the evidence found by Tanaka (2005). She discussed in her paper that compared to unpaid parental leave or parental leave without any job protection, paid leave has resulted in significantly lower mortality rates. Although infant mortality is often explained by low birth weight, the paper of Tanaka (2005), showed that improved parental leave policies did not result in higher birth weights. This suggests that these lower mortality rates are a result of something else. However, this is not discussed in her paper.
As was discussed in the previous section, it was shown that the length mothers provide their children breastmilk is determined by the length of the parental leave. This, in turn, is closely related to health. In other words, if longer leave is provided then mothers increase the length of breastfeeding, which results in health benefits. This is also in line with previous studies done on this topic.
Baker and Milligan (2008) find evidence that longer maternity leave entitlements result in mothers taking longer leave and an increase in the length that mothers breastfeed their children. They believe that breastfeeding and health are positively correlated with each other. In other words, longer length of breastfeeding period results in better health. In their paper, they mention that enormous research has been done on this topic, where it is found that breastfeeding reduces “nose/throat infections, ear infection, asthma, wheeze, allergies, bronchitis and chronic condition” when looking at the first two years of child's life. These are also the variables that the authors have used for measuring child's health, as they are closely related to breastfeeding. Moreover, there is evidence that breastfeeding results in better health for mothers. Baker and Milligan (2008a) did find an improvement in the general health of mothers. Nevertheless, they did not find a significant improvement in health for children in the above-mentioned measures when longer maternity leave is provided.
providing longer parental and/or maternity leave.
3
Empirical Strategy
To show the effect of the reform a difference-in-difference method is used. The main reason for this is that it enables to compare (mothers of) children born before 2007 with (mothers of) children born after 2007. To apply the difference-in-difference (DiD) method, a control group and treatment group needs to be identified from the data. The former should be a sample of individuals who were not affected by the reform, while the latter does. This is also known as the common trend assumption. Note that the control and treatment group can have different characteristics, but the differences between them do not change over time. For this paper, similar control group and treatment group as Kottwitz et al. (2014) are used. They did research on the effects of the 2007 reform on breastfeeding behavior, for which they found no change on the breastfeeding initiation of mothers who breastfeed their child for at least six months. They do, however, find evidence for mothers who breastfeed their child for at least four months.
The control group consists of the group of mothers who were eligible for 300 euro per month before and after the reform. In the data, these are the mothers whom have a household income less than 30,000 euro and are not working. Hence, this sample group does not face much change after the reform. The treatment group are mothers who did not receive any
benefit or a benefit below 300 euro before the reform. After the reform, this group is
eligible for a benefit that is based on earnings, which can range from 300 to 1800 euro per month. In the data, these are the working and non-working mothers who have a household income more than 30,000 euro. Hence, a new variable will be created from the data called treatmentdummy. By using the DiD method, time trends are controlled through the inclusion of control groups in the regressions. This means that the effect of the reform is separated from other changes that are occurring at the same time.
Secondly, a variable needs to be created that divides the above-mentioned sample into a group before and after the reform. That is a dummy variable equal to one if the child is born after 2007 and equal to zero if the child is born before 2007. This will be referred to as reform2007. The effect of the reform is then estimated by the interaction of the variables treatmentdummy and reform2007.
The regression function to estimate the effect on mother's health can be written as:
Health M other = β0+ β1treatmentdummy + β2ref orm2007
+ β3treatmentdummy × ref orm2007 + x0βx+ ,
Table 1: Correlation Health Measures
PCS MCS
Health Satifaction 0.6170 0.3052
Current Health State 0.7072 0.2701
MCS -0.0025 1.000
Notes to Table 1. The table shows the correlation between health satisfaction and current health state with physical and mental health state, respectively. Also, the correlation between physical and mental health is shown in this table.
where for the dependent variable different measures are used. These measures are health satisfaction, current health state, physical health and mental health. Values for health satisfaction ranges from zero to ten, where values closer to ten indicates that the individual is satisfied with their health. For the current health state, the values range from zero to four, where zero is when the individual indicates that he/she has a poor current health state, while four indicates that the individual finds that he/she has excellent current health state. Mental and physical health ranges from zero to 100, where a higher score indicates better mental or physical health. These two measures are created from twelve questions (SF-12v2) that are asked in the health section of the questionnaires. The questions are divided into “eight subscales” and then into “two superordinate dimensions of physical and mental health”. The latter is known as the PCS and MCS, which refers to “Physical Component Summary Scale and “Mental Component Summary Scale”, respectively (Andersen et al., 2007).
The measures health satisfaction and current health state are highly correlated with physical health, indicating that individuals when answering such questions are referring to their physical health rather than the mental health. The correlation of health satisfaction and current health state with physical health are 0.617 and 0.707, respectively. The correlation between health satisfaction and current health state, and mental health are 0.305 and 0.270, respectively. See also Table 1. The correlation between physical and mental health is -0.0025. The independent variables are a variable that indicates the treatment group and control group, a variable that indicates the reform, and the interaction of these variables. The latter measures the effect of the reform. The control variables, which is presented in the vector of x, are age, age-squared, the log of household income, number of years of education, dummy variable of whether the individual is married2, number of children in the household, body
mass index (BMI), dummy variable of whether the individual smokes3, dummy variable of
whether the individual does sports activities4, dummy variable of whether the individual
2The dummy variable is equal to one if the individual is marrried. 3The dummy variable is equal to one if the individual smokes.
4The dummy variable is derived from the questionnaire in which individuals are asked about how often
Table 2: Regression Function: E(health variable)
Treatmentdummy Reform2007 Coefficients
Control Before (CB) 0 0 β0
Treatment Before (TB) 1 0 β0 + β1
Control After (CA) 0 1 β0 + β2
Treatment After (TA) 1 1 β0 + β1 + β2 + β3
Notes to Table 2. This table shows the parameters of the control and treatment group, before and after the reform.
follows a healthy diet5, difficulties the individual has with daily activities6, and percentage of disability.
The regression function to estimate the effect on child's health can be written as:
Health Child = β0+ β1treatmentdummy + β2ref orm2007
+ β3treatmentdummy × ref orm2007 + x0βx+ ,
(2)
where the dependent variable is the extent the mother worries about the health of the child. The value ranges from one to four, which stands for that the mother “does worry” to “does not worry” about the health of their child. Also, the number of hospital visits in the first three months after childbirth will be used as a dependent variable. The effect of the reform on child's health is measured similarly as mother's health. For this, the same treatment and reform dummy are used in the regression.
The covariates, which is presented in the vector of x, are age of the mother, the log of household income, months of parental leave, squared of months of parental leave, the gender of the newborn, the week in which the child is born and the number of child the newborn is in the household.
As discussed above, the interaction term between reform2007 and treatmentdummy shows the effect of the reform. The regression function, E(health outcome), for the dif-ferent groups before and after the reform can be found in Table 27. The coefficient β3 is
the DiD, which is the difference between the treatment group before (TB) and after (TA) the reform minus the difference between the control group before (CB) and after (CA) the
activities, or exercises rarely or at least once a month. The dummy variable is equal to one if the individuals exercise at least once a week, or everyday.
5The dummy variable is derived from the questionnaire in which individuals are asked about their diet.
The dummy variable is equal to zero if the individual follows little or no healthy diet, and equal to one if the individual follows a strong or very strong healthy diet.
6The variables range from 0 to 2, with 2 indicating that the individual does not have any difficulties
with daily activities
7Ignoring the vector of x for the moment, in order to show the effect of the reform by using the
reform. This can be written as follows:
β3 = (T A − CA) − (T B − CB)
= (β0+ β1+ β2 + β3− β0− β2) − (β0+ β1− β0)
(3)
For both the mother and child, it is expected that the β3 is positive. In other words, the reform results in better health. Hence, this would also mean that there is more inequality between the control and treatment group.
4
Data and Sample Selection
The longitudinal/panel data is drawn from the German Socio-Economic Panel (SOEP), which consists of approximately 30,000 respondents in 11,000 households. The first interview was in 1984 and is still ongoing. Data is available until 2016. The “Individual Question-naire” consists of the following topics: current life situation, current job, life situation in the past calendar year, income in the past calendar year, health and illness, attitudes and opinions, and family situation and background. In addition to this, since 2003 information is gathered of newborns (“Mother and Child Questionnaire”), which consists of questions regarding general information about the newborn, health about the newborn and mother after childbirth, and life changes after childbirth8.
A period of 4 years before the reform is compared to a period of 4 years after the reform. Hence, the data from 2003 until 2010 is collected from SOEP. The short period before and after the reform is chosen, because the differences between the individuals that is caused over time are the smallest in this period. For instance, it is difficult to compare a child born in 1990 with a child born in 2010.
The raw data contained many missing values due to low retention rates and questions that were not asked every year. As a result of this, the regressions that are done are left with only few observation. To solve this problem, missing values are replaced with the value that was given in the previous calendar year. For example, individuals were asked every two years whether they were smoking or not. Hence, the data missed values in the years between. Due to the reason that it is very unlikely that individuals will stop smoking in the following year, the missing values are replaced with the answers that they gave in the previous year. Approximately, half of the missing variables for this smoking variable are replaced by using this method.
8More information regarding the data and documentation (i.e. questionnaires, codebooks and survey
Similar method was used for the following variables: several health variables (e.g. physical and mental health), education years, how often the individual involves in sports activities, the extent for which the individual follows a healthy diet, difficulties the individual has with daily activities, BMI, and household income. Nevertheless, by using this method still large amount of missing values exists, which reduces the number of observations to approximately 1300. The missing values are, especially, present by mothers who also have to fill out the “Mother and Child Questionnaire”.
Table 3 shows the descriptive statistics of the variables that are used in equation (1) and (2), which is divided into the control and treatment group. This table shows that there is a clear difference between the treatment and control group, which could be explained by the difference in household income. See also the previous section for more information about these groups. The values of the mean are slightly higher for the treatment group. In addition to this, the treatment group has a healthier lifestyle. Individuals in the treatment group are less likely to smoke, follow a healthier diet, do more sports activities and have a lower BMI. In Appendix A, the descriptive statistics of the full sample is shown. For most of the variables, the averages of the full sample are between the averages of the treatment and control group. The reason for this is that the control group is the most disadvantaged group
Table 3: Descriptive Statistics
Control Group Treatment Group
Mother N=899 Mother N=1,826
Dependent Variables Mean St. Dev. Min Max Dependent Variables Mean St. Dev. Min Max Health Satisfaction 7.588 2.043 0 10 Health Satisfaction 7.597 1.949 0 10 Current Health State 2.770 0.911 0 4 Current Health State 2.791 0.866 0 4
Physical Health 54.169 7.262 22.170 71.313 Physical Health 54.29 7.027 24.070 70.332 Mental Health 48.563 9.859 10.924 67.194 Mental Health 49.010 9.368 10.924 70.766 Control Variables Mean St. Dev. Min Max Control Variables Mean St. Dev. Min Max
Age 32.350 5.908 15 64 Age 33.968 5.934 15 56
log Household Income 7.712 0.508 5.635 9.903 log Household Income 7.948 0.493 5.635 9.903 Household Income (e) 2552.733 1518.127 280 20000 Household Income (e) 3184.369 1679.149 280 20000
Education Years 12.505 2.855 7 18 Education Years 13.100 2.891 7 18
Married 0.684 0.465 0 1 Married 0.774 0.418 0 1
No. of Children 1.823 0.894 1 8 No. of Children 1.961 1.010 1 11
Health Activities 1.670 0.533 0 2 Health Activities 1.668 0.543 0 2
Smoker 0.265 0.441 0 1 Smoker 0.156 0.363 0 1
Diet 0.556 0.497 0 1 Diet 0.644 0.479 0 1
Sport Activities 0.394 0.489 0 1 Sport Activities 0.549 0.498 0 1
BMI 24.689 4.838 14.533 47.323 BMI 24.091 4.201 16.225 47.323
Disability (%) 0.823 7.135 0 100 Disability (%) 0.719 6.389 0 100
Newborn N=1,324 Newborn N=1,930
Dependent Variables Mean St. Dev. Min Max Dependent Variables Mean St. Dev. Min Max
Health Worry 3.406 0.843 1 4 Health 3.422 0.805 1 4
Hospital Visits 1.500 7.361 0 99 Hospital visits 1.548 7.359 0 99
Control Variables Mean St. Dev. Min Max Control Variables Mean St. Dev. Min Max
Gender 0.496 0.500 0 1 Gender 0.517 0.500 0 1
No. of Newborn 1.626 0.883 1 8 No. of Newborn 1.660 0.949 1 12
Birth Week 39.141 2.314 26 46 Birth week 39.124 2.309 25 46
Parental Leave 3.619 4.700 0 12 Parental Leave 3.801 4.692 0 12
(i.e. lowest household income group), while the latter is the most advantaged group (i.e. highest household income group). In appendix B, graphs can be found of the different health measures, which shows the mean averages over the years (from 2003 until 2010). Graphically, there is no clear identification of the reform.
Aitken et al. (2015) did extensive review of previously done research on the impact of maternity an parental leave on mother's health. He found in his analysis that there are several problems with how the research is executed, which are also present in this paper. Firstly, there is selection bias, as there are often missing data; plus for the interviews done on cohorts, the retention rates are very low. Secondly, for most of the dependent variables, researchers use a health measure that is self-reported. This is problematic, because it results in “information bias” by not responding with true values. Thirdly, there might be problems in the observational designs. In other words, the observed effect that researchers find might not be due to maternity leave, but due to other unobservable effects. Aitken et al. (2015) give a very good example; that is that women who, for instance, receive paid leave are dif-ferent to women who receive unpaid leave. Thus, the difference in health is not explained by having paid/unpaid leave. Lastly, which was also explained by Chatterji and Markowitz (2005), there is an endogeneity problem, more specifically reverse causation. This endogene-ity problem refers to the fact that mothers who have poor health during pregnancy or after childbirth, might decide to stay longer home. The latter two problems will be solved by using a difference-in-difference method, which is discussed in previous section.
5
Results
5.1
Paired t-test: Before and After Reform
Table 4 shows that mothers experience lower health after the reform, which holds for all the different health outcomes (i.e. health satisfaction, current health state, physical health and mental health). Looking at the mean of health satisfaction, after the reform the outcome is 0.443 higher than before the reform (7.711 versus 7.268). For the average current health state, it was 0.137 higher (2.821 versus 2.684). Also for the physical and mental health, the average was higher after the reform. The difference was 0.855 and 0.768 higher respectively (53.778 versus 54.633 and 48.168 versus 48.936, respectively). Conducting a paired t-test showed that these differences before and after reform are statistically significant, implying that the difference is non-zero.
Table 4: Mean of Health Variables Before and After Reform
Mean Mean Mean Mean Mean Mean
HS CHS PCS MCS NBH NHV
Before Reform 7.268 2.684 53.778 48.168 3.419 1.914
After Reform 7.711 2.821 54.633 48.936 3.381 1.505
Difference 0.443*** 0.137*** 0.855*** 0.768* -0.038 -0.409 Notes to Table 4. *p<0.10, **p<0.05, ***p<0.01. HS, CHS, PCS, MCS, NBH, and NHV stands for the variables health satisfaction, current health state, physical health, mental health, extent of which mother worries about child's health, and the number of hospital visits. It shows the mean of the different health variables. For the all the mother's health outcomes, the mean is higher after the reform. For the the chid's health outcomes, the mean is lower. Conducting a paired t-test showed that these differences in health before and after the reform is statistically significant for the mother, which implies that the difference is non-zero. However, for child's health, the differences are not statistically significant.
(1.505 versus 1.914). However, when again conducting a paired t-test, it shows that these differences in child's health before and after reform are not statistically significant.
Nevertheless, this paired t-test is not feasible evidence to show the effect of the reform on the health outcomes. It required to control for factors that have effect on these health outcomes other than the reform itself, which can be done through the regression estimation that was discussed in the previous section.
5.2
Empirical Analysis
The empirical results from estimating the health of the mother are shown in Table 5 and Table 6. Column (1) and (4) are the regressions excluding the variables that measure health and the reform. Column (2) and (5) are the regressions including health variables, but excluding reform variables. Column (3) and (6) includes all variables. The empirical results from estimating the health of the child are shown in Table 7. Column (1) and (4) are regression excluding the reform variables.
5.2.1 Mother's Health
The first column of Table 5 and Table 6 presents the coefficients of the estimation that does not include health variables or the variables that measure the reform.
Table 5: OLS Regression: Health Satisfaction and Current Health State
Health Satisfaction Current Health State
(1) (2) (3) (4) (5) (6) age -0.055*** -0.067*** -0.051 -0.017*** -0.018*** -0.068* (0.008) (0.010) (0.084) (0.003) (0.004) (0.036) agesquared 0.000 0.001*** 0.000 -0.000* 0.000* 0.001 (0.000) (0.000) (0.001) (0.000) (0.000) (0.000) lnHH 0.433*** 0.221*** -0.084 0.192*** 0.100*** 0.102 (0.029) (0.034) (0.184) (0.011) (0.014) (0.078) educ years 0.062*** 0.012** -0.008 0.033*** 0.010*** 0.003 (0.005) (0.006) (0.020) (0.002) (0.003) (0.008) married 0.114*** 0.090** 0.100 0.010 -0.024 0.008 (0.031) (0.037) (0.143) (0.012) (0.016) (0.060) no children HH 0.055*** 0.020 0.030 0.005 -0.006 -0.004 (0.015) (0.019) (0.051) (0.006) (0.008) (0.022) BMI -0.024*** 0.004 -0.009*** -0.001 (0.003) (0.012) (0.001) (0.005) smoker -0.145*** -0.307** -0.086*** -0.152*** (0.034) (0.137) (0.014) (0.058) sport dummy 0.126*** 0.053 0.108*** 0.138*** (0.031) (0.106) (0.013) (0.045) diet dummy 0.182*** 0.474*** 0.061*** 0.193*** (0.030) (0.105) (0.013) (0.045) health activities 1.200*** 1.092*** 0.521*** 0.470*** (0.026) (0.093) (0.011) (0.039) disability percentage -0.018*** -0.017** -0.009*** -0.007** (0.001) (0.008) (0.001) (0.003) treatmentdummy 0.254 0.020 (0.195) (0.083) reform 2007 0.157 0.071 (0.410) (0.174) DiD 0.112 0.020 (0.435) (0.185) Contant 4.789*** 5.393*** 6.851*** 1.407*** 1.587*** 2.312*** (0.255) (0.330) (1.985) (0.100) (0.140) (0.844) Observations 25,183 15,354 1,339 28,751 15,364 1,339 R-squared 0.056 0.209 0.137 0.062 0.226 0.156 F-Test 250.73 338.39 13.98 317.39 373.46 16.25
Table 6: OLS Regression: Health Satisfaction and Current Health State
Physical Health Mental Health
(1) (2) (3) (4) (5) (6) age 0.090*** 0.088*** -0.006 -0.148*** -0.243*** -0.448 (0.031) (0.030) (0.254) (0.041) (0.054) (0.438) agesquared -0.004*** -0.002*** -0.000 0.002*** 0.004*** 0.006 (0.000) (0.000) (0.004) (0.001) (0.001) (0.006) lnHH 1.507*** 0.315*** -0.000 2.289*** 1.680*** 0.944 (0.112) (0.102) (0.558) (0.144) (0.181) (0.963) educ years 0.314*** 0.121*** 0.018 -0.065** -0.134*** 0.030 (0.021) (0.018) (0.061) (0.027) (0.033) (0.105) married -0.647*** -0.608*** -0.449 1.476*** 1.224*** 1.335* (0.127) (0.113) (0.434) (0.163) (0.202) (0.749) no children HH 0.266*** 0.159*** 0.271* -0.534*** -0.706*** -0.546** (0.063) (0.056) (0.155) (0.081) (0.100) (0.267) BMI -0.136*** -0.103*** 0.095*** 0.182*** (0.010) (0.036) (0.017) (0.062) smoker -0.280*** -0.182 -1.203*** -0.310 (0.101) (0.417) (0.181) (0.719) sport dummy 0.206** 0.434 -0.103 0.421 (0.094) (0.322) (0.169) (0.556) diet dummy -0.178* 0.441 1.443*** 1.729*** (0.091) (0.319) (0.164) (0.551) health activities 9.020*** 8.592*** 2.610*** 1.943*** (0.077) (0.282) (0.138) (0.487) disability percentage -0.059*** -0.046* -0.044*** -0.031 (0.004) (0.023) (0.008) (0.041) treatmentdummy 0.520 0.231 (0.593) (1.024) reform 2007 0.774 1.988 (1.244) (2.148) DiD -0.002 -2.461 (1.318) (2.276) Contant 40.208*** 37.961*** 46.633*** 33.972*** 34.463*** 40.405*** (1.008) (0.993) (6.020) (1.299) (1.779) (10.393) Observations 23,743 15,337 1,337 23,743 15,337 1,337 R-squared 0.080 0.551 0.442 0.022 0.059 0.036 F-Test 345.45 1565.70 69.70 88.95 80.10 3.31
The second column of Table 5, includes the health variables, but excludes the variables that measure the reform. The signs are the same as in column (1). Moreover, all variables are statistically significant, except for no children HH. Body Mass Index (BMI), whether the individuals smokes (smoker) and percentage of disability negatively affect the health satisfaction. Also, agesquared does have a significant and positive effect now. This is not as expected. From the variable age, there is indeed a decrease in health as people get older. However, the positive sign of agesquared indicates that from a certain age the health satisfaction increases again9. The variables of whether the individual does sports activities (sport dummy), whether the individual follows a healthy diet (diet dummy), and the extent of having difficulties by doing daily activities (health activities) positively affects health satisfaction of the mother.
Column (3) includes all variables. Only the variables smoker, diet dummy, health activities and disability percentage are statistically significant. The coefficients of these variables have the same signs as discussed above. Moreover, the treatment dummy (treatmentdummy), reform dummy (reform 2007), and DiD are positive. Although, these are not statistically significant, indicating that the reform measured by the variable DiD, did not have an effect on the health satisfaction of mothers.
When using current health state as the dependent variable, it shows that most of the vari-ables (excluding health and reform varivari-ables) have similar signs as the regression of health satisfaction. Nevertheless, married and no children HH are not statistically significant, while agesquared is now positive and significant at 10 percent significance level. The latter coin-cides more with expectations. When the health variables are included, the sign of agesquared changes to positive. Moreover, married and no children HH are still not statistically signifi-cant. All the health variables are significant at 1 percent significance level and have the same signs as the regression of health satisfaction excluding reform variables (column (2)). When including the variables that measure the reform, only the following variables are significant: age, smoker, sport dummy, diet dummy, health activities and disability percentage. The signs do not change after including the reform variables. The treatmentdummy, reform2007 and DiD are again positive and not statistically significant. As before, this indicates that the reform did not have an effect on the current health state.
The coefficients from the regression of physical health excluding health and reform vari-ables, show similar results as for health satisfaction and current health state. All variables
9Taking the derivative of equation (1) with respect to age results in the following: -0.067 + 0.001 x age
are statistically significant. The variable age is now positive, which is not in line with expec-tation. However, with agesquared being negative; it suggests that health increases as you get older, but decreases again after a certain age10.
In addition to this, the estimator of married is negative, which is not what is expected. It is expected that being married results in better health. However, it could be the case that, for example, the husband/wife of the individual is a heavy drinker, which in turn can also shape the behavior of that individual. Including health variables, the signs remain the same. The health variables are significant and similar to what was found in the regressions of Table 5. However, the diet dummy is now negative. This indicates that if the individual follows a healthy diet, their physical health decreases. This, of course, is very unlikely. By includ-ing all variables, only no children HH, BMI, health activities, and disability percentage are statistically significant. Of those variables, the coefficients are similar to previous regression and to what is expected. The treatmentdummy and reform dummy is still positive and not significant. For the DiD, the sign is negative, but still not statistically significant. Hence, the reform has no effect on physical health.
The regression, excluding reform and health variables, of mental health showed the follow-ing. The variables agesquared, lnHH and married positively affected the mental health, while age, educ years, and number of children had a negative effect. The variables agesquared, educ years and no children HH, compared to previous regressions show different signs. The reason for this is that previous regressions measure the physical health. That mental health increases after a certain age is not strange. For example, children moving out of the house or individuals going into pension could explain the better mental health later in life. Moreover, higher education negatively affecting mental health could be explained by individuals finding the need to maintain their status, which can be mentally challenging. Also, more children in the household could decrease mental health, due to higher household expenditures.
By including the health variables, the coefficients are the same as in column (4). All health variables are statistically significant at 1 percent significance level, except for sport dummy. This variable does not have a significant effect. The signs are the same as in the previous regressions. For BMI, however, there is a negative effect on mental health. This indicates that individuals who have higher BMI have lower physical health, but better mental health.
After including the reform variables, the variables married, no children HH, BMI, diet
10Taking the derivative of equation (1) with respect to age results in the following: 0.090 -0.004 x age
dummy, and health activities are statistically significant. The signs are the same as in the regression of excluding and including health variables. There is no significant effect of the reform variables. In other words, the reform has no effect on the mental health.
5.2.2 Child's Health
The first column of Table 7 shows the coefficient of the estimation excluding the re-form variables. According to the regressions, it shows that that only gender of newborn, number of newborn in the household (nb in HH) and the week in which the child was born (birth week) has significant effect. The former has a negative effect, while the latter two variables positively affect child's health. This is in line with expectation.
When including the reform variables, the variables age (age) and birth week are signif-icant, which both have a positive sign. The treatmentdummy is significant at 5%, reform dummy at 1%, and DiD is significant at 10%. The latter indicates, that child's health in-creases as a result of the reform. After the reform, there is an increase of 0.318 in child's health. Looking at the descriptive statistics discussed in the previous section, this is an in-crease of approximately 9 percent of the average health outcome as a result of the reform11.
Nevertheless, results show opposite signs for the treatmentdummy and reform2007, which can be explained by the health outcome that is used to measure the effect. Compared to the treatment group, the control group have indicated that they worry more about the health of their child.
As can be seen from the regression, not many variables affect child's health. This could be because the child is still very young, and therefore health does not depend on many factors.
Common Trend Assumption It is important to examine whether the significant effect of the reform found in this subsection is correctly identified. For this, a placebo reform dummy is created. Instead of the reform being in 2007, it is assumed that the reform is in 2006. Hence, the new dummy will be equal to 1 if the child is born after 2006 and equal to zero if the child is born before 2006. It is expected, when doing this regression with the test dummy, that the DiD will not be statistically significant anymore. This will indicate that the “reform” will have no effect.
The regression results of the estimation function with the test dummy, shows that the coefficient DiD is indeed not significant anymore. See column (3) in Table 7. Hence, this
11Average health of a child born after 2007 (after the reform) and whose mother were in the treatment
Table 7: OLS Regression: Child's Health
Newborn Health Hospital Visits
(1) (2) (3) (4) (5) age 0.004 0.013** 0.013*** 0.021 -0.028 (0.003) (0.004) (0.004) (0.025) (0.037) lnHH 0.012 0.063 0.078 0.069 0.117 (0.034) (0.068) (0.068) (0.278) (0.564) married -0.039 0.059 0.056 -0.202 -0.569 (0.035) (0.053) (0.053) (0.285) (0.440) pl months 0.011 -0.002 -0.003 0.067 0.124 (0.012) (0.018) (0.018) (0.099) (0.145) pl months squared -0.001 -0.000 -0.000 0.001 -0.001 (0.001) (0.002) (0.001) (0.008) (0.012) nb gender -0.093*** -0.035 -0.037 0.720*** 0.452 (0.030) (0.041) (0.041) (0.241) (0.341) nb in HH 0.013** 0.006 0.008 -0.172 0.0434 (0.006) (0.022) (0.022) (0.138) (0.178) birth week 0.013** 0.016* 0.017* -1.355*** -1.429*** (0.006) (0.009) (0.009) (0.053) (0.075) treatmentdummy -0.282** -0.212 0.507 (0.113) (0.130) (0.931) reform 2007 -0.342*** -0.241** 0.444 (0.106) (0.118) (0.867) DiD 0.318*** 0.190 -0.586 (0.006) (0.118) (0.133) (0.971) Contant 2.276*** 2.107*** 1.931*** 53.111*** 56.769*** (0.336) (0.597) (0.597) (2.726) (0.012) Observations 3,222 1,614 1,614 3,216 1,611 R-squared 0.009 0.022 0.019 0.176 0.191 F-Test 3.50 3.30 2.76 85.34 34.33
Table 8: OLS Regression: Child's Health
Child's Health Hospital Visits
(1) (2) (3) (4) age 0.006* 0.008* 0.005 0.027 (0.003) (0.004) (0.014) (0.018) lnHH 0.133*** 0.174** -0.611*** -0.202 (0.041) (0.069) (0.174) (0.295) married 0.056 0.093 -0.219 -0.260 (0.045) (0.060) (0.188) (0.257) gender -0.074** -0.077* 0.130 0.239 (0.036) (0.044) (0.148) (0.185) nb in HH 0.018 0.042* -0.106 -0.241** (0.021) (0.024) (0.083) (0.097) weight 0.016** 0.024** -0.056*** -0.061** (0.008) (0.010) (0.017) (0.024) disorder -0.456*** -0.446*** 0.972*** 0.792*** (0.036) (0.045) (0.149) (0.186) treatmentdummy -0.103 -0.448 (0.094) (0.370) reform 2007 -0.148* -0.020 (0.087) (0.388) DiD 0.071 0.137 (0.101) (0.433) Contant 2.078*** 1.590*** 6.044*** 2.765*** (0.303) (1.233) (1.233) (2.117) Observations 2,277 1,380 3,209 1,972 R-squared 0.082 0.100 0.024 0.022 F-Test 28.76 15.26 4.46 4.76
shows that by assuming that the reform is in a different year, the regression will lead to different results. This suggests that the common trend assumption is satisfied. Also, when using other years for the placebo reform dummy, there was no significant effect found.
Column (4) and (5) shows the regressions where the dependent variable is the number of hospital visits in the first three months after birth. Excluding the reform variables show that only nb gender and birth week are statistically significant, which was also the case in the previous regressions described above. Also, similar results are obtained. In other words, being male and later birth week results in fewer hospital visits. By including the reform variables, only the birth week has a significant effect. The variable DiD is not significant, which indicates that there is no effect of the reform on the number of hospital visits.
Hence, the regressions for estimating the effect of the reform on child's health showed mixed results. Using the extent in which the mother worries about the health of their child, there was a positive effect of the reform. However, no significant effect of the reform on the number of hospital visits was found.
Long-run Effects on Child's Health As discussed above, when measured by the extent
of mothers worry about child's health, a positive effect of the reform was found. Therefore, also regressions for children aged 2 and 3 years old are done to analyze whether there is a long run effect. Similar equation will be used as in Section 3, where the dependent variable is the extent to which the mother worries about child's health and the number of hospital visits in the last 12 months. The independent variables are the treatmentdummy, reform2007, DiD, and a vector of control variables. The control variables are the age of mother (age), log of household income (lnHH), whether the mother is married (married), gender of the child (gender), number of children in the household (no children HH), weight of the child (weight) and whether the child has any disorders (disorders).
The regressions are presented in Table 8. The results showed that there is no effect of the reform on any of these health outcomes. In other words, there is no long-run effect on child's health.
6
Discussion and Conclusion
This paper found an increase in child's health after the reform, when measured by the extent of worry. After the reform, there is an increase of 0.318, which is statistically signifi-cant at 10 percent significance level. However, this should be interpreted with care. Firstly, the effect is only significant at 10 percent significance level. Moreover, the question that is used to measure the child's health is very person related. The question is to what extent the mother worries about the health of the child. For example, mothers who are very protective over their child or who experience stressful life events, are more prone to indicate that they worry about the health of their children. Thus, this means that there is another factor that explains child's health, which indicates that there is a possible measurement error. This is similar to the well-known example of individuals having higher wages, which could be due to having more years of education or the ability. Moreover, results are not consistent. There was no evidence of a decrease in the number of hospital visits after the reform.
It is crucial to note that the results are affected by the uncertainty of the identification of the control and treatment group. It largely relies on the common trend assumption. In other words, it is assumed that the control group does not experience any change after the reform, while the treatment group does experience change after the reform.
Another issue is that there are variables omitted in this paper. Some control variables are not taken into account in this paper, because many of these were not asked in the survey. Omitting these variables could change the results of this research. For example, for mothers more variables about the parental leave could be included. In other words, higher parental leave money could lead to better mental health. The reason for this is, that there is higher replacement rate of income, which results to a better state of mind. For children, inclusion of more characteristics of mother could have been done. As was mentioned above, protective mother or mothers with stressful life events could play a role.
In addition to this, the low retention rates, questions that were not asked every year, and individuals not filling out the questionnaire completely resulted to missing values in the dataset and a reduction in the number of observations. When doing the regressions, the number of observations is approximately 1300 individuals. Some might argue, that this is not representative for a population of around 80.5 million, with approximately 51% women12
and 8.6 newborns every 1,000 population13.
While this research showed that the parental leave benefit reform in 2007 resulted in no significant effect on health for mothers, there were health benefits found for children. Hence,
12See Trading Economics, Germany - Population:
https://tradingeconomics.com/germany/ population-female-percent-of-total-wb-data.html
13See Index Mundi, Germany Demographics Profile 2018: https://www.indexmundi.com/germany/
Appendix A
Descriptive Statistics
Table A.9: Descriptive Statistics
Mother N=74,445
Dependent Variables Mean St. Dev. Min Max
Health Satisfaction 6.593 2.248 0 10
Current Health State 2.353 0.972 0 4
Physical Health 49.120 10.296 10.106 78.113
Mental Health 49.153 0.264 1.269 77.675
Control Variables Mean St. Dev. Min Max
Age 40.345 22.286 1 101
log Household Income 7.729 0.595 2.996 11.513
Household Income (e) 2713 1940.774 0 99999 Education Years 11.998 2.620 7 18 Married 0.572 0.495 0 1 No. of Children 0.870 1.131 0 11 Health Activities 1.325 0.725 0 2 Smoker 0.245 0.430 0 1 Diet 0.594 0.491 0 1 Sport Activities 0.532 0.499 0 1 BMI 25.054 4.778 11.073 76.208 Disability (%) 6.181 19.814 0 100 Newborn N=2085
Dependent Variables Mean St. Dev. Min Max
Health Worry 3.411 0.811 1 4
Hospital Visits 1.612 7.503 0 99
Control Variables Mean St. Dev. Min Max
Gender 0.509 0.500 0 1
No. of Newborn 1.660 0.979 1 12
Birth Week 39.087 2.339 25 46
Parental Leave 4.004 3.950 0 12
Appendix B
Graphs of Mean Health Outcomes
Figure B.1: Health Satisfaction
Figure B.3: Physical Health
Figure B.5: Child's Health
References
[1]Altonji, J. G., and Blank, R. M. (1999) “Race and gender in the labor market”, Handbook of labor economics, 3, pp. 3143-3259.
[2]Andersen, H.H., M¨uhlbacher, A., N¨ubling, M., Schupp, J., and Wagner, G.G. (2007) “Computation of standard values for physical and mental health scale scores using the SOEP version of SF-12v2”, Schmollers Jahrbuch, 127(1), pp. 171-182.
[3]Aitken, Z., Garrett, C.C., Hewitt, B., Keogh, L., Hocking, J.S., and Kavanagh, A.M. (2015) “The maternal health outcomes of paid maternity leave: A systematic review”, Social Science and Medicine, 130, pp. 32-41.
[4]Avendano, M., Berkman, L. F., Brugiavini, A., and Pasini, G. (2015) “The long-run effect of maternity leave benefits on mental health: evidence from European countries”, Social Science and Medicine, 132, pp. 45-53.
[5]Baker, M., and Milligan, K. (2008a) “Maternal employment, breastfeeding, and health: Evidence from maternity leave mandates”, Journal of Health Economics, 27(4), pp. 871-887.
[6]Baker, M., and Milligan, K. (2008b) “How does job-protected maternity leave affect mothers's employment?”, Journal of Labour Economics, 26(4), pp. 655-691.
[7]Berger, L.M., Hill, J., and Waldfogel, J. (2005) “Maternity leave, early maternal employment and child health and development in the US”, The Economic Journal, 115(501), pp. F29-F47.
[8]B¨utikofer, A., Riise, J., and Skira, M. (2018) “The impact of paid maternity leave on maternal health”, NHH Dept. of Economics Discussion Paper, 04/2018, Available at:https://ssrn.com/abstract=3139823 or
http://dx.doi.org/10.2139/ssrn.3139823
[9]Chatterji, P., and Markowitz, S. (2012) “Family leave after childbirth and the mental health of new mothers', Journal of Mental Health Policy and Economics, 15(2), pp. 61-76.
[11]Dustmann, C., and Sch¨pnberg, U. (2012) “Expansions in maternity leave coverage and children’s long-term outcomes”, American Economic Journal: Applied Economics, 4(3), pp. 190-224.
[12]Han, W.J., Ruhm, C., and Waldfogel, J. (2009) “Parental leave policies and
parents'employment and leave?taking”, Journal of Policy Analysis and Management,
28(1), pp. 29-54.
[13]Hewitt, B., Strazdins, L., and Martin, B. (2017) “The benefits of paid maternity leave for mother's post-partum health and wellbeing: Evidence from an Australian
evaluation”, Social Science and Medicine, 182, pp. 97-105.
[14]Hyde, J.S., Klein, M.H., Essex, M.J., and Clark, R. (1995) “Maternity leave and women's mental health”, Psychology of Women Quarterly, 19(2), pp. 257-285. [15]Kottwitz, A., Oppermann, A., and Spiess, C.K. (2016) “Parental leave benefits and
breastfeeding in Germany: Effects of the 2007 reform”, Review of Economics of the Household, 14(4), pp. 859-890.
[16]Lalive, R., and Zweim¨uller, J. (2009) “How does parental leave affect fertility and return to work? Evidence from two natural experiments”, The Quarterly Journal of Economics, 124(3), pp. 1363-1402.
[17]Oddy, W.H., Kendall, G.E., Li, J., Jacoby, P., Robinson, M., De Klerk, N.H., Silburn, S.R., Zubrick, S.R., Landau, L.I., and Stanley, F. J. (2010) “The long-term effects of breastfeeding on child and adolescent mental health: a pregnancy cohort study followed for 14 years”, The Journal of Pediatrics, 156(4), pp. 568-574.
[18]Rossin, M. (2011) “The effects of maternity leave on children’s birth and infant health outcomes in the United States”, Journal of Health Economics, 30(2), pp. 221-239. [19]Ruhm, C.J. (1998) “The economic consequences of parental leave mandates: Lessons
from Europe”, The Quarterly Journal of Economics, 113(1), pp. 285-317.
[20]Ruhm, C.J. (2000) “Parental leave and child health”, Journal of Health Economics, 19(6), pp. 931-960.
