Paid family leave Policy in California
- First, ever implemented in the United States of America
A Bachelor’s Thesis in Economics 15 Hp
AUTHORS: Hannah Lööf Björnram Ravneet Singh
SUPERVISOR: Maksym Khomenko
UNIVERSITY: University of Gothenburg
School of Business, Economics and Law Department of Economics 7t h of June 2018
List of contents
1. Introduction………....4
2. Disposition and limitations………5
3. Background………5
4. Literature review………7
5. Theoretical framework………..10
6. Data description………12
7. Empirical strategy……….12
7.1 Difference-In-Difference………13
7.2 Standard Errors………...14
7.3 Placebo Test………...14
7.4 Synthetic Control Method………..15
7.5 Threats to external validity……….17
8. Result………18
8.1 Difference-In-Difference Result……….19
8.2 Placebo Result 2000………20
8.3 Placebo Result 2002………21
8.4 Synthetic Control Method………...22
9. Analysis and Discussion………...23
9.1 Limitations………..26
10. Conclusion………27
11. Bibliography……….28
Abstract
In this essay, we examine the effects of a gender-neutral policy that was implemented in California in 2004. The policy, Paid family leave, is the first policy in the US that offers a financial compensation to both parents and consists of 6 weeks of paid leave. The purpose of the thesis, is to examine the policy’s effect on fathers’ hours worked per week last year after the implementation. We use March CPS dataset, which is a cross-sectional micro-level data, between the years of 2002 to 2006 to estimate the effects.
A Difference-In-Difference method have been used to explore the effects of the policy.
Though, in order to conduct the method, the key assumption of parallel pre-time trends must hold and is not very likely to be realistic, and therefore we perform a Synthetic Control method. The Synthetic Control method contains of a weighted averages of possible donor pool states, which creates a Synthetic California. The Synthetic California shows a
hypothetical situation of the outcome if the policy had never been implemented. The results from the Difference-In-Difference shows a negative and significant effect on hours worked per week last year by the father in California compared to the control. Due to the weak pre- trends in both the Difference-In-Difference and in the Synthetic California, we cannot conclude that the negative result only depends on the policy.
Key words: California, Paid parental leave, Difference-In-Difference, Synthetic control, Gender equality
Acknowledgement
It has been a great collaboration doing the thesis during these 2 months. We want to give a shout-out to our good buddies: Victor Fingal and Joel
Canderhed, thank you for your great support!
1. Introduction
Today many industrialized countries have legislated universal paid parental leave policies.
The United States is an exception from this and is the only advanced nation that does not yet provide for a nationwide paid family leave (PFL) program. Paid parental leave acts allow fathers and mothers to take paid time off from work to bond and care for their newborns.
(Rossin-Slater, Ruhm and Waldfogel, 2013). Paid parental leave has mainly been outlined as a matter for working mothers, and throughout history policies have been focused on
maternity-leave. These types of policies might ease the struggle that new parents go through when it comes to harmonizing family and work responsibilities (Bartel et.al, 2015). Thus, paid leave is also equally crucial for working fathers. There is a need for policies that supports and advocates fathers’ leave-taking, so they can have an opportunity to stay at home and prioritize their family responsibilities. Several weeks of paternity leave can improve child bonding, the wellbeing of families and gender equality both at home and in the workplace (United States Department of Labor, 2015).
Regardless of the many positive outcomes that paternity leave entails, fathers still struggle with social and economic barriers. Obstacles, such as attitudes about who should be the provider and inadequate accessibility to paid parental leave. Nevertheless, the social view has started to shift, and fathers are not seen as the sole breadwinner. Paid parental leave policies are on its way forward, encouraging fathers to stay at home for caregiving purposes and dismantling gender stereotypes (United States Department of Labor, 2015).
In 2004, California was the first state to implement a paid family leave policy, where mothers and fathers can get a financial compensation to take paid time off from work for care giving purposes (Andrew Chang & Company, 2015). To date, only a few studies have empirically evaluated the effect of California’s paid family leave policy (CA-PFL) on fathers. The absence of knowledge of the effects of such policies on fathers have left a missing gap in the literature (Bartel et.al, 2015). This thesis tries to fulfill that gap, by evaluating the effects of CA-PFL policy on fathers’ hours worked. By observing the effects, we acquire more insights and we strive to spread awareness about the policy.
To estimate the effect of CA-PFL, a Difference-In-Difference method (DiD) have been used to compare the outcome of a treated- and control- group. The treated group is exposed to the treatment (implemented policy) while the control group is not. A synthetic control method has also been conducted, this method provides us with a weighted average of potential control states that creates a Synthetic California.
2. Disposition and limitations
The background section intends to give the reader brief information about the policies regarding family leave in the US, and the mechanisms behind the paid family leave policy in California. The literature review is mainly focused to previous studies about CA-PFL. The theoretical framework describes the theories, which we base our research on. The empirical strategy segment explains the approach to how we estimate the effect of CA-PFL on fathers’
hours worked per week last year. We present our findings in the result section and discuss the findings in the analysis and the study ends with a conclusion that sums up the paper.
The scope of the thesis is limited to examining the effect of Paid family leave in California on fathers’ hours worked per week last year and we have excluded same-sex marriage and single parents, and limit ourselves to married couples with newborns. We are aware that this
constellation is not representative for the entire population and cannot be generalized.
3. Background
In contrast to other OECD countries, America is far behind regarding paid parental leave. The US do not provide for a legislated universal paid family leave program (Rossin-Slater, Ruhm and Waldfogel, 2013). However, the nation-wide policy, the Family and Medical Leave Act (FMLA) which was reformed in 1993 permits each state to provide for 12 weeks of unpaid leave for parents. This was an important act that improved parents leave taking opportunities (Gault et.al, 2014). Though FMLA, women and men can take unpaid time off from work for bonding and caregiving purposes, such as an ill family member, personal illness and a newborn (United States Department of Labor, 2012). The act is one of few policies that promote fathers’ leave-taking for caregiving purposes (Gault et.al, 2014).
The FMLA law entitles employees with job-protected unpaid leave for a maximum of 12 weeks per year. To be qualified for the FMLA employees must at least have worked 12 months or a minimum of 1250 hours during a 12-month period. The policy does not cover
firms fewer than 50 workers, but they must ensure job-protection (United States Department of Labor, 2012).
Before the implementation of California’s paid family leave policy, mothers can take time off from work with pay due to pregnancy issues through the Temporary Disability Insurance (TDI) (Gault et.al, 2014).
In 2004, California became the first state in the U.S. to implement a paid family leave program, where both parents can apply for paid leave up to 6 weeks. The CA-PFL act is supplementary to the TDI. Through the CA-PFL fathers and mothers are compensated with up to 55 percent of their usual payment up to a ceiling of 1,104 dollars per week in 2015
(Andrew Chang & Company, LLC, 2015). On January 1, 2018, the financial compensation was extended to 60-70 percent of wage replacement in California
So far, only four states in the US have implemented the paid family leave program: California (2004), New Jersey (2009), Rhode Island (2014) and New York (2018) (Brinle, 2018). The strength of the CA-PFL is that, it allows parents to return from the leave without any major career disturbances. It also gives the parents an opportunity to bond and care with their newborns. However, the policy does not require job-protection (Andrew Chang & Company, LLC, 2015).
The main differences between FMLA and CA-PFL, is that FMLA is unpaid parental leave and parents can take up to 12 weeks of leave. Where employers must guarantee job- protection. The CA-FPL offers paid parental leave up to 6 weeks, but does not provide for job-protection. However, when undertaking the CA-PFL one can get job-protection if the paid leave is simultaneously covered by the FMLA (Bartel et.al, 2015). The CA-PFL is integrated with the TDI insurance.
To be eligible for the policy, employees must at least have worked 300 hours during a period of 5 to 18 months before applying for paid family leave (Bedard & Rossin-Slater, 2016). The paid family leave program is financed through a payroll tax levied on the workers, which the employees must submit to the State Disability Insurance (SDI) program that funds CA-PFL.
The SDI program is administered by the California Employment Development Department (EDD). The program is not directly funded by the employers, however, the may face other costs (Andrew Chang & Company, 2015).
After the first year of the enactment of the paid family leave program the claims grew steadily. 153 000 claims were made between the years of 2005 and 2006, and in 2008
approximately 188 000 claims were paid. During the financial crisis the requests for CA-PFL somewhat decreased and is likely a result of the recession. Although, the claims recovered between the years 2012 - 2013 and around 202 000 issues were paid. The total claims that were issued, 88 percent were due to bonding with a newborn (Bartel et.al, 2014).
The age distribution of biological mothers and fathers that claim CA-PFL benefits are presented in the figure below:
4. Literature review
Studies that have been made to analyze the impact of paid family leave in California have been very little regarding the father. The extensive research that we have come across is,
“Paid Family Leave, Fathers’ Leave-Taking, and Leave-Sharing in Dual-Earner Households”
by Bartel et.al (2015). The purpose of Bartel et.al (2015) paper was to learn more about fathers’ response to CA-PFL, by analyzing how American fathers would react to the implementation of the policy. This allow both parents to get a financial compensation, and how this would affect the composition of the family roles. The team of researchers conducted a Difference-In-Difference (DiD) method and studied census years of 2000-2013 in the state of California.
They reserachers concluded that, Paid Family leave in California had raised “the share of fathers of infants” who were on care-giving leave by 0,9 percentage units. They discovered that before CA-PFL was implemented in 2004, the mean of leave-taking rate in California was 2 percent, i.e. this represented a significant 46 percent increase. In their research, they used a large sample from American Communities Survey (ACS). They saw that paid family leave in California expanded both father-only leave-taking and joint leave-taking when both parents in the household were working. They also saw that fathers are more likely to stay at home when he works in an environment where the labor force is largely part women. In the paper the authors draw the conclusion, that it is less stigmatized for men to stay at home for caregiving purposes in a more female environment compared to a male (Bartel et.al. 2015).
When the researchers are looking for heterogeneity in the pattern of who stays at home with the newborn, they discovered a considerable heterogeneity in both mother's and father’s behavior when it comes to leave-taking. They saw that probability for fathers to stay at home was higher if they had a son, after the implementation of paid family leave. This was also true for mothers, who are likely to stay at home if they have a daughter. From the results they concluded that, the birth-order matters for fathers and not for mothers. Fathers were to stay at home with their firstborn to a larger extent compared to stay at home with their second, or third child when undertaking the policy (Bartel et.al. 2015).
Due to the Paid family leave program in California, employers have reported an increase in men taking leave. The median length of leave was higher for men who used the policy
compared to those who did not (Appelbaum and Milkman, 2011). Different surveys that have been conducted regarding the CA-PFL, shows that men are more likely to take the leave if it is paid (Andrew Chang & Company, 2015).
The graph is retrieved from the United States Department of Labor
New research findings, indicate that increasing paternity leave could also evolve in
longstanding cultural norms regarding the gender, work and household duties. When the use of paternity leave increases, time studies show that the work within the household becomes more gender neutral over time between fathers and mothers. With other words, men share the household chores and takes care of children (Patnaik, 2015).
Data from Employment Development Department State of California (EDD), shows result of females in lower income groups are one of the larger groups of claimants of bonding through the CA-PFL. Reports from the EDD data shows, that women with lower income and lower education is the subgroup within the society that benefits the most from the financial compensation that is offered. Regarding men, the data shows the opposite where men from higher income groups have the most claimants (Andrew Chang & Company, 2015).
Fathers with 60 percent earnings over 36 000 dollars and for women 47 percent earnings under 36 000 dollars.
5. Theoretical framework
We base our theoretical framework on different theories that try to explain gender equality through a labor market perspective. Therefore, we try to get some perception on men’s participation in the labor force when they become fathers. The research is based on the following theories: New home economics, A theory of Social Interaction and The Separate Spheres model.
Becker, the author of the theory New Home Economic, conducted a study about families who prefers to maximize utility as one unit, and the time allocation between work in the household and work in the labor force is based on a price (Becker, 1991). In the theory Becker states, that the spouse who have the highest income will have an advantage over the spouse that have a lower income. If men earn more, than they will allocate more of their time in the workforce while women allocate their time in the household. The paid family leave provides a financial compensation to either of the parents when taking leave for caregiving purposes. In most cases trough out history men have earned more than women. Though in recent year the wage gap has decreased. Therefore, if the family strive to maximize their utility, and the mother
earns more or equal to the father, the advantage of being the provider will decrease for the man. This could lead to a decline in participation for men in the labor force and the advantage for the man will be to stay at home to maximize the family’s utility (Becker, 1991).
In 1974, Becker described in another model “A theory of Social Interaction”, about the allocation of resources in the household was constructed through a bargaining power between the household members. Within the household the father is often seen as the main provider, while the mother is seen as the caregiver. Based on these stereotypical gender roles a
bargaining power emerges. The financial compensation that one can claim in CA-PFL, could give the mother more bargaining power, and therefore remain in the work force. While the father could stay at home bonding with the child. Based on this, we believe that hours worked per week last year by the father will decrease (Becker, 1974).
In contrast from the Social interaction theory, where the decision-making was based on the family’s preferences, the Separate Spheres model describes the decision-making based on the individual’s preference (Lundberg & Pollak, 1993). The model emphases the time-allocation between hours of labor and household work between the spouses, and how a relative
bargaining position emerges. Hence, the individual in the household who have the highest income will therefore have a stronger bargaining positive relative to the person with a low- income. This model discusses the individual’s choice of maximizing their utility, and if the father has a lower income and gets a financial compensation, maybe he will be the one that stays at home and bonds with the child. The relative bargaining power between the spouses now comes down to, who earn the most in the family.
To date, the traditional gender roles in families have shifted and the father might not be the main provider to the same extent to what he was. With the financial compensation, we assume an even stronger relative bargaining power for the mother and the pressure for the father to stay at home will increase. We then predict that hours worked per week last year by the father will decrease due to a stronger bargaining position for the mother, if she is the one in the household with the highest income (Lundberg & Pollak, 1993).
In conclusion, our predictions are based on the different theories presented in the text above.
We suggest that hours worked per week last year by the father will decrease. As the families try to maximize their utility, and for the first time in US history being offered a financial compensation for paternity leave, the roles within the family might shift. Where the father is not going to be a natural candidate for being the breadwinner, and perhaps a relative
bargaining struggle is going to emerge between the spouses.
6. Data description
In this thesis, the March Current Population Survey (CPS) dataset from years 2002 to 2006 have been used to estimate the effects of California’s paid family leave policy. It is a cross- sectional individual data and the dataset can be retrieved for free from CEPR data. The dataset consists of observations on individuals and households in the United States where California is the main state of interest (Center for Economic and Policy Research, 2018).
The paid family leave policy was implemented in California in 2004, we chose to analyze the effects with a Difference-In-Difference method between the years of 2002 to 2006. We decided to narrow down the time to 2006, to avoid big macroeconomics chocks from the financial crisis that erupted at the end of 2008.
After appending the years 2002 to 2006, the dataset consists of 1, 066 ,094 observations. In the dataset the variable female existed, but we choose to exclude it since we only observe men. We have dropped observations for individuals who do not have children and are unemployed. By dropping these observations, we can argue for, that the individuals are fathers and employed, which is required for the CA-PFL study we are conducting.
We narrow the research to married fathers, and exclude adoptive-parents and same-sex marriage to estimate the relative changes. We are fully aware that the results cannot be generalized for the entire society.
When conducting the synthetic control method (SCM), we append the years between 1990 to 2003 and collapse by mean. We collapse the same variables that we have in our DiD model.
To conduct a SCM, we convert the cross-sectional individual data into a panel data. The probability of getting a better matching pre-time trend is higher if one uses a longer time period, therefore we choose the append the years 1990-2003. A more detailed explanation
about the Difference-In-Difference model is presented in the section below and SCM will be discussed in section 7.4.
7. Empirical strategy
7.1 Difference- In- Difference
To evaluate the effects of California’s paid family leave act on fathers’ hours worked per week last year, we perform a Difference-In-Difference method (DiD). The DiD-model is the difference in average outcomes before and after a policy change for a group affected to a group not affected. By matching the average outcome of the treated state, California, with a control state (the rest of the USA) where the treatment did not occur. The DiD method assumes that the following key assumption holds: “absent of the policy change the trend change in our outcome would have been the same for the treatment and control state” (Card
& Krueger, 1994). In terms of our outcome variable, hours worked per week last year, the pre- time trend must be parallel between California and the control state. The key assumption is questionable since trends are likely to fluctuate. To makes this problem less likely and try to hold the assumption, a Synthetic Control method has been conducted (Abadie & Gardeazabal, 2003).
The regression model has the following structure:
yti=α+ β1Ti+β2t+β3
(
Ti∗t)
+β4Xi+εiWe create our model by generating two dummies, Treatment and time which take the values of one and zero. The dummy variable Treatment ( Ti ) is equal to one, if the treated state is California and is equal to zero for otherwise. The main purpose of this variable is to control for certain features that should be true for the treated state, and not interfere with the
enactment of the policy. The variable time ( t ), takes on the values of one, if it is equal to after 2004 and equal to zero if it is before 2004. By combining the Treatment and time dummy variables, we get the coefficient ( β3 ). The variable becomes an interaction term and takes on the value of 1 if the state is California after the implementation, and equals zero for otherwise. The very purpose of this variable is that, if the key assumption holds for the DiD, then the interaction term will capture the causal effect of the policy. The interaction term is thus the main variable of interest in our DiD regression. By generating these steps, a
The control variables that have been selected are: married, educational level, full-time schedule, firm size and total family income ( Xi ). They are all included in the X-variable.
The variable educational level has been decoded into a dummy, that takes on the value of one, if the individual has a college degree or higher and equal to zero for lower education. The variable married is a dummy that shows if the individual is married or not. We consider these control variables to be the mediator in our equation. In the dataset they are the ones most related to the outcome and we want to remove their effects in the DiD equation. The notation ( i ), indicates the individual and the ( t ), is the census years.
We choose the outcome variable, hours worked per week last year. The reason why we chose this outcome variable is because we believe that it shows an immediate effect of the policy.
Because of the short time period observed, we believe that this outcome variable measures the fastest response of fathers to the policy. The outcome variable is lagged, and we are aware of this when analyzing the policy’s effect on the outcome.
7.2 Standard Errors
A limitation with the DiD-method, are the standard errors of the estimate. The standard errors in a DiD-method are often derived from an OLS regression in a cross-section data, that consists of data on individuals from a control and treated group before and after an intervention. If one does not cluster the standard errors in a DiD, will lead to a serial
correlation resulting into incorrect estimates. There are three certain aspects that makes serial correlation in a DiD-method an issue. The model depends on estimations over long time periods and this research paper have a fairly short time period (2002-2006). The dependent variable is generally highly positively serial correlated. The binary variable “Treatment”
rarely changes within the group and over time. These three issues reinforce each other and results into uncertain estimates. By clustering the standard errors into a pre-period and post- period eliminates this issue (Bertrand, Duflo Sendhil & Mullainathan, 2004).
7.3 Placebo Test
To validate the findings when conducting the Difference-In-Difference method, we check for the robustness of the model by performing two placebo tests. The purpose with the placebo tests is to see how the implementation of the policy in the treated state (California) behaves when we change the year of the enactment of the paid family leave act in California. The first placebo test is conducted in 2000, and the second test is conducted in 2002, two years before
the implementation of CA-PFL. The placebo regression copies the DiD-analysis for a group of people or point in time that appear like the ones in the treatment group or in the treatment period, but were not actually treated. One can also make a placebo test for the Synthetic control method, but by eyeballing the pre-time trend in the DiD model we believe to have a better pre-time trend. This is further shown in the result section.
7.4 Synthetic Control Method (SCM)
The key assumption about parallel time trends is crucial for any DiD estimations. As
mentioned, the assumption is unrealistic to hold. Which means that the treated state’s outcome path is not parallel with the control state’s outcome path before the implementation of the policy. Identifying control groups whose outcome path matches the treated group’s pre- intervention outcome trajectories can be difficult. Though without these control groups, can become problematic to distinguish the intervention’s effect from the effects of the non-policy variables. The synthetic control method (SCM) addresses the issue with finding a control group that match the pre-trend for the treated state’s trajectories. SCM studies have become an increasingly popular method to conduct for policy evaluations (McClelland & Gault, 2017).
The Synthetic Control method creates a weighted average of possible donor states chosen from a pool of possible states. By matching the dependent variables and the outcome variable before the implementation of the policy in the donor states to the corresponding variables in the treated state, creates the weighted average. An ideal synthetic control method requires that the outcome path in the treated state perfectly match the outcome path in the control state before the implementation of the policy. If this is fulfilled, then comparing the outcome paths after the intervention can provide understandings about the policy’s effect in the treated state.
If the trajectories between the treated state and the control state differ after the intervention, then the policy in the treated state can presumably have caused the deviation in outcome paths (McClelland & Gault, 2017).
To uphold the parallel trend assumption for the Difference-In-Difference model, we create a control group which is more likely to match the pre-intervention outcome path for our treated state (California). We create our control state by combining a weighted average of our
explanatory variables, and outcome variable for potential states in the US that corresponds to the variables in our treated state. The Synthetic California shows a counterfactual, meaning
that it demonstrates how the outcome would have behaved in California if the act was never implemented.
A synthetic control method relies on data over several years before the implementation of the policy. In this thesis, we observe data for years 1990 to 2006. There are three assumptions in a SCM that needs to be fulfilled to get an effective use of the model. The first assumption is that the treated state should be the only unit affected by the policy in the pretreatment period and after the implementation. The second assumption, the policy change should not have an effect before it is implemented. The third assumption, the control state’s outcome is created by a weighted sum of potential donor states (McClelland & Gault, 2017).
Table 1.
Explanatory Treated Synthetic Average of all states
Married 0.3953 0.4068 0.4319
Firm size 2.2755 2.3234 2.3548
Full-time schedule 0.7978 0.7979 0.7838
Education level 2.3813 2.4208 2.4734
Total family income 72808.63 71963.86 73797.0284
Hours worked per week last year 38.6216 38.6606 38.7141
* Table 1 demonstrates the unique weights for the Synthetic controls for California with the outcome variable
“Hrslyr” = Hours worked per week last year.
In table 1, we present an example of matched explanatory variables. The table represents the matched explanatories for hours worked per week last year, our outcome variable. By observing the table, one can see that the synthetic weighted averages are better compared to the average of all the states. This is one way to create a counterfactual control unit, where we now can compare the actual value of the outcome in California after the implementation of the policy.
Table 2.
State Unit Weight
New York 0.619
New Jersey 0.066
South Dakota 0.107
Alabama 0.154
Mississippi 0.055
Total 1.000
* Table 2, shows an example of the unique weights for the synthetic controls for California with the outcome variable hours worked per week last year.
Table 2, demonstrates the results of the weighted averages of states that creates the Synthetic California. In this table New York and New Jersey have been selected as unique weights for the counterfactual California. We do not see this as problem, since the policy has not been implemented in either of states at that time.
7.5 Threats to external validity
There are some threats to external validity that needs to be addressed concerning generalizing the results of the policy to other settings. The threats are: Experiment Specificity and Non- representative program/policy. The Experiment specificity, is an important threat to discuss in this research paper, which addresses the issue with generalizing results to other locations. The paid family leave program in California might not have similar results in other states which make it difficult to generalize any results. Non-representative program/policy, CA-PFL program could be considered as a small-scale project that only takes place in California. This makes it problematic to legislate California’s paid family leave on a state-level, since
California does not represent the entire nation (Mitrut, 2018).
8. Result
Figure 1.
* Figure 2 shows the plotted pretreatment period for California and the rest of US. The black line indicates California’s fitted values before the implementation of the CA-PFL and the dotted-line shows the fitted values for the rest of the US before the implementation of the policy.
Figure 1, shows the plotted mean values for the outcome variable in California prior to the implementation of the policy and the rest of US. One can argue that it is almost a parallel trend between California and the rest of US, but to perform a Difference-In-Difference method the key assumption must hold. Which it does not in this case, and therefore we conduct a Synthetic Control Method. In this segment, we present the results of the main variable of interest: Hours worked per week last year by the father. In table 1, the result from the model is conducted for California. The outcome variable reflects the change of the family fathers’ decision of staying at home because of the effect of the paid family leave policy. The interaction terms purpose in the regression is to show the causal effect of the policy on fathers’ hours worked per week last year. We have included control variables such as:
married, firm size, education level, total family income and full-time schedule to see how these affect hours worked per week last year by the father, when the policy is implemented in treated state California. To validate the results from the Difference-In-Difference regression, we check for the robustness in the form of two placebo tests.
8.1 Difference-In-Difference result
Table 3: The effects of CA-PFL on fathers’ hour worked per week last year
(1) A
VARIABLES hrslyr
time -0.2286***
(0.0016)
Treatment 0.3899**
(0.0224)
did -0.4672**
(0.0139)
married -0.3273
(0.0907)
firmsz 0.6177**
(0.0313)
rincf_all -0.0000*
(0.0000)
fulltimely 27.0319***
(0.2005) educ_highlevel -7.5361
(1.4760)
Constant 21.7503**
(1.4826)
Observations 9,548
R-squared 0.5433
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Our main result from table 3 revolves around the interaction term, which represents the causal effect of the policy. We find a decrease in hours worked per week last year by the father in California. It shows a significant negative effect of 0.46 hours less in California compared to the control state. It is significant on a 5 percent significance-level. This entails, that after the implementation of the policy, fathers’ hours worked per week last year have decreased in California (treated state) compared to the rest of US (control states). The p-value for the interaction term is 1,9 percent, this could mean that there is only a 1,9-percentage chance that the connection between the policy’s casual effect and fathers’ hours worked per week last year is a result of chance. We are aware that the outcome variable is lagged, therefore the
The control variables that are included in the regression, demonstrate the effects of the
economic and social mechanisms that also drive the change in fathers’ hours worked per week last year. The control variables, full-time scheduled and firm size shows a positive sign on our outcome by fathers in California and control state. The effect of the binary variable
“educ_highlevel” on our outcome variable is negative and insignificant. We thought that education-level might have a significant effect on hours worked by the father, but in the result table above it showed the opposite, which surprised us.
The results from the DiD regression shows a negative effect on hours worked per week last year by the father, but we cannot say that the implementation of the policy is the sole reason behind this. The possible reasons to why we think the policy have a negative effect on the main outcome variable will be further discussed in the analysis and discussion section.
8.2 Placebo result 2000
Table 4: The effects of CA-PFL on fathers’ hours worked last year (1)
B
VARIABLES hrslyr
time2000 -0.0180
(0.0067)
Treatmentp 0.0272
(0.0046)
didplacebo 0.7135***
(0.0107)
married 2.6025**
(0.2022)
firmsz 0.7149**
(0.0245)
rincf_all -0.0000*
(0.0000)
fulltimely 25.7138***
(0.1420) educ_highlevel -9.6027
(2.2008)
Constant 24.1008*
(2.2199)
Observations 14,647
R-squared 0.5472
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
To check for the robustness of the outcome in the original Difference-In-Difference
regression, we perform the placebo test (shown in table 3). The time of the implementation of the policy was changed to 2000. In the regression, the interaction term “didplacebo” shows the causal effect of the policy. The policy is significant and has a positive effect of 0,71 hours worked per week in California compared to the control. Since the outcome variable is lagged the positive and significant effect that occurs comes from the year 1999. The placebo test becomes significant at a 1 percent significance level.
8.3 Placebo result 2002
Table 4: The effects of CA-PFL on fathers’ hours worked per week last year (1)
C
VARIABLES hrslyr
time2002 -0.3066**
(0.0091)
Treatmentp02 0.2482***
(0.0021)
didplacebo02 0.2316**
(0.0082)
married 2.5949*
(0.2128)
firmsz 0.7133**
(0.0246)
rincf_all -0.0000**
(0.0000)
fulltimely 25.7109***
(0.1430) educ_highlevel -9.8133
(2.1010)
Constant 24.3797*
(2.1240)
Observations 14,647
R-squared 0.5473
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
In the second placebo test the year of the enactment of the policy in California was changed to 2002. The main variable of interest, “didplacebo02” indicates a positive and significant effect of 0,23 hours in California compared to the control. It is significant at a 5 percent significance level. The outcome variable is lagged which means that the effects we observe is from the
The results we got from our original DiD regressions indicated a negative and significant effect on hours worked per week last year by the father in California compared to the control state. The two placebo tests show the opposite sign but also significant. The magnitude between the placebo tests differs quite a lot. It differs with 0,48 hours worked per week by the father in California before the implementation of the policy compared to the control.
8.4 Synthetic Control Figure 2.
*Figure 2 demonstrates the effect of the paid family leave policy. The black line indicates the treated state’s trend and the plotted line shows the synthetic control, for the main outcome of interest: Hours worked per week last year, by the father. The dotted vertical line indicates the time of the intervention.
An ideal synthetic control method requires that the treated unit’s and the control unit’s outcome’s trajectories match prior to the intervention. By eyeballing one can see that the outcome path for California do not match the outcome path for the Synthetic California, before the implementation of the CA-PFL (see fig. 2). There is an apparent gap between treated California and Synthetic California in 1980, suggesting that there is some disturbance in California compared to the Synthetic California. One can see a small gap between
California and Synthetic California before the implementation of the policy. Furthermore, there is a split after the enactment of the policy between California and the Synthetic
California. The Synthetic California indicates that, fathers’ hours worked per week last year, after the implementation would have sustained on a steady path. On the contrary, treated California shows a decline in hours worked per week last year by the father. However,
fathers’ hours worked per week last year in California increases one year after the enactment of the policy.
The Difference-In-Difference estimate indicated that fathers’ hours worked per week last year in California decreased after the intervention. Since the key assumption for the DiD method was violated, the synthetic control method was conducted. After performing the synthetic model, one can observe a difference between the outcome paths between treated California and Synthetic California after the CA-PFL was implemented. Though the difference is not very large. The weak trajectories between California and Synthetic California prior to the intervention is questionable, since they do not match. Which makes us believe, that the results we retrieve from both the SCM and DiD models are not to be trusted and will be further discussed in the analysis section. Once again, the outcome variable is lagged and the observed effects are therefore delayed.
9. Analyzes and Discussion
In this thesis, we strive to evaluate the effects of California’s paid family leave on fathers’
hours worked per week last year. The DiD estimations indicates a significant and negative effect on fathers’ hours worked per week last year after the implementation of the policy. The effect of the policy is quite small on the outcome variable, hours worked per week last year by the father. Since the key assumption in the DiD model was not fulfilled, we choose to do a synthetic control method. This method addresses the issue with finding a control group and to match a pre-trend. The sole purpose of this method is to show a parallel time trend between the treated and control state before the implementation of the policy. Nonetheless, we cannot draw the conclusion that the intervention is the only indicator behind the decreased hours worked last year by the father in California. This is because both methods show a weak pre- trend, which makes us question our results.
Becker’s theory, New home economics, states that a family maximize their utility as one unit and that the spouse with the highest earning will have an advantage. Regarding this theory, we predicted that fathers’ hours worked per week last year will decrease. We believed that the father would allocate more of his time at home bonding with his newborn. The result from the Difference-In-Difference model, shows that hours worked per week last year by the father in California decreases after the implementation. This could entail that, fathers’ advantage of
might be because of the alternative for the father to stay at home instead of the mother, in order to maximize the family’s utility. We cannot say that our estimates match Becker’s theory, but we can see a decrease in our outcome after the implementation. This could be pure speculations, but perhaps the families do maximize their utility as one unit.
Furthermore, this makes us believe that the policy could have softened the traditional gender roles in the household, where fathers now can become the caregivers. Though the change in our outcome is not very big, it might entail that the traditional views of who is being the caregiver in the household is being questioned. Hence, we cannot conclude that the policy itself is the reason behind the negative effect on fathers’ hours worked last year. We can also discuss that the policy could have a positive effect on gender equality within the household.
The effects that the policy has on gender equality could lead to a “new level of maximization”
within the families, and that the advantage between the spouses becomes more even.
However, we cannot confirm entirely that the policy affects the gender equality between spouses in the household. We can only refer to the United States Department of Labor in 2015 research, where they discuss that the paid paternity leave can promote gender equality in the household.
In the theory, A theory of social interaction, Becker further discuss the bargaining power within the households. According to this theory, the conventional gender roles entail that the mother is the caregiver and the father is the breadwinner. We predicted that mothers
bargaining power would increase, giving them the advantage to remain in the workforce as a response of the policy. We assumed that fathers’ hours worked per week last year would decrease. Yet again, the results from the DiD regression, tells us that, fathers’ hours worked per week last year declines after the implementation of the policy. From our estimates we cannot really say anything about the bargaining power between the spouses. Based on the theory we concluded that the bargaining power might change between the spouses due to the financial compensation. Hence, the implementation of the policy might affect fathers’
bargaining power to decrease, which could entail that they allocate more of their time at home.
In the light of the, Separate spheres model, where the focus lays on the individual’s maximization. We assumed that the relative bargaining power for the father would
decrease, and indicating that his hours worked would decline as a response to the policy. Due to the decrease in hours worked per week last year by the father, this could entail that the mother might maximize her utility. If it exists a power struggle between the spouses, then it
comes down to who earns the most in the family. If this theory holds for our research, then perhaps this could tell us that families do not maximize their utility as one unit, rather on an individual preference. We are aware that this contradict the statement from the New home economics conclusion, but in this model the mother's bargaining power is encouraged if she is the one with the highest earning. Our empirics do not say anything about this, so we cannot say that the bargaining power exist between the mother and the father. We do not observe any wage variable for the mother or the father, so we cannot say that any of these affects the decrease in hours worked per week last year. We can only speculate from the predictions from the theories, A theory of Social Interaction and Separate spheres model, and maybe get some understandings behind the reasons to a decline in our outcome.
These three different theories presented above, describes the bargaining power between the spouses in the households that could emerge due to the financial compensation. Our aim with the theories was to get a better insight on fathers’ participation in the labor force. Our
Difference-In-Difference estimates shows a decline in hours worked per week last year by the father after the intervention. Whether this is an indicator that the traditional gender roles are disrupted is hard to tell. Nevertheless, based on our results makes it difficult to draw the conclusion, that the policy is the main reason behind why we can see a reduction in hours worked by the father. If the financial compensation has a declining effect on our outcome, this might create a bargaining power between the spouses. Let’s say hypothetical that the mother earns the most, then she will have an advantage to remain in the workforce. In this research paper we do not look at the mother’s work share, but this is something we think could be further investigated.
Previous research, have shown that fathers are more likely to stay at home with their firstborn and if they have sons (Bartel et.al. 2015). This is an interesting finding, which we cannot see in our dataset. Nevertheless, this would be interesting to learn more about since it could indicate that fathers’ leave-taking is an obligation rather than a need to bond with their child.
This contradict the entire purpose of the policy of equal responsibility, because the father is more likely to stay at home with the first born and if they have sons.
Overall, we can see that fathers’ hours worked per week have decreased after the
implementation of the paid family leave policy in California. Perhaps this is an indicator that the policy has made it more acceptable and the financial compensation works as a carrot for fathers’ leave-taking. Maybe within the American society the traditional gender roles are still
this, we draw a strong and a personal opinion, that the policy might have dismantled the gender stereotypes. This could be one of the reasons for the decline in hours worked.
From the two placebo tests conducted, we get a positive and significant effect on our
outcome. We are hesitant to draw any elaborate conclusions about these tests. The purpose of the placebo tests was to check for the robustness for DiD model. The tests came out positive which was expected, since we have changed the time of the implementation of the
intervention in California. If the tests had shown a negative effect would have instead raised some concerns about our DiD model.
9.1 Limitations
In a perfect world and if the assumptions regarding the models would have been fulfilled, then this is how we would have concluded our predictions. But since the key assumption for the Difference-In-Difference do not hold, and the pre-time trend is weak for the synthetic, we cannot conclude too much about our estimates. This, makes us hesitant to rely on the results.
Since the assumption is not fulfilled in the DiD model, tells us that the interaction term in the model does not show us the causal effect of the policy.
The aim of the thesis was to evaluate the effects of CA-PFL on fathers’ hour worked and spread awareness about the policy. The DiD model indicates a negative and significant effect on our outcome, but due to the key assumption being violated makes the estimates
questionable. Meanwhile, the estimates are uncertain and thus makes it difficult to draw any conclusion from them.
In figure 1, the Synthetic Control’s pre-trend is weak and causing the parallel trend to be doubtful. For the parallel assumption to hold, the slope of the trend lines for California and Synthetic California must be linear. The only purpose with a synthetic control method is to match the affected unit’s outcome before the intervention and control for the treated unit’s following enactment. If the trajectories between California and Synthetic California match prior to the intervention, then the difference in outcome between California and Synthetic California can reveal the interventions effectiveness. However, this is not something that we can conclude because of the weak outcome paths before the implementation of CA-PFL.
Furthermore, this makes it difficult to interpret the graph from the synthetic method.
The reason behind the weak pre-trend could be the results of the following: Short pre-treated time and a mismatch of the covariance in the dependent variables in California and Synthetic California. The short-pretreated period could disturb the linear relationship between treated-
and control unit, this is important to have in order to get a match in the pre-trend. Therefore, in a shorter period a higher trend could cause an eruption and result to misleading estimate.
The covariance of a dependent variable could lead to a disturbance that interrupt the linear pre-trend and therefore the parallel trend assumption is violated (Kaul et al. 2018).
It is rather difficult to generalize the results from our Difference-In-Difference study. The paid family leave policy has been constructed on a state-level, meaning that the policy is designed for California. This makes it problematic to generalize the results on a nation-level.
Since we decided to exclude same-sex marriage, adoptive families and single-parents, makes the estimations of our regression rather difficult to generalize.
10. Conclusion
California’s paid family leave program was the first gender neutral policy in America that promotes fathers’ leave-taking for caregiving purposes. To achieve equality between mothers’
and fathers’ in the workforce and in the household, there must exist policies that endorse fathers’ family responsibilities. In this thesis, we show that the policy in California have a negative effect on fathers’ hours worked. Although, we cannot determine that the policy is the sole reason behind these effects. We believe that the policy has questioned the roles of whom should be the provider and the caregiver in household. The only thing we can see that deviates from previous policies, is the financial compensation that is offered. Which we believe works as a motivation for fathers to stay at home for caregiving purposes. Since we know that fathers are more inclined to take paternity-leave if it is paid. Therefore, California’s paid family leave policy could have challenged the spouses’ mindset about who should stay at home. Perhaps this policy has paved the way to dismantle institutionalized gender roles in California.
11. Bibliography
Books:
Becker, G.S. (1991). A Treatise on the Family. Cambridge: Harvard University Press.
Card, D. Krueger, A. (1994) Minimum Wage and Employment: A Case Study of Fast-Food Industry in New Jersey and Pennsylvania. American Economic Review, vol. 84, s. 772-784.
Gujarati, N. Dawn. Porter, C. Dawn. (2009). Basic Econometrics. Fifth Edition. New York:
McGraw-Hill Education.
Lundberg S. & Pollak R A. (1993). Separate Spheres Bargaining and the Marriage Market.
Journal of Economic Perspectives, Vol. 101, No. 6, s. 988-1010. Chicago: University of Chicago Press.
Streissler, E (2008). Wieser, Friedrich Freiherr, (Baron) von (1851–1926). The New Palgrave Dictionary Of Economics, New Palgrave Dictionary of Economics Online. Retrieved 4 April 2018.
Electronic dissertations and thesis:
Appelbaum, Eileen. (2011). Leaves that Pay. Employer and Worker experiences with Paid Family leave in California. Professor., Rutgers University. Retrieved 25 April 2018 from:
http://cepr.net/documents/publications/paid-family-leave-1-2011.pdf.
Bartel, Ann, Rossin-Slater, Maya, Rulm, Christopher. Stearns, Jenna and Waldfogel, Jane.
(2015). Paid Family Leave, Fathers Leave-Taking and Leave-Sharing in Dual-Earner Households. The Institute for the Study of Labor, DP No. 9530, s.2-41. Retrieved 16 April 2018 from: http://ftp.iza.org/dp9530.pdf .
Becker,G S.(1974). A Theory of Social Interactions. National Bureau of Economics Research, No.4, s. 2-54. University of Chicago. Retrieved 16 April 2018 from:
http://www.nber.org/papers/w0042.pdf
Bertrand Marianne, Duflo Ester and Mullainattha Sendhil. (2004). How much should we trust difference-in-difference estimates? The Quarterly Journal of Economics, February 2004; 119 (1): 249-275. Retrieved on 15 April 2018 doi: https://doi.org/10.1162/003355304772839588 Gault, Barbara, Hartmann, Heidi, Hegewisch, Araine, Milli, Jessica & Reichlin, Lindsey.
(2014). Paid Parental Leave in the United States: What the Data Tell Us about Access, Usage, and Economic and Health Benefits. Cornell University ILR School. Retrieved on
April 8, 2018 from: https://digitalcommons.ilr.cornell.edu/cgi/viewcontent.cgi?
article=2608&context=key_workplace
Kaul Asok, Klößner, Stefan, Gregor Pfeifer and Schieler Manuel (2018). Synthetic Control Methods: Never Use All Pre-Intervention Outcomes Together With Covariates Saarland University and University of Hohenheim. Retrieved on May 5, 2018, from:
http://www.oekonometrie.uni-saarland.de/papers/SCM_Predictors.pdf
Marks, Jaime. Chun Bun, Lam and McHale M, Susan. (2009). Family Patterns of Gender Role Attitudes. Sex roles. US National Library of Medicine National Institutes of Health, 61(3-4):22-234. doi:10.1007/s11199-009-9619-3.
McClelland, Robert. Gault, Sarah. (2017). The Synthetic Control Method as a Tool to Understand State Policy. Urban Institute. March 2017. Retrieved 15 May 2018 from:
https://www.urban.org/sites/default/files/publication/89246/the_synthetic_control_method_as _a_tool_0.pdf
Milkman, Ruth. (2011). Leaves that Pay. Employer and Worker experiences with Paid Family leave in California. Professor., City University of New York City Graduate Center.
Retrieved 25 April 2018 from: http://cepr.net/documents/publications/paid-family-leave-1- 2011.pdf.
Mitrut Andreea; university lectorate at the department of economics and statistics, Gothenburg university. 2018. Policy Evaluation II, lecture Mars 5.
Patnaik, Ankita. (2015). Reserving Time for Daddy: The Short and Long Run Consequences of Fathers’ Quotas. SSRN Working Paper. Retrieved 5 may, 2018 from:
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2475970&download=yes
Rossin-Slater, Maya, Ruhm, Christopher and Waldfogel, Jane. (2013). The Effects of
California’s Paid Family Leave Program on Mothers’ Leave-Taking and Subsequent Labor Market Outcomes. Published in J Policy Anal Manage. 2013: 32(2): 224–245. Retrieved on April 2, 2018 from:
Electronic Websites:
Andrew Chang & Company, LLC. (2015). Paid Family Leave Market Research. December 14. Retrieved on April 6, 2018 from:
www.edd.ca.gov/Disability/pdf/Paid_Family_Leave_Market_Research_Report_2015.pdf
Brinle Morgan. (2018). Which States Have Paid Family Leave? New York Rang In 2018 With Big Changes. January 2, 2018. Retrieved on May 6, 2018 from:
https://www.bustle.com/p/which-states-have-paid-family-leave-new-york-rang-in-2018-with- big-changes-7744450
Center for economic and policy research 2018. March CPS Data. Retrieved on May 5, 2018 from: http://ceprdata.org/cps-uniform-data-extracts/march-cps-supplement/march-cps-data/
Report for the California Employment Development Department 2016 The Economic and Social Impacts of Paid Family Leave in California October 13. Retrieved on April 6, 2018 from: http://www.edd.ca.gov/Disability/pdf/PFL_Economic_and_Social_Impact_Study.pdf
United States Department of Labor: Wage and Hour Division. 2012. Fact Sheet #28: The Family and Medical Leave Act. Retrieved on April 7, 2018 from:
https://www.dol.gov/whd/regs/compliance/whdfs28.pdf
United States Department of Labor. 2015. Why parental leave is so important for working families. Retrieved on May 7, 2018 from:
https://archive.org/details/DOLPaternityLeavePolicyBrief
State of California: Employment Development Department Am I eligible for Paid family leave benefits? Retrieved on April 7, 2018 from:
http://www.edd.ca.gov/Disability/Am_I_Eligible_for_PFL_Benefits.htm
