Trapped at Home: The Effect of Mothers' Temporary Labor Market Exits on Their Subsequent Work Career

This paper investigates how mothers’ decision to stay at home with young children affects their subsequent work careers. Identification is based on the introduction of the Cash-for-Care program in Norway in 1998, which increased mothers’ incentives to withdraw from the labor market when their child was one and two years old. Our estimates demonstrate that the program had effects on earnings and full-time employment even when the child was no longer eligible for Cash-for-Care at ages four and five. However, from age six, we can no longer see any effects. The effects seem to dissipate because most mothers remained attached to the labor force through part-time employment.


Introduction
Despite a dramatic increase in mothers' labor force participation since the 1960s, many mothers still choose to exit the labor market temporarily in order to stay at home with their young children. In the US, the labor market participation rate among mothers with their youngest child below age three is 54 percent, rising to 63 percent for mothers with their youngest child between ages three and five, and further to 73 percent among mothers with their youngest child above age five. In the OECD, the corresponding numbers for the three age groups are 51, 63 and 66 percent. 1 In this paper, we investigate how mothers' decision to stay at home with young children affects their subsequent work careers. In particular, we address the concern that such temporary exits may lead to long-run deterioration of women's post-birth careers (Lalive and Zweimüller 2009). This is important to countries considering policies that either encourage or discourage mothers to work while their children are young. The Scandinavian countries, for example, encourage female labor force participation by providing high quality publicly subsidized day care. In addition to allowing mothers to combine family life with work while the children are young, these family policies may have consequences for mothers' long-term labor force participation. High female labor force participation is considered important to maintain economic growth and sustainable pension systems (Burniaux et al. 2003). Moreover, 1 Maternal employment by age of youngest child in 2008 (OECD 2011).
female labor force participation secures family income and may hence prevent the detrimental effects on child outcomes of growing up in poverty (Yeung et al. 2002, Carneiro et al. 2009, Duncan et al. 2010. There are many reasons as to why a mother's decision to stay at home with young children could affect her long-term labor force participation. According to theories of human capital (Mincer and Polachek 1974), a mother who exits the labor market while the children are young will accumulate less work-related human capital; the exit may even lead to skill depreciation. 2 Moreover, opportunities in the labor market are network dependent (Rees 1966, Calvó-Armengol andJackson 2004). Presumably, a woman who is working while her children are young is more likely to develop a relevant network for her career than a mother staying at home with her children. Thus, when considering returning to work, a mother who temporarily exited the labor market might be less productive and have a more limited network than a mother who chose not to exit. This will likely be reflected in lower earnings and fewer job opportunities, which may discourage her from reentering the labor market. Mothers' reentry into the labor force could be additionally discouraged if staying at home with young children increases the accumulation of human capital related to home production (Becker 1991).
2 Some restoration of human capital may take place if the mother chooses to reenter the workforce after a temporary exit. In US data, Mincer and Ofek (1982) find a relatively rapid growth in wages for women reentering the labor force, suggesting a restoration of previously eroded human capital.
Identifying a causal effect of a mother's temporary labor market exit while her children are young on subsequent career development is difficult because of omitted variable bias. In particular, women who choose not to work while their children are young may have lower career aspirations than mothers who choose to work in ways that we cannot observe. It should not be surprising then if mothers who choose to stay at home while their children are young have less successful careers than mothers who choose to work. For identification, we study a unique, natural experiment in Norway that increased mothers' incentives to withdraw from the labor market in order to stay at home with their young children. The program, Cashfor-Care, was universal and paid any parent a significant allowance if they did not utilize a publicly subsidized child care for their one or two year old child. The subsidy substantially decreased the labor force participation of mothers of one-and two-year-olds (Schøne 2004, Drange 2012). Our paper investigates how this labor market exit when the child was one and two years old affected the mothers' subsequent career development, after they were no longer eligible for the Cash-for-Care subsidy.
The analysis utilizes a comprehensive, longitudinal register database containing annual records for every person in Norway. We estimate difference-in-difference models that exploit differences in individuals' exposure to the program among families with similar structures and within similar birth cohorts. We find that the Cash-for-Care subsidy decreased full-time employment among mothers of two-year-olds by about four percentage points. Following the development in mothers' labor supply as their child grows older, we can see that there is still a significant reduction in full-time employment when the child is four years old and no longer eligible for the subsidy. The estimates suggest that about 50 percent of the mothers who exited full-time employment when their child was two years old and eligible for the subsidy have still not returned to full-time employment at age four, the year after subsidy eligibility expired. At age five, there is still a significant negative effect. However, at ages six and seven, we can no longer see any effect of the subsidy on mothers' full-time labor market attachment.
Looking at earnings, we find similar results.
Further analysis reveals that the effects of the Cash-for-Care subsidy seem to dissipate because most mothers who exit full time employment while the children are young remain attached to the labor force through part-time employment. Norway is known for a very flexible labor market with a high share of part-time employment. Such part-time employment may limit the loss in human capital and relevant network when exiting full time employment. This paper is related to the literature studying the long-term effects of parental leave policies on women's labor market outcomes. Several studies show that job protection 3 associated with parental leave increases the likelihood that mothers return to the labor market when the children are older, and that it increases the job continuity with the pre-birth employer (Ruhm 1998, Berger et al. 2004, Baker and Milligan 2008. Notably, these studies study the effect of labor market exits while the children are younger than one year. There are two recent contributions that investigate the effects of parental leave extensions when the child is older than one year. Lalive and Zweimüller (2009) take advantage of an increase in the duration of paid and job-protected parental leave in Austria from one to two years. In a regression discontinuity analysis, they demonstrate that even if most mothers exhaust the full duration of their leave, there is no effect on their employment and earnings one year after the parental leave has been exhausted. Similar findings are reported in Schönberg and Ludsteck (2011), which studies five different German parental leave extensions. Utilizing a differencein-difference approach, this study demonstrates that the extensions, in which the job protection period is as long as the maternity benefit period, have no effect on mothers' employment or earnings six years after childbirth. However, the one reform that extended the maternity benefit period from 10 to 22 months, without extending the job protection period, had a large negative effect on mothers' employment and earnings six years after childbirth.
To our knowledge, our study is the first investigation of causal effects of mothers' labor market exits on long-term labor market outcomes when the exit is not associated with parental leave. This is an important distinction because, in contrast to parental leave 4 , the Cash-for-Care was not contingent on labor force participation pre-birth and had no formal restrictions on participation post-birth. Moreover, parental leave extensions are typically during the child's first year of life, and extensions in paid leave are typically connected to extensions in job-protection. 5 Our study investigate the causal effects of mothers' labor market exits lasting up to three years post-birth, which extends beyond the two year job protection period after child birth in Norway. 4 A benefit associated with employment, which provides either paid or unpaid time off work. 5 As mentioned above, Schönberg and Ludsteck (2011) and Lalive and Zweimüller (2009) investigate the effects of parental leave extensions when the child is older than one year. Moreover, in one of the reforms investigated in Schönberg and Ludsteck (2011) the extension in paid leave is not connected to extensions in job-protection.
The paper proceeds as follows. Section 2 describes the Cash-for-Care program and gives an overview on the Norwegian institutional setting. Section 3 presents the empirical strategy, and Section 4 describes the data. The empirical results are reported in Section 5, and Section 6 concludes.

Norway's Cash-for-Care Program
In August 1998, the newly elected Christian Democratic government began awarding cash allowances to parents who did not use publicly subsidized child care programs. Any family with a one-or two-year-old toddler could claim this allowance. The government stated that the main goals of the allowance were to give families financial freedom to stay at home with their young children, to allow families themselves to choose what kind of care they wished for their children, and to equalize public transfers to families-regardless of what kind of care the family wanted or had access to for their child.
At the time of introduction, the Cash-for-Care allowance constituted a significant part of family earnings even for high-income families. The annual allowance was NOK 36,000, and the average annual fee for publicly subsidized child care was about NOK 34,600 with some price subsidies for low-income families. Bettinger et al. (2011) demonstrate that for a family in the bottom income quartile, the effective after-tax price of a full-time day care slot for a one-or two-year old constituted about 40 percent of average family earnings. For the third and fourth income quartiles, the Cash-for-Care allowance constituted 15 and 10 percent respectively.
While the Cash-for-Care program was implemented simultaneously throughout Norway, there is variation in time and the ages of eligible children. Starting in August 1998, all one-year-old children were eligible for the Cash-for-Care allowance, from the month after they turned one year old. From January 1999, both one-and two-year-old children were eligible. 6 As a consequence, all children born from 1998 onward were eligible for 24 months of the Cash-for-Care allowance. For these children, eligibility started at the end or close to the end of maternity leave. We will refer to these children as fully treated. Children born prior to 1996 are not affected by the Cash-for-Care allowance. Children born in 1996 or 1997 could be eligible for as little as one month and as many as 24 months of the Cash-for-Care allowance. We will refer to these children as partly treated. Figure 1 below describes the nature of the treatment. Each cell represents the age of a child in a given year. Each cohort of children can be followed diagonally in this matrix. The darkly shaded cells represent fully treated children, whereas the lightly shaded cells represent partly treated children. The numbers in each cell denote how many months the mother of a child at a given age in a given year was eligible for the subsidy. Note that we have also shaded the cells of some of the older children not eligible for the Cash-for-Care subsidy. As we can see, these cells illustrate children who were treated as one or two-year-olds. If the 6 There was an exemption from this rule for all children who turned two years old after August 1, 1998. This exemption ensured that no children had a break in eligibility from the Cash-for-Care allowance. shaded cells represent fully treated children, whereas the lightly shaded cells represent partly treated children. The first cohort that is partly treated was born in 1996. The first cohort that is fully treated was born in 1998.
The uptake of the Cash-for-Care program was substantial. Figure

Norwegian Parental Leave, Child Care and Female Labor Market Participation
In the decade prior to the introduction of the Cash-for-Care subsidy, there was a substantial increase in female labor market participation in Norway. In 1991, 74 percent of married or cohabitant mothers with children aged 0-15 years were working. The corresponding figure for single mothers was 59 percent. When the Cash-for-Care subsidy was introduced in 1998, these figures had increased to 81 percent among married or cohabiting mothers and to 65 percent for single mothers (Kjeldstad and Rønsen 2002). 7 We assume that most families receiving the subsidy for a two-year-old child also received it when the child was one.
However, in spite of high participation rates among Norwegian women, the incidence of part-time employment is above the OECD average, and the share of women with managerial responsibilities is below. 8 (OECD 2011). Women are in general overrepresented in the public sector, particularly in occupations related to health and social work, where the prospect of obtaining a part-time job is good (Tronstad 2007).
The Norwegian government introduced the Cash-for-Care program at a time of extensive use of publicly subsidized day care. About 40 percent of children aged one or two years used publicly subsidized day care, 9 and there was a short supply of these day care programs. Moreover, at the time of its introduction, parents were entitled to 42 weeks of parental leave with full compensation, or alternatively 52 weeks with 80 percent wage compensation, 10 in addition to one year of unpaid job protection for each parent in connection with child birth. The Cash-for-Care program made it less costly to extend the period at home with the child before returning to work. However, if a mother chose to stay home with her children until age three her job would not be protected.
If a family wanted to receive the Cash-for-Care allowance, they would either have to take care of their child themselves or use informal care (e.g., relative, neighbor, or homebased day care). In Norway, formalized care consists almost exclusively of public and publicly subsidized private day care centers. The two types of centers are regulated by the same law; they basically offer the same type of program, have the same price schedule for parental pay and are equally subsidized. As there were very few private day care centers that did not run publicly subsidized programs, Cash-for-Care recipients in practice did not have the option of using private formalized care.
The Cash-for-Care program gave families strong incentives to reduce labor supply and substitute formal care with parental care, or to substitute formal care with informal care. Rønsen (2001) demonstrates that the program increased both parental time at home and time in informal day care. Moreover, the Cash-for-Care allowance decreased eligible mothers' fulltime labor force participation by about four percentage points across the whole population but had no effect on fathers' labor force participation (Schøne 2004;Drange 2012).

Other Family Reforms
During the years prior to the introduction of the Cash-for-Care allowance in 1998, Norway implemented several work-family-related policies. In particular, there was a large extension in paid parental leave between 1986 and 1993. In 1986, Norwegian parents were granted 18 weeks of paid parental leave, but during subsequent years, leave rights were gradually extended to 35 weeks in 1992 and to 42 weeks in 1993. Moreover, in 1993, Norway introduced a paternity quota of paid parental leave. Of the 42 weeks of paid parental leave, four weeks were reserved exclusively for the father.
Notably, the parental leave policies were initiated at least three years prior to the introduction of the Cash-for-Care allowance, and at least five years before the birth of the first cohort that was fully entitled to the allowance. As such, these policies should not be of any concern for our identification strategy if uptake was immediate. The uptake of the expansions in parental leave was immediate. However, the paternity quota was not extensively used until two years after implementation. The slow uptake of the paternity quota may raise a concern for our analyses because it implies that our partially treated younger siblings were fully affected by the paternity quota, whereas our control group was only partially affected by the quota. In an investigation of the paternity reform, Rege and Solli (2010) demonstrate, however, that the paternity quota had a small effect on fathers' labor market attachment but did not affect mothers' labor supply.
Another reform relevant to our study is the 1997 school reform that changed the school starting age from age seven to age six. All children in our sample started school at age six, so even if mothers' labor supply might be affected by their children starting school earlier, 11 the same effect should be prevalent for both the comparison and treatment groups.
The school reform may still be of concern because it led to an increase in the cover of publicly subsidized day care slots in 1997, because six year olds were no longer in need of child care slots. This increase in day care availability could possibly increase mothers' labor force participation while having young children 12 and thereby bias our estimates downward.
However, considering the development in child care slots for the age groups in question, there seems to be little evidence of a spike in child care attendance for five-year-olds in 1997, 11 Gelbach (2002) finds that mothers' labor supply increases when their oldest child starts school.
suggesting that child care for this age group was not rationed at the time. 13 For the two-yearolds, however, there is an increase in enrolled children in 1997. As such, the cohorts in our analysis (all children of ages two and five in 1997 and 2000) should be similarly affected by the introduction of an early school start.
Finally, at the time of our study, the government implemented a reform giving single mothers stronger incentives to reenter the labor market. The reform was gradually implemented, starting January 1 st 1998, and entailed work requirements and higher in-work benefit levels. According to Mogstad and Pronzato (2009), this led to an increase in single mothers' labor force attachment. Given the gradual phase-in of this policy change, it is hard to determine whether and how it would affect our treatment and control groups. We include a covariate capturing whether the family is a single-mother household in all analyses unless otherwise stated. Moreover, an analysis excluding the group of single mothers provides similar results as the main analysis. These analyses are not reported in the paper but are available from the authors upon request.

Empirical Strategy
To estimate the effects of the Cash-for-Care program on mothers' long-term labor market outcomes, we exploit variation across similar families over time in a difference-indifference analysis. The shading in Figure 1 illustrates the nature of the treatment. We can see from Figure 1 that children born in 1995 represent the latest cohort that is not treated, and children born in 1998 represent the first cohort that is fully treated. We estimate the following difference-in-difference coefficient for the effect of the Cash-for-Care subsidy when the children are a years old (for a = 2, 3, 4, 5 and 6 years old): (1) ) ( ) (   95  ,  3  98  ,  3  95  ,  98 , a a a a a a a a a ls ls ls ls where y a ls , denotes the labor force participation rate of mothers of children age a in year y.
The difference-in-difference coefficient a  measures the changes in full-time attachment between year 1995+a and year 1998+a for mothers of a year olds compared with mothers of a+3 year olds.
To understand better our difference-in-difference approach, let us start by considering how the introduction of the Cash-for-Care subsidy affected eligible mothers of two-year-olds.
By substituting a=2 in Equation (1), we get: The children from the 1995 and 1998 cohorts are two years old in 1997 and 2000, respectively. Thus, we can look at how Cash-for-Care eligibility affected mothers of twoyear-olds by examining the difference in the labor supply of mothers of two-year-olds between 1997 and 2000, which is the first difference in Equation (2). Of course, there are many factors aside from the Cash-for-Care subsidy that may affect mothers' labor force participation between 1997 and 2000. Note from Figure 1 that in 2000, five-year-olds are the youngest cohort who never received the Cash-for-Care subsidy. Thus, we can control for these other factors by subtracting the difference in labor force participation of mothers of fiveyear-olds between 1997 and year 2000, which is the second difference in Equation (2). If trends in labor force participation are identical for mothers of two and five-year-olds, then 2  will capture the causal effect of the Cash-for-Care subsidy on mothers of two-year-olds.
When investigating how the Cash-for-Care subsidy affected mothers of three year olds, we follow the same cohorts and look at the same difference-in-differences one year forward. This will give us the 3  difference-in-difference coefficient. Similarly, we obtain the 4  coefficient by following the same cohorts and look at the same difference in differences two years forward, and so on.
In our empirical analyses, we estimate a  by restricting the sample to mothers of two and five-year-olds in 1997 and 2000, respectively, and then estimate the following differencein-differences model: where i ls is mother i's labor market outcome when the child is a years old, is a dummy indicating whether the year is 1998+a, and X is a vector capturing a rich set of observable characteristics of the child, mother and father that may influence the mother's labor market outcome. These variables are all observed prior to the introduction of the Cash-for-Care subsidy and are specified in Section 4.
The difference-in-difference model in (3) will provide an unbiased estimate of a  if the trends in labor supply of mothers of a year-olds and a+3 year-olds would have been the same in the absence of the Cash-for-Care subsidy. There are several ways in which this identifying assumption might be violated. For example, we know that mothers increased their labor market participation during the 1990s. This increase might not necessarily have been the same for mothers of children of different ages. Moreover, we could imagine that changing labor market conditions affected mothers differently across time and across the age of the children. Another concern is changing trends in fertility, which may lead to compositional changes among the different groups of mothers. The richness of our data allows us to perform numerous specification and robustness tests that allow us to address the validity of our identifying assumption.
Another concern of our empirical strategy is that the Cash-for-Care subsidy may have affected fertility. 14 This should not directly affect the treated children in the sample, as they are all born prior to the introduction of the subsidy. A change in fertility may, however, affect the sample indirectly, as we exclude those with a new sibling up until the year they turn seven years old, in order to ensure that the older children in the control group do not have treated siblings. 15 We address this concern by demonstrating that our results are robust to the inclusion of children with younger siblings. 14 There is a work-in-progress paper on this issue by Pål Schøne and Ines Hardoy, but the results are not conclusive. 15 Our empirical analyses address this concern by demonstrating that our results are robust to the inclusion of children with younger siblings.
We utilize registry data called FD-trygd, which is a combination of several Norwegian registry databases provided by Statistics Norway. Our dataset contains records for every Norwegian resident from 1992 to 2005. The data provides individual demographic information (marital status, spouse identifier, sex, age, number of children), socioeconomic data (years of education, income, wealth), current employment status (full time, part time, minor part time, self-employed), indicators of participation in any of Norway's welfare programs and geographic identifiers for county, municipality and neighborhood of residence.
Importantly, the data contain identifiers for mother and father, which allow us to match the children to their parents.
As described in our empirical analysis section, our main analytic sample consists of all children of two and five years old in 1997 and 2000, respectively. We further restrict the sample to children who have no younger siblings at age seven. We make this restriction in order to ensure that the older children in the control group do not have treated siblings. 16 We analyze how the Cash-for-Care subsidy affected mothers' long-term labor market attachment by using several different outcome variables. All outcome variables are constructed for each of the years when the child is 2-7 years old for the children aged two years in 1997 and 2000 and for each of the years when the child is 5-10 years old for the 16 As discussed in Section 3, this restriction is potentially endogenous if the reform affected fertility.
However, our empirical analyses address this concern by demonstrating that our results are robust to the inclusion of children with younger siblings. children aged five years in 1997 and 2000. Our key outcome, which we refer to as full-time employed, is a variable denoting whether the mother is working more than 30 hours per week.
Inspection of the data reveals that some individuals are recorded as full-time employed despite very low recorded earnings or even zero earnings. This is likely because of lags in firms' submissions of employer information. We address this by coding everybody with earnings that precludes employment as nonemployed. 17 We also use an employment variable capturing whether the mother was either full-time or part-time employed, which we refer to as employed. A mother is coded as employed if she is registered as working minor part time, part time or full time and has earnings above a certain threshold. 18 Finally, we use earnings as an outcome measure. 19 We avoid using log earnings because all mothers are included in the earnings analyses, as well as those who are not working and have zero earnings. All mothers have to be included in the earnings analysis, as the sample of working mothers is endogenous to the Cash-for-Care reform.
Our data allows us to construct several control variables capturing important child, father and mother characteristics that we include in our regression analyses. In order to ensure that covariates are not endogenous to the reform, all covariates are collected from a baseline 17 Everybody with an income less than 2 G is coded as not being full-time employed. G is set by the Norwegian Labour and Welfare Administration every year and is included in most formulae for welfare transfers. In 1997, G was NOK 42,000. year, prior to the introduction of the Cash-for-Care subsidy. For the children aged two years in 1997 and 2000, we collect covariates from the year prior to their birth in 1994 and 1997. For the children aged five years in 1997 and 2000, we collect covariates when these children are two years old in 1994 and 1997.
Our control variables include the following: child gender, number of children (0, 1, 2, 3, 4, 5), 20 mother's age at birth of youngest child (<20, 20-24, 25-29, 30-34, 35-39, 40-44, 45), age at birth of oldest child (<20, 20-24, 25-29, 30-34, 35-39, 40-44, 45), father's age at birth of youngest child, parents' education (completed high school, completed college), linear and quadratic controls for parents' earnings, parents' employment status (minor part time, part time or full time), indicator for parents receiving any social welfare benefits, indicator for parents living in a densely populated area (city), indicator for parents' immigration status, indicator for being a single mother, and municipality-specific unemployment rates interacted with the age of the children. Finally, we include municipalityfixed effects.   between post-and pre-treatment. The unadjusted difference-in-difference estimate is significant and consistent with an effect of the Cash-for-Care subsidy on mothers' full-time employment when the child is two years old. There is also some evidence consistent with persisting effects of the Cash-for-Care subsidy at age four, after subsidy eligibility has expired. We can see that mothers of four-year-olds increased their full-time employment between post-reform and pre-reform, but less so than mothers of seven-years-olds. The unadjusted difference-in-difference estimate is significant. However, this effect seems to have dissipated when the treated children reached age seven, when the unadjusted difference-indifference estimate is smaller and no longer statistically significant. Not surprisingly, looking at differences in earnings we see the same pattern, reflecting the changes in full time employment for the different groups of mothers.

Summary Statistics
When considering the covariates in Table 1, Panel B and C, there seem to be a few changes among parents of treated children that do not correspond to changes among the parents of children in the comparison group. There seems to be an increase in the number of families with two children in the comparison group post-reform. Furthermore, we see that mothers and fathers with children from later cohorts are more likely to have finished college than are mothers and fathers of the older children. This possibly relates to general trends in education of the population during the period. 23 We also note that while the share of single mothers is stable in the treatment group, it decreases somewhat in the comparison group. 23 Statistikkbanken, Statistics Norway. Note: Mean or share of indicated variable with differences. Earnings are inflation adjusted with 1997 as base year (in NOK). +, * and ** denote significance at the 10, 5 and 1 percent levels respectively (two-sided t-test) and are reported for the difference-in-difference estimates.
Finally, there is a barely significant decrease in the share of parents who lived in densely populated areas. This might influence the labor force attachment if there is a higher unemployment rate in rural areas.
As described in our empirical strategy, we account for possible observable changes in the composition of the post-and pre-reform groups by including a rich set of parental and child characteristics (described in Section 4). We also add covariates sequentially in order to investigate if differential trends or observed changes in the composition of the groups affect our estimates. Moreover, we address the concern of differential effects of different labor market conditions by including controls for the local unemployment rate interacted with the age of the child. and significant effect of the Cash-for-Care reform on mothers' labor force participation in the years when the child turns two and three years old. This is the age at which the parents are fully or partly eligible for the subsidy. The estimate suggests that the subsidy decreased fulltime employment among mothers of two-year-olds by about four percentage points, which is similar to the estimates in other studies (Schøne 2004, Drange 2012.

Mothers' Labor Supply
Following the development in mothers' labor supply as their child grows older, we observe that when the affected child is four years old and no longer eligible for the subsidy, there is still a significant effect on mothers' full-time employment. The subsidy decreased full-time employment among mothers of four year olds by about two percentage points. This suggests that about 50 percent of the mothers who exited full-time employment when their child was two years old and eligible for the subsidy were still not working full time in the year after the subsidy expired.
When the treated child is five years old, we can see that the difference-in-difference estimate is smaller than at age four and only marginally significant. At age six, the estimate is no longer significant. This suggests that the Cash-for-Care reform had an impact on mothers' labor supply in the first two years after the child lost eligibility but that the effect faded during subsequent years. Most mothers return to the labor market by the time the child turns six years old, which is the age at which children start school in Norway. This finding is in line with studies suggesting that school enrolment affects mothers' labor supply (e.g., Gelbach 2002).
As discussed in the empirical strategy section, one concern for the validity of our identifying assumption could be compositional changes among the different groups of mothers. In Models 2, 3, and 4, we investigate robustness by stepwise adding covariates capturing the child-and parental characteristics described in Section 4. We can see that adding child-, mother-and father-specific characteristics has very small impacts on the estimates. This indicates that our estimates are not biased by compositional changes among the different groups of mothers.
Another concern of our identifying assumption is unemployment shocks that affect mothers differently depending on the age of their children. We investigate this concern in Models 5 and 6 by adding controls for local unemployment rates, linearly and interacted with the child's age. We can see that the effect estimates are robust to these inclusions, suggesting that the estimates are not biased by local unemployment shocks that affect mothers with children of different ages differently.
As discussed in Section 4, we restrict the sample to children who have no younger siblings at age seven in order to ensure that the older children in the control group do not have treated siblings, which may bias our estimates downward. This restriction may raise an endogeneity concern if the Cash-for-Care reform affected fertility. In Model 7, we demonstrate that our results are robust to the inclusion of children with younger siblings. As expected, the estimates are somewhat smaller, but we still see negative and significant effects on full-time employment among mothers of children 2-5 years old. Notes: +, * and ** denote significance at the 10, 5 and 1 percent levels respectively. Estimates are based on OLS on equation (2) with the dependent variable being whether the mother has a full-time attachment to the labor market. We follow the cohort of fully treated children born in 1998 as two year olds in 2000, three year olds in 2001, etc. Model 1 is run without covariates. In the following five models, we add the following variables: sibling's characteristics, mother's characteristics, father's characteristics, municipality-specific unemployment rate in the year when the treated children turned two, and in Model 6, the same unemployment rate interacted with the age of the child. In Model 7, we report results from an estimate on the full sample of children (i.e., without restricting to the youngest child). Robust standard errors (in parenthesis) are clustered on the child's mother and account for heteroscedasticity and nonindependence of residuals across mothers' labor force participation observed at different points in time. All specifications include municipality fixed effects.

Specification Analysis
In the specification analysis of  Table 2 are due to the Cash-for-Care reform, then we should see a pattern in Table 3 that is similar to the shading in Figure 2. We should not see any effects in the non-shaded cells, as these are mothers of non-treated children. We may see some effects in the lightly shaded cells, as these are mothers of partly treated children. We should see effects in the darkly shaded cells, at least for mothers of 2-3 year olds, as these are mothers of fully treated children. If the reform has persisting effects, then we should also see an effect on older children in darkly shaded cells. 24 Additionally, the robustness matrix has a different sample selection criterion. In order to include all ages of children across a large number of years, we include all children with no younger siblings in the outcome year (Table 2 uses children with no younger siblings at age seven). Notes: +, * and ** denote significance at the 10, 5 and 1 percent levels respectively. Standard errors (in parentheses) are clustered on the child's mother and account for heteroscedasticity and nonindependence of residuals across mothers' labor force participation observed at different points in time. Estimations are based on OLS. Changes in all years are measured relative to 1996, and the reference age is eight. Because of data constraints, this matrix is run without covariates. A child will be excluded from the sample in the year when the mother gives birth to a new sibling. Source: Administrative registers: FD Trygd.
We can see that the estimates appear largely consistent with the introduction of the Cash-for-Care reform. All the coefficients in the darkly shaded cells with fully treated children are large and significant, even at ages four and five. We can also see that several coefficients in the lightly shaded cells are significant, in particular for the cohort of children born in 1997, among whom many were close to being fully treated. Important for our identifying assumption, we can see that in most of the cells with no shading, the difference-indifference estimates are small and insignificant. This suggests that trends in labor supply between mothers of children of different ages are similar prior to the reform, supporting our identifying assumption. Notably, there is a weakly significant difference-in-difference estimate at age two in 1997. This may reflect behavioral changes in expectation of the Cashfor-Care subsidy.

Subsample Analysis and Alternative Dependent Variables
In Table 4 we investigate the differential effects of Cash-for-Care across different levels of education and pre-birth earnings. All subsample analyses use our preferred model from at baseline. Mothers with college education and mothers with above median earnings at baseline are back in a full-time position to the same extent as the non-treated mothers. Notes: +, * and ** denote significance at the 10, 5 and 1 percent levels respectively. Estimates are based on OLS of equation (2) with the outcome being whether the mother has a full-time attachment to the labor market. We follow the cohort of fully treated children born in 1998 as two year olds in 2000, three year olds in 2001, etc. The covariates described in Section 4 are included. Robust standard errors (in parentheses) are clustered on the child's mother and account for heteroscedasticity and nonindependence of the residuals across mothers' labor force participation observed at different points in time. All regressions include municipality fixed effects. Source: Administrative registers: FD Trygd.
In This is somewhat surprising, as close to four percentage points of these mothers exited fulltime employment when eligible for the Cash-for-Care subsidy. If the labor market withdrawal affected their human capital development, we should expect to see this reflected in earnings.
In Model 2, we see significant effects of the subsidy on earnings for less-educated mothers of four-and five-year-old children. This reduction in earnings may reflect a relative loss in human capital in addition to the reduction in full time employment that we can see in Model 2 of suggests that those who were closely attached to the labor market prior to birth and who chose to exit at ages one and two did not experience a reduction in human capital. The reduction in earnings for those who had a weaker attachment to the labor market may reflect a relative loss in human capital in addition to the reduction in full-time employment. Notes: +, * and ** denote significance at the 10, 5 and 1 percent levels respectively. Estimates are based on OLS of equation (2) with the outcome being mothers' linear earnings. Earnings are inflation adjusted with 1997 as the base year (in NOK), and censored at the 99 th percentile.
We follow the cohort of fully treated children born in 1998 as two year olds in 2000, three year olds in 2001, etc. The covariates described in Section 4 are included. Robust standard errors (in parentheses) are clustered on the child's mother and account for heteroscedasticity and non-independence of residuals across mothers' labor force participation observed at different points in time. All regressions include municipality fixed effects Source: Administrative registers: FD Trygd.
To understand the differences in long term effects across subsamples better, we now turn to Table 6, where we investigate the effect of the Cash-for-Care subsidy on mothers' employment. This table has the same structure as Tables 4 and 5; however, the dependent variable is replaced with employment, which includes minor part-time and part-time in addition to full-time employment.
We can see that Table 6 displays a quite different picture compared to Tables 4 and 5.
The Cash-for-Care subsidy has very small effects on employment already by age three in all sub-groups. This suggests that most mothers remained attached to the labor market through part-time employment, even if they withdrew from full time employment for an extended period. In Models 1 and 4 we can see that this is particularly true for mothers with college education and high earning mothers. This part time attachment to the labor force may provide one explanation for why these mothers did not experience a persistent decrease in earnings (see Table 5), despite the fact that many exited full time employment when eligible for the Cash-for-Care (see Table 4).
In Models 2 and 3 we can see that also mothers with low education and mothers with low earnings remain attached to the labor force through part-time employment, but less so than high educated and high earning mothers. This may provide an explanation for why we see a more persistent effect of the Cash-for-Care subsidy on these mothers labor supply (see Tables 4 and 5). Notes: +, * and ** denote significance at the 10, 5 and 1 percent levels respectively. Estimates are based on OLS of equation (2) with the dependent variable being whether the mother works at all. We follow the cohort of fully treated children born in 1998 as two year olds in 2000, three year olds in 2001, etc. The covariates described in Section 4 are included. Robust standard errors (in parentheses) are clustered on the child's mother and account for heteroscedasticity and nonindependence of residuals across mothers' labor force participation observed at different points in time. All regressions include municipality fixed effects Source: Administrative registers: FD Trygd.

Conclusion
In this paper we address the concern that temporary labor market exits while the children are young may lead to long-run deterioration of women's post-birth careers. We investigate how mothers' decision to stay at home with young children affects their subsequent work careers. For identification, we utilize the introduction of a Cash-for-Care subsidy, which was universal and paid any parent a significant allowance if they did not utilize a publicly subsidized child care for their one or two year old child. The subsidy substantially decreased the labor force participation of mothers of one-and two-year-olds (Schøne 2004, Drange 2012. We demonstrate that the program had effects on earnings and full-time employment even when the child was no longer eligible for Cash-for-Care at ages four and five. However, from age six, we can no longer see any effects.
The effects seem to have dissipated by age six because most mothers remained attached to the labor force through part-time employment. Further analysis reveals that most mothers remained attached to the labor market through part-time employment, even if they withdrew from full time employment for an extended period. Norway is known for a very flexible labor market with a high share of part-time employment. Such part-time employment may limit the loss in human capital and relevant network when exiting full time employment.
Our findings are consistent with several studies that have emphasized the importance of a continued attachment to the labor market in securing labor supply post-birth. 26 26 See, for instance, Berger and Waldfogel (2004), Baker and Milligan (2008) and Schönberg and Ludsteck (2011).