ESTUDIOS SOBRE LA ECONOMÍA ESPAÑOLA

Rocio Sánchez-Mangas Virginia Sánchez-Marcos

EEE 195

new version November 2007

fedea

Fundación de Estudios de Economía Aplicada

ISSN 1696-6384

Las opiniones contenidas en los Documentos de la Serie EEE, reflejan exclusivamente las de los autores y no necesariamente las de FEDEA.

The opinions in the EEE Series are the responsibility of the authors an therefore, do not necessarily coincide with those of the FEDEA.

Rocio S·nchez-Mangas Universidad AutÛnoma de Madrid

Virginia S·nchez-Marcosy Universidad de Cantabria and Fedea

October 2007

Abstract

The aim of this paper is to measure the potential e§ect of a family policy introduced in Spain in 2003 that provides working mothers with a monthly cash beneÖt of 100 euros per child aged under 3 years. We explore the e§ect of the policy on eligible womenís labour market participation. In the tradition of the policy evaluation literature we use a di§erence-in-di§erences-in-di§erences (DDD) estimation approach. Our results support a small but signiÖcant positive e§ect of the policy. We Önd that since the implementation of the policy the labour market participation rate for mothers of children aged under 3 has risen by 3 percentage points compared to the rate for non-policy-eligible females. This represents 5% of their average labour market participation in 2002, the year before the policy was implemented. This overall policy e§ect is dominated by the e§ect of the policy among high school educated females.

Acknowledgments: We would like to thank VÌctor Aguirregabiria, Alfonso Alba, CÈsar Alonso, Carlos Bethencourt, Raquel Carrasco, Cristina Fern·ndez, Sara de la Rica, participants in the Society of Labor Economics 2005 Conference and the 2006 Econometric Society European Meeting and two anonymous referees for helpful comments. R. S·nchez-Mangas is thankful for research funding from the Spanish Ministry of Science and Technology, project SEJ 2006-04957/ECON, and from Universidad AutÛnoma de Madrid - Comunidad de Madrid, project CCG06-UAM/HUM-0452. V. S·nchez-Marcos is thankful for research funding from the Spanish Ministry of Science and Technology, project SEJ 2006-10827/ECON. This paper previously circulated with the title "Reconciling female labour participation and motherhood: the e§ect of cash beneÖts for working mothers"

yCorresponding author: V. S·nchez-Marcos, Dpto. de EconomÌa, Universidad de Cantabria, Avenida de los Castros s/n, 39503 Santander, SPAIN. Tel:+ 34 942202025. e-mail: sanchezv@unican.es

JEL Codes: J13, J18, C51

Keywords: female participation, child care beneÖt, policy evaluation, di§erencein-di§erences-in-di§erences estimation

I. Introduction

At the Lisbon Summit held in June 2000 the EU governments announced that one of their primary goals was to reach a total employment rate of 70% by the year 2010. Policies aimed at increasing mothersílabour force participation are considered crucial to achieving this target. However, as is reported by Del Boca and Pascua (2002), substantial di§erences exist between southern countries, like Spain or Italy, and northern countries, like Norway or Sweden, in terms of female participation in the labour market and in terms of family policies attempting to improve work incentives for parents.

In the case of Spain there is reason to believe that women experience di¢culties in reconciling female participation and motherhood. Firstly, the employment rate for mothers is lower in Spain than in most OECD countries. GutiÈrrez-Domenech (2005) reports that Spain, together with West Germany, are the countries with by far the highest long-term decline in employment among women following the Örst birth. Secondly, Spanish regulations concerning maternity leave establish 16 weeks of paid leave after birth, a shorter period than in most of the European countries where the average is about 25 weeks, according to OECD (2001). Thirdly, the use of formal child care arrangements for children aged under 3 years is very rare in Spain. As is reported in OECD (2001), the proportion of children aged under 3 years using formal child-care arrangements is only 5%, in sharp contrast to the average 25% for the European countries. Fourthly, the Spanish Labour Population Survey (2004) clearly indicates that family responsabilities are an important reason why Spanish females do not participate in the labour market. For the group of people aged under 45 years, less than 1% of males report family responsibilities as the main reason for not participating in the labour market; however, this Ögure is 29% in the case of females. Finally, the fertility rate in Spain is one of the lowest in the EU-15 countries (Eurostat (2007)).

Encouraging mothers to participate in the labour market is a key mechanism to reduce the gender gap labour force participation. As a group, mothers of young children have among the lowest labour participation rates, see for example OECD (2002). In addition, spells out of the labour market for maternity reasons can erode human capital and reduce the accumulation of labour market experience, and thus result in a decrease of future potential wages and in a reduction of the incentives to join the labour market later in life.

Empirical and theoretical studies support the prediction that increasing child care cost lowers the female participation probability. Several studies provide evidence suggesting that the lower rate of participation shown by mothers of preschoolers in comparison to all other females is closely related to high child care costs faced by these females (see, for example, Blau and Robins (1988), Connelly (1992), Del Boca and Pascua (2002), Ferrero and Iza (2004), Attanasio et al. (2004)). At the same time there is a political debate regarding family policies trying to encourage mothers to enter the labour market. Recent measures adopted by the Spanish government are an example of this kind of initiative. SpeciÖcally, in January 2003 the Spanish government passed a new family policy aimed at reconciling motherhood and labour market participation. The policy provides working mothers of children aged under 3 years with a monthly cash beneÖt of 100 euros per child.

The objective of this paper is to measure the direct e§ect of this policy on the participation of mothers of children aged under 3 years, and thus, to contribute to the discussion about whether to increase the amount of public resources devoted to promoting the participation of mothers in the labour market. We focus on this direct e§ect given the short period of time that has elapsed since the introduction of the policy. However, there may be an indirect e§ect on participation of mothers of children aged over 3 years resulting from a stronger attachment to the labour market during the early maternity ages. Finally, the policy may also have e§ects on fertility decisions.

Following the tradition of the policy evaluation literature this paper uses the ëdi§erence-in-di§erences-in-di§erencesí(hereafter DDD) estimation approach to measure the potential e§ect of the policy. The essence of this approach is to estimate the rate of change in the probability of participation of eligible women (in our case, mothers of children aged under 3 years) before and after the policy and compare it to the rate of change in the estimated participation probability of non-eligible women. First, we provide evidence of a sizeable growth rate di§erential in labour market participation between eligible and non-eligible women after the policy implementation. We then proceed to estimate an econometric model to perform the DDD approach. Based on our estimates, we Önd that the potential e§ect of the cash beneÖt policy on the labour market participation of eligible females has been small but positive. SpeciÖcally, the increase in labour market participation for mothers of children aged under 3 years was 3 percentage points higher than the increase experienced by noneligible females. This represents 5% of the before-policy average participation rate of the policy-eligible women. Concerning the potential heterogeneity of the e§ect across education levels, we Önd a positive and signiÖcant e§ect of the policy for high school educated females, that we do not Önd for females with elementary or college education. Thus, the overall e§ect of the policy is dominated by the e§ect stemming from the high school educated group.

Our paper is related to others in the literature analyzing family policies. Schone (2004) and Naz (2004) examine the e§ect of a reform in Norway which provides mothers of children aged under 3 years with a cash beneÖt, irrespective of their working status. They Önd that the policy increases intra-household specialization of work and hence reduces the labour market participation of women. Francesconi and Van der Klaauw (2007) focus on the e§ect of ëin workíbeneÖts to single mothers in the United Kingdom and Önd that the reform led to a substantial increase in female employment rates. We contribute to this literature by providing new evidence on the labour market e§ects of a cash beneÖt for working mothers of young children during the period 2003-04 and new estimates of the elasticity of labour supply to childcare costs. Furthermore, there is an additional contribution derived from the fact that Spain is one of the European countries with the lowest labour participation of mothers of young children, as reported by GutiÈrrez-DomÈnech (2005).

The remainder of the paper is organized as follows. In Section II we explain the details of the cash beneÖt policy under evaluation. In Section III we describe the data set and show empirical evidence for recent trends of female labour participation rates and, in particular, for those women potentially a§ected by the policy. In Section IV we formulate the econometric model and perform the DDD analysis. In Section V we show our estimation results. Finally, Section VI o§ers our conclusions.

II. The cash beneÖt for working mothers

According to Eurostat, public expenditure on policies related to family and children was below 1% of GDP in 2002 in Spain, in contrast to an average of 2% in the EU-15 countries.¹ In January 2003 several changes concerning family policy were undertaken in Spain (Law 46/2002). Some of these changes were aimed at promoting fertility and others at helping to reconcile family and work.

With the aim of promoting fertility, a reform of the income tax was implemented, with two main changes concerning the treatment of families with children: an increase in the tax deduction regarding the number of children and an increase in the tax deduction for each child aged under 3 years. The Örst of these changes was implemented as follows: before the policy, families could reduce their annual taxable income by 1200 euros each for the Örst and second child and by 1800 for the third child and subsequent children. Since the policy was introduced, families have been able to reduce their annual taxable income by 1400 euros for the Örst child, 1500 euros for the second child, 2200 for the third child and, Önally, 2300 for each sub sequent child. The second change raises the additional tax deduction for having a child under 3 years from 300 to 1200 euros per child. Both of these innovations are aimed at promoting fertility.

¹See Instituto de PolÌtica Familiar (2006) for a comparison across European countries.

With the aim of reconciling family and work, a monthly cash beneÖt for working mothers of children aged under 3 years was introduced. The monthly cash beneÖt amounts to 100 euros per child aged under 3 years. To be eligible, working mothers must fulÖll certain conditions in relation to the number of hours worked. These conditions di§er for full-time and part-time working mothers. In particular, fulltime female workers must work at least 15 days per month. For part-time female workers the equivalent Ögure is 20 days. Furthermore, part-time female workers are eligible only if they work at least 50% of full-time hours. There are two alternative ways in which the beneÖt can be received. One possibility is that eligible females apply to the Public Administration to receive the cash beneÖt via bank transfer. The other possibility is to deduct from their annual payable taxes an amount equal to the corresponding annual cash beneÖt. There is an upper limit to the cash beneÖt given by the annual social security payroll taxes. However, there is no an income test to determine eligibility for the beneÖt. The analysis of the policy e§ect on labor market participation of mothers of children aged under 3 is the main objective of this paper.

How much child care can the cash beneÖt buy and how much does it represent in terms of femalesíearnings? The cash beneÖt covers, on average, more than one third of the cost of private day-care centers in Spain, so it can Önance a substantial amount of the child care costs faced by working mothers.² According to the average female earnings by education level reported by the Spanish Statistical O¢ce in 2002, a cash beneÖt of 100 euros per month represents 13% of the earnings for females with elementary education, 8% if the woman has completed high school education and 5% in the case of college educated females. However, if instead of considering observed wages, we consider the potential wages of the whole female population these numbers would be higher given that the process of selection into the labour market is expected to exclude those women with lower potential earnings.

In 2003, when this family policy was introduced, there were 625000 working mothers who collected the cash beneÖt; as mentioned above, all working mothers with children under 3 years were eligible. In 2004, the Ögure was about 664000 mothers. According to the Institute of Family Policy, approximately six per cent of families in Spain beneÖt from the policy.

The simultaneous implementation of the cash beneÖt for working mothers of children aged under 3 years and the increase in the tax deduction for mothers of children aged under 3 years irrespective of their working status prevents us from isolating the e§ect of the separate policy reform components on female labour market participation. As stated in the introductory Section, previous research for other countries shows that cash beneÖts for mothers, irrespective of their working status, can reduce female labour market participation by increasing intra-household specialization of work. So, in the case of the policy reform we are analyzing, the potential positive e§ect of the cash beneÖt on female labour market participation could be lessened by the other component of the policy reform, the increase in the tax deduction for all mothers of children aged under 3 years. However, in the case of the particular policy passed in Spain the tax deduction available to all mothers is not expected to have a strong e§ect on female labour market participation for two reasons. First, the tax deduction is not e§ective until one year after the earnings are made, which is especially important if there are restrictions on borrowing against future disposable income in the credit markets. Second and more important, the tax deduction implies a small increase in the annual disposable income for females.³

²In 2005, the average cost of private day-care centers in Spain was 254 euros per month (Consumer (2005)). Thus, a monthly cash beneÖt of 100 euros represents 39% of that cost.

III. Data and sample

The dataset we use in this paper is from the Labour Population Survey (EPA) conducted quarterly by the Spanish Statistical O¢ce (INE). These data consist of pooled cross-sections of more than 150000 individuals, covering the period 1996-2004, i.e., seven years before and two years after the policy implementation.⁴ We restrict our attention to married females aged under 45.⁵ Our focus is on this group of females, as single female labour participation rates are similar to male participation rates. Furthermore, the proportion of single mothers is still very low in Spain compared to other countries.⁶ Left out of the analysis are those women belonging to the oldest cohorts, who will certainly have di§erent labour behaviour and are not likely to be eligible for the policy.

³Marginal income tax rates in Spain ranged between 15% and 45% in 2003 and the average marginal tax was around 20%. For a female whose marginal income tax rate was 20%, the increase in the tax deduction from 300 to 1200 euros represented an increase in annual disposable income of 180 euros, i.e, 15 euros per month.

⁴In 2005 some changes were introduced in the Labour Population Survey. First, instead of providing the individualís age as the number of years or months since birth, Age is coded into intervals, the Örst of them being 0-4 years. This does not allow us to use data from 2005 onwards for our purposes since we cannot identify the targeted group of the policy, mothers of children aged under 3 years. Second, several changes were made in the questions aimed at determining the employment and participation status of the individual. As a result of this last change, the number of employed people increased and the number of unemployed people decreased in 2005 with respec to 2004.

⁵In fact, we understand as married females all females living in biparental households and being the head or the partner of the head of the household, irrespective of whether they are married or not.

⁶ The average participation rate of married females aged under 45 is 66% for the period 2000- 2004, whereas the average participation rate of single females is 90%. If we focus on mothers, the average participation rate of married females is 61%, whereas the average participation rate of single females is 87%. The average proportion of single mothers for the period 2000-2004 is 6%.

As an initial approach to the question at hand we provide a descriptive analysis of the labour participation behaviour of our sample of married Spanish women. Since there are substantial di§erences in participation across education groups, we distinguish three groups of individuals. The Örst group, which includes elementary educated females, represents 15% of our sample. The second group, composed of females having completed their high school education, is the most numerous one, representing 65% of the sample. The remaining 20% corresponds to the third group, those females who have completed college education.

Table 1 provides the sample mean and standard deviations for continuous variables for several individual characteristics in 2004: the number of children, the fraction of women who worked in the previous year, the fraction of women whose spouse is working, womanís age and the participation rate. Finally, the last column of Table 1 includes the sample sizes. This information is reported by educational level for several groups of females: married females, married females who have no children, married mothers, married mothers of children aged 3 to 6 years old and Önally, the targeted group of the policy, married mothers of children aged under 3 years. As expected, there is a negative relationship between the number of children and the educational level of the mother. As regards previous year employment status, the higher the educational attainment of the women, the higher the fraction of women who worked during the previous year. This fraction is very di§erent across motherhood status, those exhibiting the lowest fraction being the mothers of young children, and especially so if they only have elementary education. With respect to the spouseís working status, no substantial di§erences are observed across motherhood status. However, across educational groups, we observe that, the higher educational attainment, the higher the fraction of women whose spouse is working, reáecting marital sorting by education. Femaleís age is positively related with the age of the youngest child and mothers of children younger than 6 are older the higher their educational level, a consequence of the di§erences in maternity ages across educational groups. Finally, we observe a strong and positive correlation between education and participation and a negative correlation between motherhood and participation. More speciÖcally, the lowest participation rates are found in the group of mothers of children aged under 3 years. However, di§erences in participation across motherhood status are more striking the lower female educational attainment.

TABLE 1: Summary statistics

Note: Authorsícalculations using data from the Labour Population Survey (EPA) conducted by the Spanish Statistical O¢ ce (INE). Sample married females under age 45. Standard deviations for continuous variables in parentheses.

Table 2 shows the average annual growth rate of participation for a period of six years before the policy implementation (1997-2002) and for a period of two years after the policy implementation (2003-2004). We observe an increase in the participation rates of married females throughout the period under consideration. This is part of the trend observed in Spain since the beginning of the eighties. The increase in participation of married females is primarily due to the increase in the participation of mothers, as increases in non-mothersíparticipation are minor, except for the elementary educated group in the period 1997-2002. Focusing on mothers, those who have children aged 3 to 6 years present a similar evolution before and after the policy implementation: the annual average participation growth rate is around 2% in the before-policy and after-policy periods. However, the group targeted by the policy, mothers of children aged under 3 years, shows a very di§erent pattern: whereas the average annual participation growth rate is around 2% in the beforepolicy period, this growth rate reaches 5% in the after-policy period. Conditioning on educational attainment, the di§erences are more striking for those with elementary or high school education than for college educated mothers. In order to know if the di§erence between the average participation growth rate in the before and after policy periods is signiÖcant, we run a regression of the annual participation growth rate on a dummy variable controlling for the after policy period. We Önd the di§erence to be signiÖcant at 5% in those cases that we mark with an asterisk in Table 2. In particular, this is the case for mothers of children aged 3 to 6 years with elementary education and for mothers of children under 3 years, both for the whole group and for those with high school education⁷. This empirical evidence points to a potential positive e§ect of the policy on the participation rate of the targeted group of females. As a Önal remark, for mothers of children aged 3 to 6 years with elementary education, the average annual participation rate experiences an important increase in the afterpolicy period. This may be related to the fact that a fraction of these mothers, those whose youngest child is 3 years old in 2004, were policy-eligible in 2003. Thus, their participation in 2004 could be indirectly a§ected by the policy through their decision to participate in the previous period, when they were policy-eligible.

⁷This test should be taken with caution as we only have 8 observations to perform it. If panel data were available, a di§erent approach could be adopted as we could have used individual information. Unfortunately, this is not the case, and the test can only exploit aggregate information.

TABLE 2: Variation in Participation Across Years

Note: Authorsícalculations. The annual participation growth rate for all women may be higher than those observed for each educational group because the weights corresponding to each education group change from one year to another. We mark with an asterisk those cases in which the annual participation growth rate after the policy implementation is signiÖcally di§erent (at 5%) than in the before-policy period.

IV. Econometric speciÖcation

In this Section we formulate an econometric model in order to measure the e§ect of the change in policy on the labour force participation rate of those women a§ected by its implementation.

We estimate a probit model of the participation decision and we measure the potential e§ect of the policy based on the DDD estimator, broadly used in the literature of natural experiments or pseudo-experiments. This kind of analysis, which has a rich tradition in psychology, has been used in economics to study the e§ects of public policies on individual behaviour. The introduction of a policy change provides a natural experiment that allows the evaluation of the policy e§ect on an outcome of interest if we observe individual behaviour before and after its implementation. As there can be other factors, such as macroeconomic condition changes or any kind of trend that ináuence the outcome of interest, the usual approach consists of examining this outcome not only for those groups a§ected by the policy change, but also for those similar groups that are not eligible for the policy but would presumably be subject to the other ináuences as well. In particular, in the case we are analyzing two issues reinforce the importance of these considerations: (i) the cash beneÖt policy for working mothers is introduced simultaneously with increases in income tax deductions for children that beneÖt mothers in general; (ii) a decrease in the unemployment rate is observed in Spain over the period. The decrease in unemploy ment could a§ect participation of females in di§erent directions. On the one hand, it could encourage them to search for a job. On the other, the lower probability of their husbands being unemployed could induce them to take a spell out of the labour market.

In the natural experiments literature, the policy-eligible group of the population is called the treatment group, and a suitable subset of the non-eligible group is used as a control group. The comparison of the change in the behaviour of the treatment group relative to the change experienced by the control group provides what is known as the ëdi§erence-in-di§erencesí(DD) estimator, which can be interpreted as an estimator of the policy e§ect. However, as pointed out by Meyer (1995), it may well be the case that the outcome of interest systematically evolves di§erently for the treatment and control groups and hence, the omission of a speciÖc trend for the treatment group would bias the estimation of the policy e§ect. For this reason, we allow for di§erent trends in the two groups and we build a DDD estimator by comparing changes in the behaviour of the treatment group with changes for the control group correcting for their di§erent underlying trends. This identiÖcation strategy allows us to capture the e§ects of the policy. Meyer (1995) provides an excellent survey of this kind of methodology, which has been applied in a large variety of contexts⁸.

⁸See for example Card and Krueger (1994), Meyer, Viscusi and Durbin (1995), Cummins, Hassett and Hubbard (1994) and more recent works such as Stephens and Ward-Batts (2004).

In this paper, the public policy to be evaluated is the introduction of a cash beneÖt for working mothers of children aged under 3 years, who constitute the treatment group. In our analysis we consider as the control group mothers of children aged under 6 years that are non-eligible for the policy, i.e., mothers of children aged 3 to 6 years. We restrict the control group to include only mothers of children aged under 6 as they are expected to have a closer behaviour to the treatment group. The time and care demands of children younger than 6 are very di§erent from those of older children, given that compulsory education in Spain starts at 6 years. Furthermore, we eliminate from the control group those non-policy eligible mothers of children aged under 6 years who beneÖted from the policy in a previous period. In particular, these are mothers whose youngest child was 3 years old in 2004. The inclusion of these women in the control group could underestimate the direct e§ect of the policy on the participation of mothers of children aged under 3. This will be the case if there is an indirect e§ect of the policy such that the potentially higher participation of policy-eligible mothers translates into a higher participation once their children get over the age of 3 and they become non-policy-eligible. In fact, this could be an important positive e§ect of the policy, but one that we do not measure here.⁹

Let be a binary variable which takes on the value 1 for those married females who participate in the labour market and 0 otherwise.

\[\begin{array}{c} \operatorname * {P r} (y _ {i t} = 1 | A f t e r _ {t}, T r e a t m e n t _ {i}, x _ {i t}) = \\ \Phi (\alpha + \beta_ {1} A f t e r _ {t} + \beta_ {2} T r e a t m e n t _ {i} + \beta_ {3} A f t e r _ {t} * T r e a t m e n t _ {i} + \beta_ {4} t + \beta_ {5} t * T r e a t m e n t _ {i} + x _ {i t} ^ {\prime} \gamma) \end{array} \tag {1}\]

for and ; where i indexes individuals, t indexes time and stands for the cumulative normal distribution function. The variable is a dummy variable which equals 1 after the introduction of the policy, T reatment_i is a dummy variable which equals 1 for mothers of children aged under 3 years, i.e., the treatment group, and thus, the interaction Treatment can be interpreted as the ëpolicy variableí. The vector contains explanatory variables related to socioeconomic and family characteristics. We allow a di§erent linear time trend for the treatment and control groups so that we do not ascribe to the policy an e§ect that is in fact related to di§erences in the evolution of the behaviour between the treatment and the control group over time. Our identiÖcation assumption is that there are no other contemporaneous shocks, except for the policy under analysis, that could a§ect the relative behaviour of the treatment and control groups over time.¹⁰

⁹Furthermore, we recognize that even those females that are not mothers of children younger than 3 years could potentially be a§ected by the policy as they can decide to have a newborn. However, we think it is too early to Önd evidence of such e§ects. Thus, we have concentrated on the e§ect of the policy on participation decisions and postpone the analysis of its e§ect on fertiliy decisions for future work.

¹⁰Our speciÖcation in this regard is similar to the one in Francesconi and Van der Klaauw (2007).

The variables and T reatment_i, as well as the interaction variable, T reatment_i; have a useful interpretation in terms of labour participation for the treatment and control groups before and after the policy. The variable controls for di§erences in labour participation before and after the implementation of the policy for both groups. There are many possible changes in the socioeconomic environment that can a§ect female behaviour, such as changes in other dimensions of the family policy scheme, in the returns to labour market experience or in husbandís earnings or employment¹¹. In particular, in this case the simultaneous introduction of the cash beneÖt for working mothers and the increase in the income tax deductions for children justiÖes the inclusion of this variable in the model speciÖcation. The variable Treatment controls for systematic di§erences in the participation behaviour of the treatment and control groups. The interaction variable controls for the relative change in labour participation in the treatment group after the policy. As noted above, we include a time trend common to both groups and a speciÖc trend for the treatment group. The justiÖcation for considering a speciÖc trend for mothers of young children is based on evidence for other countries like the US where recent changes in female labour participation are concentrated among this group of mothers, see for example Attanasio et al. (2004).

Among the explanatory variables in , educational attainment is controlled by the binary variables College education; which equals 1 for females with college ed ucation and High School education; which equals 1 for females with high school education completed. Elementary educated females are the reference category. The age of the female in years is measured by the variable : We also include a quadratic term of this variable in order to allow for nonlinear e§ects. The presence of children in the household is represented by the variable Number of Children. Female employment status in the previous year is controlled by a binary variable, Work ₁; which equals 1 if the female worked.¹² As is argued by De la Rica and Ferrero (2003) and others, past employment is likely to a§ect current participation positively through di§erent channels; for example, females who have already been in the labour market do not have to incur the costs derived from entering it (such as the costs related to searching for a job). Finally, the employment status of the spouse is controlled by the binary variable Spouse s work, which equals 1 if he is working.

Our speciÖcation treats fertility as exogenous with respect to participation. However, it could be the case that decisions about fertility and participation are made simultaneously, suggesting the use of instruments to account for the endogeneity of fertility. The main problem that arises in this context is the di¢ culty of Önding appropriate instruments for the fertility in the participation equation, i.e. instruments that are correlated with fertility but not correlated with unobservable factors a§ecting participation. Studies that treat fertility as endogenous do not yield consistent results. For example, Carrasco (2001) and De la Rica and Ferrero (2003) Önd that the estimated e§ect of fertility on participation is downward biased when fertility is assumed to be exogenous, whereas Angrist and Evans (1998) and Angrist (2001) Önd an upward bias. Given these di¢culties, we have followed other papers in the literature that analyze female participation decisions assuming exogenous fertility.¹³ However, as reported later, our results do not appear to be sensitive to this choice.

¹¹See for example Olivetti (2003) and Attanasio et al. (2004, 2005) for the importance of these issues for the determination of female participation.

¹²The inclusion of the variable W ork ₁ could be enough to control for the indirect e§ect of the policy on participation of mothers of children aged over 3. However, this variable is an imperfect way of controlling the e§ect of previous labour market experience on the participation decision (total labour market experience is not available in the survey). For this reason, we choose to exclude from the control group those who are mothers of 3 years old children in 2004, who beneÖted from the policy in 2003.

Finally, there is a last methodological issue we should mention. We use pooled cross sections, as panel data are not available for the period of interest. Then, we cannot control for unobserved heterogeneity through the use of individual-specifc Öxed e§ects.

Construction of the DDD estimator

The nonlinearity of the model, which comes from the outcome of interest being a binary variable, makes the construction of the DDD estimator slightly di§erent from the linear case. The e§ect of the policy is not given by the estimate of a single parameter, as it would be in a linear model, but by di§erences in estimated probabilities. Let be the change in the participation probability experienced by the treatment group periods after the policy with respect to this probability one year before the policy. Let be the analogous change in the participation probability for the control group. Let be the change in the participation probability experienced by the treatment group one period before the policy with respect to this probability periods before the policy. Let _bef be the analogous change in the participation probability for the control group. The available data allows us to perform the analysis for ; 2: We build the DDD estimator as follows:

\[D D D = (\Delta p 1 \_ a f t - \Delta p 1 \_ b e f) - (\Delta p 0 \_ a f t - \Delta p 0 \_ b e f)\tag{2}\]

We perform the analysis considering ; which implies a comparison of changes in participation probabilities from 2002 to 2004 with respect to changes from 2000 to 2002 for both the treatment and the control group. Thus, we are comparing for both groups the evolution of the participation probabilities in periods of the same length.

¹³For example, Wellingtonës (2001) investigation of self-employment as a strategy to balance family and career and GutiÈrrez-DomÈnechís (2005) study of women¥s transitions from employment to nonemployment after Örst birth in several countries of the EU both assume exogenous fertility.

V. Estimation results

In this Section we present the results for the model speciÖcation that was explained above in order to isolate the e§ect of the cash beneÖt policy introduced in Spain in 2003. We consider data covering the years 1996 to 2004, i.e., seven years before and two years after the implementation of the policy.

The estimation results of equation (1) are shown in Table 3. All the explanatory variables we include are signiÖcant at the 5% level, except for variable Trend and the Age related variables, which are non-signiÖcant at the usual levels. As regards the variables related to the structural behaviour of females, the signs of the coe¢- cients are supported by microeconomic theory.¹⁴ First of all, the variable Number of Children has a negative e§ect on participation as expected. Furthermore, the variable controlling for children aged under 3 years, T reatment; has a negative sign coe¢cient, reáecting the negative e§ect of the presence of young children on womanís participation. Concerning the educational level, since elementary education is the reference category, the sign of the coe¢cients of the high school and college edu cation variables are the expected ones. That is to say, the higher the education the higher the participation probability. This reáects the fact that the opportunity cost of being at home is an increasing function of educational level. The variable controlling for previous year employment status, W ork ₁; has a positive e§ect on participation, along the lines of De la Rica and Ferrero (2003). The variable related to the employment status of the spouse, Spouse0s work, has a negative coe¢cient that can be related to the so-called ëadded worker e§ectí, that is, females married to unemployed males are more likely to participate in the labour market. The negative coe¢cient of the variable Trend Treatment reáects that the trend in participation is áatter for the treatment group. The dummy variable Af ter has a positive e§ect on female labour participation. This reáects the relatively higher participation rates of all women in the years after the policy implementation. Finally, we focus on the variable that provides us with some insights about the question we pose in this work. The ëpolicy variableí, Af ter T reatment; has a positive and signiÖcant e§ect on the participation decision. This means that there is a particular behaviour of the targeted group, mothers of children aged under 3 years, that manifests itself via a relatively higher increase in participation with respect to the increase for the control group after the policy implementation.

¹⁴See, for example, Becker (1965) and Browning (1992).

Note: Robust-heteroskedasticity standard errors in parentheses. Treatment group: mothers of children aged under 3 years. Control group: mothers of children aged 3 to 6 years, excluding those whose youngest child was 3 years old in 2004. Variables Trend, Age and are divided by 10 in the estimation.

From our probit estimates, we compute the estimated participation probabilities for the treatment and control groups and we obtain the DDD estimate. As mentioned above, in order to calculate the DDD we compare for both groups the change in the participation probabilities from 2002 to 2004 with the change from 2000 to 2002. The data from 1996 to 1999 that are also included in our sample only serve the purpose of estimating the time trend. The results can be seen in Table 4. The change in the estimated participation probability for the control group in the two years before the policy implementation, ; is below 1 percentage point, while this change in the period 2002-2004, ; is around 2 percentage points. For the treatment group, whereas participation falls in the period 2000-2002, represents a reduction of 1 percentage point), the change in the period 2002-2004, af t; is around 4 percentage points. These results give an estimate of the DDD of around 3 percentage points, an increase which we ascribed to the family policy. This variation represents 5% of the average labour market participation of the treatment group in 2002, the year before the policy implementation. Thus, since the cash beneÖt represents 39% of the average child care cost, the implied elasticity of labour market participation of the treatment group with respect to child care prices is -0.07, assuming that there is no e§ect of the cash-beneÖt on the equilibrium price of child care services. This number is in the range estimated by Anderson and Levine (1999).¹⁵

It is interesting to compare the estimated e§ect of the policy in Spain with the e§ect of similar policies in other countries. Francesconi and van der Klaaw (2007) found that an increase in the generousity of "in work" beneÖts for low income families with children in the United Kingdom in 1999 led to an increase in the single mothers employment rate of about 5 percentage points. Although the magnitude of the credit increase was not homogeneous across mothers,¹⁶ the reform implied an average beneÖt increase of 8 pounds per week, representing around 13% of the weekly child care cost, generating a response substantially higher than in the policy we analyze in this paper. We can think of two possible explanations. First, according to several studies,¹⁷ a higher e§ect of this kind of policies is expected among single mothers than among married mothers. Second, the targeted group in Francesconi and van der Klaaw (2007) is low income families only, for which participation is more likely to be responsive.¹⁸

TABLE 4: Estimated changes in participation probabilities and DDD estimate

Note: Standard errors in parentheses computed through the Delta method

The descriptive analysis in Section III shows important di§erences in participation probabilities across educational levels so we want to explore the potential heterogenous e§ect of the policy across education groups. We show DDD estimates by educational attainment in Table 5.¹⁹ We cannot reject the null hypothesis of no policy e§ect for elementary and college education groups. For high school educated females we Önd a positive and signiÖcant e§ect of the policy. The DDD estimate is 3.06 percentage points, which represents an increase of the labor participation of the treatment group in 2002 of around 6%. The overall e§ect that we report in Table 4 is dominated by the e§ect of the policy for high school educated females.

¹⁵They found that for mothers of children aged under 13, the elasticity of labor market participation with respect to the market price of child care services ranges between -0.05 and -0.35.

¹⁶The reason is that it a§ected several parameters of the program at the same time: the amount of weekly credit to which the family was entitled depending on the age of the children and the income threshold to be entitled to the policy, among other issues.

¹⁷See, for example, Blau and Robbins (1988), Connelly (1992) or Connelly and Kimmel (2000).

¹⁸Unfortunately, we can not undertake the analysis conditioning on household income as the Spanish Labour Force Survey does not include such information.

TABLE 5: DDD estimates by education level Note: Standard errors in parentheses computed through the Delta method.

As a check on the sensitivity of our Öndings, we have also performed the analysis without excluding from the control group those women in the sample whose youngest child was 3 years old in 2004 (i.e, women that were policy-eligible in 2003). We Önd that the estimated DDD is very similar with this extended sample, around 5% of the average labour market participation of the treatment group in 2002, the year before the policy implementation.²⁰ The indirect e§ect of the policy on the group of women that are not currently policy-eligible could be partially controlled by the variable Work ₁. However, we think it would be necessary to have more after-policy observations to further investigate this indirect e§ect of the policy.

As stated above, the potential endogeneity of the fertility decision could a§ect the estimation results. In spite of the di¢ culties in Önding appropriate instruments for the fertility decisions that was explained in Section IV, as a sensitivity check, we follow the approach suggested by Manski et al. (1992) to estimate a bivariate probit model for the participation and fertility decisions.²¹ They propose a bivariate model in which the equation of interest and the equation for the variable for which the endogeneity is being treated are estimated jointly and apply this strategy to analyze the probability of high school graduation as a function of the family structure. In our bivariate probit model, we have considered the number of children in the household apart from the newborn and the interaction of education and age variables as instruments. We Önd that the potential positive e§ect of the policy is robust to this alternative speciÖcation and indeed a bit larger than if exogenous fertility is assumed. In any case, this result should be taken with caution given the di¢culties we mention above.²²

¹⁹These numbers are based on separate labor force participation equations estimated for the di§erent education groups.

²⁰ Estimation results are available upon request.

²¹ Variables related to the innate ability of females to reproduce, such as the number of brothers and sisters of the female (as a proxy for her genetic ability to reproduce), or variables related to religion could be used as instruments. However, these variables are not usually reported in the Labour Force Surveys.

VI. Conclusions

The objective of this work is to analyze the e§ects of a public policy introduced in Spain in 2003 aimed at reconciling family and work. The policy provides working mothers of children aged under 3 years with a monthly cash beneÖt of 100 euros per child. Other studies have investigated the e§ect of similar policies in Norway (see Schone (2004) and Naz (2004)) and United Kingdom (see Francesconi and Van der Klaauw (2007)). Our study is a contribution to that literature, with the additional interest of it concerning one of the countries with the lowest labour participation of mothers, as reported by GutiÈrrez-DomÈnech (2005).

The methodological approach we use in this work is based on the recent natural experiments literature, which has been broadly used in the evaluation of public policies in a wide variety of contexts. We formulate an econometric model of labour participation and measure the potential e§ect of the policy by means of a DDD estimator.

Our estimation results point to a small positive e§ect of the policy in increasing labour participation of the targeted group, mothers of children aged under 3 years. After the policy was implemented, this group of mothers increases its labour market participation 3 percentage points more than the non-policy-eligible females. The largest point estimate and the largest implied proportional e§ect is that for women with a high school education, the only group for which the estimated e§ect is statistically signiÖcant.

We conclude from our analysis that the observed growth in participation of mothers of children aged under 3 years could be partially ascribed to the cash-beneÖt policy that came into force in 2003. Our results have implications for governments seeking to design family policies aimed at increasing labour market participation of mothers of young children. As a future research project it will be interesting to measure the indirect e§ect of the policy on the participation of mothers of children aged over 3 resulting from their stronger attachment to the labour market during the early maternity ages. To perform such an analysis it will be necessary to observe policy-eligible women for several years after the period in which they beneÖt from the policy.

²² Estimation results of the bivariate model are available upon request.

References

1. Anderson, P.M. and Levine, P.B. (1999): Child Care and mothersí employment decisions. NBER WP 7058.
2. Angrist, J.D. and Evans, W.N. (1998): Children and their parents labour supply: evidence from exogenous variation in family size, American Economic Review, Vol. 88: 450-477.
3. Angrist (2001): Estimation of limited dependent variable models with dummy endogenous regressors: simple strategies for empirical practice, Journal of Business and Economic Statistics, Vol. 19: 2-16.
4. Attanasio, O., Low, H. and S·nchez-Marcos, V. (2004). Explaining Changes in Female labour Supply in a Life-Cycle Model, University of Cambridge Working Papers 2004, 04/51.
5. Attanasio, O., Low, H. and S·nchez-Marcos, V. (2005). Female labour supply as insurance against idiosyncratic risk, Journal of the European Economic Association P&P, Vol. 3(2-3): 755-764.
6. Becker, G. (1965). A theory of the allocation of time, Economic Journal, Vol. 75: 493-517.
7. Blau, D. and Robins, P. (1988). Child-care cost and family labour supply, The Review of Economics and Statistics, Vol. 60: 374-381.
8. Browning, M. (1992). Children and household economic behaviour, Journal of Economic Literature, Vol. 30: 1434-1475.
9. Card, D. and Krueger, A. B. (1994). Minimum wages and employment: a case study of the fast food industry in New Jersey and Pennsylvania, American Economic Review, Vol. 84: 772-793.
10. Carrasco, R. (2001). Binary choice with binary endogenous regressors in panel data: estimating the e§ect of fertility on female labour participation, Journal of Business and Economic Statistics, Vol. 19: 385-394.
11. Connelly, R. (1992), The E§ect of Child Care Costs on Married Womenís labour Force Participation, Review of Economics and Statistics, Vol. 74 (1): 83-90.
12. Connelly, R. and Kimmel, J. (2000), Marital Status and Full-time/Part-time Work Status in Child Care Choices, Applied Economics, Vol 35 (7): 761-77.
13. Consumer (2005): Siguen faltando plazas en las guarderÌas p˙blicas, Revista Consumer Eroski, no. 91.
14. Cummins, J. G., Hassett, K. A. and Hubbard, R. G. (1994). A Reconsideration of Investment behaviour Using Tax Reforms as Natural Experiments, Brookings Papers on Economic Activity, Vol. 2: 1-59.
15. De la Rica, S. and Ferrero, D. (2003). The e§ect of fertility on labour force participation: The Spanish evidence, Spanish Economic Review, Vol. 5: 153-172.
16. Del Boca, D. and Pasqua, S. (2002), Labour market participation of mothers in Italy: facts, studies and public policies, Working Paper CHILD 04-2002.
17. Eurostat (2007): Living Conditions in Europe, 2002-2005.
18. Ferrero, D and Iza, A. (2004). Skill premium e§ects on fertility and female labour force supply, Journal of Population Economics, Vol. 17: 1-16.
19. Francesconi, M. and Van der Klaauw, W. (2007): The socioeconomic consequences of "in-work" beneÖt reform for British lone mothers, Journal of Human Resources, Vol. 42: 1-31.
20. GutiÈrrez-Domenech, M. (2005): Employment after motherhood: a European Comparison, Labour Economics, Vol. 12: 99-123.
21. Instituto de PolÌtica Familiar (2006), EvoluciÛn de la familia en Europa.
22. Manski, C., Sandefur, G. D., McLanahan, S. and Powers, D. (1992). Alternative estimates of the e§ect of family structure during adolescence on high school Graduation, Journal of the American Statistical Association, Vol. 87: 25-37.
23. Meyer, B. D. (1995). Natural and quasi-experiments in economics, Journal of Business and Economic Statistics, Vol. 13: 151-161.
24. Meyer, B. D., Viscusi, W. K. and Durbin, D. L. (1995). Workersícompensation and injury Duration: Evidence from a Natural Experiment, American Economic Review, Vol. 85: 322-340.
25. Naz, G. (2004). The impact of cash-beneÖt reform on parentsílabour force participation, Journal of Population Economics, Vol. 17: 369-383.
26. OECD (2002), Women at work: who are they and how are they faring?, OECD Employment Outlook 2002.
27. OECD (2001), Balancing work and family life: helping parents into paid employment,
28. OECD Employment Outlook 2001.
29. Olivetti, C. (2006), Changes in Womenís Aggregate Hours of Work: The Role of Returns to Experience. Review of Economic Dynamics, Vol. 9 (4): 557-587.
30. Sch¯ne, P. (2004). Labour supply e§ects of a cash-for-care subsidy, Journal of Population Economics, Vol. 17: 703-727.
31. Spanish Statistical O¢ce (2002): Encuesta de Estructura Salarial.
32. Spanish Statistical O¢ce (2004): Encuesta de PoblaciÛn Activa, 1996-2004.
33. Stephens, M. and Ward-Batts, J. (2004), The impact of separate taxation on the intra-household allocation of assets: evidence from the UK, Journal of Public Economics, Vol. 88 (9-10): 1989-2007.
34. Wellington, A.J. (2001), Self-employment: the new solution for balancing family and career?, Labour Economics 13: 357-386.
35. Yelowitz, A.S. (1994). The Medicaid Notch, labour Supply and Welfare Participation: Evidence from Elegibility Expansions, UCLA Economics Online Papers, No. 77.