Introduction

In the last two decades, a new feature has received increasing attention in the countries included in the Organization for Economic Cooperation and Development (OECD). It is the so-called motherhood wage penalty; that is to say, the wage gap between women with children and those without children. This wage penalty can be explained as part of the effects felt by the interruption of a woman’s employment to have children, and the resulting responsibilities she assumes in raising them.Footnote 1 A number of recent articles have shown the existence of this motherhood penalty in different countries such as the U. S. (Anderson et al. 2002, 2003; Budig and England 2001; Edwards 2005; Korenman and Neumark 1992; Lundberg and Rose 2000; Waldfogel 1997, 1998a, 1998b), the UK (Joshi et al. 1999; Waldfogel 1995, 1998a), Canada (Phipps et al. 2001), and Germany (Kunze and Ejrnaes 2004); although in other economies, basically those of Northern Europe such as Sweden (Albrecht et al. 1999) and Denmark (Datta-Gupta and Smith 2002; Nielsen et al. 2004; Rosholm and Smith 1996), studies have found no evidence for this.

Against this background, a series of cross-country studies, such as those by Harkness and Waldfogel (2003), Sigle-Rushton and Waldfogel (2007a, 2007b), and Todd (2001) present comparable international evidence showing a great deal of variation. Overall, it is observed that the United Kingdom and Australia are the countries that show the greatest losses in wages by mothers, with the average being more than 10% for women who have two or more children, followed by the US and Canada. In Germany, the penalty is around 8%, in the Netherlands less than 5%, whereas the wage penalty is almost imperceptible in Sweden, Norway and Finland.Footnote 2 These authors have related these results to the classical division of welfare states in industrialized countries. Thus, the liberal Anglo-American countries show the largest motherhood wage penalties, followed by the Continental conservative countries, with the Nordic social democratic countries exhibiting no penalty at all.

More recently, the sociological literature has questioned the reduced typologies of welfare state regimes and suggested the need to include additional possibilities. One is the Mediterranean (or Latin) rim, which reflects traits from both the liberal and conservative models, as explained below. In this context, our research uses data from Spain, which has rarely been considered in the literature, in order to provide evidence and explanations which help us to describe the motherhood wage penalty in this representative Mediterranean country.

In recent years, Spain has witnessed profound changes in the social structure of the labor market. These include an increase in the education level of women, with a resulting increase in the number of women participating in the workforce and, at the same time, a decrease in the birth rate. Despite the lack of certainty that the traditional wage gap between men and women has been closing (de la Rica et al. 2005; López et al. 2001), we present here the first single-country study providing specific empirical evidence of the motherhood wage penalty for Spanish women. Our method took into account, not only observed characteristics, but also unobserved individual heterogeneity. To this end, a stepwise procedure using panel data was followed in order to control for a series of sets of determinants of this wage penalty and to assess the relative influence of each of these sets on the degree of the wage penalty.

Specifically, while many observed characteristics, related to individuals, the family, human capital and the workplace, derived from economic rationality, are explicitly considered, the effects of unobserved individual heterogeneity are dealt with through the inclusion of time-invariant individual fixed effects. In this way, and once all these factors are controlled for, the gap in pay observed between mothers and non-mothers can only be attributed to discrimination (or, for example, to time-varying unobserved characteristics; such as, the degree of effort exerted in the workplace). A better understanding of the explanatory factors is needed in order to identify the relevant social policy measures that would combat potential discrimination against mothers.

The estimation results show, first, that unobserved heterogeneity plays an important role in disentangling the lack of evidence on the motherhood wage penalty observed in the raw data. A double-sided interpretation can be put forward: Either mothers look for jobs that are better paid, or women who are better paid decide to become mothers (inverse causality). Secondly, as expected according to the discussion presented in the next section, the motherhood wage penalty in Spain is at least as large as in the conservative countries, even reaching the levels observed in liberal countries, in which social security is not as extensive as in the other types of welfare state regimes.

The following section is dedicated to a brief background on both the characterization of the different welfare state regimes and the explanations for the existence of a motherhood wage penalty. The section following presents the database and describes some statistics of the sample. We then show the stochastic formulation and the empirical results. The paper ends with a summary of the most important conclusions.

Background

Esping-Andersen (1990) typically considered three regime models: Anglo-Saxon, Continental European and Nordic European. These are associated, respectively, with liberal, conservative and social democratic countries. As a first approximation, in the Anglo-Saxon model, in which a fairly high reliance on means-tested public assistance programs exists, individualism and the free market determine the provision of care and personal services. In contrast, the Continental countries, shaped both by the Catholic legacy and statism, and the Nordic countries, exhibit universal insurance programs, the latter providing for individualized, citizenship-based entitlements.

Focusing on the aspects related to gender differences, the Nordic countries have developed strong family and equal opportunity policies, instrumented by generous maternity leave provisions and extensive child care. In Anglo-American countries the opposite is true, where there are poorly-developed family leave and child-care policies. The Continental countries fall in the middle, with generous maternity leave provisions, but less extensive child-care coverage. These differences are reflected in the fact that, in liberal countries, such as Ireland and the U. K., the state favors the functioning of the market, such that women are encouraged to participate in the labor force (particularly, in the service sector; see Arts and Gelissen 2002), resulting in high levels of employment which largely take the form of part-time work (Trifiletti 1999). However, these female workers are insufficiently protected by the state, with this being reflected in a wide gender pay gap, and an even greater gap between mothers and non-mothers. Other international evidence on gender wage differences appears in Firestone et al. (1999), Swanberg (2005) and Urban and Olson (2005). In the Scandinavian countries, women, irrespective of whether they are mothers or not, are also encouraged to participate in the labor market, especially in the public sector, so that few differences in employment rates between mothers and non-mothers are observed, and, simultaneously, the gender pay gap is the narrowest. Finally, in the conservative countries, labor market participation by married women is strongly discouraged with a redistributive policy of allocating resources to families with children. As a consequence, significant differences in employment rates exist between men and women, and between mothers and non-mothers, as do differences in pay between genders.

The contributions of Harkness and Waldfogel (2003), Sigle-Rushton and Waldfogel (2007a, 2007b), and Todd (2001) have been to extend the characterization of the three regime types by considering the motherhood wage penalty across countries. Thus, the largest penalties are observed in the liberal countries, followed by the conservative, and finally, the social democratic countries, in which negligible gaps are found. However, recent contributions of Arts and Gelissen (2002), Korpi (2000), and others have suggested additional new typologies, with one being the so-called Mediterranean model. The countries included in this regime type; such as, Italy, Greece, Portugal and Spain, are characterized by relatively little state intervention in the welfare sphere. This is reflected in the lack of an articulated social minimum and right to welfare, with some benefit levels being very generous (e.g., old age pensions), and a highly fragmented social security system, with health care institutionalized as a right of citizenship (Arts and Gelissen 2002). Other important features in the characterization of the Mediterranean model are the gender division of paid and unpaid work, and the fact that social care tends to be privatized within the family.

These latter characteristics are not considered by some authors to be sufficient to differentiate Mediterranean countries from the Continental model. For example, Katrugalos (1996, p. 43) classifies them as “immature, underdeveloped species of the Continental model.” Others (see e.g., Trifiletti 1999), highlight two features that distinguish Mediterranean countries as a different regime type from the Continental. First, the subsidiary role of the state typical of the Continental countries is modified, since in Mediterranean countries, it only covers those social risks which the family cannot protect itself against. Second, whereas in Continental countries most benefits devoted to the family are instrumented through the male breadwinner irrespective of whether the wife works, in Mediterranean countries such benefits are allocated only to families in which the wife is in the labor market. Thus, the traditional family role of the non-working mother is not especially protected.

Along these lines, Trifiletti (1999) constructed a typology of welfare states across two dimensions, one describing whether women are treated by the state as wives and mothers or as workers, and a second shaping whether or not the state protects women within the labor market. The Mediterranean welfare state regime type is then characterized both by considering women fundamentally in their family role, wives and mothers, and by not protecting them within the labor market. Therefore, this Mediterranean type combines traits from both the Continental conservative and the liberal models. Specifically, in Continental countries, policy measures treat women as wives and mothers; whereas, in liberal countries women are not protected within the labor market. As a counterbalance to this, in the Continental model, women are effectively protected in the labor market through the social benefits received via their husbands; whereas, in liberal countries women are essentially considered as workers (for more on this characterization, see Trifiletti 1999). In these circumstances, a reduced proportion of women work, but they work mainly full time and only receive benefits and access to social services through their status as workers. This is the main way in which they are protected within the labor market. This situation can be attributed to limited aid from the welfare regime, the existence of help resources from extended family networks, and the necessity to work within a labor market whose conditions favor men.

Having characterized the different welfare state regime types, we now discuss the economic rationality for the existence of a motherhood wage penalty. The most common explanation for mothers earning less than non-mothers is that the loss of individual skills, as well as the depreciation of experience, is associated with the period spent out-of-work resulting from childbearing and child caring. This leads to a reduction in productivity related to the level of education attained by the mother (Mincer and Polachek 1974; Ruhm 1998; Waldfogel 1998a).

The earlier applied studies on this matter emerged from the existence of wage gaps between married and single women (Becker 1985). From this initial research, the first rigorous econometric analysis of the wage penalty comes from Korenman and Neumark (1992), who, using data from the United States, did not generally find significant effects on wages from having a first child, although there were effects in the case of a second child. Subsequently, Waldfogel (1998b), also in the United States, found that having a first child did have an effect on wages, and that this increased with the arrival of the second child. She also found that the existence of maternity leave, which covers the out-of-work period and maintains seniority/tenure, helped to reduce such a penalty. This prompted us to consider alternative explanations for the existence of the motherhood wage penalty.

The first of these has to do with the fatigue experienced by a woman who cares for her children at home, leading to less effort being dedicated to her job activity (VandenHeuvel 1997). The greater effort dedicated to home activities decreased as the child grew older, and increased as a higher level of education was required at work (Becker 1985). Another important factor is that women show a preference for jobs that allow them to combine household schedules with their work schedule, in exchange for a lower wage. For example, part-time employment was associated with a large wage penalty in the U. K. (Waldfogel 1995). In this sense even when human capital-related variables were controlled for, confirmed the existence of penalty evidence in the United States (Anderson et al. 2003; Budig and England 2001). This, they argued, is not due to less work effort as a consequence of maternity, but rather to the choice of a flexible work schedule (an evaluation on the satisfaction of spouses with their leisure time in a set of EU countries can be seen in García et al. 2007).

Additional explanations are related to discrimination. On the one hand, statistical discrimination may explain why firms assume that all women will interrupt their working career at some time, although they may not subsequently have children, in such a way that firms tend to place them in jobs that have a lesser human capital requirement. These positions require less training, and consequently pay lower wages (Becker 1991). On the other hand, there is a pervasive wage discrimination against women, which can be interpreted in two ways. The first interpretation relies on a more traditional explanation, according to which women subordinate their professional careers to those of their husbands, accepting lower paid jobs (Anderson et al. 2003). The second refers to wage discrimination against mothers. This source of wage differentials is difficult to measure and can only be approximated as the residual in wage equations (Oaxaca and Ransom 1994).

Finally, in pursuing the research question of our work, that is to say, to test the existence of a wage penalty for Spanish working mothers and to assess the impact of its determinants, we used panel data in order to estimate econometric models, such as fixed effects and/or instrumental variables, which allowed us to control for unobserved individual heterogeneity, thus providing more efficient and robust estimates (Amuedo-Dorantes and Kimmel 2005; Anderson et al. 2002; Datta-Gupta and Smith 2002).

Data

The Spanish data used in this paper come from the panel formed by the eight waves of the European Community Household Panel (ECHP) which correspond to the years 1994–2001. We here briefly outline Spanish social policies regarding maternity. Specifically, the rights recognized for maternity and child-care in Spain are fundamentally regulated by articles 46 and 48 of the Workers’ Statute. Maternity leave gives the right to 16 weeks away from work, immediately before or after the birth, paid for by the National Health System in the amount of 100% of the last wage. Subsequent to this leave, the mother is entitled to return to her job, maintaining her seniority. This leave can be shared with the father, although with certain restrictions: The first 6 weeks are exclusively for the mother and the rest of the time can be shared, although not simultaneously. An exception is made for cases of international adoption, only when the mother expressly gives up part of her leave. In addition, it is possible for either the mother or the father to take an optional leave for child-care, available from the end of the maternity leave to a maximum of 36 months. Although this is not paid, during this optional leave the same job is reserved during the first year and within the same professional group thereafter. There also exists a paternity leave that gives the father two additional weeks. The period of maternity leave in Spain is around the average for Europe, although the percentage of payment is among the highest. The leave for child-care in other European countries is often paid, although it is of much shorter duration. On the other hand, publicly-financed kindergartens are rare in the 0–3-year-old range, but quite common in the 3–5-year-old range. Higher education, including university, is fully subsidized.

Given our objective, women between the ages of 24 and 45 who had a paid job, either salaried or self-employed, were selected for the sample. Those who were in job training, or who worked in family businesses, were excluded. Women under 24 were ruled out, to avoid correlations between their education level and their hours of work, as were women older than 45, to avoid those who had no children who were minors at home, but could have had them in the past. For each of these waves, an average of 1,400 sample observations were used, while each female worker was observed, on average, more than three times.

Table 1 shows the wage gap by expressing wages and earnings of women (mothers, non-mothers, and all combined) as a fraction of men’s wages and earnings for the selected individuals in the most recent sample year, 2001. The first noticeable result is that women received 77% of the earnings of men. When we focus on hourly wages, and ignore the fact that women work, on average, fewer hours than men, the wage gap was reduced to 10%. We also confirm that this gender gap exists, whether or not the woman works in the private or the public sector, or whether or not she has a full-time or part-time contract. Only in the case of single women, do they receive slightly higher wages than single men.

Table 1 Women’s earnings and wages expressed as a percentage of men’s for different categories (ECHP-2001)
Full size table

When looking at the differences between mothers and non-mothers in hourly wages, it appears initially that there is a motherhood wage advantage in the aggregate, and that in several kinds of work (public sector, permanent contract and salaried work) mothers made a larger fraction of men’s wages than did childless women. However, in several categories, such as among full-time workers, temporary contract workers and single mothers, there is evidence of a motherhood wage penalty (see Harkness and Waldfogel 2003, for a similar evidence in several OECD).

This requires a careful treatment of all the characteristics influencing wages, in order to investigate the extent to which wage differences between women who are mothers and those who are not, can be explained by economic factors. As a consequence, controlling for demographic, human capital, family and job-related variables, provided a more refined measurement of the motherhood wage penalty. Thus, information about age, educational level, experience and seniority at work, size and resources of the family, the type of contract, sector of activity, and firm size, was necessary to determine if differences in pay could be attributed to the characteristics studied. To the extent that panel data were available and time-invariant unobserved characteristics; such as, motivation, desire for a professional career, the wish to be a mother, the interest in completion of tasks and exerting the appropriate effort, could also be controlled for, we could conclude that the family penalty finally observed could be interpreted as a proxy for discrimination against mothers, or some time-varying unobserved characteristics.

Therefore, the need arises for a rigorous regression analysis in order to calculate the effects on the wages of working mothers, controlling for both observed variables and non-observed individual characteristics. In the next section, we carried out a pool estimation, and we then took advantage of the panel structure to conduct a fixed-effects estimation. Prior to this, we devote some space to enumerating the variables taken into account in the analysis, and presenting some brief descriptive statistics in Table 2 for the last period of the sample, 2001.

Table 2 Definition of variables and characteristics of sample group (ECHP-2001)
Full size table

The dependent variable in the wage equation, Wage, is expressed by the logarithm of real wage per hour.Footnote 3 Regarding the exogenous variables, we included the bearing of children, as well as several socio-demographic and other economic characteristics. In gathering the facts regarding having children, we considered two possible measures. The first, Birth, records if there had been a birth in the household during the current year. The second, Number of Children, is a set of qualitative variables that indicate whether the woman had no children, one, two, three or more. We chose these two measures since the period available only covered 8 years and, therefore, there could have been women in the sample with children, but who had not had a child during the period under study. This means that the effect on wages of having children could have been felt some time ago and would not be included in the estimates of the Birth variable.

Additionally, we considered a broad group of explanatory variables with triple justification. First, to increase the reliability and strength of the estimations, we included as many controls as possible so that the variables, Number of Children and Birth, were effectively capturing the influence of motherhood. The second reason was to be able to provide some explanation of why the existence of a motherhood wage penalty in Spain is masked by the rough data presented in Table 1, as demonstrated below. Accordingly, including a large set of regressors helped us to identify which individual and job characteristics of women counteract the negative effect on wages of having a child. The third aim was to reduce the influence of any bias in the selection of the sample women, in terms of their participation in the labor market. It is worth mentioning that the correction of the selection bias in a framework of panel data is currently subject to some debate. In spite of the existence of interesting proposals (see Kyriazidou 1997 or Wooldridge 1995), there is no unanimity regarding the most suitable approach for its application (see Dustmann and Rochina-Barrachina 2007; Jensen et al. 2002; Wooldridge 2002).

Therefore, in our study, the variables finally included are the following: with respect to the social-demographic variables, some are considered to be strictly individual (Age, Marital Status), others are related to human capital (Education, Experience, Seniority, Over-education), others are related to work at home (Child-care, Elder-care), and still others are regional or time factors (Region, Cycle). Finally, with respect to the economic variables, we included the resources available to the family (Wage Income, Non-wage Income, Family Income, Household Size), as well as others linked to the woman’s job (Part Time, Sector, Occupation, Activity, Company Size, Contract Type, Self-employment).

Beginning with the dependent variable, Table 2 shows that the average wage of mothers was slightly higher, compared to women who had not had children. We confirmed that 5.5% of the women had a child during the current year; 26% had one child, 14% had two and only 2% had three or more; that is to say, almost 60% of working women sampled were childless. Sixty-five percent lived with a partner. Regarding education level, 28% completed only primary school, 21% completed only secondary school and 51% completed university. The women in the sample had an average of 13 years of work experience and almost 6 years in their current position. Nine percent worked part-time, 73% worked in the private sector, 72% had a permanent contract, and 9% were self-employed. Sixty percent of the female workers said their job required less education than they actually had. The average number of hours devoted to child-care was 14.2 per week, while the care of other adults took less than 1 hour a week.

When comparing average values between mothers and non-mothers, the main differences observed in Table 2 are related to demographic and human capital variables. Thus, mothers were, on average, 3 years older than non-mothers, most of them lived with a partner, and had almost 4 years more work experience and 18 months more seniority. If age, experience, and tenure are expected to be remunerated, then mothers should receive higher wages than non-mothers, which partially explains the average higher wage of the former. Additional factors explaining higher hourly wages for mothers are that they were more likely to be enrolled in the public sector, with permanent and part-time contracts (which pay higher hourly wages), in skilled or manual jobs, or in the service sectors. Furthermore, non-wage income of mothers was higher than that of non-mothers (probably due to transfers from the government). By contrast, non-mothers had a higher level of education, since half of them had a university degree, and worked two hours more per week than mothers. Finally, non-mothers obtained higher family income, their household size was smaller, and they devoted fewer hours to childcare.

Model Specification and Empirical Results

Empirical Specification

Using the traditional model of human capital (Mincer 1974), the wage equation for women takes the following form:

$$\ln\omega_{it}=\mu+\beta z_{it}+\alpha_{i}+u_{it}\quad\quad{\text{i}}=1,\ldots,{\text{N}}\quad{\text t} = 1,\ldots,{\text{T}}$$
(1)

where the vector of parameters β corresponds to the set of explanatory variables z it . μ and α i are constant terms, with the first representing the average population, and the second, the individual deviation from this average. When pool estimation is carried out, α i is assumed to be zero; whereas, in the panel estimation, α i is specific to each woman in the sample. Finally, u it represents the error terms assumed to be independent, with a null average and a constant variance.

Estimation Procedure

After an initial estimation of the regression Eq. 1 in pool form by Ordinary Least Squares (OLS), we used the panel structure of available information to carry out a second set of regressions using an estimator of fixed-effects. We chose this particular method since the relevant variables, whose coefficients we wished to estimate in a consistent way, correspond to the number of children of the working woman. The fact of having children, as well as their number, can be related to individual unobserved characteristics of the female worker (i.e., preference for a professional career, a desire to be a mother), and thus we must consider an estimator of fixed-effects, which gives consistent estimations, even under the hypothesis of correlation between individual and unobserved effects.

In both pool and panel estimations, we proposed a sequential process, progressively including new explanatory variables. We began by considering, in addition to the variables related to having children, a series of fixed time and regional effects, as well as the Marital Status of the female worker. This would be only a rough measure of the wage penalty, since only demographic variables were included. The results from this estimation capture the information coming from the raw data shown in Tables 1 and 2. In the second stage of the estimation, the set of variables related to human capital was added (Education, Experience, Seniority, Over-education and Age). Note that the variables Age and Experience were introduced in squared terms to allow for the possibility of decreasing returns over time. In this way, we took into account the possibility of differences in education and in experience between mothers and non-mothers. In the third phase, additional factors were included which gave information about the job characteristics of the position held by the women surveyed (Occupation, Activity, Company Size, Part Time, Contract Type, Sector, Self-employment). Thus, the possibility of mothers choosing more appropriate jobs was explicitly controlled for. We added to the set of factors variables that attempted to collect characteristics related to resources, both monetary and human, in the home (Child-care, Elder-care, Household Size, Non-wage Income, Family Income). Thus, we considered that households could differ between mothers and non-mothers because of their composition, and the resources available to them, which may indicate how decisions made by women about the hours they work, or the wage they earn in a specific job, could be influenced by these resource-related variables.

It should be noted that in the available statistical data base there was no way of knowing if the female worker who had given birth had taken maternity leave or not, nor did we know the amount of time she had taken, or whether she had extended it. Therefore, this possibility was not considered in our study. Nevertheless, given the existing legislation in Spain (a period of 16 weeks absence from work compensated at a rate of 100% of her previous wage), the usual case is that mothers use up their maternity leave, especially if they work in the public sector, or have a permanent contract. This situation could be different in the case of self-employed women or those who work with a temporary or a fixed-term contract, but there was no available data to test this possibility.

Empirical Results

The results from the pool estimation related to the variables of special interest to this study, Number of Children and Birth, are shown in Table 3, while the other exogenous variables appear in Table 5 of the Appendix.Footnote 4 Significant evidence of wage penalty was only obtained when controls related to human capital and the workplace were included (see column 3), quantifying the penalty as 3% for two children. When the Birth variable was used, the wage penalty was somewhat higher, at 6%. With respect to the other exogenous variables (see Appendix Table 5), we confirm in this pool estimation that the fixed regional effects, as well as the Marital Status variable, was not significant. Wage premiums existed for higher education, broader experience, and greater seniority. Occupation, as well as the sector of Activity and Company Size, were significant variables in determining the wage per hour of the female worker. Also significant were the Part Time, Contract Type and Sector variables. Regarding the variables dealing with resources, only a greater Non-wage Income had a negative effect on the wage level, while the other variables were not significant.

Table 3 Pool estimation
Full size table

Based on these results, it seems that weak evidence for the motherhood wage penalty was found. Raw data displaying the wage premium for mothers have been imperceptibly modified by the pool estimation, which shows slightly higher wages for non-mothers, after controlling for human capital and job related characteristics. Compared with the international literature, the case of Spanish working mothers would be very close to that of the Scandinavian countries. However, these initial results must be treated with caution, since we did not include a relevant factor related to both the wages of working women, and the fact of having children, that is to say, the existence of unobserved characteristics. As stated earlier, this can be a source of an observed wage penalty and, consequently, it must be controlled for. In this sense, the ECHP is extremely useful since it permits us to control the influence of these variables.

The results of the panel estimation of fixed-effects, presented in Table 4, show evidence of a motherhood wage penalty in Spain. In directly comparing Tables 3 and 4, we observe that the estimated coefficient of the variable of children was always more negative and more significant when it was estimated with panel data, with this being a first important finding in our study. The empirical evidence of fixed-effects, given that the Lagrange Multiplier (LM) rejected the hypothesis that all individual effects were equal and therefore a strong individual heterogeneity existed, indicates to us that, in fact, the wage penalty does exist in Spain for female workers who have children. Thus, the wage penalty suffered by working mothers was masked in the raw data by the existence of some characteristics of mothers, some observed and others not, that cause them to be perceived as receiving a higher wage than women without children. That is, by controlling for unobserved characteristics of women, we detected that mothers tend to earn less than non-mothers. In this sense, mothers are rewarded more for the unobserved characteristics (skill, desire to work, effort and concern for completion of tasks, etc.), which results in higher wages. Alternatively, one could think in terms of inverse causality in the sense that the women who earn more money are those who decide to have children. That is to say, there would exist a double causation between wages and having children. In this case, the advisable thing would be to estimate with variables instrumental for those related to having children. Nevertheless, this is hard to achieve given the difficulty of finding suitable instruments for these variables. By ignoring that the variable of children is possibly endogenous, the individual fixed-effects may be reflecting the existing correlation between the perturbation and the variables, thus underlining the importance of this inverse causal relationship.

Table 4 Fixed-effects estimation
Full size table

As for the observed characteristics, when a comparison is now made, within Table 4, column by column, we observe that in the most simple specification, with none of the controls, signs of the wage penalty are only found when the woman had three or more children, although the fact of having a birth in the family carries with it an 8% wage loss. From a comparison with the estimates in column 1 of Table 3 and the figures in Table 1, it can be deduced that mothers are rewarded more due to observed and unobserved characteristics. Once the latter are taken into account through time-invariant individual fixed effects, we can investigate the extent of the influence of the observed characteristics, which are entered in groups. Thus, by controlling for the variables related to human capital, the penalty extends to any number of children, progressively increasing from 5% with just one child, to more than 20% when there are three or more children. Accordingly, when there is a birth in the family, the woman loses 9% of her wage relative to not having a child, with this being interpreted as a rough average of the coefficients of Number of Children. This shows that, on average, mothers accumulate more human capital than non-mothers, and they are compensated appropriately. This becomes more evident when variables describing the job are included. Thus, the fact that there is a birth in the family during the current year means that the woman loses 9% of her wage, compared to if there had not been a birth. Also, it can be interpreted as mothers working at firms who pay higher wages than those employing non-mothers. Additionally, having one child living in the household means a significant loss in wages of 6%; having two children, almost 14%; and having three or more, more than 15%. The addition of variables describing family resources somewhat diminishes the existence of the motherhood wage penalty, with this being discussed later in more detail.

One final noteworthy result is that, as we progressively included controls, the motherhood wage penalty became more apparent, showing that mothers tended to allocate themselves to better paid jobs or, alternatively, that women with better paid jobs were more likely to become mothers. This is opposite to the case of the U. S. (Anderson et al. 2003; Budig and England 2001), in which the motherhood wage penalty progressively decreased as controls were introduced in the estimation. This indicates that, in the U. S., the raw data show a clear penalty for working mothers, which decreases as additional characteristics are controlled for.

Discussion

Taken together, our results show that some characteristics of Spanish mothers, their level of accumulation of human capital, and the qualities of their job, allow them to receive higher wages, thus masking any wage penalty. In other words, in the case of two women with the same education level and job attributes, one being a mother and the other not, the first would receive a lower wage. Consequently, two conclusions can be immediately stated. First, mothers are not looking for more child-compatible jobs in exchange for lower wages. Rather, the opposite seems to hold. Mothers are enrolled in jobs in which they can get higher wages. Second, the remaining explanations for the un-revealed wage penalty in Spain rely either on some kind of wage discrimination, statistical or not, against mothers, or on the fact that mothers exert less effort at work than non-mothers. The impossibility of obtaining information about these two circumstances makes testing for them fruitless, and so they should be considered simple conjectures. In this context, if we believe that discrimination underlies the observed wage penalty, it may be interpreted as a specific feature of a probable Mediterranean model, as suggested by many authors, according to which, mothers’ employment has been traditionally discouraged, and only in very recent years has women’s participation begun to increase. This accords with the existence of a traditional gender division of paid and unpaid work, especially care and housework, in the Mediterranean countries. Only in recent years has this changed, allowing for an increase in women’s participation rates that has, however, not ignored the fact that childcare and elder care are still primarily provided by women (García et al. in press).

Nevertheless, one final possibility is that causation is acting in the opposite way. That is, women earning more are the ones who decide to have children. Some of the results observed so far support this view. First, when fixed effects are included, the wage penalty becomes clear. Second, we have seen that mothers enjoy greater accumulation of human capital and tend to opt for jobs which allow them to obtain higher wages. Furthermore, and just as in the case of the pool estimation, when the variables related to the existing resources in the home were introduced, the wage penalty was considerably reduced (see columns 4 and 8). However, this cannot be interpreted as an indication of a reduction of the wage penalty. Rather, it indicates that the family group compensates in large measure for the lost wages resulting from having a child. In particular, the importance of the Non-wage Income variable leads us to conclude that the distribution of alternate sources of income is related to a lower wage per hour, which can be interpreted that those who have more financial support out-of-work can afford to have a child. Again, this could be associated with a special welfare state characterized by the existence of maternity benefits only for working women, and with reduced benefits in the case of child-care. Both are typical of the Mediterranean typology, as is the relevance of the family as a form of protection against labor risks. This suggests that, while benefits for parenthood, motherhood and childcare are not as extensive as in the conservative or social democratic countries, policies enhancing the role of the family as the economic center should be implemented, since the family is a significant secondary provider of welfare to the individual.

In conclusion, despite our approach being limited to some extent by the fact that we consider only working women, and that having children is usually planned and, therefore, endogenous, our fixed-effects estimations effectively show the existence of the motherhood wage penalty in Spain. Specifically, the penalty is close to 9% if there is a birth in the family during the current year; 6% if there is one child living in the household, almost 14% if there are two, and more than 15% if there are three or more. These figures are comparable to the Anglo-Saxon countries and greater than those of the Continental European countries. Thus, our analysis presents new empirical evidence for a Mediterranean country, illustrating some clear differences with the results observed in the literature for countries classified in the other well-established welfare state regime types. These results can be useful for future research investigating the emergence of a new welfare regime type, the Mediterranean or Latin rim, in which Spain can be interpreted as a paradigm.