The effect of poverty, gender exclusion, and child labor on out-of-school rates for female children.

Castillo, Leopoldo Laborda ; Salem, Daniel Sotelsek ; Sarr, Leopold Remi 等

In this article, the authors analyze the effect of poverty, social exclusion, and child labor on out-of-school rates for female children. This empirical study is based on a dynamic panel model for a sample of 216 countries over the period 1970 to 2010. Results based on the generalized method of moments (GMM) of Arellano and Bond (1991) and the tests of causality and zero autocorrelation to the panel data show a negative and significant relation between contributing family workers (female) and number of primary school-age children out of school (female) in Europe and Central Asia region. However, the authors cannot find empirical evidence between primary school-age children out of school rates (female) and the variables used to analyze the effect of poverty and social exclusion (poverty headcount ratio at national poverty line and total vulnerable employment). Moreover, the article identifies effects of other variables like proportion of seats held by women in national parliaments. In addition, this article examines geographic regions separately, with the anticipation that differentials in livelihood strategies and opportunities could be reflected in female child schooling decisions.

Keywords: child labor, poverty, social exclusion, out-of-school rates

**********

Education is a human right that ought to be accessible to everyone, without any discrimination. According to Miles and Singal (2010), all children must be able to go to school and thereby benefit from the same opportunities to build a future. Education also must be free so that children from disadvantaged environments are able to enjoy this right. (1) However, according to Lewis and Lockheed (2006), nearly three fourths of the 60 million girls still not in school belong to ethnic, religious, linguistic, racial, or other minorities. Before this disheartening fact became known, many authors (e.g., Herz & Sperling, 2004; Lewis & Lockheed, 2006; Tembon & Fort, 2008) tried to analyze female child education.

Herz and Sperling (2004) consider that policymakers will need to make special efforts to address the economic, social, and cultural barriers that keep even larger proportions of girls in poor countries out of school. In a similar way, Lewis and Lockheed (2006) point out that enrolling and retaining excluded children, in particular girls, in school requires new strategies, and reaching them can be costly because it requires new methods tailored to each group.

Taking into account these recommendations, Tembon and Fort (2008) suggest several strategic directions for advancing gender equality in education, such as improving the quality of not only primary but also secondary education and focusing on the most vulnerable groups. On the other hand, it is important to improve the relationship between public sector, private sector, and nonprofit organizations. All this implies improving studies regarding gender, poverty, and the development and presentation of policymakers' information. (2)

Taking into account the above considerations, we propose an empirical study based on some of the patterns of inequalities in educational attainment proposed by Filmer (2006). These include significant differences between and within countries and inequality associated with economic status, gender, child labor, and so on. (3)

To conduct the empirical analysis, we employ a dynamic panel model for a sample of 213 countries over the period 2003 to 2007. With this objective in mind, the work is organized as follows: In the next section, we present a brief empirical background about the effect of poverty, social exclusion, and child labor on out-of-school female children rates. In the third section, we develop the methodology used for the empirical analysis. In the fourth section, we present the main empirical results obtained. In the fifth section, we present a summary of the main conclusions. Finally, the last section ends with some policy implications.

BACKGROUND: OUT-OF-SCHOOL FEMALE CHILDREN RATES FACTORS

According to Yu (2007), each country has its own educational policies and goals/functions and these system-wide differences in educational goals emphasize that the criteria for judging/determining school effectiveness should take into account the contextual factors within which each school/nation operates.

Taking into account the above consideration, the main object of this section is to find some empirical support that allows a satisfactory understanding of the role that poverty, social exclusion, and child labor could play in the out-of-school female children rates. With this idea in mind, we propose a systematic revision of the most relevant empirical studies (according to our objectives) conducted at the country level.

The Effect of Poverty

To analyze the effect of poverty in education, we selected some studies country by country in the developing world, focusing on the works published by several important authors, and considered their main conclusions about this issue.

To begin with, we can see that family structure and a child's relationship to the head of the household are found to be significant influences and important factors on investments in children's education, according to Shapiro and Tambashe (2001), along with the place of residence and the economic status of the household. Jonathan Morduch (1999) believes that poverty can be related to vulnerability, as numerous poor households are exposed to many risks, including economic ones, that carry negative implications for the oncoming generation, reducing rates of children in school.

In regard to family structure, by the examination of the possibility that parents obtain informal income insurance by letting their children work, Gubert and Robilliard (2006) conclude that transitory income affects children's school dropout behaviors significantly but not their school entrance. The probability of school entrance nevertheless appears to be sensitive to shocks in the demographic structure of the household, because it is negatively correlated with the death or moving out of elderly household members. Monetary and nonmonetary costs of education are a great burden on the poorest households and act as a significant barrier to education. The indirect costs are also considerable, with seasonal variations relating to the demand for labor, according to Boyle, Brock, Mace, and Sibbons (2002). As a result, this seasonal cycle of opportunity costs impacts attendance patterns, which, in turn, influence permanent premature removal from school.

As we find that increased economic well-being translates into greater investments in children's education for females and males, Shapiro and Tambashe (2001) determine that improved economic status does not necessarily result in reduced gender differences in school outcomes. As well, Brown and Park (2002) use direct measures of credit limits and women's empowerment, finding evidence of gender bias in which academically weak girls are more likely to drop out in primary school whereas most boys continue on to junior secondary school.

Analyzing the areas of location of the household, Mugisha (2006) explores patterns of school enrollment comparing urban slum, urban nonslum, and rural children. The results suggest that school enrollment is higher in urban nonslum areas than in urban slum areas and is higher in slums than in rural areas at younger ages. Factors contributing to these results point to the poor quality of primary schools in slums, limited access to secondary schools for slum children, increased vulnerability to coercion into sexual activity and other ills that hinder school participation, disabling environment at home, and increased child labor.

Another subject, studied by Deon Filmer (2005), is the relationship between whether young people have a disability, the poverty status of their household, and their school participation. According to Filmer, youth with disabilities are almost always substantially less likely to start school and, in some countries, have lower transition rates resulting in lower schooling attainment. The order of magnitude of the school participation disability deficit is often larger than those associated with the other characteristics we have read before, such as gender, area of residence, or economic status differentials.

A very important matter we have to consider as we deal with the effect of poverty on households and children's education are the possible benefits of conditional cash transfers (CCT). As the authors agree that conditional transfers can provide a high and positive impact in increasing school attendance, acting as safety nets for the schooling of the poor, certain kind of conditions (e.g., requiring enrolling children into public schools) and right strategies are necessary in order to make these transfers reach their real purpose.

De Janvry, Finan, Sadoulet, and Vakis (2006) found that conditional transfers helped protect enrollment but did not prevent parents from increasing child work in response to shocks. In this sense, to reduce the poverty headcount ratio by increasing incomes among poor households, cash transfers would have to be sizeable. Even then, an increase in income, by itself, would not suffice to significantly increase school attendance. Higher impacts at lower cost could be achieved by making transfers targeted and conditional (Kakwani, Soares, & Son, 2006). Extending this conclusion, Son and Florentino (2008) found that the targeted CCT program would lead to greater school attendance and poverty reduction. Designing a program with a weak or nonexistent targeting strategy not only reduces the transfer cost per beneficiary but also leads to leakages to the nonpoor, driving down its impact and effectiveness. Using a similar simulation methodology, Kumara and Pfau (2011) find that cash transfer programs targeting poor children would be the most cost-effective way to reduce child poverty and encourage school attendance. Their findings suggest that even a limited program budget can provide significant impacts. On the other hand, Bourguignon, Ferreira, and Leite (2003) evaluated the Bolsa Escola program in Brazil, with results that suggest approximately 60% of poor children from age 10 to 15 not in school enroll in response to the program. They conclude that the targeting of this CCT is accurate, but that poverty reduction, though effective, does not cause a large positive effect. Therefore, governments must transfer cash in an intelligent and efficient way, substantially increasing the amounts if they hope to cause bigger effects in reducing poverty.

Gender Exclusion

Reviewing some studies country by country in the developing world, conducted by some significant authors, we continue to analyze the effect of gender exclusion in education. One of the main factors found to be a cause for the gender gap in child education is cultural, taking an elevated importance in the household decisions.

Related to this, Colclough, Rose, and Tembon (2000) argue that poverty is associated with an underenrollment of school-age children, but that the gendered outcomes of such underenrollment are the product of cultural practice, rather than of poverty per se. Their study shows the variety and extent of adverse cultural practice that impedes the attendance and performance of girls at school, relative to boys. Taking into account these findings, the authors warn that gender inequalities in schooling outcomes, measured in qualitative and quantitative terms, will not necessarily be reduced as incomes rise. Defining the cultural aspect in rural areas, the presence of a public school for girls in the village makes an enormous difference for girls in primary enrollment, given parents' reluctance for girls to travel far from home. Lloyd, Mete, and Sathar (2005) find that girls' enrollment in public primary school is particularly responsive to improvements in some aspects of school quality, in particular whether the teacher resides in the village. This would suggest that school quality is important not only for retention but also for enrollment. As a proof, Sawada and Lokshin (2009) discovered serious supply-side constraints that might arise from a village-level lack of demand for primary schools for girls.

It is also important to take into account the education level of the head of the household, as Rose and Al-Samarrai (2001) did by showing that the probability of attending and completing school was lower for girls than for boys in economically constrained households with illiterate parents but was equal for girls and boys in better-endowed households.

Furthermore, the economic changes also affect the rates of schooling in rural areas. Kajisa and Palanichamy (2010) found that the initial high correlation between children's attainment of basic schooling and the household's assets for farming disappeared during the mid-1980s. However, even after that time, the attainment of advanced education is still affected by rainfall and thus by farm income, indicating the lack of insurance markets and the segregation of poor households under agriculturally unfavorable conditions from advanced education. Meanwhile, segregation based on gender and adult members' education has been disappearing.

Analyzing the gender gap in child education, some authors locate the larger difference between boys' and girls' attendance in the primary school-age group. For example, Sawada and Lokshin (2009) revealed a high education retention rate and observed that school progression rates between male and female students after secondary school are comparable. In particular, they find gender-specific and schooling-stage-specific birth-order effects on education. Their overall findings are consistent with the implications of optimal schooling behavior under binding credit constraints and the self-selection of education-friendly households. On the other hand, Aslam and Kingdon (2008) investigated whether the intrahousehold allocation of educational expenditure favors males over females, finding that in middle and secondary school ages, evidence points to significant promale biases in the enrollment decision as well as the decision of how much to spend conditional on enrollment. However, in the primary school, only the former channel of bias applies. Their results suggest that the observed strong gender difference in education expenditure is within rather than across household phenomena.

Child Labor

Finally, focusing on the works of some authors, country by country in the developing world, we can analyze the effect of child labor in education, taking it into account as a clearly huge obstacle for children's education. As we have previously read in some conclusions, children from poor households and households with parents with a low level of education are less likely to attend school and, by consequence, are more likely to be engaged in child labor, giving strength to the hypothesis that poverty is the root cause of child labor, according to Friedrich Huebler (2008). As well, Jensen and Nielson (1997) investigate what affects school attendance and child labor and find with the empirical analysis that economic and sociological variables are important determinants for the choice between school attendance and child labor. In particular, Jensen and Nielson find some support for the hypothesis that poverty forces households to keep their children away from school.

The labor force participation is nontrivial among those below the legal working age or supposed to be in school, and these working children contribute significantly to total household income, according to Psacharopoulos (1997). The fact that a child is working reduces his or her education attainment by about 2 years of schooling relative to the control group of nonworking children. Also, Psacharopoulos found that grade repetition is closely associated with child labor.

Nevertheless, poverty does not appear alone among the causes that enhance child labor. The empirical analysis by Canagarajah and Coulombe (1998), analyzing the determinants of child labor in conjunction with decision to school, shows, though not very convincingly, proof that poverty is the main culprit of child labor; however, it is correlated significantly with attendance and enrollment in school. They show a significant negative relationship between going to school and working. According to the authors, the high cost of schooling and the low quality and weak relevance of education has also pushed many children into work. On the other hand, family characteristics have a big role to play in the decision about children going to school or work, and the parents' education has a significant negative effect on child labor and the effect is stronger for girls than boys. In addition, Moyi (2011) finds that socioeconomic status and structure of the household have a strong effect on child labor.

In regard to the household structure, Patrinos and Psacharopoulos (1997) analyzed the effects of being indigenous, number of siblings, sibling activities, and sibling age structure on child schooling progress and child nonschool activity. Their analysis shows that family size is important. However, the analysis also demonstrates the importance of taking into consideration the siblings' activities. The age structure of siblings is important, but in conjunction with their activities. According to these authors, having a greater number of younger siblings implies less schooling, more age-grade distortion in the classroom, and more child labor.

Paying attention to the gender differences inside child work, Canagarajah and Coulombe (1998) show some clear gender-based distinctions in the type of tasks performed by a girl and boy worker (girls do more household chores, while boys are in the labor force). As work, broadly defined, substantially reduces schooling for boys and girls, Assaad, Levison, and Zibani (2007) present evidence that lower rates of school attendance for girls are caused by a substantial burden of household work. Although market work is a serious impediment to schooling for boys, a much larger proportion of girls than boys engage in substantial hours of work, when work is defined to include both the labor force and domestic work.

By studying rural areas in particular and comparing them with urban areas, Ersado (2005) finds that though strong evidence shows that poverty drives child labor in rural areas, there is a general lack of support for the poverty hypothesis in urban areas. This suggests that policies such as a ban on child labor in rural areas could have an adverse effect, as child labor decisions are more likely a response to poverty and subsistence requirements. Similarly improving access to credit has greater potential for alleviating child labor and enhancing school enrollment in rural than urban areas. On the other hand, the availability of alternative child care options appears to considerably decrease child labor and create conditions for higher school attendance rates in urban than in rural areas. Finally, the evidence indicates that efforts to bolster adult education levels and wages will help curb the prevalence and intensity of child labor and improve the likelihood that children stay in school.

Concerning the actions taken to reduce child labor and increase children's schooling, Ravallion and Wodon (1999) studied the effects on children's labor force participation and school enrollments of the pure school-price change induced by a targeted enrollment subsidy. In their theoretical model, these authors predict that the subsidy increases schooling, but its effect on child labor is ambiguous. Their empirical findings indicate that the subsidy increased schooling by far more than it reduced child labor. (4) Moreover, Hazarika and Bedi (2002) draw a distinction between child labor within the household (intrahousehold) and child work in the labor market (extrahousehold), and examine the separate effects of schooling costs on the two types of child labor. The authors found that extrahousehold child labor and schooling costs are positively related, whereas intrahousehold child labor is insensitive to changes in schooling costs. Lastly, their results suggested that reduction in schooling costs will have limited success in the abatement of child labor.

It is also important to indicate that a large proportion of working children attend school. According to Moyi (2011), if the consequence of working is to hinder educational attainment, then policymakers need to focus on the dimension of education inequality between students who combine work and school and those who do not.

METHOD

To solve some methodological problems related with the evaluation of the models by traditional methods, (5) we propose one method (based on Arellano & Bond, 1991) that consists of obtaining consistent estimators.

As a result of the above consideration, we propose an empirical study based on a dynamic panel model for a sample of 216 countries over the period 1970 to 2010. With this objective in mind, in the next subsections we develop the methodology used for the empirical analysis and present the main empirical results obtained.

Econometric Model

Arellano and Bond (1991) proposed an extension of GMM introduced initially by Hansen (1982), to the case of panel data for a simple model AR (1):

[y.sub.it] = [alpha] x [y.sub.it-1] + [[mu].sub.i] + [v.sub.it] where [absolute value of [gamma]]<0 (1)

As we noted, our sample has 216 countries over the period 1970 to 2010. For this reason, we consider the case where temporal dimension is small or medium (T = 41), whereas individual dimension (N = 216) is important. Also, we consider that individual effects are stationary and we assume traditional hypotheses of residues. Difference models (1) can be written as below:

[DELTA][y.sub.it] = [alpha] x [DELTA][y.sub.it-1] + [u.sub.it] where [absolute value of [gamma]]<0 (2)

Where [u.sub.it] = [v.sub.it] - [v.sub.it-1]. The gait of Arellano and Bond, in presence of the exogenous variables, consists in estimating the model in difference (6):

[DELTA][y.sub.it] = [p.summation over (k=1)] [[alpha].sub.k][DELTA][y.sub.i(t-k)] + [beta]'(L)[X.sub.it] + [DELTA][v.sub.it] (3)

The preceding dynamic model can be rewritten for each individual in the following form:

[y.sub.i] = [w.sub.i] x [delta] + [[tau].sub.i] x [[mu].sub.i] + [v.sub.i] (4)

where r is a vector of parameter and vv, is a matrix that contains the retarded dependent variable and explanatory variables (7).

To have previous value GMM, it is necessary to pass by a first stage that consists of making wished transformation (first difference or orthogonal deviation) to find and to use instruments matrix and to achieve a first evaluation, named "evaluation of first stage." This stage corresponds to an evaluation that allows for providing estimated residues after transformation. The objective of transformation is, as at Anderson and Hsiao (1982), to eliminate individual heterogeneity of the model. (8)

Finally, Arellano and Bond (1991) propose a test verifying the absence of autocorrelation of first and second order. (9)

We specify a dynamic model characterized by presence of one endogenous variable delayed among explanatory variables. Our specified model is a dynamic panel model given by:

[y.sub.it] = [alpha] x [y.sub.it-1] + [beta]' x [x.sub.it] + [[mu].sub.i] + [v.sub.it] (5)

where [y.sub.it] is the endogenous variable, [y.sub.it-1] symbolizes endogenous variable appearing in the regression as being a retarded explanatory variable. X represents the vector of exogenous variables, ([alpha], [beta]) designate parameters to estimate, and [[mu].sub.i] is the specific effect of country (i). This specific effect can be a stationary or uncertain effect and constitute individual heterogeneity as: [[mu].sub.i], i.i.d ~ N[(0,1)]; [v.sub.it] is stochastic term as: [v.sub.it] ~ i.i.d N[(0,1)]. The bias is positive and increases with the variance of the specific effect. Indeed, [y.sub.it] is a function of [v.sub.it] as is [v.sub.it-1] x [y.sub.it-1] is an explanatory variable correlated with stochastic term. It introduces a bias in the value of ordinary least squares. Even as putting forth the hypothesis that stochastic terms are not correlated, this value is nonconvergent. (10)

In our model, [y.sub.it] is the rate of primary school-age female children out of school ([COSf.sub.it]) in period (t) and country (i). This rate is explained by (1) ([COSf.sub.it]) of the period (t - 1); (2) family workers contributing, (11) female ([FWCf.sub.it]) (% of females employed); (3) total vulnerable employment (TVEit) (% of total employment); (4) poverty headcount ratio at national poverty line (12) ([PHR.sub.it]) (% of population); (5) proportion of seats held by women in national parliaments ([SWP.sub.it]) (%); and (6) total population ([TP.sub.it]) like control variable.

Under another form, one can write our model as below:

[COSf.sub.it] = [alpha] x [COSf.sub.it-1] + [[beta].sub.1] x [CFWf.sub.it] + [[beta].sub.2] x [TVE.sub.it] + [[beta].sub.3] x [PHR.sub.it] + [[beta].sub.4] x [SWP.sub.it] + [[beta].sub.5] x [TP.sub.it] + [[mu].sub.i] + [v.sub.it] (6)

Data and Variables

The statistical sources used for this analysis are the World Bank's World Development Indicators Database (13) and the World Bank's Education Statistics Database (14).

The primary World Bank collection of development indicators is compiled from officially recognized international sources. It presents the most current and accurate global development data available and includes national, regional, and global estimates.

The World Bank compiles data on education inputs, participation, efficiency, and outcomes. Data on education are compiled by the United Nations Educational, Scientific, and Cultural Organization (UNESCO) Institute for Statistics from official responses to surveys and from reports provided by education authorities in each country (for more details about the data, see The World Bank EdStatsQuery (15)).

These databases provide more than 800 development indicators, with time series for 209 countries and 18 country groups from 1960 to 2011. Table 1 presents a summary of the key variables used to empirically validate the dynamic models proposed.

From the World Bank's World Development Indicators Database and World Bank's Education Statistics Database, we have temporal observations (T = 41) by countries for the period 1970 to 2010. We are able to form a balanced panel data; see descriptive statistics of variables in Table 2.

EMPIRICAL RESULTS

Starting with a descriptive analysis of data, Table 3 shows large differences between regions when we compare the current rates (total and only female) of primary school-age children out of school for 1970 to 2010 in low-income countries. We can underline the rates of East Asia and Pacific (50.6 and 50.1, respectively) and Europe and Central Asia (7.50 and 7.03, respectively). The other regions have values of 39.27 and 37.72 (Sub-Saharan Africa), 26.98 and 24.14 (South Asia), and 34.4 and 29.6 (Latin America & Caribbean).

If we are considering countries with lower middle income, taxes will be substantially reduced in East Asia and Pacific, going from 50.1 to 12.68 (one fourth), similar to what occurs in Latin America and the Caribbean, reducing female dropout rate to one third.

Taking into account the current rates (total and only female) of contributing family workers, we can highlight for 1970 to 2010 the cases of East Asia and Pacific (43.97 and 56.15, respectively, in low-income countries) and 11.90 and 11.20 (Latin America & Caribbean). The other regions have values of 30.74 and 40.45 (Sub-Saharan Africa), 15.78 and 21.53 (Europe and Central Asia), and 24.56 and 54.30 (South Asia).

Finally, when comparing the proportion of seats held by women in national parliaments (%) by income level, we can see large differences between regions that are reduced as we compare them to samples of countries with higher incomes within each region. Focusing on low-income level countries, we can underline the cases of East Asia and Pacific (15.06), South Asia (13.83), Sub-Saharan Africa (13.17), Europe and Central Asia (10.30), and Latin America and Caribbean (3.88).

On the other hand. Table 4 shows small differences between regions when we compare the current percentages (by income level) of the population that is female. These differences are higher when we compare the rates of the population between age 0 to 14 as a percentage of the total population. The low-income East Asia and Pacific region has a value of 35.45. The other regions have values of 39.54 (Europe and Central Asia), 40.89 (Latin America & Caribbean), 42.86 (South Asia), and 45.14 (Sub-Saharan Africa).

In Table 4, we also can see large differences between regions when we compare the poverty headcount ratios at national poverty line (in terms of % of population). Focusing on low income level, we can highlight the cases of Latin America and Caribbean (77.00), Europe and Central Asia (59.67), Sub-Saharan Africa (54.40), South Asia (42.14), and East Asia and Pacific (36.98).

Finally, taking into account the current rates (total and only female) of Vulnerable employment (in terms of % of total employment) (see Table 4), we can emphasize for 1970 to 2010 the cases of East Asia and Pacific (84.90 and 87.05, respectively, in low-income countries). The other regions have values of 82.69 and 88.58 (Sub-Saharan Africa), 75.60 and 84.28 (South Asia), and 49.62 and 49.54 (Europe and Central Asia).

The evaluation that we present in Table 5 corresponds to the GMM evaluation of Arellano and Bond (1991). In Table 5, the empirical evaluations show a negative relation between contributing family workers (female) and primary school-age children out of school rate (female), in Europe and Central Asia region. In this case for GMM method in first difference, the variable contributing family workers (female) is negative and significant (-.5869098), (T-Stat = .3082361). The result is consistent with the findings of Canagarajah and Coulombe (1998) who report some clear gender-based distinctions in the type of tasks performed by a girl and a boy worker; girls do more household chores, while boys are in labor force.

However, we cannot find empirical evidence between primary school-age children out of school rate (female) and the variables used to analyze the effect of poverty and social exclusion (poverty headcount ratio at national poverty line and total vulnerable employment, respectively). In other words, like in Colclough et al. (2000), we do not find sufficient evidence that poverty is associated with an underenrollment of school-age female children. In other words, the gendered outcomes of such underenrollment could be the product of cultural practice, rather than of poverty per se. Under this point of view, and according to these authors, gender inequalities in schooling outcomes, measured in qualitative and quantitative terms, will not necessarily be reduced as incomes rise.

Finally, identification of effects of other variables is far from being obvious according to different evaluations. For example, a positive and significant value on proportion of seats held by women in national parliaments (.2914276) with a (T-Stat = .1309497) in Europe and Central Asia region. This result is not consistent with the findings of Brown and Park (2002). According to these authors, women's empowerment reduces the likelihood of dropping out.

SUMMARY AND CONCLUSIONS

The literature on poverty, gender exclusion, and child labor and schooling is large and continually growing. This article considered the impact of each of these factors, while controlling others at the same time. It examined geographic regions individually and separately, with the anticipation that differentials in livelihood strategies and opportunities could be reflected in female child schooling decisions.

Given the data, it is noteworthy that in most low-income countries the male dropout rate is higher than the female rate, which may give us a first indication of the relationship between child labor (more men than women) and dropouts.

Another finding of note is that though in poor countries in East Asia and the Pacific, the proportion of women contributing to family workers is higher than the total (56.9 vs. 43.9), in terms of dropout percentage coincides (50.1 and 56.1), in regions such as Latin America and Caribbean, this percentage of contribution of family workers is much lower compared to the female dropout rate (29.6 and 11.2). In Europe, it is the opposite; this relation is inverse, with a 21.6 female family contributions and 7.03 female dropout rate.

It seems that some correlation exists between female dropouts and contributing family workers and the level of development of the region. As we analyze the poorest regions (in global terms), contributing family workers and dropout have similar rates (women leave school and go to work on their own, the economy has high rates of informal employment that can withstand it); this is the case in East Asia and Pacific, where the dropout rate is much lower than the proportion of contributing family workers in the most developed region, whereas school dropout is much higher in less-developed regions with high levels of inequality income--though the number of women working on their own is lower, like in Latin America and the Caribbean.

Our econometric analysis shows a negative and significant relation between contributing family workers (female) and primary school-age children out of school rate (female), in Europe and Central Asia region. However, we cannot find empirical evidence between primary school-age children out of school rate (female) and the variables used to analyze the effect of poverty and social exclusion when we analyze other regions (poverty headcount ratio at national poverty line and total vulnerable employment).

Moreover, identification of effects of other variables, like amount of seats held by women in national parliaments, is far from being evident according to different evaluations.

Even though our results were obtained considering aggregate primary school-age children out of school rate as the explanatory variable, we think it will be a very useful disaggregate (data by family) approach, to evaluate if our findings are robust.

We believe that efforts in these directions could be very productive for a deeper understanding of the factors that may affect the education of socially excluded girls in developing countries.

POLICY IMPLICATIONS

As for the policy implications that can be associated with our analysis, we start by considering that education systems in many regions leave out a significant number of people, especially girls, forcing us to recommend that if investments in education are discriminate then the inversions should focus on resolving structural impediments that go beyond education. Consequently, it is noteworthy that education policies will not solve the problem of discrimination as there are many factors and sectors involved in this relationship and in any case the level of development determines any action.

Taking into account the last consideration, we judge especially relevant all research about programs that work to mitigate the effects of social exclusion, child labor, poverty, and education, to offer policy recommendations and unanswered research questions. In this line of work, we recommend the excellent work of Tembon and Fort (2008), where the authors have an interesting list of interventions that have worked to improve girls' education.

According to Tembon and Fort (2008), and taking into account what has been said so far, the first thing to consider is that an increase in demand should reduce tuition fees but mostly promote conditional transfer programs to women's education.

Another important issue regards concentrating research efforts on gender inequality above all factors and strengthening community action policy as a way to improve the social and cultural limitations associated with access of female children to education.

Finally, it is necessary to promote female secondary education because it involves improving returns on investment in female education. Attending to educational models, including diffusion and awareness of this gender discrimination is linked to education.

DOI: 10.1080/02568543.2014.884028

ACKNOWLEDGMENTS

The authors thank the editor and two anonymous reviewers for their constructive comments and kindness in helping us improve the manuscript. Also, the authors thank Diana C. Prieto for her inspiration to conduct our research on gender issues.

REFERENCES

Anderson, T. W., & Hsiao, C. (1982). Formulation and estimation of dynamic models using panel data. Journal of Econometrics, 18, 47-82.

Arellano, M., & Bond, S. (1991). Some test of specification for panel data: Monte Carlo evidence and application to employment equations. Review of Economic Studies, 58, 277-297.

Aslam, M., & Kingdon, G. (2008). Gender and household education expenditure in Pakistan. Applied Economics, 40, 2573-2591.

Assaad, R., Levison, D., & Zibani, N. (2007). The effect of child work on schooling: Evidence from Egypt (Minnesota Population Center Working Paper 2007-4). Minneapolis, MN: University of Minnesota Press.

Bourguignon, F., Ferreira, F., & Leite, P. (2003). Conditional cash transfers, schooling and child labor: Micro- simulating Brazil's Bolsa Escola program. The World Bank Economic Review, 17(2), 229-254.

Boyle, S., Brock, A., Mace, J., & Sibbons, M. (2002). Reaching the poor: The "costs" of sending children to school. Synthesis report. London, UK: Department for International Development (DFID).

Brown, P., & Park, A. (2002). Education and poverty in rural China. Economics of Education Review, 21(6), 523-541.

Canagarajah, R. S., & Coulombe, H. (1998). Child labor and schooling in Ghana (Policy Research Working Paper 1844). Washington, DC: World Bank.

Colclough, C., Rose, P., & Tembon, T. (2000). Gender inequalities in primary schooling: The roles of poverty an adverse cultural practice. International Journal of Educational Development, 20(1), 5-27.

De Janvry, A., Finan, F., Sadoulet, E., & Vakis, R. (2006). Can conditional cash transfer programs serve as safety nets in keeping children at school and from working when exposed to shocks? Journal of Development Economics, 79, 349-373.

Ersado, L. (2005). Child labor and schooling decisions in urban and rural areas: Comparative evidence from Nepal, Peru, and Zimbabwe. World Development, 33(3), 454-480.

Filmer, D. (2005). Disability, poverty, and schooling in developing countries: Results from 11 household surveys. Washington, DC: World Bank.

Filmer, D. (2006). Gender and wealth disparities in schooling: Evidence from 44 countries. International Journal of Educational Research, 43(6), 351-369.

Gubert, F., & Robilliard, A. (2006). Do household income shocks affect school attendance in rural areas?: A case study of Madagascar (Working Paper). Paris, France: Developpement, Institutions & Analyses de Long Terme.

Hansen, L. P. (1982). Large sample properties of generalized method of moments estimators. Econometrica, 50, 1029-1054.

Hazarika, G., & Bedi, A. S. (2002). Schooling cost and child labor in rural Pakistan (ZEF Discussion Paper on Development Policy, No. 52). Bonn, Germany: ZentrumfurEntwicklungsforschung.

Herz, B., & Sperling, G. B. (2004). What works in girls' education. Evidence and policies from the developing world. New York, NY: Council on Foreign Relations.

Huebler, F. (2008, September). Child labour and school attendance: Evidence from MICS and DHS surveys. Paper presented at Seminar on Child Labour, Education and Youth Employment. Understanding Children's Work Project, Universidad Carlos III de Madrid, Madrid, Spain.

Jensen, P., & Nielson, H. S. (1997). Child labor or school attendance? Evidence from Zambia. Journal of Population Economics, 10, 407-424.

Kajisa, K., & Palanichamy, N. V. (2010). Schooling investments over three decades in rural Tamil Nadu, India: Changing effects of income, gender, and adult family members' education. World Development, 38(3), 298-314.

Kakwani, N., Soares, F., & Son, H. H. (2006). Cash transfers for school-age children in African countries: Simulation of impacts on poverty and school attendance. Development Policy Review, 24, 553-569. doi: 10.1111/j.14677679.2006.00347.x

Kumara, A. S., & Pfau, W. D. (2011). Impact of cash transfer programmes on school attendance and child poverty: An ex ante simulation for Sri Lanka. Journal of Development Studies, 47(11), 1699-1720.

Lewis, M. A., & Lockheed, M. E. (2006). Inexcusable absence: Why 60 million girls still aren't in school and what to do about it. Washington, DC: Center for Global Development.

Lloyd, C. B., Mete, C., & Sathar, Z. A. (2005). The effect of gender differences in primary school access, type, and quality on the decision to enroll in rural Pakistan. Economic Development and Cultural Change, 53(3), 685-710.

Miles, S., & Singal, N. (2010). The Education for All and inclusive education debate: Conflict, contradiction or opportunity? International Journal of Inclusive Education, 14(1), 1-15.

Morduch, J. (1999). The microfinance promise. Journal of Economic Literature, 37(4), 1569-1614.

Moyi, P. (2011). Child labor and school attendance in Kenya. Educational Research and Reviews, 6(1), 26-35.

Mugisha, F. (2006). School enrollment among non-slum, slum and rural children in Kenya: Is the urban advantage eroding? International Journal of Educational Development, 26(5), 471-482.

Patrinos, H. A., & Psacharopoulos, G. (1997). Family size, schooling and child labor in Peru: An empirical analysis. Journal of Population Economics, 10, 387-405.

Psacharopoulos, G. (1997). Child labor versus educational attainment: Some evidence from Latin America. Journal of Population Economics, 10(4), 377-386.

Ravallion, M., & Wodon, Q. (1999). Does child labor displace schooling? Evidence on behavioral responses to an enrollment subsidy (Working Paper 2116). Washington, DC: World Bank.

Rose, R, & Al-Samarrai, S. (2001). Household constraints on schooling by gender: Empirical evidence from Ethiopia. Comparative Education Review, 45(1), 36-63.

Sawada, Y., & Lokshin, M. (2009). Obstacles to school progression in rural Pakistan: An analysis of gender and rivalry using field survey data. Journal of Development Economics, 88, 335-347.

Shapiro, D., & Tambashe, O. B. (2001). Gender, poverty, family structure, and investments in children's education in Kinshasa, Congo. Economics of Education Review, 20(4), 359-375.

Son, H. H., & Florentine, J. (2008). Ex-ante impact evaluation of conditional cash transfer program on school attendance and poverty: The case of Philippines (ADB Economics Working Paper Series, No. 142). Manila, Philippines: Asian Development Bank.

Tembon, M., & Fort, L. (2008). Girls' education in the 21st century: Gender equality, empowerment, and economic growth. Washington, DC: World Bank.

Yu, G. (2007). Research evidence of school effectiveness in sub-Saharan countries (EdQual Working Paper, 7). Bristol, UK: University of Bristol Press.

Leopoldo Laborda Castillo

The World Bank and Institute of Latin American Studies, University of Alcala, Spain

Daniel Sotelsek Salem

Institute of Latin American Studies, University of Alcala, Spain

Leopold Remi Sarr

The World Bank SASHD, Washington, District of Columbia

Submitted May 5, 2012; accepted April 19, 2013.

Address correspondence to Dr. Leopoldo Laborda Castillo, Institute of Latin American Studies, University of Alcala, c/ Trinidad no. 1, Colegio de Trinitarios, Alcala de Henares, 28801, Madrid, Spain. E-mail: llabordacatillo@gmail.com

NOTES

(1.) According to the 2012 Education for All Global Monitoring Report, many young people the world--especially the disadvantaged--are leaving school without the skills they need to thrive in society and find decent jobs. The report will focus on skills development, emphasizing strategies that increase employment opportunities for marginalized groups.

(2.) To achieve these, we need to solve some methodological problems related with the estimation of models by traditional methods, such as Ordinary Least Square (OLS) and Least Squares Dummy Variable (LSDV), that led to ad hoc results. In order to solve this problem, we propose one method (based on Arellano & Bond, 1991) that consists of obtaining consistent estimators.

(3.) According to Filmer (2006), the gender gap is still substantial in countries in South Asia and North and West Africa. In countries where the gender gap is large among youth in the poorest quintile, it is not nearly as large in the richest quintile.

(4.) A plausible explanation, according these authors, is that substitution effects helped protect current incomes from the higher school attendance induced by the subsidy.

(5.) Ordinary least square (OLS) and least squares dummy variable (LSDV) give biased and non convergent values because of inter-relationship between retarded endogenous variable and individual heterogeneity. Under these circumstances, our models should not be estimated by the method of OLS and LSDV due to the fact that estimating by these methods led to ad hoc results.

(6.) Previously, we tested for every individual of the linear restrictions of type:

E[([DELTA][Y.sub.it] = [alpha] x [DELTA][y.sub.it-1])[y.sub.it]] = 0 for j = 2,..., t; t = 3,...t

(7.) The method proposed by these authors permit a GMM in two stages, written in the following form:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

(8.) The number of instrument increases in the time for every individual. In the case where explanatory variables exist, [x.sub.it], is the model correlated with heterogeneity individual [[mu].sub.i];.

(9.) Thus, if distribution is non auto-correlated, this test gives a value of residues differentiated negative and significant to first order and non significant to the second order. This test is based on auto-covariance of residues following a normal law N(0,1) under hypothesis [H.sub.0].

(10.) The evaluation of the models by traditional methods (OLS and within) gives biased and non convergent values because of inter-relationship between retarded endogenous variable and individual heterogeneity. In this context, our models should not be estimated by OLS and LSDV methods due to the fact that estimating by these methods led to ad hoc results. We propose one method that consists of obtaining consistent estimators.

(11.) Originally, we used other variables to approach child labor, like total economically active children (% of children age 7-14) or economically active children, work only (% of economically active children, ages 7-14). Unfortunately, the available data for these variables do not allow applying the proposed methodology.

(12.) Also, we used other variables, like GINI Index, in order to approach the inequality. One more time, the available data for these variables do not allow applying the proposed methodology.

(13.) World Development Indicators Database is the primary World Bank database for development data from officially recognized international sources.

(14.) Education Statistics Database provides data on education from national statistical reports, statistical annexes of new publications, and other data sources.

(15.) The World Bank EdStats Query holds around 2,500 internationally comparable education indicators for access, progression, completion, literacy, teachers, population, and expenditures. The indicators cover the education cycle from pre-primary to tertiary education. The query also holds learning outcome data from international learning assessments (Programme for International Student Assessment [PISA], Trends in International Mathematics and Science Study [TIMSS], etc.), equity data from household surveys, and projection data to 2050. EdStats website: http://go.worldbank. org/ITABCOGIV1.

TABLE 1
Key Variables

Variable Definition

Rate of primary school-age Rate of primary school-age
children out of school. Total children out of school. Total is
 the total number of out-of-school
 children as a percentage of all
 primary school-age children.

Rate of primary school-age Rate of primary school-age
children out of school. Female children out of school. Female is
 the total number of female
 out-of-school children as a
 percentage of all female primary
 school-age children.

Contributing family workers, total Contributing family workers are
(% of total employed) those workers who hold
 "self-employment jobs" as
 own-account workers in a
 market-oriented establishment
 operated by a related person
 living in the same household.

Contributing family workers, Contributing family workers are
female (% of females employed) those workers who hold
 "self-employment jobs" as
 own-account workers in a
 market-oriented establishment
 operated by a related person
 living in the same household.

Vulnerable employment, total (% of Vulnerable employment is unpaid
total employment) family workers and own-account
 workers as a percentage of total
 employment.

Vulnerable employment, female Vulnerable employment is unpaid
(% of female employment) family workers and own-account
 workers as a percentage of total
 employment.

Poverty headcount ratio at national National poverty rate is the
poverty line (% of population) percentage of the population
 living below the national poverty
 line. (a)

Proportion of seats held by women Women in parliaments are the
in national parliaments (%) percentage of parliamentary seats
 in a single or lower chamber held
 by women.

Population, total Total population is based on the
 de facto definition of
 population, which counts all
 residents regardless of legal
 status or citizenship--except for
 refugees not permanently settled
 in the country of asylum, who are
 generally considered part of the
 population of their country of
 origin. (b)

Population, female (% of total) Female population is the
 percentage of the population that
 is female. (c)

Population ages 0-14 (% of total) Population between the ages 0 to
 14 as a percentage of the total
 population. (c)

Source: Author's elaboration using World Bank's World Development
Indicators Database (2011) and World Bank's Education Statistics
Database (2011). http://datacatalog.worldbank.org/

(a) National estimates are based on population-weighted subgroup
estimates from household surveys. (b) The values shown are
midyear estimates. (c) Population is based on the de facto
definition of population.

TABLE 2
Descriptive Statistics

 Variable Observations Mean

Total Sample
 Primary school-age children 2,568 19.87
 out of school rate. Female
 Contributing family workers, female 1,718 9.93
 Vulnerable employment, total 1,494 25.65
 Poverty headcount ratio at national 537 37.09
 poverty line
 Proportion of seats held by women 2,561 13.84
 in national parliaments
 Population, total 8717 24,600,000
East Asia & Pacific
 Primary school-age children out of 296 9.31
 school rate. Female
 Contributing family workers, female 237 15.73
 Vulnerable employment, total 211 28.20
 Poverty headcount ratio at national 60 24.72
 poverty line
 Proportion of seats held by women 368 10.62
 in national parliaments
 Population, total 1476 49,300,000
Europe & Central Asia
 Primary school-age children out of 613 5.63
 school rate. Female
 Contributing family workers, female 782 6.72
 Vulnerable employment, total 679 16.21
 Poverty headcount ratio at national 101 24.37
 poverty line
 Proportion of seats held by women 720 18.27
 in national parliaments
 Population, total 2,358 14,300,000
Latin America & Caribbean
 Primary school-age children out of 434 11.56
 school rate. Female
 Contributing family workers, female 422 6.27
 Vulnerable employment, total 394 32.37
 Poverty headcount ratio at national 216 40.82
 poverty line
 Proportion of seats held by women 465 14.88
 in national parliaments
 Population, total 1,625 11,100,000
Middle East & North Africa
 Primary school-age children out of 422 23.66
 school rate. Female
 Contributing family workers, female 93 17.06
 Vulnerable employment, total 93 23.41
 Poverty headcount ratio at national 21 17.09
 poverty line
 Proportion of seats held by women 250 6.19
 in national parliaments
 Population, total 838 12,300,000
North America
 Primary school-age children out of 36 3.33
 school rate. Female
 Contributing family workers, female 58 0.67
 Proportion of seats held by women 30 17.21
 in national parliaments
 Population, total 123 94,000,000
South Asia
 Primary school-age children out of 61 35.86
 school rate. Female
 Contributing family workers, female 43 38.05
 Vulnerable employment, total 46 52.96
 Poverty headcount ratio at national 25 35.66
 poverty line
 Proportion of seats held by women 103 10.29
 in national parliaments
 Population, total 328 141,000.000
Sub-Saharan Africa
 Primary school-age children out 706 38.95
 of school rate. Female
 Contributing family workers, female 83 26.19
 Vulnerable employment, total 71 56.15
 Poverty headcount ratio at national 114 51.80
 poverty line
 Proportion of seats held by women 625 13.32
 in national parliaments
 Population, total 1,969 11,100,000

 Variable SD Minimum

Total Sample
 Primary school-age children 22.03 0.06
 out of school rate. Female
 Contributing family workers, female 14.83 0.00
 Vulnerable employment, total 19.55 0.40
 Poverty headcount ratio at national 18.07 2.80
 poverty line
 Proportion of seats held by women 9.93 0.00
 in national parliaments
 Population, total 101,000,000 5450
East Asia & Pacific
 Primary school-age children out of 9.68 0.06
 school rate. Female
 Contributing family workers, female 17.06 0.00
 Vulnerable employment, total 24.68 2.00
 Poverty headcount ratio at national 13.94 2.80
 poverty line
 Proportion of seats held by women 9.49 0.00
 in national parliaments
 Population, total 186,000,000 7296
Europe & Central Asia
 Primary school-age children out of 4.97 0.12
 school rate. Female
 Contributing family workers, female 11.29 0.00
 Vulnerable employment, total 12.64 0.80
 Poverty headcount ratio at national 16.50 4.90
 poverty line
 Proportion of seats held by women 10.58 0.00
 in national parliaments
 Population, total 25,000,000 19,136
Latin America & Caribbean
 Primary school-age children out of 12.85 0.15
 school rate. Female
 Contributing family workers, female 6.61 0.00
 Vulnerable employment, total 11.15 3.90
 Poverty headcount ratio at national 14.62 9.90
 poverty line
 Proportion of seats held by women 8.53 0.00
 in national parliaments
 Population, total 27,200,000 5,450
Middle East & North Africa
 Primary school-age children out of 21.30 0.52
 school rate. Female
 Contributing family workers, female 17.93 0.00
 Vulnerable employment, total 16.88 0.40
 Poverty headcount ratio at national 9.87 3.80
 poverty line
 Proportion of seats held by women 6.56 0.00
 in national parliaments
 Population, total 16,800,000 108,401
North America
 Primary school-age children out of 2.29 0.10
 school rate. Female
 Contributing family workers, female 0.57 0.10
 Proportion of seats held by women 4.02 7.00
 in national parliaments
 Population, total 116,000,000 53,000
South Asia
 Primary school-age children out of 24.93 1.40
 school rate. Female
 Contributing family workers, female 21.19 3.40
 Vulnerable employment, total 15.21 29.60
 Poverty headcount ratio at national 11.10 15.20
 poverty line
 Proportion of seats held by women 7.94 2.00
 in national parliaments
 Population, total 281,000,000 115,772
Sub-Saharan Africa
 Primary school-age children out 24.63 0.74
 of school rate. Female
 Contributing family workers, female 25.94 0.40
 Vulnerable employment, total 31.26 3.50
 Poverty headcount ratio at national 14.49 10.10
 poverty line
 Proportion of seats held by women 9.20 0.00
 in national parliaments
 Population, total 18,200,000 36,755

 Variable Maximum

Total Sample
 Primary school-age children 95.38
 out of school rate. Female
 Contributing family workers, female 89.10
 Vulnerable employment, total 96.40
 Poverty headcount ratio at national 92.30
 poverty line
 Proportion of seats held by women 56.30
 in national parliaments
 Population, total 1,340,000,000
East Asia & Pacific
 Primary school-age children out of 45.30
 school rate. Female
 Contributing family workers, female 64.20
 Vulnerable employment, total 90.10
 Poverty headcount ratio at national 61.10
 poverty line
 Proportion of seats held by women 33.60
 in national parliaments
 Population, total 1,340,000,000
Europe & Central Asia
 Primary school-age children out of 25.43
 school rate. Female
 Contributing family workers, female 71.30
 Vulnerable employment, total 65.10
 Poverty headcount ratio at national 92.30
 poverty line
 Proportion of seats held by women 47.30
 in national parliaments
 Population, total 149,000,000
Latin America & Caribbean
 Primary school-age children out of 77.95
 school rate. Female
 Contributing family workers, female 40.20
 Vulnerable employment, total 66.10
 Poverty headcount ratio at national 77.00
 poverty line
 Proportion of seats held by women 43.20
 in national parliaments
 Population, total 195,000,000
Middle East & North Africa
 Primary school-age children out of 95.38
 school rate. Female
 Contributing family workers, female 56.10
 Vulnerable employment, total 58.10
 Poverty headcount ratio at national 40.10
 poverty line
 Proportion of seats held by women 31.50
 in national parliaments
 Population, total 81,100,000
North America
 Primary school-age children out of 9.71
 school rate. Female
 Contributing family workers, female 2.10
 Proportion of seats held by women 22.10
 in national parliaments
 Population, total 309,000,000
South Asia
 Primary school-age children out of 91.58
 school rate. Female
 Contributing family workers, female 77.30
 Vulnerable employment, total 85.00
 Poverty headcount ratio at national 56.60
 poverty line
 Proportion of seats held by women 33.20
 in national parliaments
 Population, total 1,170,000,000
Sub-Saharan Africa
 Primary school-age children out 93.17
 of school rate. Female
 Contributing family workers, female 89.10
 Vulnerable employment, total 96.40
 Poverty headcount ratio at national 81.60
 poverty line
 Proportion of seats held by women 56.30
 in national parliaments
 Population, total 158,000,000

Source: Author's elaboration using World Bank's World Development
Indicators Database (2011) and World Bank's Education Statistics
Database (2011). http://datacatalog.worldbank.org/

TABLE 3
Descriptive Statistics

 Primary school-
 age children out Contributing
 of school rate family workers

 Variable Total Female Total Female

Total sample
 High income: OECD 3.64 2.63 2.57 4.78
 High income: non-OECD 3.90 2.80 1.42 2.58
 Low income 36.14 34.60 28.82 39.61
 Lower middle income 18.94 16.79 14.44 22.89
 Upper middle income 11.21 9.22 6.68 10.70
East Asia & Pacific
 High income: OECD 5.24 10.92
 High income: non-OECD 1.00 1.91
 Low income 50.60 50.10 43.97 56.15
 Lower middle income 13.66 12.68 21.83 33.18
 Upper middle income 15.10 14.40 19.33 29.20
Europe & Central Asia
 High income: OECD 3.64 2.63 2.35 4.18
 High income: non-OECD 2.38 4.00
 Low income 7.50 7.03 15.78 21.53
 Lower middle income 21.16 19.64 10.19 12.70
 Upper middle income 12.61 10.94 9.01 16.88
Latin America & Caribbean
 High income: OECD
 High income: non-OECD 3.90 2.80 1.00 1.80
 Low income 33.40 29.60 11.90 11.20
 Lower middle income 15.86 11.48 10.59 11.57
 Upper middle income 9.36 6.99 4.14 5.53
Middle East & North Africa
 High income: OECD 0.41 0.73
 High income: non-OECD 0.05 0.17
 Lower middle income 12.50 10.32 14.60 32.09
 Upper middle income 5.06 9.26
North America
 High income: OECD 0.39 0.67
South Asia
 Low income 26.98 24.24 24.56 54.30
 Lower middle income 8.80 7.57 17.75 38.81
 Upper middle income 6.55 11.25
Sub-Saharan Africa
 High income: non-OECD 39.50 76.50
 Low income 39.27 37.72 30.74 40.45
 Lower middle income 28.35 27.22 19.83 28.79
 Upper middle income 21.57 20.54 4.35 6.01

 Proportion of
 seats in national
 parliaments
 Variable Held by women

Total sample
 High income: OECD 21.64
 High income: non-OECD 11.18
 Low income 12.88
 Lower middle income 10.11
 Upper middle income 14.12
East Asia & Pacific
 High income: OECD 16.82
 High income: non-OECD 13.54
 Low income 15.06
 Lower middle income 8.26
 Upper middle income 9.92
Europe & Central Asia
 High income: OECD 23.21
 High income: non-OECD 15.49
 Low income 10.30
 Lower middle income 11.66
 Upper middle income 13.96
Latin America & Caribbean
 High income: OECD
 High income: non-OECD 14.46
 Low income 3.88
 Lower middle income 13.20
 Upper middle income 16.07
Middle East & North Africa
 High income: OECD 12.83
 High income: non-OECD 3.58
 Lower middle income 7.32
 Upper middle income 6.05
North America
 High income: OECD 17.21
South Asia
 Low income 13.83
 Lower middle income 8.64
 Upper middle income 7.79
Sub-Saharan Africa
 High income: non-OECD 9.67
 Low income 13.17
 Lower middle income 11.30
 Upper middle income 18.90

Source: Author's elaboration using World Bank's World Development
Indicators Database (2011) and World Bank's Education Statistics
Database (2011). http://datacatalog.worldbank.org/

Note. OECD = Organisation for Economic Co-operation &
Development.

TABLE 4
Descriptive Statistics

 Population

 Variable Total Female Ages 0-14
Total sample
 High income: OECD 29,548,585 51.04 21.14
 High income: non-OECD 1,189,143 47.75 29.45
 Low income 1,4881,133 50.49 43.67
 Lower middle income 31,682,649 50.10 40.81
 Upper middle income 37,324,952 50.35 33.35
East Asia & Pacific
 High income: OECD 45,947,661 50.41 23.57
 High income: non-OECD 1,266,778 48.93 29.51
 Low income 22,776,484 51.10 35.45
 Lower middle income 20,196,383 49.32 40.69
 Upper middle income 238,100,000 49.31 32.74
Europe & Central Asia
 High income: OECD 18,554,854 51.21 20.27
 High income: non-OECD 530,138 50.88 20.30
 Low income 4,668,863 50.76 39.54
 Lower middle income 12,321,953 52.13 29.41
 Upper middle income 20,562,176 51.44 25.20
Latin America & Caribbean
 High income: non-OECD 568,787 51.31 28.94
 Low income 7,197,950 50.64 40.89
 Lower middle income 4,460,522 50.32 42.13
 Upper middle income 18,661,863 50.23 35.02
Middle East & North Africa
 High income: OECD 5,045,210 50.28 30.56
 High income: non-OECD 3,285,539 41.90 32.16
 Lower middle income 19,680,477 49.88 43.95
 Upper middle income 16,015,017 49.17 39.01
North America
 High income: OECD 140,900,000 50.68 21.92
 High income: non-OECD 58,864
South Asia
 Low income 48,747,619 48.82 42.86
 Lower middle income 245,700,000 48.61 37.67
 Upper middle income 216,408 48.24 42.19
Sub-Saharan Africa
 High income: non-OECD 402.619 49.62 40.59
 Low income 11,143,525 50.59 45.14
 Lower middle income 14,487,527 50.38 44.37
 Upper middle income 5,810,620 50.30 39.85

 Population Vulnerable
 employment
 Poverty
 headcount
 Variable ratio Total Female
Total sample
 High income: OECD 14.15 12.80 11.72
 High income: non-OECD 11.15 11.65 9.65
 Low income 52.94 77.95 82.95
 Lower middle income 42.17 47.75 52.63
 Upper middle income 29.21 29.38 29.61
East Asia & Pacific
 High income: OECD 14.50 14.71
 High income: non-OECD 7.57 5.39
 Low income 36.98 84.90 87.05
 Lower middle income 29.31 59.81 61.45
 Upper middle income 14.00 45.00 47.53
Europe & Central Asia
 High income: OECD 14.15 12.70 11.47
 High income: non-OECD 11.15 17.82 16.87
 Low income 59.67 49.62 49.54
 Lower middle income 36.78 47.05 47.29
 Upper middle income 19.64 23.46 25.65
Latin America & Caribbean
 High income: non-OECD 15.39 13.35
 Low income 77.00
 Lower middle income 52.20 44.11 48.89
 Upper middle income 35.65 30.82 29.53
Middle East & North Africa
 High income: OECD 7.31 5.16
 High income: non-OECD 6.57 3.27
 Lower middle income 22.43 35.60 44.47
 Upper middle income 9.97 19.72 20.23
North America
 High income: OECD
 High income: non-OECD
South Asia
 Low income 42.14 75.60 84.28
 Lower middle income 32.01 50.76 56.79
 Upper middle income 45.03 47.60
Sub-Saharan Africa
 High income: non-OECD
 Low income 54.40 82.69 88.58
 Lower middle income 50.18 64.35 73.96
 Upper middle income 39.70 22.22 25.63

Source: Author's elaboration using World Bank's World Development
Indicators Database (2011) and World Bank's Education Statistics
Database (2011). http://datacatalog.worldbank.org/

Note. OECD = Organisation for Economic Co-operation &
Development.

TABLE 5
Arellano-Bond Dynamic Panel-Data Estimation

 Europe &
 Total East Asia Central
 sample & Pacific Asia

Primary school-age children .5807747 -.6373511 5.237717
 out-of-school rate. Female (3.661537) (1.995393) (4.840322)
 (t-1)
Contributing family workers, .2410075 .2718573 -.5869098 *
 female (8.989808) (1.500356)
Vulnerable employment, total -.1096033 1.992803
 (8.989808) (1.500356)
Poverty headcount ratio at .0482351 -.6984193
 national poverty line (.5444332) (.8583749)
Proportion of seats held by .1320133 -.1338773 .2914276 **
 women in national parliaments (2.924875) (.3214143) (.1309497)
Population, total 8.67e-08 5.49e-09 3.93e-06
 (7.83e-07) (3.23e-07) (4.49e-06)
Constant -1.022577 5.641936 -165.2246
 (343.0793) (27.46188) (165.0323)
Wald test 21.10(6) 4.60(4) 895.64(6)
 p = 0.0018 p = 0.3303 p = 0.0000
Z1 -.29097 .24454 1.213
 p = 0.7711 p = 0.8068 p = 0.2251
Z2 .33551 -1.3442 .11813
 p = 0.7372 p = 0.1789 p = 0.9060
Number of observations 62 42 22
Number of countries 16 7 6
Number of instruments 61 43 23

 Middle
 Latin East &
 America & North North
 Caribbean Africa America

Primary school-age children .1069637 1.735705 -.7145833
 out-of-school rate. Female (.8839579) (2.522761) (1.096824)
 (t-1)
Contributing family workers, .2522918 -.8473913
 female (4.860935)
Vulnerable employment, total 1.099444
 (4.860935)
Poverty headcount ratio at -.1193502
 national poverty line (.3750561)
Proportion of seats held by -.4332201 16.8154
 women in national parliaments (1.173789) (16.90888)
Population, total 9.66e-09 -5.36e-06
 (3.17e-06) (6.25e-06)
Constant 14.01507 -20.8496 6.169449
 (60.3795) (83.19342) (4.534489)
Wald test 2.56(4) 34904.89(5) 0.42(1)
 p = 0.6334 p = 0.0000 p = 0.5147
Z1 -.10265 -1.0399 .35266
 p = 0.9182 p = 0.2984 p = 0.7243
Z2 .42416 .35869 -.59595
 p = 0.6715 p = 0.7198 p = 0.5512
Number of observations 37 33 22
Number of countries 8 5 2
Number of instruments 38 34 23

 Sub-
 Saharan
 South Asia Africa

Primary school-age children .4515872 1.946769
 out-of-school rate. Female (4.007345) (2.168624)
 (t-1)
Contributing family workers, -.7275889
 female
Vulnerable employment, total

Poverty headcount ratio at
 national poverty line
Proportion of seats held by 2.047856 5.09946
 women in national parliaments (15.88617) (4.817993)
Population, total -1.00e-07 -.0000165
 (8.25e-07) (.0000115)
Constant 3.144029 277.2494
 (583.0526) (162.2576)
Wald test 53.90(3) 1020.45(4)
 p = 0.0000 p = 0.0000
Z1 .11264 -.63949
 p = 0.9103 p = 0.5225
Z2 -.27875 .93073
 p = 0.7804 p = 0.3520
Number of observations 16 13
Number of countries 4 5
Number of instruments 17 14

Source: Author's elaboration using World Bank's World Development
Indicators Database (2011) and World Bank's Education Statistics
Database (2011). http://datacatalog.worldbank.org/

Note. Z Arellano-Bond test for zero autocorrelation in
first-differenced errors, where HO: no autocorrelation. Model
with WC-Robust Standard Error and Two-Step Results.