Child labour and educational success in Portugal

ARTICLE IN PRESS

Economics of Education Review 27 (2008) 575–587 www.elsevier.com/locate/econedurev

Child labour and educational success in Portugal Pedro Goularta,b, Arjun S. Bedib, a

CISEP-ISEG, Technical University of Lisbon, Portugal Institute of Social Studies, Kortenaerkade 12, 2518 AX, The Hague, The Netherlands

b

Received 28 February 2006; accepted 17 July 2007

Abstract The current debate on child labour focuses on developing countries. However, Portugal is an example of a relatively developed country where child labour is still a matter of concern as between 8% and 12% of Portuguese children may be classified as workers. This paper studies the patterns of child labour in Portugal and assesses the consequences of working on the educational success of Portuguese children. The analysis controls for typically unobserved attributes such as a child’s interest in school and educational ambitions and uses geographical variation in policies designed to tackle child labour and in labour inspection regimes to instrument child labour. We find that economic work hinders educational success, while domestic work does not appear to be harmful. Furthermore, after controlling for a host of socio-economic variables, factors such as a child’s interest in school and educational ambitions have a large effect on boosting educational success and reducing economic work. r 2007 Elsevier Ltd. All rights reserved. JEL classification: J23; J24; O15 Keywords: Child labour; Human capital; Educational ambitions

1. Introduction Historically, the development of countries has been associated with a long-run decline in child employment (see Cunningham & Viazzo, 1996). Accordingly, the current focus in the child labour debate is on conditions faced by children in developing countries, even though there are examples of relatively developed countries such as Portugal which still struggle with the issue of working children.

Corresponding author. Tel.: +31 70 426 0493; fax: +31 70 426 0799. E-mail address: [email protected] (A.S. Bedi).

0272-7757/$ - see front matter r 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.econedurev.2007.07.002

Since the early 1990s, child labour in Portugal has attracted considerable attention. For example, a 1992 report by Anti-Slavery International (Williams, 1992) estimated that there were 200,000 working children in Portugal. In part, as a consequence of the controversy generated by these numbers the Government conducted household surveys, in 1998 and 2001, to provide credible information on working children. These surveys revealed that about 8–12% of Portuguese children in the age group 6–15 were involved in some form of economic or domestic work (see Table 1).1 While 1 Domestic labour consists of domestic chores and economic labour refers to paid or unpaid activities performed on the family farm/enterprise or for an employer.

ARTICLE IN PRESS 576

P. Goulart, A.S. Bedi / Economics of Education Review 27 (2008) 575–587

Table 1 Incidence of child work in Portugal (%)a (absolute number of children working) 1998

2001

Economic work Economic work—outside the household Economic work—within the household

3.13 (33,792) 0.68 (7342)

3.70 (44,003) 0.73 (8689)

2.45 (26,450)

2.97 (35,314)

Both economic and domestic work Domestic work—within the household

0.86 (9285)

0.43 (5130)

7.68 (83,037)

4.05 (48,165)

Total

11.67 (126,114) 8.18 (97,298)

a

Incidence is defined as the percentage of all children in the age group 6–15 who report at least 1 hour of work per week. Estimates of the absolute number of working children working are based on weighting the sample data to obtain population figures.

this figure is lower than the 20–25% participation rate suggested by Williams (1992), it is higher than ILO (2002) estimates of the average work participation rates in developed (2%) and transition countries (4%). The resilience of child labour despite overall economic progress and efforts to tackle the issue suggests that Portugal’s economic and cultural characteristics still generate a favourable environment for child labour.2 Per se, a work incidence of 8–12 may not be a matter of concern. However, an issue of concern is, whether the work activities of Portuguese children hampers their educational performance? This paper assesses the factors that determine both these outcomes and examines whether the work activities of children has a causal impact on their educational success.3 2

The spread of education and the enforcement of compulsory education laws is a relatively recent (post-1974) phenomenon in Portugal. Children are expected to start school at the age of 6 and to continue till they are 15. Consistent with these educational requirements, minors are only allowed to work under three conditions—they are at least 16 years old, they have completed compulsory school and there is medical confirmation of their physical and psychological capabilities for that job. 3 The measure of educational success used in this paper is a binary variable that takes on a value 1 if a child has never repeated a grade in school and 0 otherwise. We focus on educational success rather than enrolment or attendance as almost all children are formally enrolled in school (98.5%) and appear to be attending school regularly (97.8% of children do not miss school more than once a week). In contrast, 25% of students have repeated a grade.

The motivation for our work stems from the potential consequences of the early labour force entry of children on their educational success. For an individual, lower educational attainment translates into a life-long handicap, leading to a lower probability of employment and access to low-paying jobs. Beyond the individual, in an enlarged and increasingly competitive environment within the European Union (EU), the ability of Portugal to compete depends on a well-educated labour force. However, with functional literacy at 52% (OECD, 2000) and low levels of educational achievement Portugal continues to lag behind its EU counterparts (OECD, 2003). There are several notable features of our study. First, for obvious reasons there is little work on developed countries. Examining this issue in the context of a relatively high-income country may provide guidance on useful policies for developing countries. Second, the bulk of the child work— educational outcome literature focuses on the correlation between these two outcomes.4 In our work, we attempt to identify the causal effect of child work on educational success. This is similar to the more recent literature on developing (Beegle, Dehejia, & Gatti, 2003; Boozer & Suri, 2001; Gunnarsson, Orazem, & Sanchez, 2003) and developed (Stinebrickner & Stinebrickner, 2003; Tyler, 2003) countries which uses an instrumental variables (IV) estimation strategy to identify the impact of child work. Third, in addition to the econometric strategy we have information on unusual educational related measures such as a child’s interest in school and educational ambitions. Thus, we are able to control for typically unobserved attributes of children and use an IV strategy. Fourth, we draw a distinction between domestic and economic work and assess the influence of these two types of work on the educational success of children. The following section provides a descriptive analysis of child labour and educational success. Section 3 lays out our analytical approach, Section 4 discusses the data and model specification. Section 5 presents estimates and Section 6 concludes.

4 Examples of this first generation literature are Patrinos and Psacharopoulos (1995), Psacharopoulos (1997), Jensen and Nielsen (1997).

ARTICLE IN PRESS P. Goulart, A.S. Bedi / Economics of Education Review 27 (2008) 575–587

577

Table 2 Incidence of child work by regions (%) North

Centre

Lisbon

Alentejo

Algarve

Azores

Madeira

1998 Economic work Domestic work

4.3 10.0

4.7 11.4

1 3.3

1.5 3.3

1.5 3.2

n.a. n.a.

n.a. n.a.

Combined

14.3

16.1

4.3

4.8

4.7

n.a.

n.a.

4.8 7.2

5.8 3.8

.4 1.0

2.8 0.9

2.3 1.3

3.4 4.4

0.29 1.06

9.6

2.4

3.7

3.6

7.8

1.35

2001 Economic Work Domestic Work Combined

12

Note: The 1998 survey did not cover the Azores and the Madeira regions. Table 3 Educational indicators by working status

Enrolment (%) Attendancea (%) School successb (%) a

Does not work

Economic work

Domestic work

Economic and domestic work

99.3 97.9 76.4

84.9 81.1 48.8

95.0 91.6 61.0

90.1 86.6 55.2

Attendance ¼ 1 if a child misses school less than once a week. School success ¼ 1 if a child has never repeated a grade.

b

2. Child labour and educational success in Portugal—descriptive analysis Table 1 presents a break-down of the incidence of child work and the absolute number of working children in four mutually exclusive categories. In 1998, about 12% or 126,000 Portuguese children were involved in some form of work while it fell to about 8% (about 97,000) in 2001. Across the 2 years, the incidence of economic work does not change sharply (3.1% in 1998 and 3.7% in 2001). However, there is a decline in the number of children involved in domestic work. While the decline seems promising, it is apparent rather than real as there was a change in the information gathering process between the two surveys.5 Most economic work is carried out in the context of a family farm or enterprise and only about 9% of 5 In 2001, the question requesting information on child work activities was adjusted from ‘‘Do you perform domestic chores?’’ to ‘‘Do you perform domestic chores in excess?’’, with the definition of excess being left to the subjective judgement of the respondent. Thus, by definition, the 4% of Portuguese children contributing excess domestic work should be treated as child workers.

child workers work outside the household in a formal employer–employee relationship. About 5% combine economic and domestic work, and accordingly, the paper focuses on children who provide either economic or domestic work. There are clear regional differences (Table 2). Both forms of labour are highest in the Northern and Central parts of the country. Northern Portugal is a rocky, mountainous region characterized by small family farms and vineyards. Typically, families in this region espouse traditional values of hard work and thrift and have lived modestly on their farms for several generations (for details, see Alves Pinto, 1998). While more heterogeneous, Central Portugal is also characterized by small- and medium-sized farms with light industry and services. The presence of small family farms and firms, and larger families, reminiscent of developing countries, probably promotes child work in these regions. Patterns of educational enrolment, attendance and school success by work status are displayed in Table 3. Children who do not work enjoy a 10 percentage point enrolment and attendance advantage as compared to working children. Our measure of educational success is a binary variable which

ARTICLE IN PRESS 578

P. Goulart, A.S. Bedi / Economics of Education Review 27 (2008) 575–587

indicates whether a child has never repeated a grade. While information on test scores may be a better measure of educational performance, there is a tight link between test scores, failure and repetition. Usually, based on a child’s competencies and scores on formal tests, a school level council determines whether a child should be allowed to proceed to a higher grade or should repeat a grade. The data show that 76% of non-working children have never repeated a grade while the corresponding number for working children is 55%. 3. Analytical approach Following a large educational production function literature we treat the test scores of children (Y) as a function of child (C), family (F), socioeconomic (SE), educational (E), and demand for labour characteristics (D). In addition, we include hours of economic and domestic work (W) contributed by children. That is, Y i ¼ C i bc þ F i bF þ SEi bSE þ E i bE þ Di bD þ W i bW þ i .

adequately interested in school and four variables which capture the educational ambitions of children. These variables may be viewed as proxies for a child’s educational ability and motivation. Second, we include these ability controls and use a two-stage IV strategy. Following, Vella (1993), reduced form expressions for the two types of work may be written as, W i ¼ C i bc þ F i bF þ SEi bSE þ E i bE þ Di bD þ Pi bP þ LIi bLI þ vi .

ð3Þ

In addition to the variables in (2), this specification includes a set of variables that captures geographical variation in child labour policies (P) and in the labour inspection regime (LI). In the first stage, we estimate the hours of work equations using tobit models. These estimates are used to construct generalised residuals (l) of the form: ( ) ^ fðX i bÞ ^li ¼ sð1 ^  I iÞ þ I i ðW i  X i b^ i Þ, ^ ð1  FðX i bÞÞ (4)

While we do not observe test scores, we do observe whether a child has achieved educational success, that is, whether a child has never repeated grades.6 When test scores obtained by a student cross a certain threshold we observe school success (Y ¼ 1). Thus, the probability that a child succeeds is,

where s^ and b^ are tobit maximum likelihood estimates of the parameters in the hours of work equations, Xi represents all the explanatory variables in (3), Ii indicates whether a child works or not and f(  ) and F(  ) denote the probability density and cumulative distribution function of the standard normal distribution. In the second stage the generalised residuals are included in (2), yielding,

Prob½Y i ¼ 1 ¼ Prob½C i bc þ F i bF þ SEi bSE þ E i bE þ Di bD þ W i bW þ i 40. ð2Þ

Y i ¼ C i bc þ F i bF þ SEi bSE þ E i bE þ Di bD þ W i bW þ l^ i d þ i . ð5Þ

Assuming a normally distributed error term allows estimation of (2) using a probit model. The main econometric concern about (2) is that unobserved factors that determine school success and hours worked may be correlated. Our empirical strategy to control for this correlation consists of two parts. First, we include a set of unusual educational related variables in the educational success equation. These include two variables that capture whether a child is very interested or

This augmented probit equation provides consistent estimates (Rivers & Vuong, 1988; Vella, 1993). A test of the null hypothesis that the coefficients on the generalised residuals are zero, is a (Hausman) specification test for the exogeneity of Wi. We rely on two sets of potential instruments to implement IV.7 The first set captures geographical variation in the implementation of plans to tackle child labour. Under the aegis of PETI, a programme designed to reduce the supply of child labour, multidisciplinary teams have been set up across the

ð1Þ

6

The school success–child work relationship is based on all the children in our sample and not a select sample of children who are still enrolled in school. The school enrollment rate in our sample is 98.5% and information on grade repetition is available for all children regardless of whether they are currently enrolled in school or not. School enrollment and regular school attendance are almost universal and accordingly the appropriate concern is the educational performance of children.

7 The data to fashion the instruments were obtained from PETI and from the labour inspectorate. The PETI data are for 2002 and the labour inspectorate data are for 2001. We are forced to use 2002 data for the PETI variables as we were unable to get information for 2001 or for earlier years. However, given that our survey data are from October 2001 and the PETI data are for early 2002 the use of these data should not pose a problem.

ARTICLE IN PRESS P. Goulart, A.S. Bedi / Economics of Education Review 27 (2008) 575–587

country. These teams work with children, families and local authorities to raise awareness, monitor and prevent child labour and more pro-actively, to design tailor-made training programs to help working children or children considered at risk to acquire certain competencies. These training programs are delivered through centres (PIEF centres) that are set up in co-operation with the local administration. Despite the desire to set up centres in all counties, till 2002, only about 45% of the children in our sample lived in counties with centres (see Table 4). Whether a county has set up such a centre may be interpreted as a signal of the local administration’s willingness to devote resources to tackling child labour. A second variable is the total number of children in a county divided by the number of members in the multi-disciplinary teams (the number of members ranges from 1 to 8). While the presence of a PETI centre in a county may be negatively correlated with child labour, the ratio of youngsters per PETI member is likely to be positively correlated with child labour. The second set of instruments is based on Portugal’s labour inspection regime. All counties have labour inspectors who are charged with ensuring that the labour laws are followed in the firms that lie in their territory. We gathered comprehensive information from the labour inspection office on the number of inspectors, number of firms, number of workers, the frequency of the inspection regime, the number of serious illegalities detected and the total fines charged for labour illegalities. Based on this information we created two variables that capture inter-county variation in the strictness of the labour inspection regime. These are the total number of illegalities detected per worker and the average fine per illegality.8 Both sets of instruments may be expected to influence child labour while they should have no direct bearing on educational success.9 8 The idea was to generate variables that capture the probability that an illegality is detected and the fine per illegality. To create the former, we would have liked information on the total number of illegalities. However, this is not observed and thus we use the total number of workers in the county as a proxy for the total number of illegalities. Average fine per illegality is obtained by dividing the total provisional fines by the total number of detected illegalities. The actual fine may differ as it depends on the fine imposed by the courts. The fines that may be imposed lie in a range and we use the amount that corresponds to the minimum fine for the illegality. 9 While merging the county level information with the household survey provides instruments, it comes at a cost. Information

579

4. Data, specification and descriptive statistics Our paper relies on information contained in household surveys conducted in 1998 and 2001. While both surveys are utilised to analyse the incidence of child labour, we use the 2001 survey for our econometric work. Apart from being a more recent survey, the 2001 data has wider geographical coverage. In the 2001 survey, 19,849 households were interviewed and we focus on about 26,000 respondents in the age group 6–15.10 The school success and hours of work equations are specified as functions of child, family, socio-economic, education and labour demand characteristics. The detailed list of variables along with descriptive statistics is in Table 4. The discussion here is restricted to the variables that we wish to highlight in our analysis. Amongst the educational characteristics are two sets of variables that capture a child’s interest and educational ambitions. Parents were asked to provide information on their child’s interest in school, that is, whether a child is very interested, shows adequate interest or has no interest in school (omitted category).11 The educational ambition question asked children about the educational level that they would like to achieve and is divided into five categories— tertiary, upper secondary, compulsory, less than compulsory education and children who are not sure about their school ambitions (omitted category). (footnote continued) on the implementation of child labour policies and labour inspection regimes is not available for Azores and Madeira and this reduces the number of observations from about 26,000 to 24,000. 10 It should be noted that the two data sets are independent cross-sections and do not constitute a panel. 11 Parents and children provide information on the educational interest questions. For 62% of the children the two sets of responses coincide. In terms of distribution, according to parents, 55.5% of the children are very interested in school, 33.3% are adequately interested while 11.9% are not interested. The corresponding figures based on children’s responses are 58.5%, 36.9% and 4.6%. Thus, parents are more likely to indicate that their children are not interested in school as compared to the children themselves. This pattern persists for children of all ages and the gap between parental and child responses increases with age. It is likely that time allocation decisions are taken mainly by parents and hence it is their perception of a child’s interest in school that matters in terms of determining the time that a child spends on work activities. Accordingly, we use the responses provided by parents in the models of child work. However, we also estimated regressions using the information provided by children. While there are differences, regardless of whether we use parental or child responses in the work and educational success equations, the overall flavour of the results remains unchanged.

ARTICLE IN PRESS 580

P. Goulart, A.S. Bedi / Economics of Education Review 27 (2008) 575–587

Table 4 Descriptive statistics Variables

Mean

Standard deviation

Min.

Max.

Child work Incidence of economic work ¼ 1 Incidence of domestic work ¼ 1 Hours of economic work in reference week Hours of domestic work in reference week

0.037 0.041 0.515 0.339

0.187 0.197 3.74 2.41

– – 0 0

– – 56 56

Child characteristics Sex (male ¼ 1) Age

0.513 10.89

0.499 2.794

– 6

– 15

4.412 0.231

1.312 0.421

2 –

12

0.342 0.154

0.474 0.361

– –

– –

4.429 0.137 3.943

1.628 0.344 1.209

1 – 1

7 – 10

0.291 0.623

0.454 0.485

– –

– –

0.746 0.725 13.27 0.375 0.556 0.018 0.103 0.206 0.523

0.435 0.446 10.17 0.484 0.497 0.133 0.304 0.404 0.499

– – 7.5 – – – – – –

– – 60 – – – – – –

0.448 0.632 0.125 0.083 0.085

0.497 0.482 0.331 0.278 0.278

– – – – –

– – – – –

0.446 44,464

0.497 27,145

– 12,580

– 100,321

1.97 1177.8

1.47 601.41

0 485

7 4204

Family characteristics Household size Female-headed household ¼ 1 Schooling of household head 5–9 years ¼ 1 49 years ¼ 1 Socio-economic characteristics Household income (1–7, increasing in income) Reduction in income ¼ 1 Number of rooms in dwelling Housing conditions Adequate ¼ 1 Good ¼ 1 Educational characteristics School success ¼ 1 Pre-school attendance ¼ 1 Time to reach school (in minutes) Interest in school—adequate ¼ 1 Interest in school—very interested ¼ 1 School ambition, ocompulsory ¼ 1 School ambition, compulsory ¼ 1 School ambition, upper secondary ¼ 1 School ambition, tertiary ¼ 1 Demand characteristics Backyard ¼ 1 Occupational status of household head-wage labour ¼ 1 Occupational status of household head-self employed ¼ 1 Occupational status of household head-employer ¼ 1 Household employs domestic worker Child labour policies Counties with a PIEF centre ¼ 1 Children per PETI member Labour inspection Serious illegalities per 1000 workers Fine per illegality (in Euros)

In standard economic analyses, variables such as interest and ambition fall in the category of unobserved attributes and are often ignored. In contrast, sociological examinations of educational success often include such measures.12 Given Portugal’s level of 12

The role of educational aspirations in determining attainment and the formation of such aspirations has been a lively area of research in sociology since the work of Kahl (1957). Early

economic development and the persistence of child labour, it is likely that these psycho-social factors are important in explaining observed outcomes. Accordingly, we include these variables in some of our (footnote continued) examples of empirical work which incorporate such types of information include, Sewell and Hauser (1972), Alexander, Eckland, and Griffin (1975), Otto and Haller (1979).

ARTICLE IN PRESS P. Goulart, A.S. Bedi / Economics of Education Review 27 (2008) 575–587

specifications and treat them as proxies for the unobserved abilities of a child. An immediate concern is that a child’s educational ambitions/interest and educational success are likely to be simultaneously determined. We are sensitive to this additional source of endogeneity and present estimates with and without the inclusion of these educational attributes. Several variables which may influence demand for labour are included. These are a variable indicating ownership of a backyard or small farm, a set of three variables (self-employed, employer and wage labourer) to capture the potential links between the occupation of the household head and child work and a variable indicating household employment of domestic help. Table 5 contains selected descriptive statistics conditional on working status. In terms of socio-economic

581

conditions, domestic child workers clearly belong to families that are better-off as compared to families with economic child workers. With regard to educational characteristics, while 60% of nonworking children are very interested in schooling, the corresponding number is 42% for children involved in economic work. The interest gap is not as pronounced between non-working children and domestic workers (60% versus 55%). Similarly, differences in the educational ambitions of nonworkers and domestic workers are not as sharp as the differences between non-workers and economic workers. While more than 50% of non-workers/ domestic workers aspire to reach tertiary education, less than a third of working children share the same aspirations. The labour demand variables show that farm ownership is substantially higher

Table 5 Selected descriptive statistics Variables

Child does not work

Economic work

Domestic work

Mean

Standard deviation

Mean

Standard deviation

Mean

Standard deviation

Child characteristics Sex (male ¼ 1) Age Hours of work

0.517 10.76 –

0.500 2.794 –

0.735 12.54 14.05

0.442 2.387 13.96

0.261 12.238 8.36

0.439 2.268 8.77

Family characteristics Schooling of household head, 5–9 years ¼ 1 Schooling of household head 49 years ¼ 1 Years worked by household head till age 12

0.350 0.164 0.374

0.477 0.370 1.034

0.236 0.020 0.910

0.425 0.139 1.666

0.263 0.063 0.586

0.440 0.242 1.337

4.482 0.135

1.627 0.341

3.704 0.156

1.523 0.363

3.980 0.161

1.504 0.367

0.287 0.633

0.452 0.482

0.337 0.506

0.473 0.500

0.339 0.528

0.474 0.499

Educational characteristics Pre-school attendance ¼ 1 Interest in school—adequate ¼ 1 Interest in school—very interested ¼ 1 School ambition, ocompulsory ¼ 1 School ambition, compulsory ¼ 1 School ambition, upper secondary ¼ 1 School ambition, tertiary ¼ 1

0.738 0.373 0.567 0.012 0.093 0.203 0.532

0.439 0.483 0.495 0.110 0.291 0.402 0.499

0.559 0.411 0.333 0.133 0.276 0.224 0.306

0.497 0.492 0.471 0.340 0.447 0.417 0.461

0.599 0.383 0.481 0.035 0.150 0.228 0.514

0.490 0.486 0.499 0.185 0.358 0.420 0.500

Demand characteristics Occupational status of household head-wage labour ¼ 1 Occupational status of household head-self employed ¼ 1 Occupational status of household head-employer ¼ 1 Backyard ¼ 1 N

0.641 0.120 0.082 0.432 24,006

0.480 0.325 0.275 0.495 –

0.444 0.257 0.130 0.674 968

0.497 0.437 0.337 0.468 –

0.619 0.120 0.064 0.580 1071

0.486 0.325 0.246 0.493 –

Socio-economic characteristics Household income (1–7) Reduction in income ¼ 1 Housing conditions Adequate ¼ 1 Good ¼ 1

ARTICLE IN PRESS 582

P. Goulart, A.S. Bedi / Economics of Education Review 27 (2008) 575–587

among children who provide economic work (67%) as compared to non-workers (43%). In terms of parental occupational status, parents of domestic workers and non-workers are equally likely to be wage employees while parents of economic workers are more likely to be employers or self-employed.

5. Regression results and discussion 5.1. Economic and domestic work Estimates of the probability of working are in Table 6.13 There are clear gender differences and as children age their work contributions increase. The higher educational attainment of parents is associated with a lower probability of working. Even though our specification contains parental education and family wealth variables we find that children of parents who worked in their pre-teen years are 0.2 percentage points more likely to contribute to economic work. The income and wealth variables have expected signs. Transitory income shocks have little bearing on work, suggesting that child labour is a structural phenomenon. For both economic and domestic work, children who are interested in school are 0.4–0.6 percentage points less likely to work. The educational ambition variables show marked variation across the four ambition categories and across the two types of work. Children falling in the lowest educational ambition category are 6 percentage points more likely to provide economic work as compared to those whose educational ambitions are unknown while the marginal effect for those with the highest ambition is about 0.6 percentage points. The effect of the ambition variables is much smaller among domestic workers and there are no clear patterns across ambition levels. On the demand side, family ownership of a small farm calls for additional (0.4–0.8 percentage points) labour effort. Being self-employed or an employer is associated with a 2–3 percentage point increase in the probability that a child is an economic worker while the effect of these variables on domestic work is negligible. The presence of a hired domestic worker reduces domestic and economic work. 13 Several tobit specifications of the duration of work were also estimated. Since the story emerging from the probit and tobit estimates were similar, only the probit estimates are reported in the text. The tobit estimates are available in our working paper (Goulart & Bedi, 2005).

Across all specifications the presence of a PETI centre is negatively linked to child labour. The coefficient is statistically significant and the effect ranges between 0.6 and 1.7 percentage points. The number of children per PETI member does not have a clear-cut effect. Consistent with expectations, both the labour inspection regime variables are negatively linked to child labour. While the probability of detection does not have a statistically significant effect, the effect of fine per illegality is stable across specifications and statistically significant at conventional levels. A Euro 1000 increase in fine per illegality is likely to reduce child work by about 0.3 percentage points. To serve as valid instruments these variables need to be correlated with patterns of work while at the same time they should be valid exclusions from the educational success equation. While the individual statistical significance of these variables differs across specifications, jointly they are statistically significant in all the probit models presented in Table 6 (the Chi-square test-statistics range between 23 and 134 and the p-values are always less than 0.0001). There is little reason to expect that these variables have a direct bearing on educational success and statistical tests support this view. Tests for the inclusion of the child labour policy variables in the educational success equations which exclude (Table 7, specification 1) or include the educational interest variables (Table 7, specification 3) recorded p-values between 0.92 and 0.95. While, tests for the inclusion of the child labour policy and labour inspection variables recorded p-values between 0.14 and 0.42. 5.2. School success As displayed in Table 7, male children are less likely to succeed while belonging to a female-headed household reduces the chances of educational success by 2.1–2.6 percentage points. The presence of a well-educated household head (more than 9 years) boosts educational performance by about 11–15 percentage points. The income and wealth variables have expected effects. The inclusion of the educational characteristics leads to several interesting changes. A comparison of specification 1 and 3 shows that the negative effect of being male is now lower, the effect of parental schooling and household income and wealth is also considerably lower. The interest and aspirations of children is strongly linked to their educational success. Children who

ARTICLE IN PRESS P. Goulart, A.S. Bedi / Economics of Education Review 27 (2008) 575–587

583

Table 6 Probit marginal effect (ME) estimates of the probability of working Variables

Economic CL

Economic CL

Domestic CL

Domestic CL

Total CL

ME

Std. error

ME

Std. error

ME

ME

Std. error

ME

Std. error

0.001 0.002 0.001

0.012 0.004 0.001

0.001 0.002 0.001

0.027 0.018 0.006

0.002 0.003 0.001

0.025 0.018 0.006

0.002 0.002 0.001

0.017 0.027 0.008

0.003 0.004 0.001

Child characteristics Sex (male ¼ 1) Age Age squared  10

0.018 0.001 0.001

Std. error

Family characteristics Household size Female-headed household ¼ 1 Schooling of HH—5–9 years Schooling of HH—49 years ¼ 1 Years worked by HH till age 12

0.003 0.006 0.006 0.013 0.002

0.001 0.002 0.001 0.002 0.0005

0.003 0.004 0.003 0.009 0.002

0.0005 0.002 0.001 0.002 0.0004

0.003 0.002 0.004 0.005z 0.0004

0.001 0.002 0.002 0.003 0.001

0.002 0.003 0.003 0.004 0.0003

0.001 0.002 0.001 0.003 0.001

0.009 0.012 0.007 0.014 0.005

0.001 0.004 0.003 0.005 0.001

Socio-economic characteristics Household income Reduction in income  10 ¼ 1 Number of rooms in house100 Adequate housing conditions ¼ 1 Good housing conditions ¼ 1

0.003 0.001 0.008 0.012 0.006

0.0005 0.002 0.001 0.003 0.002

0.002 0.002 0.040 0.007 0.004

0.0005 0.002 0.001 0.002 0.002

0.002 0.000 0.002 0.006 0.004

0.001 0.000 0.001 0.003 0.002

0.002 0.002 0.002 0.004 0.003

0.001 0.002 0.001 0.003 0.002

0.005 0.002 0.002 0.018 0.012

0.001 0.003 0.001 0.005 0.004

Educational characteristics Pre-school attendance Time to reach school Adequate interest in school ¼ 1 Very interested in school ¼ 1 School ambition o compulsory ¼ 1 School ambition, compulsory ¼ 1 School ambition, upper sec. ¼ 1 School ambition, tertiary ¼ 1

– – – – – – – –

– – – – – – – –

0.004 0.001 0.001 0.004y 0.062 0.028 0.011 0.006

0.001 0.001 0.002 0.002 0.019 0.006 0.003 0.003

– – – – – – – –

– – – – – – – –

0.004 0.001 0.006y 0.006 0.004 0.006z 0.005z 0.003

0.002 0.001 0.003 0.003 0.009 0.004 0.003 0.003

0.011 0.001 0.011 0.014 0.081 0.048 0.021 0.013

0.003 0.002 0.004 0.004 0.023 0.009 0.006 0.005

Demand characteristics Backyard ¼ 1 HH works as wage labour ¼ 1 HH is self-employed ¼ 1 HH is employer ¼ 1 Domestic house worker hired ¼ 1

0.008 0.001 0.025 0.033 0.010

0.002 0.002 0.005 0.006 0.002

0.006 0.000 0.023 0.031 0.008

0.001 0.002 0.005 0.006 0.002

0.004 0.002 0.001 0.001 0.008

0.001 0.002 0.001 0.001 0.003

0.004 0.002 0.001 0.001 0.007y

0.002 0.002 0.003 0.003 0.003

0.016 0.002 0.028 0.035 0.021

0.003 0.004 0.006 0.008 0.004

Pro-active child labour policies Counties with a PIEF centre ¼ 1 Children per PETI member  100,000

0.006 0.005

0.002 0.003

0.006 0.007

0.002 0.003

0.017 0.000

0.002 0.000

0.017 0.000

0.002 0.000

0.034 0.010y

0.003 0.005

Labour inspection Serious illegalities per worker  10 Fine per illegality

0.010 0.003y

0.001 0.001

0.004 0.003

0.001 0.001

0.002z 0.003z

0.001 0.002

0.001 0.003z

0.009 0.002

0.024z 0.009

0.001 0.003

N Log likelihood

24,382 3171.25

24,031 2811.12

24,382 3517.60

24,031 3369.67

24,031 5457.25

Notes: Other variables included in the specification are a set of regional indicators for the province of residence, indicators for residing in urban, semi-rural, rural areas, variables capturing county level regional unemployment and the proportion of individuals working in the primary, secondary and tertiary sectors. Standard errors are heteroscedasticity consistent.  Significant at the 1% level. y Significant at the 5% level. z Significant at the 10% level.

are extremely interested in schooling are 24 percentage points more likely to succeed as compared to those who have no interest. Children with

educational ambitions up to the compulsory level are 24 percentage points less likely to be successful than those whose educational ambitions are

ARTICLE IN PRESS 584

P. Goulart, A.S. Bedi / Economics of Education Review 27 (2008) 575–587

Table 7 Probit marginal effect (ME) estimates of the probability of school success Variables

Specification 1

Specification 2

Specification 3

Specification 4

ME

Std. error

ME

Std. error

ME

Std. error

ME

Std. error

Child characteristics Sex (male ¼ 1) Age Age squared  10

0.108 0.138 0.004

0.005 0.008 0.0004

0.109 0.142 0.004

0.005 0.008 0.0004

0.047 0.161 0.005

0.005 0.009 0.0004

0.049 0.160 0.005

0.005 0.009 0.0004

Family characteristics Household size Female headed household ¼ 1 Schooling of HH—5–9 years Schooling of HH— 49 years ¼ 1 Years worked by HH till age 12

0.039 0.026 0.090 0.152 0.007

0.002 0.007 0.005 0.006 0.002

0.037 0.025 0.089 0.151 0.007

0.002 0.007 0.005 0.006 0.002

0.029 0.021 0.063 0.112 0.006

0.002 0.006 0.005 0.006 0.002

0.028 0.021 0.064 0.112 0.006y

0.002 0.006 0.005 0.006 0.002

Socio-economic characteristics Household income Reduction in income ¼ 1 Number of rooms in house  100 Adequate housing conditions ¼ 1 Good housing conditions ¼ 1

0.036 0.027 0.022 0.106 0.061

0.002 0.007 0.002 0.010 0.008

0.035 0.026 0.022 0.103 0.059

0.002 0.008 0.003 0.010 0.008

0.027 0.020 0.015 0.076 0.046

0.002 0.007 0.002 0.010 0.008

0.027 0.020 0.015 0.076 0.046

0.002 0.007 0.002 0.010 0.008

Educational characteristics Pre-school attendance Time to reach school Adequate interest in school ¼ 1 Very interested in school ¼ 1 School ambition ocompulsory ¼ 1 School ambition, compulsory ¼ 1 School ambition, upper sec. ¼ 1 School ambition, tertiary ¼ 1

– – – – – – – –

– – – – – – – –

– – – – – – – –

– – – – – – – –

0.011 0.009 0.116 0.237 0.243 0.162 0.005 0.119

0.006 0.003 0.010 0.014 0.037 0.014 0.009 0.009

0.011 0.009 0.116 0.237 0.239 0.160 0.006 0.120

0.006 0.003 0.010 0.014 0.037 0.014 0.009 0.009

Demand characteristics Backyard ¼ 1 HH works as wage labour ¼ 1 HH is self-employed ¼ 1 HH is employer ¼ 1 Domestic house worker hired ¼ 1

0.012y 0.018 0.028 0.038 0.055

0.006 0.007 0.009 0.011 0.012

0.014y 0.019 0.031 0.039 0.054

0.006 0.007 0.009 0.011 0.012

0.023 0.010 0.034 0.036 0.036

0.005 0.007 0.009 0.010 0.012

0.024 0.011 0.034 0.037 0.036

0.005 0.007 0.009 0.010 0.012

Child work Hours of economic work Hours of domestic work

– –

– –

0.004 0.005

0.0006 0.001

– –

– –

0.003 0.004

0.0009 0.001

N Log likelihood

26,429 11493.50

26,429 11458.07

26,045 10128.08

26,045 10114.80

Notes: Other variables included in the specification are a set of regional indicators for the province of residence, indicators for residing in urban, semi-rural, rural areas, variables capturing county level regional unemployment and the proportion of individuals working in the primary, secondary and tertiary sectors. Standard errors are heteroscedasticity consistent.  Significant at the 1% level. y Significant at the 5% level.

unknown, while those with higher educational ambitions (tertiary level) are 12 percentage points more likely to succeed. Notwithstanding the large set of socio-economic controls, it is likely that the estimates presented here exaggerate the impact of ambition/interest on educational success. While

acknowledging this possibility, it would be incorrect to ignore such variables as they clearly show that, psychological factors such as the aspirations of children, however they may be formed, play a large role in determining their educational success and labour effort.

ARTICLE IN PRESS P. Goulart, A.S. Bedi / Economics of Education Review 27 (2008) 575–587

585

Table 8 School success and hours of work: marginal effect probit and instrumental variable probit estimates (Std. error)

Hours of economic work Hours of domestic work Generalised residual— economic work Generalised residual— domestic work N Log likelihood

1 Probit

2 IVPa

3 IVPb

4 IVPc

5 Probit

6 IVPa

7 IVPb

8 IVPc

0.004 (0.0006) 0.005 (0.001) .

0.002y (0.001) 0.002 (0.002) 0.0007y (0.0003) 0.0013y (0.0006)

0.002y (0.001) 0.002 (0.002) 0.0007y (0.0003) 0.0013y (0.0005)

0.002y (0.001) 0.003 (0.002) 0.0007y (0.0003) 0.0013y (0.0006)

0.003 (0.0009) 0.004 (0.001) .

0.003y (0.001) 0.0002 (0.0019) 0.0002 (0.001) 0.002y (0.001)

0.003y (0.001) 0.0003 (0.0019) 0.0001 (0.0005) 0.002y (0.001)

0.003y (0.001) 0.0003 (0.0019) 0.0001 (0.0005) 0.0016 (0.0006)

26,429 11452

24,381 10390

24,381 10390

26,429 11458

26,045 10114

26,045 10111

24,031 9185

24,031 9185

Notes: To enable comparisons, the estimates reported in column 1 are the same as the estimates in Table 7, spec. 2. Estimates presented in columns 2, 3 and 4 of include all the variables in the specification reported as Table 7, spec. 2 and two additional variables to correct for the endogeneity of hours of economic work and hours of domestic work. To enable comparisons, the estimates reported in column 5 of are the same as the estimates in Table 7, spec. 4. Estimates presented in columns 6, 7 and 8 include all the variables in the specification reported as Table 7, spec. 4 and two additional variables to correct for the endogeneity of hours of economic work and hours of domestic work. a Identification is based only on differences in functional form. b Identification is based on differences in functional form and the inclusion of the variables that capture the policies of the county towards tackling child labour, namely, whether a county has a PIEF program and the number of children per PETI member. c Identification is based on differences in functional form, the policies of the county towards tackling child labour and the labour inspection regime, that is, number of illegalities detected per worker and the fine per illegality.  Significant at the 1% level. y Significant at the 5% level.

5.3. School success and work As shown in Table 7 (specification 2), an hour of economic work reduces educational success by 0.4 percentage points while an hour of domestic work has a negative effect of 0.5 percentage points. To account for the potential correlation between unobserved characteristics that determine school success and work we re-estimated the hours of work effect with the inclusion of educational interest and ambition (specification 4). The negative effect of work is now smaller (0.3 and 0.4 percentage points for economic and domestic work, respectively). At the mean value of weekly hours of work these estimates translate into educational success reductions of 4.2 percentage points for economic work and 3.2 percentage points for domestic work. Table 8 presents several IV estimates. These specifications include all the variables used in the school success regressions reported in Table 7. To aid comparison the estimates in column 1 of Table 8 are the same as those reported earlier in Table 7, specification 2. Column 2 presents IV estimates based on functional form identification, column 3 relies on variation in child labour policies to aid identification,

while column 4 relies on variation in child labour policies and labour inspection regimes to support identification. Regardless of the identification approach, we find that there is a sharp difference between the IV and single-equation probit estimates. There is a decline in the effect of hours of economic work. The marginal effect is halved from 0.4 to 0.2 percentage points. The effect of hours of domestic work also falls from 0.5 to 0.2 percentage points and is no longer statistically significant. Across all specifications, the selection correction variables are individually and jointly statistically significant (p-values of about 0.02). Columns 5–8 report IV estimates that include proxies for the educational ability of children. Once again, regardless of the identification strategy, a comparison of the probit and the IV estimates shows that the effect of domestic work on educational success is small and no longer statistically significant. On the other hand, the effect of economic work is stable across specifications and is unaffected by changes in estimation approach and identification strategy.14 14

Since we have four instrumental variables and two endogenous variables we estimated school success using OLS and carried

ARTICLE IN PRESS 586

P. Goulart, A.S. Bedi / Economics of Education Review 27 (2008) 575–587

Overall, the IV estimates suggest that the entire negative effect of domestic work on educational success may be attributed to selection effects. As for economic work, the estimates imply that the average work contribution of an economic worker reduces educational success by 2.8–4.2 percentage points. While this is not a trivial effect, neither is it very large as compared to the 27% educational success gap between economic workers and nonworkers. Thus, at most, the economic work contribution of children accounts for about 15% of the educational gap. The smaller IV estimates and negative selection effects reported in our paper may be contrasted with the results reported for the United States by Stinebrickner and Stinebrickner (2003) and Tyler (2003). Unlike the Portuguese case, in these papers, OLS estimates tend to underestimate the negative effect of work and allowing for the endogeneity of child labour leads to a larger negative impact of work on educational outcomes. The reasons for this difference probably lie in the characteristics of child work in the two countries. In the United States, work is likely to be voluntary, paid, occurring amongst older children and only children who are able to handle both activities enter the labour market, leading to positive selection and underestimated OLS effects of work on educational outcomes. In contrast, child labour in Portugal is a structural phenomenon, starts at a younger age, it is typically unpaid and is concentrated in certain families in the Northern and Central parts of the country. Children living in these families are expected to work. However, based on their school performance, parents may adjust the hours that their children work. Additional work may be demanded from children who are not performing well leading to a negative selection effect which in turn would be responsible for the lower IV estimates of the effect of work on educational success. The negative selection effect that we detect is consistent with work on educational returns for Portugal. Vieira (1999) and Modesto (2003) find that IV (footnote continued) out tests for overidentifying restrictions. Regardless of the specification none of the four instruments were individually statistically significant and the tests were unable to reject the null hypothesis that all the instruments are not correlated with the error term in the school success equation. The computed test statistics were 4.88 (excluding the interest variables) and 4.86 (including the interest variables) as compared to the critical value of 5.99 ðw22 Þ.

estimates of returns to education are lower than the corresponding OLS estimates and argue that low investment in education may be related to low ability rather than to high marginal costs. 6. Concluding remarks This paper assessed the factors that determine child labour and educational success in Portugal and examined whether the work activities of children hinders their educational success. We found that while increases in income were associated with reduced economic work, variables that captured a household’s occupational structure played a large role in determining child labour. Child labour in Portugal is concentrated in the Northern and Central parts of the country, precisely those areas that have a strong presence of smalland medium-sized family enterprises and small land ownership. A long tradition of relying on child workers probably slows the change of habits and leads to the persistence of this norm. From the perspective of developing countries the results suggest that, while reductions in poverty are likely to reduce child labour, sensitisation programs that attempt to break entrenched norms are also required. Furthermore, our estimates highlighted the importance of treating child work as endogenous and distinguishing between types of work. The singleequation estimates displayed that both economic work and domestic work hinder educational success. However, the two-stage estimates revealed that only economic work exerts a negative effect on educational outcomes, suggesting that policy initiatives should focus on eliminating economic work. Although tentative, an interesting aspect of our work was the correlation between a child’s educational ambitions and educational and labour outcomes. We found that higher ambitions were associated with greater educational success while at the same time they reduced the likelihood of economic work. While a part of the estimated effect of these psycho-social variables is probably simultaneously determined with educational success, the magnitude of the coefficients suggests that these types of variables should not be ignored. It is possible that parents and teachers have low ambitions for some children, regardless of their performance, and such prejudices may translate into self-fulfilling prophecies. From a policy perspective,

ARTICLE IN PRESS P. Goulart, A.S. Bedi / Economics of Education Review 27 (2008) 575–587

these results indicate that standard approaches such as controlling work through labour inspections or encouraging school attendance through cash subsidies may need to be supplemented with programs that attempt to foster higher educational aspirations. Acknowledgements We thank Margarida Chagas Lopes, Ashwani Saith, Chris Elbers, seminar participants at the Institute of Social Studies and especially two anonymous referees for useful comments. SIETI a recently established government statistics unit and DETEFP the statistics department of the Labour and Training Ministry graciously allowed the use of household surveys on the ‘‘Social Characterisation of the Portuguese Household with School Age Children’’. References Alexander, K. L., Eckland, B. K., & Griffin, L. J. (1975). The Wisconsin model of socioeconomic achievement: A replication. American Journal of Sociology, 81(2), 324–342. Alves Pinto, G. (1998). O Trabalho das Crianc- as. Lisbon: Celta. Beegle, K., Dehejia, R., & Gatti, R. (2003). Why should we care about child labor? Paper presented at the northeastern universities development consortium conference. Boozer, M. A., & Suri, T. S. (2001). Child labor and schooling decisions in Ghana. Mimeo. Yale University. Cunningham, H., & Viazzo, P. P. (1996). Child labour in historical perspective. Florence: UNICEF International Child Development Centre. Goulart, P., & Bedi, A. S. (2005). Child labour and educational success in Portugal. ISS working paper series no. 412, The Hague. Gunnarsson, V., Orazem, P., & Sanchez, M. (2003). Child labor and school achievement in Latin America. Paper presented at the northeastern universities development consortium conference.

587

ILO. (2002). Every child counts. New global estimates on child labour. Geneva: ILO. Jensen, P., & Nielsen, H. S. (1997). Child labour or school attendance? Evidence from Zambia. Journal of Population Economics, 10(4), 407–424. Kahl, Joseph A. (1957). Educational aspirations of ‘‘Common Man’’ boys. Harvard Educational Review, 23(Summer), 186–203. Modesto, L. (2003). Should I stay or should I go? Educational choices and earnings: An empirical study for Portugal. Journal of Population Economics, 16(2), 307–322. OECD. (2000). Final report on the international adult literacy survey. Paris: OECD. OECD. (2003). Learning for tomorrow’s world-First results from PISA 2003. Paris: OECD. Otto, Luther B., & Haller, A. O. (1979). Evidence for a social psychological view of the status attainment process: Four studies compared. Social Forces, 57(3), 887–914. Patrinos, H. A., & Psacharopoulos, G. (1995). Educational performance and child labour in Paraguay. International Journal of Educational Development, 15(1), 47–60. Psacharopoulos, G. (1997). Child labour versus educational attainment: Some evidence from Latin America. Journal of Population Economics, 10(4), 377–386. Rivers, D., & Vuong, Q. (1988). Limited information estimators and exogeneity tests for simultaneous probit models. Journal of Econometrics, 39(3), 347–366. Sewell, W. H., & Hauser, R. M. (1972). Causes and consequences of higher education: Models of the status attainment process. American Journal of Agricultural Economics, 54(5), 851–861. Stinebrickner, R., & Stinebrickner, T. R. (2003). Working during school and academic performance. Journal of Labor Economics, 21(2), 473–491. Tyler, J. H. (2003). Using state child labor laws to identify the effect of school-to-work on high school achievement. Journal of Labor Economics, 21(2), 381–408. Vella, F. (1993). A simple estimator for simultaneous models with censored endogenous regressors. International Economic Review, 34(2), 441–457. Vieira, J. C. (1999). Returns to education in Portugal. Labour Economics, 6(4), 535–541. Williams, S. (1992). Child workers in Portugal. London: AntiSlavery International.