Remittances and child labour in Ghana: Does the gender of the household head matter?

ABSTRACT

This study investigates the effect of remittance income on child labour and the role of gender in Ghana. Data were sourced from the Ghana Living Standards Survey Round 6. In order to solve the issue of endogeneity, we adopt the instrumental variable approach. The study finds evidence that, irrespective of whoever is the head of the household, the total effect of remittances on child labour is negative. However, the negative effect is much higher for female-headed households. It is recommended that more mobile money transfer outlets be established and the cost of transferring money be reduced to the barest minimum so as to enable households to maximise the benefits of the remittances they receive. Furthermore, financial transfers should be given to female household heads upon the condition of school attendance by their children so as to reduce child labour.

KEYWORDS:

• Child labour
• household
• instrumental variable
• migrant
• remittances

1. Introduction

With the rapid pace of globalisation, and from a symptom of development failure in migrants’ communities of origin, the movement of workers both internally and internationally has come to be seen as an indication of local development opportunity because of the remittances these migrants send home (Abdul-Mumuni & Quaidoo, 2016). Thus, international and internal migrant remittances flow has been occurring in all countries. Sander (2003) defines remittances as monies sent from one individual or household to another. International remittances are those sent by migrant workers who have left their home country. On the other hand, internal (domestic) remittances are those sent by migrant workers who have left their home village or town to work elsewhere in their home country (e.g. rural–urban migration).

The World Bank estimates that international migrants alone remitted about US$429 billion to developing countries in 2016 (World Bank, 2017a). This figure is even lower than the actual figure as many migrants use informal mechanisms to send money back home. In view of this, remittances are, generally, underreported.

In Ghana, the flow of remittances is part of the reality of a significant proportion of households. From the early 1970s into the twenty-first century, international migration has been increasing steadily. As a consequence, total remittances flow into Ghana from abroad increased from US$31 million in 1999 to US$2100 million in 2016 (World Bank, 2017b). Internal migration too has been substantial. About 48.6% of the population is made up of internal migrants and so internal remittances have affected a significant proportion of households over the years (GSS, 2014). These remittances are often spent on food, health, education, housing, businesses, developmental projects, funerals and other ceremonies. The increase in remittances flow to and within Ghana could be attributed to the liberalisation of the foreign exchange market and the proliferation of money transfer institutions.

There are different transfer mechanisms available to migrants to send remittances to Ghana, namely: mobile money outlets; banks; credit unions; small and large money transfer institutions such as Western Union, MoneyGram and VIGO; hand delivery by the sender through a third party; and other informal channels. However, the choice of the method of transfer depends on several factors such as the amount sent, the legal status of the migrant, the cost of sending money through the legal channels, the existence of social networks or clubs and government regulations regarding the transfer of money, to mention but a few (Quartey, 2005).

Given the magnitude and pervasiveness of remittance income for many households, decisions relating to child labour will be affected by remittance flows. According to the International Labour Organization (ILO, 2013) there are about 168 million child labourers in developing countries of which 21% are in Sub-Sahara Africa. It is also estimated, based on the Ghana Living Standards (GLSS) Round 6 report, that around 28% of children between the ages of 7 and 14 years are involved in child labour in Ghana. In other parts of the sub-region such as Burkina Faso, Mali and Cote D’Ivoire, child labour is very common, with children normally engaged in farming and fishing activities.

Despite the region’s high rate of child labour, much of the studies on the effect of remittances on child labour tend to focus on Asia and Latin America. According to Kukwaw (2013), poverty is considered as the main driver of child labour and so higher remittances flow to poor households is expected to translate into lower incidence of child labour. The aim of this study is to explore the gender dimension of remittances. Given that migration is often recurrent, some women may be effectively left in charge of their households for long stretches of time and are likely to preside over labour decisions for their children. It is this aspect of female headship that is pertinent to the concerns of this study.

Concerning research knowledge in the subject, we are aware of only one study by Joseph & Plaza (2010) in the case of Ghana. These authors sought to investigate whether international remittance receiving households behave differently from domestic remittance receiving households with regard to the decision on sending their children to work and the number of hours they are made to work. This study is different from Joseph & Plaza (2010) since it provides a gender perspective to the problem. Aside Joseph & Plaza (2010) there are other studies that have investigated similar issues in other regions. For instance, in Latin America (Bayot, 2007; Acosta, 2011; Alcaraz et al., 2012; Binci & Giannelli, 2012; Nguyen & Nguyen, 2013), in Asia (Priyambada et al., 2005; Mansuri, 2006) labour and in Africa (Bargain & Boutin, 2014; Dimova et al., 2015). However, none of these offer a gender perspective to the problem and this provides grounds for further scrutiny. Thus, in an attempt to extend the realm of understanding, the current study fills the gap in the literature by investigating the effect of remittances and the gender of the household head on child labour in Ghana.

Child labour is a complex phenomenon and so there is no universally accepted definition of child labour. Thus, definitions of child labour differ by actors, history, context and purpose (Weston, 2005). For the purpose of this study, child labour is defined as any activity, economic or non-economic (such as household chores) performed by a child between the ages of 5 and 14 years that is either too dangerous or hazardous and that has the potential to hinder or negatively affect his/her health, education, moral or normal development. The ILO broadly defines child labour as work that deprives children of their childhood, their potential and their dignity and that is harmful to their physical and mental development (ILO, 2013).

This study uses cross-sectional data constructed from the GLSS Round 6. The GLSS Round 6 is a nationwide household survey that was carried out between 2012 and 2013 designed to generate information on living conditions in Ghana. A total of 16 772 out of a sample of 18 000 households were successfully enumerated leading to a response rate of 93.2% (GSS, 2014). The estimated total annual value of remittances received by urban households was GH¢1268.7 million and the amount received by rural households was GH¢535.2 million. Table 1 shows the contribution of remittances to total household income in the various regions of Ghana during the study period.

Table 1. Remittances received by households.

Region	Contribution of remittances to household income (%)
Western	2.0
Central	1.2
Greater Accra	1.7
Volta	2.0
Eastern	2.7
Ashanti	3.0
Brong Ahafo	2.9
North	3.7
Upper East	2.0
Upper West	1.0

Source: GSS (2014).

The study is organised as follows. The literature on remittances and child labour is reviewed in section 2. The methodology (theoretical framework and empirical strategy) is described in section 3, followed by the study’s empirical results in section 4. Section 5 concludes the study and discusses policy recommendations based on its findings.

2. Literature review

2.1. Theoretical literature review

Many theories explain the motives behind migrants’ decisions to send remittances to their relatives back home. The theoretical debate about the motives of remittances was triggered by Lucas & Stark (1985). These authors studied remittances on a household level and claimed the main motives to be ‘pure altruism’, ‘pure self-interest’ and ‘tempered altruism or enlightened self-interest’. Altruistic motives assume that an agent receives utility from an increase in utility of family members left behind. On the other hand, an agent may send remittances based on selfish motives. Solimano (2003) states that the analytical literature on the motives behind remittances can be summarised by the following four approaches, namely: altruistic motive; self-interest motive; implicit family contract I—loan repayment; and implicit family contract II—co-insurance.

Altruistic motives are associated with situations where remitting is considered as an obligation to the household. Migrants send remittances out of affection and responsibility towards their families. The migrant thus cares about poverty, shocks and other unfortunate situations that the family might face and consequently sends remittances. In this case there is a positive relationship between adverse conditions of the receiving household and remittances sent. The altruistic model asserts that sending remittances to households yields a satisfaction to the migrants out of a concern for the welfare of their families. According to the self-interest motive, at every point in time, the successful migrant saves and so the need then arises on how and where to accumulate wealth. In most cases, the migrant chooses to invest in his/her home town or country by buying property, financial assets, etc. The family then acts as a trusted agent by taking care of those assets for the migrant during the emigration period.

The loan repayment theory states that families tend to develop an implicit contract among those who stay at home. This implicit contract combines elements of investment and repayment. Thus, in the loan repayment theory, the family invests in education of the migrant and usually finances the costs of travel by the migrant to the host town or country. In the implicit family contract II model, migration becomes a co-insurance strategy with remittances playing the role of an insurance claim. For instance, if the economic risks between sending and foreign countries are not positively correlated then the family can send some of its members abroad to diversify the risks. The migrant can support his/her family during unfavourable economic times at home. At the same time, the migrant can be supported by his/her family during bad times in the foreign country.

While the above motives are considered to be separate and different they nonetheless overlap (e.g. in the reaction to shocks) and are essentially all the same motive; namely, an increase in welfare for the sender.

2.2. Empirical literature

In many developing countries including Ghana, children work and also attend school. Some of the typical child work activities include family business, farming and other part-time jobs. While these activities are beneficial in the short run for households in order to obtain some income for their upkeep, they certainly disrupt the acquisition of human capital for future generations. For example, Alcaraz et al. (2012) found that a reduction in remittances from the US to Mexican households can cause an increase in child labour and a decrease in school attendance of children. In a similar study by Ebeke (2010), there is significant evidence that an increase in remittances as a share of GDP is associated with a reduction of the share of the 10–14-year-old population actively engaged in work. The study also found that remittances have a stronger effect on the reduction of child labour in countries where credit constraints are higher and GDP growth more volatile.

Again, Acosta (2011) found that remittance receipt significantly reduces the likelihood that children will engage in paid labour in El Salvador. Joseph & Plaza (2010), also focusing on Ghana, examined whether international remittance receiving households behave differently from domestic remittance receiving households and households that do not receive remittances, with regard to the decision on sending their children to work. They found that children belonging to international remittance receiving households tend to work fewer hours than those who belong to households that do not receive remittances. Using longitudinal data from a balanced panel of 962 households from the two rounds of the Nepal Living Standards Survey, Acharya & Leon-Gonzalez (2014) found that remittances can be used to purchase labour-saving equipment, possibly decreasing the reliance on child labour. Again, using data from the 1992–3 and 1997–8 Vietnam Living Standards Surveys, Binci & Giannelli (2012) found that remittances lead to a reduction in child labour.

On the issue of gender, Ray (2000) found that children from female-headed households are more likely to be engaged in employment than their counterparts from male-headed households. Again, Badmus (2011) found evidence that, among rural households in Nigeria, female-headed households usually have higher dependency ratios and this increases the probability of sending children to work.

The underlying assumption in all of these studies is that remittances help reduce liquidity constraints and increase household income sufficiently to reduce the need for child labour. It is, however, important to note that not all studies find uniformly positive results. For example, while Acosta (2011) found that remittances are associated with a decrease in wage labour, he also found that remittances tend to increase the prevalence of domestic labour. Essentially, what this suggests is that remittances simply lead to a reallocation of child labour from paid to unpaid work. Similarly, in a study on El Salvador, Calero et al. (2009) found that remittances only reduce the incidence of paid child labour among rural and non-poor households. However, in urban households, remittances increase the incidence of domestic labour.

Other studies indicate that migrant remittances have no significant effect on child labour. For example, using data from Vietnam Household and Living Standard Surveys in 2006 and 2008, Nguyen & Nguyen (2015) estimated the effect of the receipt of international remittances and internal remittances on child labour in Vietnam. Their results show that there is no statistically significant effect of receipt of remittances on child labour. In addition, Bargain & Boutin (2014) found that, in Burkina Faso, receiving remittances does not significantly influence child labour when no distinction is made between migration and remittance effects.

All these studies tend to suggest that, while increased remittances may benefit the household, family disruption caused by the loss of a working-aged household member to migration may have negative effects on the children left behind. The above-mentioned studies do not, however, address the gender perspective of the remittances–child labour nexus. The differing results discussed could reflect genuine differences in how remittances are spent in different parts of the world and by different household heads depending on their gender.

3. Methodology

3.1. Theoretical framework

The theoretical framework for this study employs the model proposed by Basu & Van (1998) and adopted by Coon (2016). The study assumes that the parent ultimately decides the optimal child labour decision for the household. Utility is maximised by parents with respect to consumption and their children’s leisure. Thus, following Coon (2016), we have the following function:(1) where c represents consumption, e is each child’s labour effort level and . s is a parameter representing the minimum subsistence level (i.e. the household’s needed level of consumption). It is only after achieving s that the household will consider children’s education. Thus, if s is very high then child labour effort will be equal to 1 (e = 1). The household’s budget constraint is stated as follows:(2) where n represents the number of adults in the household, m is the number of children in the household and each child consumes proportion of what an adult consumes, and and represent the wages paid to adults and children in the household, respectively (Coon, 2016). The amount of remittances to the household is represented by r. For the case where c ≥ s, the optimal allocation of children’s effort is given as:(3)

The gender of the household head that receives remittances is represented by g × r. Partially differentiating equation (3) with respect to and m gives the following:(4) (5) (6) (7) (8) (9)

As adult wages and remittances to the household increase, child labour is expected to reduce and this is shown in equations (4) and (5). Equation (6) gives the change in optimal child labour effort with respect to the number of adults in the household. This implies that when the adult wage is less than the minimum subsistence level there will be a reduction in child labour and when the adult wage is greater than the minimum subsistence level, child labour increases. Furthermore, a large enough increase in remittance income received by a household takes a child completely out of the labour market. When wages are greater than the subsistence level of consumption, migration and remittances reduce child effort only if . That is, if remittance income to the household is very small, such that it does not offset the loss of the adult wage due to migration, then migration can increase child labour. Thus, a utility maximising household would not send a member abroad in this case unless they expected remittance income to offset the loss of domestic income.

The change in child labour effort with respect to an increase in the number of children is expressed using equation (7). In households where the adult income exceeds subsistence, increasing the number of children increases child labour effort.

There are both income and substitution effects determining the supply of child labour and this is indicated in equation (8). When total child subsistence is greater than adult surplus income , child labour decreases with an increase in child wages. On the other hand, when adult surplus income is greater than the child subsistence level, child labour increases with an increase in child wages. Equation (9) indicates that a female-headed household that receives remittances will influence a reduction in child labour more than a male-headed household that receives remittances.

3.2. Empirical strategy

The econometric model in this study is designed to test the preceding theoretical framework in which it has been indicated that remittances are important in shaping child labour supply by the household; hence, child labour depends on remittances. It is also assumed that the household head ultimately decides the number of hours a child in the household works. Following Bargain & Boutin (2014), the child labour supply function is stated as follows:(10) where represents the number of hours worked per week by the child labourer i in household j, and is the log of the remittance amount. The sign of the remittance coefficient directly tests the theories on household–child labour decisions. A negative coefficient suggests that remittances reduce the number of hours a child in a household works. is a vector of household characteristics related to the household head (age, gender and level of education) or to the household (income without remittances, size, location and region). It has been argued in the literature that these characteristics have a strong effect on the schooling and work patterns of children (Milligan, 2009; Binci & Giannelli, 2012). The mean zero error term captures the effects of unobserved factors common to a given household.

Equation (10) is rewritten as:(11) where is child labour, is the number of migrants, is an interactive term for gender and remittances, is log of remittances, is the gender of the household head, is the age of the household head, is household size, is a dummy variable representing employed and unemployed household head, is the number of jobs being engaged by the household head, is a dummy variable for location of the household (whether rural or urban), is a dummy variable indicating whether the head of the household is educated or not, is household income excluding remittances, represents regional dummies, ε is the error term and are the coefficients.

In Ghana there are quite a number of migrants that tend to send remittances to their households through unofficial channels. However, the majority of databases only include flows transferred through official channels. In view of this, only remittance data captured by the GLSS Round 6 were used for the study.

3.3. Potential sources of endogeneity

A problem commonly faced in estimating the causal impact of migrant remittances is endogeneity. Running a simple ordinary least squares (OLS) regression of household outcomes with remittance receipts as explanatory variables could give a biased estimate of the impact. The error term and the explanatory variables are likely to be correlated due to several reasons such as reversed causality, omitted variables or selection bias. For instance, let us assume that the true coefficient is negative (i.e. remittance receipt reduces the extent of child labour). Reverse causality could occur if, for example, a migrant remits money to the household with the specific intention of retiring children from work by financing their education. In that case, child labour determines remittance receipt and not the other way around (Amuedo-Dorantes & Pozo, 2010).

Estimates of may possibly be an upward bias in this situation. In other words, the coefficient becomes less negative than expected or could even be positive. It is also possible to have an upward bias arising from omitted unobservables. This occurs when, for instance, a specific shock forces the household to rely more on child labour and at the same time claim more remittances from its migrant. Finally, and more problematic, a downward bias may also occur and lead us to wrongly conclude about the reducing effect of migrant remittances on child labour. Such a bias may arise when, for instance, there is a common shock affecting both the migrant’s location and the household’s place of residence. In this case, a bad economic situation (such as recession) could hinder the migrant’s effort to send remittances home and increase the household’s reliance on child labour.

3.4. Instrumental variable estimation–two stage least squares (2SLS)

Given the above sources of endogeneity, the use of OLS may lead to inconsistent estimates of the parameters of the model (Wooldridge, 2006). To account for this endogeneity problem, and to identify the effect of remittances on child labour, the instrumental variable (IV) approach–two stage least squares (2SLS) procedure is employed. Another choice would have been to use panel estimations. However, this has proven not to be a valid option. For instance, children age over the panel and this tends to disturb the analysis of child labour defined according to a fixed age window. Besides, unobservables can be time-varying and, for that matter, are not taken into account by fixed effects estimators (see the discussion in Antman, 2013). To apply the 2SLS procedure, it is required that we identify a variable (z) that satisfies two main conditions: (i) the variable must be uncorrelated with the error term μ; and (ii) the variable must be correlated with the regressor being instrumented.

Given that the log of remittances is expected to be endogenous, the first stage equation is estimated as follows:(12)

The means of transferring funds are used as an instrument for remittance amount. This instrument is chosen because it correlates with remittances flow but not child labour. After estimating the first stage equation, we plug the predicted values of equation (12) into the structural equation. Thus, we have:(13) The definition and measurement of the variables used in equation (13) are shown in Table 2.

Table 2. Definition and measurement of variables.

Variable	Definition	A priori sign
	Internal and international remittances	Negative
Gender_head	Binary variable to capture the gender of the household head	Indeterminate
	A variable interacting the gender of the household head with remittances	Indeterminate
	Number of migrants	Negative
	Binary variable to capture the location of the household	Indeterminate
	Household size (number of people living in a household	Negative
Employed	Dummy variable indicating whether the household head is engaged in economic activity or not	Negative
	Education of the household head	Negative
Agey	Age of the economic head of the household	Positive
	Number of jobs of the household head	Positive
	Household income excluding remittances	Negative
Region	Categorical variable to capture regional effects	Indeterminate

Sources: Compiled from literature and theories (2017) (Milligan, 2009; Amuedo-Dorantes & Pozo, 2010).

4. Results and discussion

4.1. Effects of remittances and the gender of the household head on child labour

Table 3 shows the results of the effect of remittances and the gender of the household head as estimated by OLS and IV. Column 2 reports the OLS results, which does not control for any potential endogeneity biases. According to the OLS estimates, a 1% increase in remittances, lnrem will lead to an increase in child labour by 0.052 (i.e. 5.196 ÷ 100) of an hour per week, other factors remaining constant. However, after instrumenting for remittances, there is a negative and highly significant effect of lnrem on the number of hours a child in a household works per week indicating that OLS estimates are biased toward zero, which is consistent with results found by previous studies (Acosta, 2011; Alcaraz et al., 2012; Bargain & Boutin, 2014; Coon, 2016). According to the IV estimates (column 3), a 1% increase in remittances received by a household conditioned that the household head is a female decreases the time a child in that household works per week by about 0.11 (i.e. −38.872 + 27.162  ÷ 100) of an hour, all other factors remaining constant. This is significant at 5% level.

Table 3. Results of effects of remittances and gender of household head on child labour.

Dependent variable:	hours_chd
Variable	OLS	IV
(1)	(2)	(3)
	5.196*** (1.921)	−38.872** (16.795)
	−4.003*** (0.805)	−3.687*** (0.954)
	−3.720*** (1.428)	27.162** (11.783)
	22.734*** (7.867)	−139.544** (61.978)
	4.692*** (0.257)	4.576*** (0.305)
Agey	0.483*** (0.060)	0.457*** (0.071)
Employed	−20.983*** (3.810)	−16.484*** (4.789)
	3.304* (1.905)	−0.344 (2.629)
	1.629*** (0.055)	1.689*** (0.069)
	−4.586*** (1.292)	0.005 (2.304)
	−1.899 (1.817)	−2.574 (2.151)
	0.584 (1.747)	1.289 (2.071)
Region (base = Volta)
Western	−24.721*** (3.682)	−27.171*** (4.426)
Central	−22.766*** (4.747)	−23.836*** (5.594)
Greater Accra	−15.325** (6.094)	−1.899 (8.776)
Eastern	−25.396*** (3.238)	−25.006*** (3.809)
Ashanti	−27.676*** (3.231)	−24.305*** (4.005)
Brong Ahafo	−28.684*** (3.361)	−26.331*** (4.049)
Northern	−27.956*** (3.627)	−36.973*** (5.456)
Upper East	−28.022*** (4.035)	−25.359*** (4.848)
Upper West	−12.392*** (4.442)	−9.056* (5.371)
Constant	−0.746 (14.557)	221.75*** (85.750)
Sample size	1321	1321
Adjusted R-squared	0.647	0.513
Under identification		13.44
P-value of under identification		0.0620
Weak identification		74.36
P-value–Link test		0.000
(_hatsq)		0.11

*p < 0.10; **p < 0.05; ***p < 0.01.

Heteroskedasticity-robust standard errors in parentheses.

Source: Computed from GLSS 6, 2012–3.

Similarly, a 1% increase in remittances received by a household conditioned that the household head is a male decreases the time that a child in the household works per week by about 0.38 (i.e. 38.372 ÷ 100) of an hour all other things being equal. Based on the aforementioned results, irrespective of whoever is the head of the household, the total effect of remittances on child labour is negative. However, the negative effect is much higher when the head of the household is a male than when it is a female. This is mainly due to the fact that female heads of household particularly, in rural Ghana tend to be older and have fewer years of education than male heads of household (FAO, 2012). Thus, these household heads adopt negative coping strategies, such as an increased participation of children in productive activities. Besides, most migrants are relatively young married men with children, so their wives are likely to preside over labour decisions for their children. These wives therefore tend to engage the children to take up more household chores and economic activities such as petty trading, farm work, fishing, quarrying and mining.

The IV estimates show that an increase in household income, lninc_less_remit by 1% will lead to a reduction in the number of hours a child in that household works per week by 0.03687 (i.e. 3.687 ÷ 100) of an hour, all other factors remaining constant. This is also statistically significant at 1% and consistent with the study by Dayioğlu (2006) which found evidence of a negative effect of household income on child labour. In this study, conducted in Turkey, the author found that an increase in household income, whether in the form of an improvement in paternal earnings or non-wage income, is expected to bring about a reduction in child labour. Thus, the decision of the parent to send the child to work depends somehow on the income of the household. If the income of the household is high, there may be no need for the child to work all other things being equal.

Again, the IV results indicate that an increase in the age of the household head by a year will increase the number of hours per week a child works by 0.46. This result is statistically significant at 1%. The result suggests that elderly household heads are a financial burden and so help increase child labour.

Similarly, the size of the household (hhsize) has an effect on the number of hours a child in that household works. The IV results show that an increase in the size of the household will lead to an increase in the number of hours a child in that household will work per week by 4.58 hours, all other factors held constant and also statistically significant at 1%.This finding is consistent with the study by Bayot (2007). In particular, the author found that the presence of an additional child increases the probability of child labour.

The IV estimates also indicate that, if the household head is employed, the number of hours a child in that household works per week will reduce by about 16.48 hours (ceteris paribus) as compared to a child living in a household in which the head is unemployed. This result is also significant at 1%. The probability of a child being sent to work is high when the parent is unemployed.

Again, the IV estimates show that if the number of jobs the head of a household has increases by one, the number of hours a child in that household works per week increases by 1.69 hours, all other factors remaining the same. This is also statistically significant at 1%. This result is expected because the tendency for a child to be engaged in child labour increases if the parents are engaged in many economic activities (e.g. farming and trading). According toVillamil (2002), households with their own enterprises, which include farming and home-based small-scale manufacturing, are likely to involve their children in production work.

Furthermore, the IV results show that if the head of a household has formal education the number of hours a child in the household works per week decreases by 0.344 as compared to a situation where the household head is not formally educated. Even though this result is statistically insignificant, it is intuitive because educated heads tend to know much about the benefits of schooling and so would send their children to school rather than allowing them to engage in child labour. Better-educated parents tend to understand the importance of education for their children’s well-being.

Additionally, the IV results reveal that, per week, a child in a household located in a rural area works 1.29 hours more than a child in a household located in an urban centre. These results are statistically insignificant but they make a lot of sense. Labour demand tends to be higher in rural areas because of the need for children in farm work. Besides, the cost of hiring children tends to be lower in rural areas where monitoring and enforcement of labour laws are weak.

The study also includes dummy variables for the regional location of the household where the child belongs to capture regional differences in child labour. The base region is the Volta region. The region in which a child lives may affect the number of hours the child works per week. The IV results show that a child in a household in the Western region works 27.171 hours per week fewer than a child living in a household in the Volta region and this is statistically significant at 1%. Even though child labour is common in the mining areas of the Western region, the practice is more prevalent in the Volta region because of child trafficking, prevalent at Kpandu, Gemini, Denu and the trokosi system (ritual servitude) in the Ketu, Keta, North Tongu and South Tongu districts. Following the creation of the Volta lake, the livelihoods of communities along the lake have been severely altered leading to child labour becoming a main economic lifestyle and a coping strategy due to increasing levels of poverty (Kukwaw, 2013).

Again, the IV results show that a child in a household in the Central region works 23 836 hours per week fewer than a child living in a household in the Volta region, ceteris paribus. This result is also statistically significant at 1%. Fishing is done in both the Central and Volta regions. However, the prevalence of child labour in the Volta region is higher.

With respect to the Eastern region, the IV estimates indicate that, all other things being equal, a child in a household in this region works for fewer hours than a child in a household in the Volta region. The study indicates that a child in a household in the Eastern region works 25.006 hours per week fewer than a child in a household in the Volta region. This result is statistically significant at 1%. Just like the Western region, in the Eastern region child labour is common in the mining sector. However, it is not as prevalent as the child labour activities in the Volta region.

Additionally, the IV results show that, other factors remaining constant, a child in a household in the Ashanti region works 24.304 hours per week fewer than a child living in a household in the Volta region. This is also statistically significant at 1%. Child labour activities in the Ashanti region can be found mostly in the agricultural and services sectors. For instance, children are normally engaged on the cocoa farms instead of being sent to school. In Kumasi itself, head porter (kaayayei) business is rife and some of those engaged in it are children. All this notwithstanding, however, the prevalence of child labour in the Volta is higher than that found in the Ashanti region.

Furthermore, the IV result reveals that a child living in a household in the Brong Ahafo region works 26.331 hours per week fewer than a child in a household in the Volta region and this is statistically significant at 1%. Farming is the main activity of the people of the Brong Ahafo region. Thus, child labour is normally found in the farms of that region. However, the prevalence of child labour in the Brong Ahafo region is not as high as the child labour activities found in the Volta region.

Similarly, the IV results indicate that a child in a household in the Northern region works 36.973 hours per week fewer than a child in a household in the Volta region. Just like the Brong Ahafo region, the main economic activity in the region is farming. Some parents, especially on weekends and holidays, send their children to work on the farms. Child labour can therefore be found in the Northern region. However, when compared to the Volta region the prevalence of child labour in the Northern region is less.

Again, the IV results show that a child in a household in the Upper East region works 25.359 hours per week fewer than a child living in a household in the Volta region, ceteris paribus. This result is also statistically significant at 1%. In the Upper East region, child labour is mostly found on farms and in illegal mining activities. However, the prevalence of child labour activities is higher in the Volta region than in the Upper East region.

Finally, the IV results show that, other factors remaining constant, a child in a household in the Upper West region works 9.056 hours per week fewer than a child living in a household in the Volta region. This is, however, statistically significant at 10%. Farming activity is the main activity of the people of the Upper West region. Thus, child labour is mostly found on farms. However, as mentioned earlier, the livelihoods of communities along the Volta lake have been severely altered leading to high incidence of child labour. The prevalence of child labour in the region is therefore higher than that found in the Upper West region.

4.2. Under identification test

This test is conducted to see if the included instrument correlates with the endogenous regressor. The null hypothesis is that the equation is under identified. The Kleibergen–Paap rk LM statistic shows a chi-square value of 13.44 and a p-value of 0.0620. We therefore reject the null hypothesis at the 10% level of significance and conclude that the equation is identified. This is seen in Table 3.

4.3. Weak identification test

The weak identification test is carried out to find out if the set of instrument(s) has a weak correlation with the endogenous regressors. A weak correlation with the endogenous regressors may lead to bias estimates of the IV coefficients. In the model, the Kleibergen–Paap rk Wald F statistic is 74.36 and exceeds the Stock-Yogo weak ID critical values at 10% maximal IV size. We therefore conclude that the model does not have problems with weak identification. The results are shown in Table 4.

Table 4. Results of weak identification test.

	Model hours_chd
Kleibergen–Paap Wald rk F statistic	74.36
Stock-Yogo weak ID test critical values:
10% Maximal IV size	16.38
15% Maximal IV size	8.96
20% Maximal IV size	6.66
25% Maximal IV size	5.53

Source: Computed from GLSS 6, 2012–3.

4.3. Test of omitted variables

The specification link test was conducted for the model and the test output shows that the square of the predicted dependent variable has no explanatory power (see Table 3).

5. Conclusions and policy recommendations

In the large body of literature, studies on the specific effect of remittances and the gender of the household head on child labour in Sub-Saharan Africa and, for that matter, Ghana, remain thin on the ground. This article estimates the effect of remittances and gender of the household head, instrumented by means of transferring funds, on child labour in Ghana. In comparison to other West African countries, the problem of child labour is significant in Ghana. The study finds that, although having significantly different magnitudes, both remittance income and household income have significant effects on child labour.

The study also indicates that the supply of child labour is influenced by the gender of the household head. Agriculture and industrial sectors are the main sectors in which child labour is common and both of these sectors are dominated by men. Thus, households headed by females tend to have a lower supply of child labour than male-headed households. However, migration which is intertwined with remittances does not seem to significantly affect child labour.

At this juncture, it is important to highlight the policy recommendations of the findings from the study. Considering the child labour reducing effect of remittances, it is important that policies are put in place to increase both internal and international remittances in Ghana. To facilitate internal remittances, there is the need to encourage the establishment of more mobile money transfer outlets by improving infrastructure and strengthening social capital networks in the country.

Additionally, a policy aimed at reducing the cost of transferring remittances would encourage more Ghanaian migrants home and abroad to remit money home and this would help reduce child labour.

Finally, financial incentives such as cash transfers to poor households and female-headed households should be put in place to help reduce child labour. These transfers can be made conditional such that only poor households and female heads that send their children to school would be qualified for such incentives.

Disclosure statement

No potential conflict of interest was reported by the authors.