To work or not to work? Child stunting and maternal labour force participation in South Africa

Abstract

Stunting, which affects about one-quarter of South African children below five years, has a long-term effect on a child's development and a potential economic cost to mothers. A stunted child necessitates more attention, thereby reducing the time a mother, especially if she is the primary caregiver, has available to seize labour market opportunities. This study investigated the relationship between child stunting and maternal labour force participation (MLFP) in South Africa, using a bivariate probit regression model. Data was sourced from the 2017 South African National Income Dynamics Study (NIDS). Using maternal height and improved water supply at home as instrumental variables to control for potential endogeneity, the average marginal effect model indicates that caring for a stunted child is associated with a 20 % reduction in MLFP, on average, at a 1 % statistical significance level. In addition to mothers' intrinsic merit of protecting their children, measures to prevent childhood stunting, such as nutrition education and promotion of exclusive breastfeeding, can improve MLFP. A more flexible working arrangement could facilitate mothers' engagement in the labour force while caring for their stunted children.

Highlights

• •Stunting remains a public health challenge in South Africa and negatively affects children and their caregivers, predominantly their mothers.
• •The results suggest that, on average, caring for a stunted child is associated with a 20 % reduction in maternal labour force participation (MLFP).
• •Measures to prevent childhood stunting, such as nutrition education and promotion of exclusive breastfeeding, can improve MLFP.
• •A more flexible working arrangement could facilitate mothers' engagement in the labour force while caring for their stunted children.

1. Introduction

Stunting, which refers to impaired linear growth-for-age, is defined as having a height-for-age z score (HAZ) less than −2 standard deviation from the World Health Organisation (WHO) Child Growth Standards median (Leroy & Frongillo, 2019; WHO, 2014). According to the United Nations Children's Emergency Fund, WHO and the World Bank report (UNICEF et al., 2023), stunting remains one of the most prevalent forms of malnutrition affecting South African children under five years of age, with a prevalence of 22.8 % in 2022. Previous studies that found similar results include Devereux, Jonah and May (2019); Modjadji and Mashishi (2020); Shisana et al. (2013). Nutrition, a public health concern, significantly affects individual and societal health outcomes (Adair, 2014; Lenoir-Wijnkoop et al., 2011; Péter et al., 2015), with well-nourished children more likely to incur lower medical expenditures (Hoddinott et al., 2013).

Child health is essential for socio-economic growth and poverty reduction because it determines current and future human capital and livelihood prospects. Childhood malnutrition, especially stunting, can affect a child's education by causing delayed enrolment, absenteeism, and low academic performance in school (Hoddinott et al., 2013; Sunny et al., 2018). Thus, negatively affecting their future productivity, earnings, and economic growth, and leading to intergenerational poverty transmission (Tette, Sifah and Nartey 2015). Because healthy children require less intensive care, their caregivers can participate more in the labour market (Adair, 2014; Tambi, 2017). Childhood stunting is caused by a combination of factors, including inadequate nutrition, recurrent infections, and poor socioeconomic and environmental conditions (Agbozo, Atito and Abubakari 2016; Quamme & Iversen, 2022; UNICEF, 1991).

Insufficient intake of essential nutrients, such as proteins, vitamins, and minerals by a child (Soliman et al., 2021), and poor maternal diets before, during and after pregnancy can lead to childhood stunting (Baïz et al., 2019; Berti et al., 2016; Jama et al., 2020). Additionally, frequent bouts of illness, especially diarrhoea and respiratory infections, further exacerbate the problem by inhibiting nutrient absorption and diverting energy necessary for growth (Jones et al., 2014; Quamme & Iversen, 2022). Socioeconomic factors such as poverty (Reinhardt & Fanzo, 2014), lack of access to clean water, sanitation and hygiene (WASH) (Devereux, Jonah and May 2019; Mshida et al., 2018; Shrestha et al., 2020), limited healthcare services (Stenberg et al., 2014), and inadequate maternal and childcare practices (Saaka, 2014; Santosa, Arif and Ghoni 2022) also contribute significantly to childhood stunting. Healthy children reduce the demand for childcare, thus encouraging mothers to stay in the labour force (Fotso, 2017; Powers, 2003; Spiess & Dunkelberg, 2009; Tambi, 2017). If childcare cost depends on the child's characteristics, caregivers with higher childcare costs and therefore higher reservation wages are less likely to work (Powers, 1999). In such situations, the responsibility of female household members, especially mothers, decreases due to cultural norms that assign women the primary responsibility of caring for sick household members (Mussida & Patimo, 2021; Spiess & Dunkelberg, 2009).

Children with developmental delays or frequent illnesses can significantly affect both parents' and the child's psychological well-being (Hu & Callander, 2024; Muscara et al., 2015; Woolf et al., 2016). Parents whose children have been diagnosed with such conditions are likely to develop post-traumatic stress disorder (PTSD), which can seriously affect daily functioning (Bronner et al., 2010; Le Brocque et al., 2010). PTSD sufferers experience cognitive impairments, difficulties in daily tasks (Kaye et al., 2014) and depression (Elshafei et al., 2023; Haithar et al., 2018). These conditions decrease the likelihood of maintaining family and social relationships, increase absenteeism and career difficulties, and ultimately reduce productivity (Wilson, Guliani and Boichev 2016). Hu and Callander (2024) determined that mothers whose children are at risk of developmental delay in areas such as communication, mobility, self-care, relating, learning, coping, or behaviour had significantly higher odds of being out of the labour market and not returning, except in cases related to mobility and self-care, compared to mothers of children who are not at risk of developmental delays.

Garti, Ali and Garti (2018) in their study on the relationship between maternal working hours and child nutritional status in Northern Ghana found that the stunting rate of children whose mothers are farmers was approximately 22.1 %, while that of public service mothers was about 12.8 % due to higher incomes, more food supply at home and sociocultural factors, such as older siblings or other extended family members, external childcare and food preparation by grandparents. Nankinga, Kwagala, and Walakira (2019) note that compared with professional mothers, mothers engaged in agricultural and manual work had a higher odds ratio of having stunted children. Willey et al. (2009), using South African data from Birth to Twenty (Bt20), postulate that maternal employment is linked with a lower probability of having a stunted child. Rashad and Sharaf (2019) and Debela, Gehrke and Qaim (2021) observe that maternal employment can increase family income and improve family well-being, thus freeing the family from the threat of poverty and food insecurity. Therefore, for working mothers, the higher the income, the better the nutritional status of their children (Negash et al., 2015).

Jakaria, Bakshi and Hasan (2022), in their study on Bangladesh, found that maternal employment significantly decreases children's height-for-age Z-score. An increase in market work for women reduces the time and energy available for household production, thus giving household members the incentives to consume foods that are convenient like fast food, restaurant meals or frozen food (Gomis-Porqueras et al., 2011) and school meal programmes1 for their children's meals (Datar & Nicosia, 2012), which may be of low quality or lack dietary diversity. Cawley and Liu (2012) ascertain that the time allocated to children's diet and physical activity decreased by 127 min per day for working mothers. Datar, Nicosia and Shier (2014) observe that mothers' working hours are positively associated with children's body mass index (BMI), particularly for high-socioeconomic families. Increased maternal working hours reduce supervision, which can lead to unhealthy behaviours amongst children, including snacking and sedentary activities (Fertig, Glomm and Tchernis 2009). Eshete et al. (2017) postulate that there is no statistically significant difference in the overall nutritional status of children of working mothers and those of unemployed mothers.

Although prior research has examined the bidirectional relationship between maternal employment and child health, the specific effect of childhood stunting on maternal labour force participation (MLFP) remains underexplored, particularly in South Africa, where stunting among children under five is prevalent. Understanding this relationship is crucial when conducting a cost-benefit analysis (CBA) of interventions aimed at improving public health, reducing household poverty, and supporting maternal employment. To address this gap, this paper contributes to the empirical evidence base by investigating the relationship between child stunting and MLFP in South Africa.

2. Method

2.1. Data and study population

This study used secondary data from the 2017 National Income Dynamics Study (NIDS).2 The NIDS is the first nationally representative South African household panel dataset designed to track changes in the welfare of South Africans over time (Chinhema et al., 2016). The first NIDS was implemented in 2008 by the Southern Africa Labour and Development Research Unit (SALDRU) at the University of Cape Town. The sample for the analysis was limited to women of reproductive and working age, that is, those aged 16–45 years (WHO, 2006a), who have children aged 0–59 months and are their primary caregivers.3 It should be noted that some mothers had more than one child under 5 years. To avoid duplicating mothers’ characteristics, we controlled for the youngest child under her care, and duplicates were dropped.

2.2. Empirical strategy

A univariate or bivariate probit model could be used to estimate the effect of a stunted child (SC) in a household on MLFP. To determine which model is appropriate, a sensitivity test was done. Results were interpreted using the conventional significance levels of 1 %, 5 %, and 10 %, based on the corresponding p-values. Findings with p-values $\le$ 0.001 were considered strongly statistically significant, those with p-values $\le$ 0.05 statistically significant, and those with p-values $\le$ 0.10 weakly statistically significant.

2.2.1. Sensitivity analysis

Exogeneity of SC: To address potential endogeneity in the analysis, instrumental variables (IVs) were used. The Wald test was used to test the validity of the instrumental variables (IVs). The null hypothesis of exogeneity is $:\rho =0$ (Wooldridge, 2010). In this analysis, $\rho \ne 0$ and the bivariate probit was considered the appropriate model.

2.2.1.1. Bivariate probit model

The estimation of the parameters was based on the likelihood function of the joint distribution of (MLFP; SC) given exogenous variables $X_i$ and $Z_i$. To simplify the notations let $Y_1=MLFP$ and $Y_2=SC$: The joint distribution of $\left(Y_1,Y_2\right)$ conditional on $X$, $Z$ can be decomposed as:

$$f\left(Y_1,Y_2|X,Z,\theta \right)=f\left(Y_1|Y_2,X,Z,\theta \right)f\left(Y_2|Z,\theta \right)$$ (1)

where $\theta \text{,}$ is the vector of the model parameters. The average marginal effect was computed taking into consideration the endogeneity of stunting. To control for the endogeneity problem, valid instruments, $Z_i$, for ${SC}_i$ were needed to obtain consistent and valid estimates. For $Z_i$ to be valid: (i) It must be correlated with the suspected endogenous variable, ${SC}_i$; (ii) It must be uncorrelated with the error term (${\mu }_i$) that is, the instrument must be uncorrelated with the dependent variable $cov\left[Z_i,{\mu }_i\right]=0$. Condition (ii) of this instrumental variable framework is vital for identification purposes.

2.2.2. Key variables and definitions

MLFP: The narrow definition of MLFP is the internationally accepted definition of LFP by the ILO and is used by the South African government (Oosthuizen & Bhorat, 2005). It was used in this study for international comparability as recommended by the ILO (1995). Thus, MLFP is a dummy variable that equals one if the mother was employed or a job seeker and zero if non-economically active over the last 4 weeks. Given that prior research (Kingdon & Knight, 2004; Oosthuizen & Bhorat, 2005) suggests that the broad definition of LFP (which includes discouraged unemployed) is appropriate for SA, it was used to check for the robustness of the results. Given the relatively small size of the informal sector in South Africa (SA) compared to most other African countries (Kingdon and Knight 2007), and the absence of informal sector variables in the dataset, the analysis did not differentiate between participation in the formal and informal labour markets.

Stunting: Children aged 0–59 months whose height-for-age z-score (HAZ) fell below −2 standard deviation (SD) from the WHO's Child Growth Standards median were considered as stunted (Rachmi et al., 2016; Rashad & Sharaf, 2019; WHO, 2006b). Stunting values that fall outside the range of ±6 SD around the cut-off points from the WHO Anthro Survey Analyser (version 3.2.2, 2011) for the Child Growth Standards were flagged as biologically implausible. We controlled for mother, child and household characteristics.

2.2.2.1. Other correlates

Other potential correlates controlled for within this study include age, gender, race, marital status, area of residence (rural vs urban), education, household access to governmental grants, educational level of the mother and health status, the presence of an employed male and other adult females within the household.

Age (mother and child): Age was estimated as a continuous variable. Squared age was analysed to account for the potential non-linear effect of age through a quadratic relationship (the effect could increase with age up to a certain point and then decrease).

Gender of the child: In most literature, stunting significantly affects boys more than girls (Bukusuba, Kaaya and Atukwase 2017; Cruz et al., 2017; Manggala et al., 2018). As such, the child's gender was controlled for in the regression analysis. A dummy variable was generated with 1 if the child was a boy and 0 otherwise. The reference group is a girl.

Race of the mother: Although in computing the population-level analysis Asians4 and Whites were included, the two groups were excluded in the sub-analysis and regression analyses due to their small sample size. A dummy variable was generated with 1 if the mother is Coloured and 0 otherwise. The reference group was African.

Marital status: The maternal marital status variable was derived from a five-level question regarding the individual's current marital status: Possible responses were 1: Married, 2: Living with partner, 3: Divorced or Separated, 4: Widow/Widower and 5: Never Married. From these, two categories were created to ease analysis. “Married” which combined 1 and 2, and “Single” which combined 3, 4 and 5. A dummy variable was generated with 1 if the mother was married and 0 otherwise. The reference group was single. This variable is important because the theory of marriage/cohabitation and labour market participation, suggests that married women are less likely to be economically active than their single counterparts (Becker, 1981).

Area of residence: The area of residence variable was derived from a four-level question regarding where the individual stays. The responses were from rural, tribal, urban and urban informal areas. The rural and tribal areas were grouped to form the variable rural areas and, the urban and urban informal areas were jointly classified as urban areas, for simplification and international comparison. A dummy variable was generated with 1 if the mother or child lives in the urban area and 0 otherwise. The reference group was rural.

Grants: Grants received by a household member from the government were used as a proxy for non-labour income. A dummy variable was created taking the value of 1 when a household member receives money from the government and “0” otherwise. The reference group was a household member who did not receive money from the government.

Education: The education variable was derived for mothers with at least 12 years of schooling and was called Matric+. A dummy variable was generated taking the value 1 if she had Matric+ and 0 otherwise.

Employed male: An employed male in a household may positively or negatively affect MLFP. In some cases, matching may occur between similarly employed husband and wife. On the other hand, men with assured jobs may discourage their wives or relatives from working because they feel their income will be sufficient to run the house (Majumder, 2012). The variable was constructed as a continuous variable.

Female adult at home: Adult females in a household can provide additional childcare. The number of women aged 18–85 years who are not sick in a household was controlled for in the regression analysis. Additionally, the strong correlation (0.77) between household size and the number of adult females aged 18–85 years in the household also motivated the inclusion of the number of adult females in the analysis. The variable was included as a continuous variable.

Health status: The health status variable was derived from a five-level question regarding how the mother rates her health, with 1 = Excellent, 2 = Very good, 3 = Good, 4 = Fair, and 5 = Poor. The health status was further categorised into two groups: either sick or healthy. Sick represents those who reported their health is fair or poor, while healthy are respondents who declared their health is excellent, very good, or good. A dummy variable was generated taking the value 1 if the mother is healthy and 0 otherwise. The reference group was sick.

2.2.3. Instrumental Variables (IVs)

Although finding the appropriate instrument using NIDS was difficult, we selected a couple of potential variables that could be instruments.

Mother's height: Respondents' height was measured in centimetres (cm). It is widely accepted that there is a positive correlation between parental short stature, particularly that of mothers and child stunting (Berhe et al., 2019; Kamiya et al., 2018; Khan, Wu et al., 2021; Zaheer and Safdar 2019). This indicates that it will pass the first condition test. Nonetheless, there is no guarantee that it will pass the second condition test of instrument validity. If a particular job has certain criteria, for instance, modelling, which requires a certain height, then LFP and maternal height might be correlated. In such situations, maternal height may not pass the second condition test. In this analysis, it is assumed that there is no direct relationship between maternal height and MLFP, as adult height is biologically fixed and does not inherently determine employment participation. Furthermore, the dataset does not include height-specific occupational categories, which supports the validity of this assumption. The mother's height was defined as a dummy variable, which takes the value of 1 when the mother's height is < 150 cm and 0 otherwise.

Improved water source: Respondents were asked about their main source of water. Some studies have found a negative relationship between the water source and childhood stunting (Adekanmbi, Kayode and Uthman 2013; Fink et al., 2011). A dummy variable was created taking the value of 1 when the household has piped tap water in the dwelling, on site or in the yard and 0 otherwise.

Given that these IVs are treated as directly affecting stunting, there is no compelling reason to include them directly in the MLFP equation with stunting as a covariate. It is therefore assumed that it affects MLFP indirectly through their influence on stunting. The broad definition of MLFP was used to determine the robustness of the analysis, given its applicability in SA. In addition, the robustness of the results was tested using a reduced regression model that excluded being a male child and the presence of an adult female and male in the household.

3. Result

3.1. Descriptive analysis

Table 1 presents the descriptive statistics of the variables used in the childhood stunting-MLFP analysis. Approximately 57.4 % of the mothers participated in the LF, and 20.7 % of the children were classified as being stunted. The average age of the mother and child was 29 years and 26 months, respectively. Half of the children in the dataset were boys, with less than half of the mothers classified as coloured, married, having matric and above, living in a household with at least an employed male, and short in stature. More than half of the mothers reside in urban areas and a household with a government grant recipient or improved water supply, and are healthy. Table 1 shows the statistical summary of the variables used in the regression.

Table 1.

Descriptive statistics of variables.

Variables	N	Value
Mother's age (years- mean)	2439	29.0
Labour force participation (%)	2435	57.4
Stunted child (%)	1880	21.1
Child's age (months – mean)	2439	26.0
Boy child (%)	2438	50.6
Healthy (%)	2437	94.9
Coloured (%)	2437	10.2
Urban (%)	2439	58.0
Married (%)	2436	37.8
Grant (%)	2437	75.6
Matric+ (%)	2426	47.9
At least one Employed Male in Household (%)	2439	0.3
Number of adult females (mean)	2439	0.7
Mother's short stature (%)	2428	6.1
Improved water source (%)	2438	72.9

Source: Own calculations using NIDS 2017 and post-stratification weight

The results in Table 2 show that for the narrow (strict) definition of LFP, 20.7 % of mothers with stunted children participated in the LF as opposed to 21.7 % who did not. A difference of 1 %. Likewise, for the broad definition, the difference between mothers of stunted children who did not participate in the labour market was 0.5 %. The results indicate a positive correlation between better nutritional status of children and MLFP, although the difference is statistically insignificant.

Table 2.

Childhood stunting-MLFP status in 2017.

LFP status	Child Stunted (%) [95 % CI]	Child Not Stunted (%) [95 % CI]	Row Total(N)
Strict LFP	20.7[17.0–24.9]	79.3[75.1–83.0]	1032
Strict Non-LFP	21.7[18.0–26.0]	78.2[74.1–82.0]	845
Broad LFP	20.9[17.3–25.0]	79.1[75.0–82.7]	1084
Broad Non-LFP	21.4[17.6–25.8]	78.6[74.2–82.4]	793

Source: Own calculations using NIDS 2017 and post-stratification weight

Table 3 provides preliminary evidence for the non-trivial correlation between stunting and the IVs. Amongst children whose mothers have normal stature, only 19.5 % [95 % CI: 16.7–22.5] reported being stunted, while 48.0 % [95 % CI: 35.6–60.6] have short mothers. A similar finding was obtained for an improved water source. The summary statistics of the distribution of the instruments across the child's stunting status are presented in Table 3.

Table 3.

Distribution of instruments across stunted children categories.

Instruments	Has Condition?	Stunted (%) [95 % CI]	Not Stunted (%) [95 % CI]	Raw Total (N)
Mother short stature
	Yes	48.0[35.6–60.6]	52.0[39.4–64.4]	146
	No	19.5[16.7–22.5]	80.6[77.5–83.3]	1729
Improved water source
	Yes	19.3[16.1–23.0]	80.7[77.0–83.9]	1195
	No	25.6[21.1–30.6]	74.4[69.4–78.9]	684

Source: Own calculations using NIDS 2017 and post-stratification weight

The analysis above shows substantial evidence for including maternal short stature and improved water source in the MLFP determination. It has been shown that (in Table 2) non-labour force participants, when compared with participants, are 1 % likely to report the presence of a stunted child. Determining the relationship between childhood stunting and MLFP is the focus of the subsequent analysis.

3.2. Empirical results

As shown in Table 4, the Wald test $\rho$ = 0 is 0.39, statistically significant at the 5 % significance level. Chi2 (1) = 4.25, prob > Chi2 = 0.04. Therefore, the null hypothesis was rejected and $\rho$ was significantly different from 0. Indicating SC is endogenous in this context.

Table 4.

Average marginal effects of child stunting on maternal labour force participation: Strict and broad MLFP models (with controls).

Variables	Strict MLFP (i)	Broad LFP (ii)
Stunted Child	−0.20∗∗∗ (0.06)	−0.18∗∗ (0.07)
Mother's Age	0.06∗∗∗ (0.01)	0.05∗∗∗ (0.01)
Mother's Age Square	−0.00∗∗∗ (0.00)	−0.00∗∗∗ (0.00)
Child's Age	0.00∗ (0.00)	0.01∗ (0.00)
Child's Age Square	−0.00 (0.00)	−0.00 (0.00)
Male	−0.01 (0.02)	−0.01 (0.02)
Healthy	0.00 (0.04)	0.00 (0.04)
Coloured	0.12∗∗∗ (0.03)	0.12∗∗∗ (0.03)
Urban	0.08∗∗∗ (0.02)	0.06∗∗∗ (0.02)
Married	−0.02 (0.02)	−0.04 (0.02)
Grant	−0.00 (0.02)	−0.00 (0.02)
Matric+	0.16∗∗∗ (0.03)	0.15∗∗∗ (0.03)
Number of adult females	0.02∗∗ (0.01)	0.02 (0.01)
Employed Males	−0.00 (0.02)	0.01 (0.02)
Rho	0.39∗∗ (0.17)	0.36∗ (0.18)
N	1858	1858

Source: Own calculations using NIDS 2017, robust standard errors in parentheses: ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1

The results in Table 4 (i) show that the coefficient for stunted children is −0.20 and is statistically significant at a 1 % significance level (p-value = 0.001). This implies that, on average, a 1 % increase in the rate of stunting while holding other factors fixed will decrease the probability of MLFP by 20 % (based on the strict definition of MLFP). The magnitude and robustness of the result indicate a strong and substantive association between childhood stunting and reduced maternal labour force participation. The variables, child's age square, and being married, were found to be statistically insignificant. The presence of another adult female in the household was seen to be positively associated with MLFP at 5 % level of statistical significance. This implies that a 1 % increase in the number of adult females in the household increased the probability of MLFP by 2 %, on average, while holding other factors fixed. Given p-value = 0.04, this indicates moderate evidence of the effect. Table 4 presents the regression results of the bivariate probit model and covariates on both the strict and broad MLFP probability, reporting the average marginal effect.

For the broad MLFP in Table 4 (ii), it is also seen that there is a negative relationship between caring for a stunted child and MLFP. The result indicates statistical significance, as the p-value = 0.004. Therefore, caring for a stunted child can potentially decrease a mother's decision to work, indicating that the main finding is robust to alternative model specifications. For the other control variables, a similar outcome was observed as in Table 4 (i) except for the presence of an employed male in the household, where a statistically insignificant positive outcome was observed. Also, the presence of an adult female in the household was statistically insignificant. Table 5 presents another robustness check of the results using a reduced regression model that excludes being a male child and the presence of an adult female and male in the household.

Table 5.

Average marginal effects of child stunting on maternal labour force participation: Strict and broad MLFP models (with controls).

Variables	Strict MLFP (i)	Broad LFP (ii)
Stunted Child	−0.20∗∗∗ (0.06)	−0.17∗∗ (0.07)
Mother's Age	0.06∗∗∗ (0.01)	0.05∗∗∗ (0.01)
Mother's Age Square	−0.00∗∗∗ (0.00)	−0.00∗∗∗ (0.00)
Child's Age	0.00∗ (0.00)	0.01∗ (0.00)
Child's Age Square	−0.00 (0.00)	−0.00 (0.00)
Healthy	0.00 (0.04)	0.00 (0.04)
Coloured	0.12∗∗∗ (0.03)	0.12∗∗∗ (0.03)
Urban	0.08∗∗∗ (0.02)	0.06∗∗∗ (0.02)
Married	−0.03∗ (0.02)	−0.04 (0.02)
Grant	0.00 (0.02)	0.00 (0.02)
Matric+	0.16∗∗∗ (0.03)	0.16∗∗∗ (0.03)
Rho	0.38∗∗ (0.17)	0.36∗ (0.18)
N	1858	1858

Based on the reduced model, the results in Table 5 are similar to those in Table 4, indicating that the main findings are robust to the exclusion of household composition variables and being a male child.

4. Discussion

In evaluating the regression of the presence of a stunted child on MLFP, given the assumption that childhood stunting is endogenous, the instrumental variables used (mother's short stature and improved water source) passed both the first and second tests of instrument validity. This aligns with other findings, such as Berhe et al., 2019; Khan, Zaheer and Safdar 2019; Wu et al., 2021), who observed a positive correlation between parental short stature, particularly that of mothers and child stunting. The Wald test of exogeneity of stunting was rejected, and the bivariate probit model was considered the valid model. However, the univariate model revealed that, on average, caring for a stunted child reduces MLFP by 2 % but the result was statistically insignificant at the conventional level of significance. The results suggest that, on average, the presence of a stunted child is a significant determinant of MLFP. Thus, reducing MLFP by 20 % at a 1 % level of statistical significance, while holding other factors fixed at their sample means. Given that childhood stunting is an indicator of poor child health, this finding is consistent with those of Tambi and Nkwelle (2013) and Tambi (2017), who observed a negative relationship between child health and MLFP. Conversely, Zimmer (2007) observed no significant effect of child health on MLFP.

Childhood stunting may negatively affect a mother's LFP because children who are sick require more attention. This may force their mothers to reduce their time for market work or exit the labour market completely to care for them. These mothers are more likely to engage in time-intensive than goods-intensive activities, especially if they are the secondary earners within the family (Kleven, Kreiner and Saez 2009). In South Africa's patriarchal family settings, where many fathers are absent, the caregiving burden falls disproportionately on mothers. This may cause their reservation wages to increase, as increased health-related costs make them less willing or able to accept low-paying jobs. Nonetheless, this may not hold universally. In some contexts, mothers facing their children's nutritional or health challenges may increase their labour supply out of necessity. Cidav, Marcus and Mandell (2012) show that mothers may enter the labour force or extend their working hours to meet their child's medical and other financial demands. Moreover, cultural norms often place full responsibility for child health and nutrition on mothers, and social pressures to avoid stigma associated with malnourishment can further compel them to seek paid work (Kyanjo et al., 2025).

For some of the control variables, the a priori expectations were met. For instance, the findings on maternal age for both models indicate that MLFP increases up to a certain age and later decreases at an older age, which is similar to that observed by Ntuli and Wittenberg (2013) in their study on the determinants of black women's LFP in Ntuli and Wittenberg (2013) found that those aged between 15 and 19 years have a significantly higher LFP when compared with those 55–59 years old. The likelihood of participating in the labour market decrease with age (Sefiddashti et al., 2016) and some researchers (Ahmed, Andersson and Hammarstedt 2012; Watermann Fasbende, Klehe 2023) hypothesised that this decrease also results from age discrimination by employers.

Racially, Coloureds are more likely to participate in LF than Africans. This means that in post-Apartheid SA, race still plays an important role in determining MLFP. This is similar to other research findings (Leibbrandt et al., 2010; Gradín, 2019) and could be attributed to the socio-economic disparities between the different sub-populations in South Africa. Those living in urban areas are more likely to participate in the labour market than those in rural areas. This is similar to the findings of Wittenberg (2002); Mbatha and Roodt (2014); and Chatterjee, Murgai and Rama (2015) and can be linked to labour market opportunities available in urban areas. Contrary to these findings, Biswas and Banu (2023) found that rural women in India are more economically empowered, measured based on work participation, than their urban counterparts despite the greater job opportunities and availability of higher remuneration in urban areas.

Having at least a matric+ (≥12 years of schooling) is associated with increased LFP. Completion of higher education is more likely to result in a higher likelihood of LFP (Borjas, 2009; Branson et al., 2012). In most instances, education is used by job seekers to signal their intrinsic productive characteristics that employers cannot observe (Wanka & Rena, 2019). Therefore, education signals the existence of human capital, thus solving the problem of information asymmetry between employers and job seekers (Page, 2010). In Bolivia, the correlation between education and female labour force participation (FLFP) was not statistically significant except for tertiary schooling, which affects participation positively (Klasen et al., 2021). Many factors may cause the insignificant correlation between education and FLFP, such as skills mismatch, gender norms, cultural expectations and political barriers that limit women from realising their full potential.

The findings also reveal that the presence of another adult female aged 18–85 years in the household who is not sick gives room for mothers to take advantage of labour market opportunities. This is similar to the findings of Tambi (2017), who observed a positive relationship between the presence of a female sibling in the household and maternal labour supply. This might be the case because in a family with three generations or an adult female in the household, the grandparents or adult female might care for the children, relieving the mother from childcare. This decreases the mother's value of non-market time, hence, a drive to participate in the labour force (Compton & Pollak, 2014; Garcia-Moran & Kuehn, 2017; Shen, Yan and Zeng 2016).

5. Conclusion

The findings of this study underscore the impact of child stunting on maternal labour force participation (MLFP) in South Africa. Stunting, a manifestation of chronic malnutrition, not only compromises the health and developmental prospects of children but also imposes significant economic costs on their primary caregivers, predominantly their biological mothers. The bivariate probit regression model used in this study revealed that caring for a stunted child is associated with a 20 % reduction in MLFP, highlighting the need for policies to address this multifaceted issue.

Addressing the root causes should be the priority and the suite of interventions is well known. These include adequate social grants, vitamin supplementation, promoting exclusive breastfeeding for the first six months of life, and ensuring that communities have access to clean water and proper sanitation to reduce the incidence of diarrhoea. However, additional measures will be needed that target employed or work-seeking mothers. These could include the implementation of flexible work arrangements such as part-time work, remote work options, and flexible hours. Expanding access to healthcare services, including mental health support for mothers, is also important. The stress and psychological burden of caring for a malnourished child can contribute to maternal depression, further hindering labour force participation. Finally, lifelong learning opportunities for women should be supported so that women can improve their skills and capacity to seek better-paid employment.

By implementing these policies, the government improves the health and development of children as well as empowers mothers to participate more fully in the labour market. This, in turn, will contribute to economic growth and poverty reduction. Future research should continue to explore this relationship, considering the diverse socio-economic contexts and the potential for scalable interventions.

CRediT authorship contribution statement

Fru Awah Wanka: Writing – review & editing, Methodology, Funding acquisition, Conceptualization, Writing – original draft. Maame Esi Woode: Methodology, Writing – original draft, Writing – review & editing. Coretta Maame Panyin Jonah: Conceptualization, Methodology, Writing – original draft, Writing – review & editing. Julian Douglas May: Conceptualization, Funding acquisition, Methodology, Writing – original draft, Writing – review & editing.

Ethical responsibilities of authors

The study utilized secondary data from the National Income Dynamics Study (NIDS), which is publicly available and anonymized, ensuring that no personally identifiable information was disclosed. Hence, additional ethical approval was not required. As part of a PhD, the research complies with the ethical standards of the University of the Western Cape and the guidelines of the NIDS. Preliminary results were presented at the 2019 conference of the Economic Society of South Africa, and feedback from the conference informed the revision of this paper.

Submission declaration statement

This article is not under consideration elsewhere. However, the preliminary findings were presented during the Economic Society of South Africa 2019 conference. The article will not be submitted for publication elsewhere without the agreement of the Managing Editor.

Information on rights and permissions

All protocols were observed.

Ethical statement

The study utilized secondary data from the National Income Dynamics Study (NIDS), which is publicly available and anonymized, ensuring that no personally identifiable information was disclosed. Hence, additional ethical approval was not required. As part of a PhD, the research complies with the ethical standards of the University of the Western Cape and the guidelines of the NIDS. Preliminary results were presented at the 2019 conference of the Economic Society of South Africa, and feedback from the conference informed the revision of this paper.

Declaration of competing interest

The authors declare that there are no financial and personal relationships with other people or organizations that could inappropriately influence or bias their work.

Data availability

Data will be made available on request.