Wealth Paradox in Vietnam: Landholdings, Child Age, and Agricultural Labor Participation

Abstract

Despite economic growth, child labor persists in the agricultural sector. The wealth paradox—whereby increased land size leads to an increase in child labor—remains underexplored in the Vietnamese context. This study investigates this paradox in Dak Lak province, Vietnam, using data from the Thailand Vietnam Socio-Economic Panel. We examine the nonlinear relationship between cultivated land area and child farm labor, specifically analyzing how children’s age moderates this effect. We use a probit model to estimate the probability that children (aged 5–17 years) engage in household agricultural work as their primary occupation. We find a significant inverted-U-shaped relationship between land wealth and child labor, which supports the wealth paradox hypothesis. This effect is significantly stronger for children aged 16 years and above than for those aged 15 years and under. The probability of child labor peaks at land sizes of 3.4–3.6ha, which is substantially above the sample average, implying that future scale expansion could inadvertently increase child labor. Therefore, policies encouraging farm expansion should be complemented by interventions that reduce incentives for child labor, especially among younger children. We find no significant evidence that improved proximity to schools reduces this probability.

1.  Introduction

Child labor is a critical issue for individual well-being and broader economic development. It deprives children of educational opportunities and jeopardizes their future income, which can lead to a cycle of poverty. The agricultural sector accounts for 70% of child labor globally (ILO and UNICEF, 2021), with most of it occurring within the child’s household (FAO, 2021). This trend is evident in Dak Lak province in Vietnam’s Central Highlands, where a well-developed commercial agricultural sector—driven by coffee and pepper production—coexists with a relatively high child labor rate exceeding 10% (ILO and MOLISA, 2020). Reports indicate that child labor remains prevalent in the production of Vietnamese commodities, including cashew nuts, coffee, pepper, rice, rubber, sugarcane, tea, and tobacco (USDOL, 2020).

The literature on the causes of child labor can be broadly categorized into two groups: coping strategies and structural factors. Coping strategies are responses to temporary issues, such as production shocks (Dillon, 2013), unemployment of a family member (Fatima, 2017), and price fluctuations (Beck et al., 2019). Structural factors are long-term underlying issues, such as poverty (Basu and Van, 1998), land wealth (Bhalotra and Heady, 2003; Edmonds and Turk, 2004), and labor demand (Montt and Luu, 2020).

Poverty is widely regarded as the primary driver of child labor. Since Basu and Van (1998) introduced the “luxury axiom,” child labor has commonly been expected to disappear once a country or household escapes poverty. Indeed, the share of economically active children is negatively correlated with GDP per capita (Edmonds and Pavcnik, 2005b). In Vietnam, child labor, including agricultural work, decreased by nearly 30% between 1993 and 1997, with 80% of this reduction attributed to improvements in household expenditure (Edmonds, 2005). This suggests that if agricultural land generates sufficient income, child labor can be reduced.

Paradoxically, rather than reducing child labor, land wealth may lead to its persistence. For example, Bhalotra and Heady (2003) indicated that children on large farms are more likely to engage in agricultural labor than those on small farms. They explain this “wealth paradox” through imperfect labor and land markets. Land generates income, creating incentives for landowners to employ their children’s labor under imperfect market conditions (the incentive effect). At the same time, as land wealth increases, household income rises, and the relative economic benefit of child labor declines (the wealth effect). When the incentive effect outweighs the wealth effect, children are more likely to participate in agricultural labor. Basu et al. (2010) formalized this mechanism by deriving an inverted-U-shaped relationship, in which the probability of child labor initially rises with land wealth and then declines beyond a threshold. They confirmed this relationship using their own data.

A growing body of literature has examined the determinants of child labor in Vietnam, including export price increases (Edmonds and Pavcnik, 2005a), improvements in per capita expenditure (Edmonds, 2005), remittances (Binci and Giannelli, 2018; Nguyen and Nguyen, 2015), rainfall shocks (Trinh et al., 2020), access to credit (Nguyen and Anh, 2018), primary school expansion programs (Cornelissen and Dang, 2022), and firm agglomeration (Giang et al., 2017). Edmonds and Turk (2004) found that the child labor rate increased only in the Central Highlands during the 1990s, noting that development during that period did not benefit all ethnic groups equally. Giang et al. (2021) examined the relationship between socioeconomic factors and child labor. Despite this extensive research, the wealth paradox in Vietnam remains unexplored. Therefore, our primary objective is to assess whether this paradox exists in the Vietnamese context.

We also aim to broaden the existing literature on the wealth paradox by exploring its relationship with child’s age. While most studies control for children’s age and focus primarily on identifying an inverted-U-shaped relationship, the effect of land wealth on child labor may differ systematically across age groups. By considering the interaction between children’s age and land wealth, we examine the differences in the land wealth effect across age groups.

Furthermore, given the inverted-U relationship between land wealth and child labor, child labor may naturally decrease as scale expands. However, this prediction assumes that some farmers will leave the agricultural sector (Barrett, 2008). If this assumption holds, it becomes crucial for children to spend their formative years in ways that protect them from future poverty after leaving farming. Considering these aspects, we examine how improved school proximity influences the probability of child labor.

We focus on Dak Lak province, a region with prevalent child labor, and utilize data from the Thailand Vietnam Socio-Economic Panel (TVSEP). We employ a probit model to analyze how land wealth influences the likelihood that children engage in agricultural work at home and examine whether this relationship varies with children’s age.

This study makes two key contributions to the literature. First, to the best of our knowledge, it provides the first evidence of the wealth paradox in Vietnam. Second, it analyzes the moderating effect of children’s age on the relationship between land wealth and child labor.

Unlike most studies, which have relied on the Vietnam Household Living Standards Survey (VHLSS), this study is the first quantitative analysis to use the TVSEP in the context of child labor research. The dataset has both advantages and limitations. Focusing on low-income, agriculture-dependent rural households in vulnerable environments, this survey provides a sample size twice that of the VHLSS at the provincial level. This is advantageous when analyzing difficult-to-observe phenomena such as child labor. However, the TVSEP lacks data on children’s working hours and the nature of their work, preventing us from strictly applying the International Labour Organization’s definition of child labor¹. Despite these limitations, we leverage its strengths to explore child labor in Dak Lak province.

2.  Theoretical framework

We build on the theoretical model proposed by Oryoie et al. (2017). While the assumptions and basic framework remain the same, we modify the definition of the disutility of child labor. Oryoie et al. (2017) assumed that the disutility of child labor is greater for households with larger landholdings. We extend this by allowing the disutility of child labor to decrease as the distance to school or the child’s age increases.

This extended model yields three theoretical hypotheses. First, the relationship between land wealth and child labor can be represented by either an inverted-U-shaped curve (quadratic specification) or an inverted-N-shaped curve (cubic specification). Second, a longer distance to school or an increase in the child’s age raises the incidence of child labor, shifting the inverted-U-shaped or inverted-N-shaped curve upward. Third, an increase in the child’s age nonlinearly strengthens the incentive for child labor by boosting labor productivity. Therefore, the impact of a one-unit increase in land on the probability of child labor is larger for older children. The model is described in detail in online appendix B.

3.  Methodology

(1)  Data and location

We used data from the TVSEP. The sampling was a multistage process. First, poor, remote, and agriculture-dependent provinces were purposively selected. Next, communes and villages were chosen based on population size within stratified geographic areas. Finally, ten households were randomly selected from each village (Hardeweg et al., 2013). Owing to data availability, our analysis utilized the 2016 wave for village-level data. For all other household or individual data, we used the 2017 wave.

(2)  Empirical model

We first compared two probit models to determine whether a quadratic (Eq. 1) or cubic specification (Eq. 2) is more appropriate. $Y_i$ is a dummy variable that takes a value of 1 if child i engaged in agricultural labor at home as their primary occupation and 0 otherwise. $X_h$ is the land wealth of household h. $W_{vhi}^{\prime }$ is a set of control variables representing village, household, and child characteristics. Some control variables that are strictly positive are log-transformed to reduce distributional skewness and mitigate the risk of complete separation². $\Phi$ represents the standard normal cumulative distribution function.

  

$$\begin{array}{c}P(Y_i=1|X_h,W_{vhi})\hfill \\ =\Phi ({\alpha }_0+{\alpha }_1{X}_h+{\alpha }_2{X}_h^2+W_{vhi}^{\prime }{\alpha }_3)\hfill \end{array}$$(1)

  

$$\begin{array}{c}P(Y_i=1|X_h,W_{vhi})\hfill \\ =\Phi ({\alpha }_4+{\alpha }_5{X}_h+{\alpha }_6{X}_h^2+{\alpha }_7{X}_h^3+W_{vhi}^{\prime }{\alpha }_8)\hfill \end{array}$$(2)

We then construct the interaction term based on the selected model to examine how children’s age influences the land wealth effect. For model (1), we use Eq. (3), and for model (2), we use Eq. (4). $M_i$ is the moderating variable, children’s age, constructed by creating two categories: “15 years old and under” and “16 years old and above.”³ Table 1 presents the definitions and summary statistics of all variables.

Table 1. 

Variable definitions and descriptive statistics

Variables	Definition1)	Y = 0 (N = 558)2)	Y = 1 (N = 54)2)	Test3)
ldsizeCulti	Area cultivated (0.1ha) by h	16.08 (20.48)	16.02 (11.20)	0.982
gen4)	Gender of child i
female		287 (51.4%)	24 (44.4%)	0.327
male		271 (48.6%)	30 (55.6%)
age_group4)	Age group of child i
15 and under		443 (79.4%)	13 (24.1%)	<0.001
16 and above		115 (20.6%)	41 (75.9%)
ethnicity4)	Ethnicity of child i
majority		307 (55.0%)	9 (16.7%)	<0.001
minority		251 (45.0%)	45 (83.3%)
HHhead_age	Age of household head	47.01 (10.64)	48.20 (11.98)	0.436
HHhead_edu	Schooling years of household head	5.96 (4.07)	3.46 (2.94)	<0.001
relation4)	Relationship between household head and child i
son or daughter		462 (82.8%)	49 (90.7%)	0.133
others		96 (17.2%)	5 (9.3%)
elder	Number of older adults (65–) in h	0.22 (0.52)	0.22 (0.42)	0.980
adults	Number of adults (18–64) in h	2.80 (1.13)	3.13 (1.32)	0.045
infant	Number of infants (0–4) in h	0.28 (0.55)	0.30 (0.57)	0.868
child	Number of children (5–17) in h	2.03 (0.86)	2.15 (0.94)	0.359
asset	Value of assets h owned (excluding agricultural assets)5)	39.25 (44.86)	16.91 (17.50)	<0.001
livestock	Value of livestock h owns5)	16.25 (30.24)	20.87 (33.64)	0.289
YExpCap_cat4)	Four quantiles of annual expenditure per capita
1st quantile		126 (22.6%)	27 (50.0%)	<0.001
2nd quantile		140 (25.1%)	14 (25.9%)
3rd quantile		140 (25.1%)	12 (22.2%)
4th quantile		152 (27.2%)	1 (1.9%)
DistPSC	Distance to primary school (km)	0.70 (1.02)	0.58 (0.76)	0.382
DistJSS	Distance to junior secondary school (km)	2.74 (3.85)	2.24 (1.91)	0.348
DistSSS	Distance to senior secondary school (km)	9.37 (10.70)	10.65 (12.04)	0.409
PopTotal	Village population (1,000 people)	1.02 (0.63)	1.12 (0.69)	0.243
DistPCap	Distance to provincial capital from village (km)	46.95 (29.53)	51.78 (37.57)	0.264

Source: Calculated by the authors based on TVSEP data.

1) Subscript i indicates an individual (child) and h indicates a household.

2) For continuous variables, the figures show the mean, with standard deviations in parentheses. For categorical variables, the figures show frequency, with percentages in parentheses.

3) The “Test” column reports p-values from t-tests (continuous variables) or Pearson χ2 tests (categorical variables).

4) The baseline category is reported in the first row.

5) The unit is Million VND.

  

$$\begin{array}{c}P(Y_i=1|X_h,M_i,W_{vhi})\hfill \\ =\Phi \{{\beta }_0+{\beta }_1{X}_h+{\beta }_2{X}_h^2+{\beta }_3{M}_i+\hfill \\ {{\beta }_{4}X}_h{M}_i+{{\beta }_{5}X}_h^2{M}_i+W_{vhi}^{\prime }{\beta }_6\}\hfill \end{array}$$(3)

  

$$\begin{array}{c}P(Y_i=1|X_h,M_i,W_{vhi})\hfill \\ =\Phi \{{\gamma }_0+{\gamma }_1{X}_h+{\gamma }_2{X}_h^2+{\gamma }_3{X}_h^3+{\gamma }_4{M}_i+\hfill \\ {{\gamma }_{5}X}_h{M}_i+{{\gamma }_{6}X}_h^2{M}_i+{{\gamma }_{7}X}_h^3{M}_i+W_{vhi}^{\prime }{\gamma }_8\}\hfill \end{array}$$(4)

(3)  Child labor

This study defines child labor as children (5–17 years old) who spend most of their time engaging in agricultural work at home. According to TVSEP data, “Student/Pupil” is the largest primary occupation category for children, followed by “Household farm labor,” which encompasses children who engage in family agricultural labor. The dataset includes 612 children from 360 households in Dak Lak province. Among them, 54 fit our definition of child labor. Of these children, 3 never enrolled in school, 1 attended school but spent less time studying than working, and the remaining 50 had left school: 4 from elementary, 26 from junior secondary, and 20 from senior secondary school⁴.

Although the TVSEP also covers Ha Tinh and Thua Thien Hue provinces, most child labor in Ha Tinh province occurs in the non-agricultural sector. In contrast, the proportion of child labor in Thua Thien Hue province is relatively low (Table A1, online appendix). Furthermore, when addressing the potential endogeneity issues discussed in the following subsection, we must consider Dak Lak’s unique history of land clearing and its distinct land market. Therefore, this study focuses on child labor in Dak Lak province.

(4)  Land wealth

Land wealth is defined as the area of cultivated land. This definition differs from that used in some previous studies; however, we consider it to be more appropriate for Dak Lak province for two reasons. First, in Dak Lak province, permanent crops are often grown on residential land that is not formally registered as agricultural land. Our definition includes productive land used for farming, even if unregistered. Second, given that land wealth should be understood as the productive area that can be potentially cultivated, the possibility of double-cropping should be considered. Double-cropping is common in Dak Lak province, consistent with our dataset, which shows that the cultivated area of paddy fields is approximately 1.5 times the owned area. Thus, “area that is actually cultivated” is more appropriate than “area the households own.”

However, this definition has two drawbacks. First, it includes both leased and sharecropped land. Although the literature often does not categorize these as land wealth, this may be a minor issue, as they account for only 4.4% of the total land in the dataset. Second, cultivated area may be endogenous to child labor because decisions regarding land cultivation likely depend on the availability of child labor. Specifically, parents may expand their farmland in anticipation of their children growing up and participating in agricultural labor. While some studies assume that land wealth is exogenous because most land is inherited and difficult to sell or buy (Lima et al., 2015; Rosenzweig and Wolpin, 1985), this assumption does not apply to Dak Lak province.

After the Vietnam War ended in 1975, the Central Highlands were targeted for the resettlement of the Kinh people, Vietnam’s majority ethnic group. Following the introduction of the Doi Moi reforms in 1986, voluntary migration to the Central Highlands increased rapidly, driven by agricultural liberalization and rising coffee prices (Meyfroidt et al., 2013). Households were able to expand their landholdings through the adoption of the 1993 Land Law, which permitted the transfer, exchange, inheritance, rental, and mortgage of land (Do and Iyer, 2008). The surge in coffee prices in 1993/94 further accelerated in-migration, as many people sought to plant coffee (Dang, 2003). The Ede people, an ethnic minority native to Dak Lak, initially practiced nomadic slash-and-burn agriculture but later recognized the land’s economic value and began planting coffee (Dang, 2003). Eventually, the forest area managed by the Ede community was cleared and converted into household-owned farms. Thus, the parents of working children in 2017 may have acquired or expanded their farmland in the 1990s. In fact, in the data we used, purchasing was the most common method of acquiring land, followed by clearing forests. However, using the two-stage residual inclusion approach (Papies et al., 2017; Wooldridge, 2015), we find no evidence of significant endogeneity (online appendix C). Therefore, this study treats land wealth as an exogenous variable.

(5)  Distance to school

Three important caveats should be noted when interpreting the distance-to-school variable. First, the distance is measured from the village leader’s home. If a school is in the village, the distance is recorded as zero. Households were randomly selected; therefore, the distance from the leader’s house to individual households was considered random, mitigating systematic bias. However, interpreting the non-zero distances still requires caution, as they indicate the absence of a school within the village. In such cases, children must attend school in another village, and ethnic minority students may face significant challenges due to a lack of linguistic, cultural, or instructional support (World Bank, 2009). Many villages in the dataset are ethnically homogeneous; thus, the non-zero distance may indicate a lack of support for minority students, potentially reducing the disutility of choosing child labor over schooling. However, because we control for ethnicity, the distance-to-school variable mainly reflects the disutility of child labor owing to school proximity⁵.

Second, as students advance to higher levels of education, the geographic area served by a single school expands, leading to longer commutes. This creates a correlation between distance to school and children’s age, complicating the interpretation of whether the shift in the inverted-U-shaped curve is due to age or distance. Therefore, to eliminate this correlation, we treat the distances to elementary, middle, and high schools as separate variables in the model. This approach aligns with our theoretical model (online appendix B), which treats distance to school and the children’s age independently.

Third, “distance to school” as defined above refers to the proximity to the nearest school, and not the distance to the specific school in which the child is enrolled or intends to enroll.

4.  Results and discussion

Table 2 presents the estimation results. The reported figures represent the estimated coefficients. Columns (1) and (2) show the basic models without the interaction term fitted using quadratic and cubic polynomials, respectively. Based on the significance of the coefficients for land (ldsizeCulti), there is no statistical evidence for including the cubic term for land in the model, suggesting that a quadratic polynomial is appropriate. Furthermore, we plotted the land wealth effect on the probability of child labor (Fig. 1), which also shows an inverted-U rather than an inverted-N curve. Thus, we determined that Eq. (3) is more appropriate than Eq. (4) to examine the moderating effect of children’s age.

Table 2. 

Estimation results¹⁾

	(1)		(2)		(3)
Variables	Y	SE	Y	SE	Y	SE
ldsizeCulti	0.074***	(0.024)	0.084**	(0.033)	0.063**	(0.027)
ldsizeCulti2	−0.001**	(0.000)	−0.002	(0.001)	−0.001	(0.001)
ldsizeCulti3			0.000	(0.000)
DistPSC	−0.159	(0.115)	−0.160	(0.114)	−0.159	(0.115)
DistJSS	−0.009	(0.027)	−0.009	(0.027)	−0.009	(0.026)
DistSSS	−0.005	(0.009)	−0.005	(0.009)	−0.005	(0.009)
age_group2)	1.518***	(0.209)	1.518***	(0.209)	1.352***	(0.321)
age_group#ldsizeCulti					0.019	(0.035)
age_group#ldsizeCulti2					−0.000	(0.001)
gen2)	0.256	(0.188)	0.253	(0.187)	0.250	(0.184)
ethnicity2)	0.552**	(0.264)	0.555**	(0.265)	0.547**	(0.267)
relation2)	−0.415	(0.405)	−0.414	(0.404)	−0.404	(0.408)
HHhead_age	0.015	(0.015)	0.015	(0.015)	0.015	(0.015)
HHhead_edu	−0.011	(0.031)	−0.012	(0.031)	−0.011	(0.031)
YExpCap_cat2)3)
4th quantile	−1.475***	(0.526)	−1.459***	(0.522)	−1.492***	(0.539)
3rd quantile	−0.293	(0.343)	−0.286	(0.345)	−0.288	(0.340)
2nd quantile	−0.354	(0.277)	−0.352	(0.277)	−0.354	(0.275)
asset_log3)4)	−0.169*	(0.087)	−0.172**	(0.087)	−0.166*	(0.086)
livestock3)	0.002	(0.003)	0.002	(0.003)	0.002	(0.003)
elder	0.021	(0.240)	0.018	(0.239)	0.020	(0.242)
adults	−0.033	(0.089)	−0.035	(0.088)	−0.033	(0.087)
infant	−0.211	(0.170)	−0.210	(0.170)	−0.212	(0.169)
child	0.064	(0.130)	0.065	(0.129)	0.069	(0.128)
PopTotal_log4)	0.239	(0.172)	0.237	(0.171)	0.240	(0.171)
DistPCap_log4)	0.108	(0.175)	0.106	(0.177)	0.106	(0.172)
Constant	−3.277***	(0.995)	−3.304***	(0.992)	−3.196***	(1.015)
Observations5)	600		612		600
Pseudo R2	0.381		0.381		0.381

Source: Created by the authors based on TVSEP data.

1) Reported figures are coefficients. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively. Standard errors clustered at the household level are reported in parentheses.

2) Baselines for categorical variables are shown in Table 1.

3) With a maximum VIF of 3.54 among wealth-related variables, multicollinearity is not a serious concern.

4) The “_log” suffix indicates that the variables from Table 1 were log-transformed to mitigate distributional skewness.

5) From 612 observations, 12 that caused complete separation were excluded before the estimation in (1) and (3).

Fig. 1 

Relationship between land wealth and child labor¹⁾

Source: Created by the authors based on TVSEP data.

1) The left and right panels are based on Eqs. (1) and (2), respectively. Shaded areas represent 95% confidence intervals calculated using the delta method. The histogram shows the density of land size (ldsizeCulti) for households with children.

Column (3) presents the estimation results of Eq. (3). The statistical significance of the coefficient on the linear term of land area and the insignificance of the coefficient on the squared term indicate that, in the baseline age group category (15 and under), land wealth and child labor exhibit a linear relationship rather than a quadratic one. However, whether this holds for the non-baseline group (16 and above) cannot be determined based on Table 2 (Brambor et al., 2006). To address this limitation, we complemented our analysis by calculating the predicted probability (Fig. 2) and the average marginal effect by age category (Table 3). Fig. 2 shows an inverted-U relationship for the age group 16 and above⁶. When the land area is 3.4ha, the predicted probability of child labor reaches its maximum value of 0.44 and then declines. This is considerably higher than that of the age group 15 and under, which has a maximum value of 0.08 when the land area is 3.6ha. Their respective confidence intervals do not overlap; thus, the difference between the predicted probabilities is statistically significant. What is critical about these peak locations (3.4ha and 3.6ha) is that these values are more than double the average (Table 1)⁷. Therefore, if scale expansion proceeds while other conditions are kept constant⁸, it may contribute to increased child labor.

Fig. 2 

Changes in the land wealth effect by age group¹⁾

Source: Created by the authors based on TVSEP data.

1) The black lines with dark gray shaded areas and dashed lines with light gray shaded areas show the predicted probability with a 95% confidence interval at the “15 and under” and “16 and above”, respectively. Confidence intervals are calculated using the delta method. The histogram shows the density of land size (ldsizeCulti) for households with children.

Table 3. 

Average marginal effect of land wealth on child labor by age group¹⁾

	Average Marginal Effect	Standard Error	95% Confidence Interval
15 and under	0.0020**	0.0009	0.0003–0.0037
16 and above	0.0115***	0.0036	0.0044–0.0186
Observations	600

Source: Calculated by the authors based on TVSEP data.

1) Reported figures are the average marginal effect of land wealth (ldsizeCulti) on the probability of child labor. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively.

Although the slope of the inverted-U-shaped curve appears steeper for children aged 16 and older, given the width of the confidence intervals, it is difficult to determine visually whether the difference in slopes is statistically significant. Table 3 presents the marginal effects of land wealth on the probability of child labor by age group. Based on Williams (2011), the calculations were performed by first computing the marginal effect for each observation at its observed values and then averaging them. Consequently, this strongly reflects the marginal effect within the range where the observations are concentrated. The marginal effect of land on the probability of child labor was 0.0020 (p=0.019) for the group aged 15 and under and 0.0115 (p=0.001) for the group aged 16 and above. Their 95% confidence intervals do not overlap; therefore, this provides statistical evidence that increased age magnifies the effect of land wealth on child labor. This is consistent with the theoretical model, which assumes that age nonlinearly increases the incentive for child labor by boosting labor productivity.

The labor of children aged 15 and under can be regarded as harmful child labor. Thus, when adopting agricultural policies that aim to expand scale, or when natural scale expansion is anticipated, policies that reduce the incentive for child labor must be adopted simultaneously. These include policies that improve educational access (Basu et al., 2010). However, our findings suggest that improving school proximity alone, one aspect of access, may be insufficient. Therefore, it may be necessary to consider different forms of educational investments, such as improving school quality, reducing attendance costs, or providing scholarships. Given that the incentive for child labor among younger children is initially low, it may be possible to prevent child labor through small investments that slightly increase the utility or reduce the cost of education.

Regarding children aged 16 and above, the implications of their labor participation are contingent on the nature of the work—specifically, whether it is hazardous. However, this study lacks data on working hours and specific tasks necessary to make this distinction. In cases where labor is harmful, interventions to lower labor incentives are necessary, though achieving this for older children may be more challenging or costly than achieving it for younger children. In cases where labor is safe, however, it may positively contribute to agricultural development by fostering skills under parental supervision. Distinguishing between these cases is an indispensable task for future studies.

5.  Conclusion

This study examined the relationship between land wealth and child labor in Dak Lak province, Vietnam. Theoretical models suggested that the relationship between land and child labor followed either an inverted-U or an inverted-N shape. Using TVSEP data, we found an inverted-U-shaped wealth paradox. The probability of child labor increased with land size up to 3.4–3.6ha, holding other factors constant. In this scenario, policies aimed at reducing the incentives for child labor are required. Although we found no evidence that improving school proximity reduces child labor, future research should consider factors beyond physical distance, such as the quality of education or the availability of scholarships. Given the lower labor incentives for younger children, for whom the average marginal effect of land is approximately one-sixth that of older children, relatively small investments may be effective in reducing child labor among younger children. Without data on working conditions, we could not assess whether the labor performed by older children was hazardous. This represents a key question for future studies.

Acknowledgments

We sincerely appreciate the editors, anonymous reviewers and Dr. Magezi Eustadius Francis for the constructive comments. This study was supported by JST and the establishment of university fellowships for the creation of science technology innovation, Grant Number JPMJFS2102. This study relies on data from the long-term project No. 20220831434900116103, funded by the Deutsche Forschungsgemeinschaft (DFG). For detailed information, see http://www.tvsep.de.

Notes

1  The ILO considers the working environment and working hours appropriate for the children’s age.

2  The complete separation issue was not fully resolved, in Eqs. (1) and (3), mainly because of the skewed distribution of ldsizeCulti (X_h).

3  Ideally, we should divide the age groups at the age of 14, which is the boundary between junior and senior secondary school. However, of the 54 children who were working, only 7 were 14 or under (3 were 13 years old and 4 were 14 years old), which caused the problem of complete separation. Therefore, we divided the age groups at 15 years.

4  Primary school covers ages 6–10, junior secondary school covers ages 11–14, and senior secondary school covers ages 15–17.

5  Distance to school may also reflect physical fatigue and reduced study time for children, which causes them to drop out of school. However, we did not explicitly consider the problem on the children’s side because we assume maximization of household utility.

6  When the baseline category is the 16 and above, the squared term of land becomes significant (p=0.013), suggesting the statistically significant inverted-U curve in older children, as illustrated in Fig. 2.

7  Although the quadratic term for younger children was statistically insignificant, we maintain the assumption of an inverted-U-shaped relationship for this discussion. The lack of significance suggests either the absence of non-linearity (implying labor increases linearly with land) or that the relationship was not statistically detectable due to low power. The latter interpretation appears more plausible. Limited statistical power is likely driven by the low incidence of child labor and the scarcity of large landholders in the sample (see Figs. 1 and 2). Furthermore, it is theoretically implausible for younger children to increase labor supply as land wealth increases, specifically when such wealth is sufficient to allow older children to withdraw from work.

8  For example, industrial development could increase the returns to education, potentially overturning the scenario. We recognize the absence of this discussion as a limitation of our study.

References

•  Barrett,  C. B. (2008) Smallholder Market Participation: Concepts and Evidence from Eastern and Southern Africa, Food Policy 33(4): 299–317. https://doi.org/10.1016/j.foodpol.2007.10.005.
•  Basu,  K. and  P. H.  Van (1998) The Economics of Child Labor, The American Economic Review 88(3): 412–427.
•  Basu,  K.,  S.  Das, and  B.  Dutta (2010) Child Labor and Household Wealth: Theory and Empirical Evidence of an Inverted-U, Journal of Development Economics 91(1): 8–14. https://doi.org/10.1016/j.jdeveco.2009.01.006.
•  Beck,  U.,  S.  Singhal, and  F.  Tarp (2019) Commodity Prices and Intra-Household Labor Allocation, American Journal of Agricultural Economics 101(2): 436–454. https://doi.org/10.1093/ajae/aay082.
•  Bhalotra,  S. and  C.  Heady (2003) Child Farm Labor: The Wealth Paradox, The World Bank Economic Review 17(2): 197–227. https://doi.org/10.1093/wber/lhg017.
•  Binci,  M. and  G. C.  Giannelli (2018) Internal Versus International Migration: Impacts of Remittances on Child Labor and Schooling in Vietnam, International Migration Review 52(1): 43–65. https://doi.org/10.1111/imre.12267.
•  Brambor,  T.,  W. R.  Clark, and  M.  Golder (2006) Understanding Interaction Models: Improving Empirical Analyses, Political Analysis 14(1): 63–82. https://doi.org/10.1093/pan/mpi014.
•  Cornelissen,  T. and  T.  Dang (2022) The Multigenerational Impacts of Educational Expansion: Evidence from Vietnam, Labour Economics 78: 102243. https://doi.org/10.1016/j.labeco.2022.102243.
•  Dang,  D. T. (2003) Coffee Production, Social Stratification and Poverty in a Vietnamese Central Highland Community, in R. De Koninck, J. Lamarre, and B. Gendron, eds., Understanding Poverty in Vietnam and the Philippines: Concepts and Context, Laval , QC: Localized Poverty Reduction in Vietnam (LPRV) Project, 71–100.
•  Dillon,  A. (2013) Child Labour and Schooling Responses to Production and Health Shocks in Northern Mali, Journal of African Economies 22(2): 276–299. https://doi.org/10.1093/jae/ejs025.
•  Do,  Q. and  L.  Iyer (2008) Land Titling and Rural Transition in Vietnam, Economic Development and Cultural Change 56(3): 531–579. https://doi.org/10.1086/533549.
•  Edmonds,  E. V. (2005) Does Child Labor Decline with Improving Economic Status? The Journal of Human Resources 40(1): 77–99. https://doi.org/10.3368/jhr.xl.1.77.
•  Edmonds,  E. V. and  N.  Pavcnik (2005a) The Effect of Trade Liberalization on Child Labor, Journal of International Economics 65(2): 401–419. https://doi.org/10.1016/j.jinteco.2004.04.001.
•  Edmonds,  E. V. and  N.  Pavcnik (2005b) Child Labor in the Global Economy, Journal of Economic Perspectives 19(1): 199–220. https://doi.org/10.1257/0895330053147895.
•  Edmonds,  E. V. and  C.  Turk (2004) Child Labor in Transition in Vietnam, in P. Glewwe, N. Agrawal, and D. Dollar, eds., Economic Growth, Poverty, Household Welfare in Vietnam, Washington, DC: World Bank, 505–550.
•  Fatima,  A. (2017) Child labour in Pakistan: Addressing supply and demand side labour market dynamics, Economic and Labour Relations Review 28(2): 294–311. https://doi.org/10.1177/1035304617690332.
• Food and Agriculture Organization of the United Nations (FAO) (2021) Accelerating action to help to end child labour in agriculture in Asia—Regional Workshop on Ending Child Labour in Agriculture, 29 September 2021: Regional report , Bangkok: FAO. https://doi.org/10.4060/cb7135en.
•  Giang,  L. T.,  C. V.  Nguyen,  T. Q.  Tran, and  V.  Thieu (2017) Does Firm Agglomeration Matter to Labor and Education of Local Children? Evidence in Vietnam, Child Indicators Research 10(4): 1015–1041. https://doi.org/10.1007/s12187-016-9409-8.
•  Giang,  L. T.,  V. T.  Nguyen, and  T. D.  Nguyen (2021) Child Labor in Vietnam: Issues and Policy Implications, Children and Youth Services Review 130: 106222. https://doi.org/10.1016/j.childyouth.2021.106222.
•  Hardeweg,  B.,  S.  Klasen, and  H.  Waibel (2013) Establishing a Database for Vulnerability Assessment, in S. Klasen and H. Waibel, eds., Vulnerability to Poverty: Theory, Measurement and Determinants, with Case Studies from Thailand and Vietnam, London: Palgrave Macmillan, 50–79. https://doi.org/10.1057/9780230306622_3.
• International Labour Organization (ILO) and Ministry of Labour, Invalids and Social Affairs of Viet Nam (MOLISA) (2020) Viet Nam National Child Labour Survey 2018: Key findings, Geneva: ILO.
• International Labour Organization (ILO) and United Nations Children’s Fund (UNICEF) (2021) Child labour: Global Estimates 2020, trends and the road forward, New York: ILO and UNICEF.
•  Lima,  L. R.,  S.  Mesquita, and  M.  Wanamaker (2015) Child Labor and the Wealth Paradox: The Role of Altruistic Parents, Economics Letters 130: 80–82. https://doi.org/10.1016/j.econlet.2015.03.005.
•  Meyfroidt,  P.,  T. P.  Vu, and  V. A.  Hoang (2013) Trajectories of Deforestation, Coffee Expansion and Displacement of Shifting Cultivation in the Central Highlands of Vietnam, Global Environmental Change 23(5): 1187–1198. https://doi.org/10.1016/j.gloenvcha.2013.04.005.
•  Montt,  G. and  T.  Luu (2020) Does Conservation Agriculture Change Labour Requirements? Evidence of Sustainable Intensification in Sub‐Saharan Africa, Journal of Agricultural Economics 71(2): 556–580. https://doi.org/10.1111/1477-9552.12353.
•  Nguyen,  C. V. and  H. Q.  Nguyen (2015) Do Internal and International Remittances Matter to Health, Education and Labor of Children and Adolescents? The Case of Vietnam, Children and Youth Services Review 58: 28–34. https://doi.org/10.1016/j.childyouth.2015.09.002.
•  Nguyen,  V. D. and  M. N.  Anh (2018) Does Governmental Microcrediting Benefit Child Labour in the Poorest Regions? New Evidence from a Transitional Economy, Economics and Sociology 11(3): 333–344. https://doi.org/10.14254/2071-789X.2018/11-3/20.
•  Oryoie,  A. R.,  J.  Alwang, and  N.  Tideman (2017) Child Labor and Household Land Holding: Theory and Empirical Evidence from Zimbabwe, World Development 100: 45–58. https://doi.org/10.1016/j.worlddev.2017.07.025.
•  Papies,  D.,  P.  Ebbes, and  H. J.  Van Heerde (2017) Addressing Endogeneity in Marketing Models, in P. S. H. Leeflang, J. E. Wieringa, T. H. A. Bijmolt, and K. H. Pauwels, eds., Advanced Methods for Modeling Markets, Cham: Springer International Publishing, 581–627. https://doi.org/10.1007/978-3-319-53469-5_18.
•  Rosenzweig,  M. R. and  K. I.  Wolpin (1985) Specific Experience, Household Structure, and Intergenerational Transfers: Farm Family Land and Labor Arrangements in Developing Countries, The Quarterly Journal of Economics 100(Supplement): 961–987. https://doi.org/10.1093/qje/100.supplement.961.
•  Trinh,  T. A.,  A.  Posso, and  S.  Feeny (2020) Child Labor and Rainfall Deviation: Panel Data Evidence from Rural Vietnam, The Developing Economies 58(1): 63–76. https://doi.org/10.1111/deve.12215.
• United States Department of Labor (USDOL) (2020) 2020 List of Goods Produced by Child Labor or Forced Labor, https://www.dol.gov/sites/dolgov/files/ILAB/child_labor_reports/tda2019/2020_TVPRA_List_Online_Final.pdf (accessed on February 8, 2026).
•  Williams,  R. (2011) Using the Stata’s Margins Command to Estimate and Interpret Adjusted Predictions and Marginal Effects, https://www3.nd.edu/~rwilliam/stats/Margins01.pdf (accessed on February 8, 2026).
•  Wooldridge,  J. M. (2015) Control Function Methods in Applied Econometrics, The Journal of Human Resources 50(2): 420–445. https://doi.org/10.3368/jhr.50.2.420.
• World Bank (2009) Country Social Analysis: Ethnicity and Development in Vietnam (Vol. 2 of 2): Main report (English), Washington, DC: World Bank.