Unraveling the paradox of child growth in Tanzania’s breadbasket regions: impact of household food production

Abstract

Tanzania’s food supply relies heavily on crop production from its breadbasket regions (BBRs). Despite their central role in national agriculture, the 2018 National Nutrition Survey revealed a troubling paradox: five of the regions with the highest rates of child stunting and severe malnutrition are located within these BBRs. This paper investigates the underlying causes of this paradox. Using data from the 2017–2018 National Household Budget Survey, the 2020–2021 National Integrated Labor Force Survey, and the 2020–2021 National Panel Survey, instrumental variable probit models were used to assess the impact of household crop production on children’s growth status. The findings confirm that children in BBRs are more likely to experience stunting than those in non-BBRs. Overall, higher crop production is associated with a lower risk of stunting and improved height-for-age z-scores. However, these benefits appear more pronounced in non-BBRs than in BBRs. Further analysis shows a positive relationship between increased crop production and household dietary diversity, although this relationship is also weaker in BBRs. These results indicate that factors beyond food availability, such as dietary practices and household-level conditions, may contribute to the observed paradox, highlighting the need for more nuanced policy discussions.

Graphical abstract

Keywords

Stunting / crop production / dietary diversity / working status

Highlight

	● Growth of children in breadbasket regions is more likely to be impaired than in non-breadbasket regions.
	● Higher crop production reduces the risk of stunting in non-breadbasket regions, but not in breadbasket regions.
	● Higher crop production contributes to improved height-for-age in non-breadbasket regions, but not in breadbasket regions.
	● Increased crop production is associated with greater household dietary diversity in non-breadbasket regions.

1 Introduction

Tanzania’s economy is predominantly based on agriculture. The agricultural sector contributes about 31% to Tanzania’s GDP and supports the livelihoods of about 75% of the population^[1,^2]. Notably, 80% of Tanzania’s crop production is focused on food, underscoring the subsistence-oriented nature of its agriculture^[2]. Despite Tanzania’s vast agricultural potential, variations in climate and topography have made certain regions particularly well-suited for crop production, especially grain cultivation. Eight of Tanzania’s 26 administrative regions, Ruvuma, Rukwa, Mbeya, Iringa, Njombe, Songwe, Morogoro and Katavi, are recognized as Tanzania’s breadbasket regions (BBRs) due to their significant contribution to food production. According to a previous study^[1], between 2007 and 2010, these regions together accounted for over 38% of Tanzania’s maize production, with maize being the most commonly consumed cereal in the country. By 2023, the top three regions among these had been identified as the leading grain-producing areas^[3].

However, a paradox emerges when examining the relationship between crop production and nutrition in Tanzania. While increasing crop production is generally regarded as essential for improving nutrition, particularly in developing countries where undernutrition and malnutrition are prevalent, the situation in Tanzania is more complex. The 2018 National Nutrition Survey revealed that five regions with the highest rates of child stunting and severe malnutrition, Njombe, Rukwa, Iringa, Songwe and Ruvuma, are all located within the BBRs^[4]. This finding contradicts earlier studies indicating a direct correlation between higher crop production and improved nutritional outcomes. Research in Tanzania has generally found a positive association between crop production and household nutrition^[5–^8]. Similar findings have been reported in studies from other countries, including Malawi^[9], Bangladesh^[10,^11] and Nigeria^[12], along with broader cross-country analyses^[13,^14], all of which highlight positive links between agricultural production and nutritional quality.

However, other studies caution that while such relationships may exist, they are neither universally applicable nor straightforward. Some studies^[15–^18] have indicated that these associations are more nuanced. In Tanzania, the paradox of high crop production coexisting with high malnutrition rates has largely been identified through descriptive statistics. Unfortunately, such statistics often oversimplify the data, masking underlying causes. This underscores the need for a more detailed analysis to fully understand the complex relationship between crop production and nutrition. It is therefore crucial, particularly for policymaking, to determine whether the descriptive findings from Tanzania’s surveys accurately reflect real-world dynamics or merely simplify a far more complex situation.

This study sought to resolve the paradox between crop production and child stunting in Tanzania by examining the impact of household food production on children’s growth. Specifically, it investigated how household crop production affects child growth status, accounting for regional differences between families residing in breadbasket and non-breadbasket regions. Additionally, the study examined two potential mechanisms: the influence of household dietary quality and the employment status of mothers. The research had two main objectives. First, it aimed to rigorously assess whether the paradox observed in the national nutrition survey, concerning the differential impact of crop production on child growth in BBRs versus non-BBRs, is statistically significant. Second, it sought to explore possible explanations for this paradox, if they arose from the survey data or from the statistical analysis conducted in this study.

The remainder of this paper is structured as follows: Section 2 outlines the materials and methods employed; Section 3 presents the results; Section 4 discusses the findings; and the final section concludes with implications for policy-making.

2 Material and methods

2.1 Data sources

We used secondary data obtained from the National Bureau of Statistics of Tanzania, specifically drawing information from three separate surveys. The use of multiple surveys was necessary due to the difficulty of capturing all relevant variables within a single data set. The 2020–2021 Tanzania Panel Survey served as the primary data set for empirically examining the differential impact of crop production on children’s growth status between BBRs and non-BBRs. This panel survey provided two key indicators of child growth: stunting and height-for-age z-scores (HAZ). Stunting was recorded as a binary variable, where a value of 1 indicates stunting and 0 indicates its absence. HAZ was categorized into four levels: severe, moderate, mild and healthy. Additionally, the data set included information on household crop production, reported in kilograms for the season preceding the survey.

However, since the 2020–2021 Panel Survey has not included data on household dietary diversity scores (HDDS), so we supplemented it with the 2017–2018 Household Budget Survey. This survey was used to evaluate HDDS as a potential explanatory factor for the observed paradox. It provided detailed information on the variety of foods consumed by households over a 7-day period, which is used to compute the HDDS. To address the absence of data on working hours in the other two surveys, we also incorporated the 2020 Integrated Labor Force Survey. This data set allowed us to examine a second mechanism, the time mothers spend away from home, as it relates to children’s growth outcomes.

To address potential inconsistencies arising from the use of different surveys, we relied on the 2020–2021 Tanzania Panel Survey as the core data set for analyzing the differential impact of crop production on stunting and HAZ between BBRs and non-BBRs. The other two surveys were used only to supplement variables not available in the panel data. To ensure comparability, we selected households that match the socioeconomic and demographic characteristics of those in the panel data set. We also adopt standardized and harmonized definitions for key variables, including crop production, child growth indicators and dietary diversity. In addition, we conducted robustness checks by comparing household characteristics across data sets to confirm there are no systematic differences that could bias our findings. We verify that key demographic and economic indicators (e.g., household size, income and education levels) are consistent across data sets, thereby mitigating concerns about sample incompatibility.

2.2 Estimation techniques

In the first part of the study, we used the propensity score matching (PSM) technique to examine differences in children’s growth status and crop production between BBRs and non-BBRs. As given in Eq. (1), we controlled for household income and size, the gender and age of the household head, and an urban residence dummy. Subsequently, we empirically analyzed the relationship between household crop production and children’s growth status, using two measures: stunting and HAZ. A key challenge in estimating Eq. (2) is the potential endogeneity bias arising from mutual causality between child growth status and household crop production decisions. For example, households may increase food production in response to child undernutrition. To address this potential bias, we used instrumental variable (IV) probit models.

When estimating stunting, we used land ownership status (specifically, whether the household owns planted land) as an instrumental variable for crop production. For HAZ, which was categorized into four levels, we apply the two-stage residual inclusion (2SRI) method, following approaches used in prior studies^[10,^19]. Land ownership qualifies as a valid instrument for two main reasons. First, after controlling for other household characteristics such as income and size, land ownership was not expected to directly influence children’s growth outcomes, satisfying the exogeneity condition. Second, since crop production is heavily dependent on access to land, the relevance condition was also fulfilled.

Accordingly, in our empirical analysis, we used IV probit models to examine the relationship between crop production and stunting. For HAZ, we used the 2SRI approach to assess how crop production relates to children’s overall growth status.

The specific model specifications for PSM are:

(1)\begin{array}{rl}{\tau }_{ATT}& =E(\tau |D=1,P(X))=E[Y(1)|D\\ & =1,P(X)]-E[Y(0)|D=1,P(X)]\end{array}

where, {\tau }_{ATT} is an average treatment effect on the treated, D = 1 is BBRs, Y(1) and Y(0) are children’s growth status and crop production when D = 1 and D = 0 at given matching variables X, respectively.

The probit model can be specified as:

(2)P(Growt{h}_i=1)=F({\gamma }_0+{\gamma }_{1}P{r}_i+{\gamma }_{2}BBR+{\gamma }_{3}Pr\times BBR+{\gamma }_4{X}_i+{\epsilon }_i)

where, Growth_i is a binary variable representing the growth status of a child in the i^th household; {Pr}_i is crop production, measured as the total quantity produced; BRR captures the five regions identified as having higher rates of poor child growth in the national nutrition survey; X_i is a vector of control variables, including the child’s age, household income and size, gender and age of the household head, an urban dummy, and the marital status of the household head; ε_i is the error term.

The covariates were included to enhance the comprehensiveness of our analysis, ensuring that we account for all plausible explanations behind the observed paradox. From a theoretical standpoint, these variables represent hypothesized pathways discussed in the literature^[20–^22], and remain essential to understanding child growth dynamics in Tanzania’s breadbasket regions.

We also included an interaction term between crop production and BBR status to capture the heterogeneous impact of production on children’s health outcomes across BBRs and non-BBRs. Economic theory, particularly agricultural household models^[23], suggests that household production decisions affect both consumption and health outcomes through income effects and own-production effects. These effects are unlikely to be homogeneous across regions, especially where structural and institutional contexts differ.

In Tanzania, BBRs vary markedly from non-BBRs in terms of agricultural intensity, production scale and crop specialization. Households in BBRs often engage in higher levels of production and are more market-integrated. As a result, food production may not translate into improved household consumption, particularly when much of the harvest is commercialized. This aligns with agricultural transformation theory, which posits that increases in production, especially in market-oriented regions, do not automatically yield better nutrition outcomes. This disconnect may arise from commercialization, limited dietary diversity, and intra-household food allocation dynamics. If the sum of {\gamma }_1and {\gamma }_3 is less than {\gamma }_1, and statistically significant, consistent with findings from the national nutrition survey, then a deeper investigation is warranted to explain the coexistence of food abundance and poor child growth in BBRs.

We developed two hypotheses to explore this paradox. First, economic trade-offs may be at play: increased crop production may require women, typically the primary caregivers, to spend more time away from home on economic activities. Prior research emphasizes that time is a critical determinant of nutritional outcomes, closely linked to food preparation and caregiving^[24–^27]. Second, we hypothesize that increased crop production may reduce household dietary diversity scores, as commercialization and specialization may drive households to cultivate fewer food types for home consumption in favor of market-oriented crops.

To test these hypotheses, we examined whether increases in crop production are associated with (1) longer working hours for women away from home in BBRs, and (2) lower HDDS. It is important to note that dietary diversity can be measured at the individual or household level, each offering distinct interpretive insights. For this study, we used HDDS due to data availability. The recall period for HDDS varies across studies, from 24-h recall^[28,^29] to 7-day recall periods^[30,^31]. Therefore, we adopted a 7-day recall period, which better captures day-to-day variation in household food consumption^[14]. HDDS is computed based on 10 food groups: cereals; legumes, nuts and seeds; tubers and roots; meat; fish and other seafood; eggs; milk and milk products; fruits; vegetables; and cooking oils.

To test our hypotheses, we used instrumental variable regression models. Specifically, we applied a Poisson IV regression model for Eq. (3) and a standard IV regression model for Eq. (4). Poisson regression is appropriate for count data, making it suitable for modeling HDDS in Eq. (3). To focus on women’s working hours, Eq. (4) was estimated using a subsample of adult women. These equations are:

(3)HDD{S}_i={\alpha }_0+{\alpha }_{1}P{r}_i+{\alpha }_{2}P{r}_i\times BBR+{\alpha }_{3}BBR+{\alpha }_4{Z}_i+{\epsilon }_i

(4)Hour{s}_i={\beta }_0+{\beta }_{1}P{r}_i+{\beta }_{2}P{r}_i\times BBR+{\beta }_{3}BBR+{\beta }_4{M}_i+{\mu }_i

where, Z_i and M_i are a set of covariates, including household size, the age and education level of the household head and an urban dummy variable.

Additional controls in the HDDS model included household income, the gender of the household head, and dummy variables indicating the availability of a kitchen and ownership of a refrigerator for food preservation. For the working-hours model, additional controls included possession of production skills, a dummy variable for involvement in commercial agriculture, the disability and residence status of the household head, perceived benefits of the activity being performed and access to credit within the last 12 months.

In the HDDS model, if {\alpha }_2<0, this implies that as crop production increases in BBRs, dietary diversity is lower compared to non-BBRs. Similarly, in the working-hours model, if {\beta }_2<0, it indicates that increases in production in BBRs are associated with individuals, particularly women, spending more time away from home than in non-BBRs, potentially impacting childcare negatively. To assess whether BBRs experience a greater impact, we compare the interaction coefficients between models estimated separately for BBRs and non-BBRs. Also, to address potential heteroskedasticity, all models are estimated using robust standard errors. Notably, the working-hours model is estimated using a sub-sample consisting only of women.

3 Results

3.1 Descriptive statistics

Tab.1 presents the descriptive statistics of the variables used in this study, highlighting notable differences between BBRs and non-BBRs. Children residing in BBRs were more likely to have experienced stunting compared to those in non-BBRs. Also, the HAZ scores of children in BBRs were significantly lower than those of children in non-BBRs. Since HAZ was measured as a categorical variable, with higher values indicating better growth outcomes, the lower HAZ in BBRs aligns with the higher stunting rates observed. This finding is consistent with data from the national nutrition survey, which also reported a higher prevalence of poor child growth in these regions.

In addition, the level of crop production in BBRs exceeded that of non-BBRs. Among the predictor variables, significant differences are observed between the two regions in most cases. For example, children in BBRs tend to be older, and their families are generally wealthier but smaller in size. BBRs also had a higher proportion of female-headed households, with many located in more urbanized settings. In these regions, mothers are more commonly responsible for fetching water.

3.2 Differences in children’s growth status and crop production between BBRs and non-BBRs

Tab.1 indicates that children residing in BBRs were more likely to have experienced stunting and have lower HAZ scores. We observe a significant disparity in children’s growth status between the two regions. To determine whether this variation can be attributed to individual and household-level heterogeneity, we used a PSM approach. This method uses individual and household characteristics as matching variables to compare children’s growth outcomes across regions. As shown in Tab.2, our results revealed that children in BBRs continue to face a significantly higher risk of stunting compared to their counterparts in non-BBRs. However, the difference in HAZ scores was no longer statistically significant after matching.

Also, crop production in BBRs is about 50% higher than in non-BBRs. While previous studies have demonstrated a strong link between food production and improved child growth in developing countries, our findings show that despite higher levels of crop production in BBRs, children’s growth outcomes remained significantly poorer.

3.3 Impact of crop production on stunting risk of children

To examine the relationship between crop production and children’s growth status, we used instrumental variable (IV) probit models. We used stunting and HAZ as indicators of child growth and designate them as the dependent variables. The rationale for using the IV probit model was based on the results of the Wald test of exogeneity. The test yielded a χ²(2) of 8.66 (p = 0.01), leading to the rejection of the null hypothesis of no significant difference between the probit and IV probit estimates. This result confirms the presence of endogeneity, implying that estimates from the standard probit model may be biased. Therefore, we relied on the IV probit model in our analysis.

To further assess the strength and validity of our instruments, we conducted both the Anderson-Rubin (AR) and Wald tests. The AR test evaluates whether the instrument is strongly correlated with the endogenous regressor, a key condition for valid parameter identification in structural equations^[32]. It tests the null hypothesis that the instrument is either weak or invalid, i.e., it lacks sufficient correlation with the endogenous variable.

In contrast, the Wald test examines whether the coefficient of the endogenous variable is statistically different from zero. In the context of weak instruments, it indirectly assesses instrument strength by testing the significance of the IV-based estimate. The null hypothesis of the Wald test states that the endogenous regressor has no effect on the dependent variable or that the instrument is weak.

As shown in Tab.3, the results from both the AR and Wald tests rejected their respective null hypotheses, indicating that the chosen instrument, land ownership, is strongly correlated with the endogenous variable and satisfies the relevance condition for IV estimation.

Tab.4 presents the IV probit results on the relationship between crop yield and children’s stunting status. The interaction term between household crop production and BBR status had a statistically significant marginal effect of 1.18 at the 5% significance level. This indicates that the influence of crop production on child stunting is significantly more positive (i.e., less beneficial or even adverse) in BBRs than in non-BBRs.

Economically, while an increase in household crop production is generally expected to reduce the risk of stunting, this beneficial effect appears to be attenuated, or even reversed, in BBRs. In other words, BBR status moderates the relationship between crop production and nutrition, likely due to structural characteristics of these regions. BBRs are potentially more commercially driven, with households often selling a large share of their agricultural output instead of consuming it directly. This commercialization may undermine the nutritional benefits of self-produced food, particularly if the resulting income is not used to purchase a diverse array of nutrient-rich foods.

Additionally, households in BBRs may face competing demands on caregiver time due to the labor-intensive nature of farming. This may reduce the time available for childcare and feeding. These two possible explanations are examined in greater depth in the next section of this paper. Also, the findings indicate that the overall likelihood of children experiencing stunting is higher in BBRs than in non-BBRs. The analysis also revealed a negative and statistically significant association between crop yield and the probability of stunting. Specifically, marginal effects suggest that, in the total sample, a 100% increase in crop yield is associated with a 164% reduction in the probability of stunting. However, regionally disaggregated effects showed that in BBRs, a 100% increase in crop yield leads to a 13.2% increase in the probability of stunting, while in non-BBRs it results in a 104% reduction.

This weaker reduction in stunting in BBRs may explain the higher prevalence of poor child growth in those regions relative to non-BBRs. When analyzing subsamples, the negative relationship between crop yields and stunting remains strong and statistically significant in non-BBRs, but is not statistically significant in BBRs. These findings reinforce the paradox observed in the national nutrition survey. The results are consistent with previous studies conducted in Nigeria^[15], a cross-country study^[16] and prior research in Tanzania^[33]. Among the control variables, child age, household size, and the marital status of the household head are positively associated with the risk of stunting.

3.4 Robustness test: results from HAZ

The results for HAZ, obtained from the 2SRI model, are presented in Tab.5. These results are consistent with those of the IV probit model used to estimate stunting. First, the findings indicate that an increase in crop yield was associated with a higher probability of children attaining a healthy HAZ score. Specifically, the results suggest that increased crop production significantly reduced the likelihood of children falling into the severe, moderate and mild malnutrition categories, while increasing the probability of being in the healthy category. A 100% increase in crop yield was associated with a 9.0% reduction in the probability of severe malnutrition, a 5.9% reduction in moderate malnutrition and a 1.2% reduction in mild malnutrition. Conversely, it increased the probability of a child being in good nutritional health by 16.2%.

The marginal effects of the interaction term between crop production and BBR status followed a similar trend as the main effects, negative for all malnutrition levels and positive for healthy status. However, none of these interaction effects were statistically significant. This lack of significance indicates that the positive impact of crop production on children’s HAZ scores did not differ meaningfully between BBRs and non-BBRs. In other words, despite expectations that households in BBRs, due to better agroecological conditions, would experience greater nutritional benefits from increased crop production, the results do not support this. The lack of significant interaction terms indicates that being located in a breadbasket region neither amplifies nor diminishes the effect of crop production on child nutritional outcomes, at least within the framework of this model.

In addition, child age, household size, marital status of the household and female-headed households was negatively associated with a healthy HAZ score. In contrast, household income, urban residence, and the age of the household head were positively associated with healthier HAZ outcomes.

3.5 Potential mechanisms for BBRs and growth of unhealthy children

The findings of this study align with the national nutrition survey, which indicates that children in BBRs are more prone to growth impairments than those in non-BBRs. To investigate this paradox, we explored whether the number of hours women spend working away from home may be a contributing factor. The core hypothesis posits that higher crop production in BBRs may compel women to engage more actively in out-of-home economic activities. As a result, the time available for household care and child nutrition management is reduced. To test this hypothesis, we used an instrumental variable (IV) regression approach. A key concern is the potential simultaneity between farm production and women’s working hours. From one perspective, increased farm yields may enhance household food security and income, reducing the need for women to seek off-farm employment. From another, longer hours spent by women in external work may reduce their availability for farm labor, thereby affecting yields. This endogeneity issue would render ordinary least squares estimates biased and inconsistent.

To address this, we used farm ownership (a binary variable indicating whether the household owns or rents the farm) as an instrument for farm production. Theoretically, farm ownership is strongly associated with production levels but is unlikely to directly affect women’s working hours, thus making it a suitable instrument. The validity of the IV approach over ordinary least squares was confirmed through endogeneity tests. Specifically, Durbin’s score (Chi-square) test yielded a statistic of 7.22 (p = 0.03) and the Wu-Hausman F-test returned an F-statistic of 25.3 (p = 0.00). Since both p-values are below the 5% threshold, we reject the null hypothesis of exogeneity, thereby confirming that farm production is endogenous and justifying the use of IV regression. Consequently, the IV estimates can be considered to provide more consistent and reliable insights into the causal relationship between farm production and women’s working hours.

We estimated a system of three equations. In the first model, using the full sample and including an interaction term between crop yields and BBR status, we assessed whether higher yields in BBRs were associated with increased time women spend working outside the home. However, as shown in Tab.6, the marginal effect of this interaction term was not statistically significant. This finding does not support our hypothesis: higher crop yields in BBRs did not significantly increase women’s out-of-home working hours compared to non-BBRs. Thus, agricultural productivity in BBRs does not appear to have driven women’s time away from the home. To explore regional heterogeneity, we estimated the model separately for BBRs and non-BBRs. In BBRs, the marginal effect of crop yield was negative but not statistically significant, suggesting that higher agricultural production may slightly reduce women’s time spent away from home, possibly due to improved food security and income, but the evidence is inconclusive.

Conversely, in non-BBRs, the marginal effect was positive but also not statistically significant. This may indicate a weak tendency for increased agricultural production to create economic opportunities that draw women away from the home. However, the absence of statistical significance prevents firm conclusions. Overall, these findings suggest that the number of hours women spend working away from home is unlikely to explain the paradoxical coexistence of high agricultural production and poor child growth outcomes in BBRs.

Another possible explanation for the paradoxical finding of higher stunting rates and poor HAZ scores among children in Tanzania’s BBRs relates to household dietary diversity. We hypothesized that higher levels of crop production in BBRs may, paradoxically, lead to lower HDDS. This may be driven by increased specialization and commercialization of agriculture. As households become more commercially oriented, focusing on the cultivation of specific cash crops, they may deprioritize the production and consumption of a wide variety of food items essential for a balanced diet. Such dietary patterns can negatively affect children’s nutritional outcomes, offering a plausible explanation for the paradox.

To test this hypothesis, we used an IV-Poisson regression model rather than the standard Poisson regression, due to concerns about potential endogeneity. Such endogeneity arises from a possible bidirectional relationship between household crop production (measured by yield) and HDDS. On one side, higher crop production may improve dietary diversity by increasing food availability. On the other, households with more diverse dietary preferences may choose to cultivate a wider range of crops. Ignoring this simultaneity may lead to biased and inconsistent estimates. To address this issue, we use farm ownership (a binary variable equal to 1 if the household owns the farm and 0 if it rents) as an instrumental variable. Farm ownership affects a household production capacity but, after controlling for socioeconomic factors, is plausibly exogenous to HDDS. It thus satisfies both the relevance and exclusion criteria for a valid instrument.

The results, as shown in Tab.7, indicate a notable divergence in marginal effects between the standard Poisson and IV-Poisson regression models, confirming the presence of endogeneity. The significant shift in marginal estimates after instrumenting highlights the importance of accounting for the bidirectional relationship. Since HDDS is a count variable, the Poisson regression is appropriate for this estimation. We conducted three sets of estimations, on the full sample and separately for BBRs and non-BBRs. For the full sample, we included an interaction term between crop yield and BBR status. A negative and statistically significant coefficient on this interaction term would support the hypothesis that crop yield in BBRs is associated with lower dietary diversity. The results confirmed this: the interaction term is negative and significant at the 10% level, indicating that rising crop yields in BBRs are linked to less diverse household diets relative to non-BBRs. Although modestly significant, this result offers compelling preliminary evidence that commercialization and crop specialization in BBRs may have contributed to nutritional deficiencies.

The subsample analyses provided further insight. In both BBRs and non-BBRs, the marginal effects of crop production on HDDS were positive, indicating that increased yields generally improve dietary diversity. However, the strength and significance of this relationship differed by region. In non-BBRs, the effect is statistically significant, indicating that higher crop yields could have led to more diverse diets and, likely, improved nutritional outcomes. In contrast, in BBRs, the marginal effect, though positive, was not statistically significant. This supports the earlier interaction effect and implies that agricultural productivity in BBRs had not meaningfully enhance dietary diversity. This may be due to commercialization (with households selling rather than consuming diverse foods), market access limitations or cultural food preferences that restricted dietary diversity, even in highly productive regions.

In combination, these results offer a plausible explanation for the nutritional paradox observed in BBRs. In non-BBRs, higher crop production generally enhances dietary diversity and nutritional outcomes. As noted in previous studies^[34,^35], smallholder farmers in less-productive, non-BBR regions often rely more heavily on market purchases due to lower food self-sufficiency. In such contexts, increased yields can raise household income through market sales, enabling the purchase of a wider variety of foods and thereby improving dietary diversity. Conversely, BBRs are typically more commercially oriented, with households specializing in the production of a few staple crops. Despite higher overall production, this specialization may reduce dietary diversity. Greater market integration in BBRs often promotes cash crop cultivation, leading households to prioritize sales and non-food expenditures over diverse food consumption.

Previous studies have documented a weakened or even negative relationship between crop production and dietary diversity in high-production areas, primarily due to commercialization effects^[14,^36]. Our findings are consistent with this pattern: the positive impact of crop yields on dietary diversity is more pronounced in non-BBRs than in BBRs. These results underscore the importance of contextual factors in shaping the relationship between agricultural production and dietary outcomes.

4 Discussion

This study highlights a troubling paradox in Tanzania’s BBRs. Despite their critical role in national food production, children in BBRs exhibit worse growth outcomes, particularly higher stunting rates, than those in non-BBRs. These findings are consistent with prior region-specific studies^[37,^38].

To investigate potential mechanisms, we examined whether maternal out-of-home working hours mediate this relationship. Contrary to expectations, crop production did not significantly influence the number of hours mothers spend working outside the home in either BBRs or non-BBRs. This may be attributed to the commercial nature of agriculture in BBRs, where higher yields do not necessarily demand increased maternal labor. This result aligns with some studies indicating that women’s economic participation does not always compromise childcare^[39–^42], but contrasts with others that link higher stunting to children of farming mothers^[43], possibly due to differing methodological scopes or regional contexts.

We also tested the hypothesis that higher crop production reduces HDDS due to agricultural specialization. While the results show a significant positive correlation between crop production and HDDS in non-BBRs, this relationship was not statistically significant in BBRs. This finding contrasts with studies reporting an inverse relationship between production and nutrition in highly commercialized agricultural settings^[15,¹⁶,³³,^35]. The analysis of control variables provides additional insights into the determinants of child growth. Larger household sizes are associated with poorer nutritional outcomes, likely due to resource dilution effects^[44]. Similarly, children in households where mothers are responsible for fetching water are more prone to stunting, echoing previous findings linking long water-fetching distances to malnutrition^[45]. Importantly, female-headed households and urban residence are positively associated with healthier child growth outcomes.

In summary, the coexistence of high agricultural productivity and poor child growth in BBRs indicates that increasing food production alone is insufficient to improve children’s nutritional outcomes. Addressing broader socioeconomic determinants, such as water access, household composition, and women’s caregiving capacity, is crucial. Targeted community development and nutrition-sensitive interventions are urgently needed to ensure equitable improvements in child health.

5 Conclusions

This study highlights a paradoxical relationship between crop production and child growth outcomes in Tanzania’s BBRs. Although these regions are vital to the national food supply, children in BBRs had poorer growth outcomes, particularly higher stunting rates, compared to those in non-BBRs. While increased crop production is generally associated with improved child nutrition, this positive effect is statistically significant only in non-BBRs.

The findings also challenge the assumption that agricultural work reduces the time mothers can devote to childcare. We find no significant impact of crop production on maternal working hours in either BBRs or non-BBRs. Additionally, in BBRs, HDDS does not show a significant relationship with crop production, further complicating the link between agricultural productivity and child health.

These results underscore the complex interplay of factors shaping child nutrition in high-production regions. They indicate that policies should go beyond increasing food production to also address broader socioeconomic challenges that affect child well-being. When promoting agriculture toward specialization and marketization, it should be noted that in poor areas, the underdevelopment of food markets may lead to regional-level food shortages. Specialization (or monoculture) will further affect the dietary diversity of specialized farmers, thereby impacting their dietary quality. This is an issue that policymakers need to focus on. Importantly, the study was unable to explore other potential pathways underlying the crop-nutrition paradox. Future research should aim to fill this gap using targeted surveys and more comprehensive data.

References

[1] Arce C E, Caballero J. Tanzania: agricultural sector risk assessment. Washington, DC: World Bank, 2015

[2] Food, Agricultural Organization of the United Nations (FAO). Small Family Farms Country Factsheet: Tanzania. Rome: FAO, 2018. Available at FAO website on July 20, 2025

[3] Kilimo Kwanza. Ministry of Agriculture Recognizes Top-Producing Regions in Tanzania. Kilimo Kwanza, 2023. Available at Kilimo Kwanza website on July 20, 2025

[4] United Republic of Tanzania (URT). Tanzania National Nutrition Survey 2018: Summary of Key Findings. Dar es Salaam: URT, 2019

[5] Ngongi A M, Urassa J K. Farm households food production and households’ food security status: a case of Kahama district, Tanzania.Tanzania Journal of Agricultural Science, 2014, 13(2): 40–58

[6] Slavchevska V. Agricultural production and the nutritional status of family members in Tanzania.Journal of Development Studies, 2015, 51(8): 1016–1033

[7] Kim J, Mason N M, Snapp S, Wu F. Does sustainable intensification of maize production enhance child nutrition Evidence from rural Tanzania.Agricultural Economics, 2019, 50(6): 723–734

[8] Chegere M J, Stage J. Agricultural production diversity, dietary diversity and nutritional status: panel data evidence from Tanzania.World Development, 2020, 129: 104856

[9] Jones A D, Shrinivas A, Bezner-Kerr R. Farm production diversity is associated with greater household dietary diversity in Malawi: findings from nationally representative data.Food Policy, 2014, 46: 1–12

[10] Deb U, Bayes A. Crop diversity, dietary diversity and nutritional outcome in rural Bangladesh: evidences from VDSA panel household surveys.Leveraging Agriculture for Nutrition in South Asia, 2018, 31: 1–28

[11] Uddin M R. Crop diversification, dietary diversity and nutrition: evidence from rural Bangladesh.Bangladesh Development Studies, 2019, 42(4): 111–133

[12] Gbenga O, Opaluwa H I, Adedeji S O, Abdulrahaman S. Effect of crop diversity on rural farming households’ dietary diversity.Journal of Asian Rural Studies, 2020, 4(2): 218–229

[13] Pellegrini L, Tasciotti L. Crop diversification, dietary diversity and agricultural income: empirical evidence from eight developing countries.Canadian Journal of Development Studies, 2014, 35(2): 211–227

[14] Sibhatu K T, Qaim M. Review: meta-analysis of the association between production diversity, diets, and nutrition in smallholder farm households.Food Policy, 2018, 77: 1–18

[15] Dillon A, McGee K, Oseni G. Agricultural production, dietary diversity and climate variability.Journal of Development Studies, 2015, 51(8): 976–995

[16] Sibhatu K T, Krishna V V, Qaim M. Production diversity and dietary diversity in smallholder farm households.Proceedings of the National Academy of Sciences of the United States of America, 2015, 112(34): 10657–10662

[17] Sibhatu K T, Qaim M. Farm production diversity and dietary quality: linkages and measurement issues.Food Security, 2018, 10(1): 47–59

[18] Isbell C, Tobin D, Thiede B C, Jones K, Reynolds T. The association between crop diversity and children’s dietary diversity: multi-scalar and cross-national comparisons.Food Security, 2024, 16(4): 883–897

[19] Wooldridge J M. Control function methods in applied econometrics.Journal of Human Resources, 2015, 50(2): 420–445

[20] Terza J V. Two-stage residual inclusion estimation: a practitioners guide to Stata implementation.Stata Journal, 2017, 17(4): 916–938

[21] Altare C, Delbiso T D, Mutwiri G M, Kopplow R, Guha-Sapir D. Factors associated with stunting among pre-school children in southern Highlands of Tanzania.Journal of Tropical Pediatrics, 2016, 62(5): 390–408

[22] Mrema J D, Elisaria E, Mwanri A W, Nyaruhucha C M. Prevalence and determinants of undernutrition among 6- to 59-months-old children in lowland and highland areas in kilosa district, Tanzania: a cross-sectional study.Journal of Nutrition and Metabolism, 2021, 2021: 6627557

[23] de Janvry A, Sadoulet E. Development Economics: Theory and Practice. London: Routledge, 2021

[24] Frumence G, Jin Y, Kasangala A A, Mang’enya M A, Bakar S, Ochieng B. A qualitative exploration on perceived socio-cultural factors contributing to undernutrition among under-fives in the southern Highlands of Tanzania.International Journal of Public Health, 2023, 68: 1605294

[25] Johnston D, Stevano S, Malapit H J, Hull E, Kadiyala S. Review: time use as an explanation for the agri-nutrition disconnect: evidence from rural areas in low and middle-income countries.Food Policy, 2018, 76: 8–18

[26] Komatsu H, Malapit H J L, Theis S. Does women’s time in domestic work and agriculture affect women’s and children’s dietary diversity Evidence from Bangladesh, Nepal, Cambodia, Ghana, and Mozambique.Food Policy, 2018, 79: 256–270

[27] Rao N, Raju S. Gendered time, seasonality, and nutrition: insights from two Indian districts.Feminist Economics, 2020, 26(2): 95–125

[28] Vemireddy V, Pingali P L. Seasonal time trade-offs and nutrition outcomes for women in agriculture: evidence from rural India.Food Policy, 2021, 101: 102074

[29] Korkalo L, Erkkola M, Heinonen A E, Freese R, Selvester K, Mutanen M. Associations of dietary diversity scores and micronutrient status in adolescent Mozambican girls.European Journal of Nutrition, 2017, 56(3): 1179–1189

[30] Amirhamidi Z, Omidvar N, Eini-Zinab H, Doustmohammadian A, Esfandiari S, Azadi R, Haidari H. Association of weight status with dietary intake and dietary diversity score in 10–12-year-old children in Tehran: a cross-section study.Iranian Journal of Pediatrics, 2019, 29(4): e85317

[31] Cordero-Ahiman O V, Vanegas J L, Franco-Crespo C, Beltrán-Romero P, Quinde-Lituma M E. Factors that determine the dietary diversity score in rural households: the case of the paute river basin of azuay province, Ecuador.International Journal of Environmental Research and Public Health, 2021, 18(4): 2059

[32] Khandoker S, Singh A, Srivastava S K. Leveraging farm production diversity for dietary diversity: evidence from national level panel data.Agricultural and Food Economics, 2022, 10(1): 15

[33] Wooldridge J M. Econometric Analysis of Cross Section and Panel Data. Cambridge: MIT Press, 2002

[34] Anderson T W, Rubin H. Estimation of the parameters of a single equation in a complete system of stochastic equations.Annals of Mathematical Statistics, 1949, 20(1): 46–63

[35] Rajendran S, Afari-Sefa V, Shee A, Bocher T, Bekunda M, Dominick I, Lukumay P J. Does crop diversity contribute to dietary diversity Evidence from integration of vegetables into maize-based farming systems.Agriculture & Food Security, 2017, 6(1): 50

[36] Carletto C, Corral P, Guelfi A. Agricultural commercialization and nutrition revisited: empirical evidence from three African countries.Food Policy, 2017, 67: 106–118

[37] Jones A D. Critical review of the emerging research evidence on agricultural biodiversity, diet diversity, and nutritional status in low- and middle-income countries.Nutrition Reviews, 2017, 75(10): 769–782

[38] Pingali P, Sunder N. Transitioning toward nutrition-sensitive food systems in developing countries.Annual Review of Resource Economics, 2017, 9(1): 439–459

[39] Ikelegbe O O, Edokpa D A. Agricultural production, food and nutrition security in rural Benin, Nigeria.African Journal of Food, Agriculture, Nutrition and Development, 2013, 13(5): 8388–8400

[40] Shilugu L L, Sunguya B F. Stunting in the context of plenty: unprecedented magnitudes among children of peasant’s households in bukombe, Tanzania.Frontiers in Nutrition, 2019, 6: 168

[41] Rashad A S, Sharaf M F. Does maternal employment affect child nutrition status New evidence from Egypt.Oxford Development Studies, 2019, 47(1): 48–62

[42] Boyd C M. Rainfall, mothers’ time use, and child nutrition: evidence from rural Uganda. Population and Environment, 2023, 45(3): 17

[43] Margolies A, Colantuoni E, Morgan R, Gelli A, Caulfield L. The burdens of participation: a mixed-methods study of the effects of a nutrition-sensitive agriculture program on women’s time use in Malawi.World Development, 2023, 163: 106122

[44] Nordang S, Shoo T, Holmboe-Ottesen G, Kinabo J, Wandel M. Women’s work in farming, child feeding practices and nutritional status among under-five children in rural Rukwa, Tanzania.British Journal of Nutrition, 2015, 114(10): 1594–1603

[45] Mbwana H A, Kinabo J, Lambert C, Biesalski H K. Determinants of household dietary practices in rural Tanzania: implications for nutrition interventions.Cogent Food & Agriculture, 2016, 2(1): 1224046

RIGHTS & PERMISSIONS

The Author(s) 2025. Published by Higher Education Press. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0)