Effect of agricultural diversification on dietary diversity in rural households with children under 5 years of age in Zambia

Abstract Micronutrient deficiencies in low‐income countries are associated with the monotonous consumption of nutrient‐deficient crops, contributing to childhood stunting with far‐reaching socioeconomic consequences. To promote nutrition sensitive agriculture, policy makers in such countries have embarked on policy initiatives that encourage agricultural diversification in smallholder farming systems. This paper investigates the link between agricultural diversification and two key indicators of food and nutrition security among children under 5 years in rural Zambia. Data from the 2015 Rural Agricultural Livelihoods Survey and regression models are used to explain household dietary diversity and months of inadequate household food provisioning among 7934 households. Factors associated with the key outcome variables include land cultivated, household size, total livestock units, household head education, households receiving extension information, and use of productivity‐enhancing inputs such as fertilizers. Although the results demonstrate that agricultural diversification is positively associated with the household dietary diversity score, the relationship is not statistically significant. Further, the study findings illustrate that agricultural diversity is negatively associated with months of inadequate household food provisioning but that this relationship is also not statistically significant. The implication for policy is that other interventions such as productivity enhancement and behavioral change communication need to be scaled up.

than the national prevalence (GNR, 2020). Micronutrient deficiencies are also noted in young children and can lead to stunting, which can be up to four times higher in vulnerable communities (Mulmi et al., 2017;UNICEF, 2020).
According to the 2018 Zambian Demographic Health Survey (ZDHS), approximately 35% of Zambian children are reportedly stunted (Central Statistical Office [CSO], 2019). In 2019, the Global Hunger Index (GHI) ranked Zambia 113th out of 117 qualifying countries, with a score of 38.1, signaling alarming levels of hunger (GHI, 2019). The GHI ranks countries based on undernourishment, child mortality, child wasting (low weight for height), and child stunting (low height for age). In Zambia, the underfive mortality rate between 2014 and 2018 was estimated at 61 deaths per 1000 births (CSO, 2019). Acute and chronic micronutrient deficiencies are highly prevalent, particularly in rural areas. For instance, the prevalence of some degree of anemia among children under 5 years was approximately 58% based on hemoglobin levels in grams/decilitre, with 29% of children classified as mildly anemic (10.0-10.9 g/dl), 28% were moderately anemic (7.0-9.9 g/dl), and 2% were severely anemic (<7.0 g/dl) (ZDHS, 2019). The consequence of micronutrient deficiencies is poor cognitive development, which has far-reaching social and economic consequences (Mulmi et al., 2017). To avoid this, children should be fed a diverse diet consisting of various nutrientrich foods (Mulmi et al., 2017). It has been proposed that promoting agricultural diversification among farming households can improve dietary diversity and hence the nutritional status of children (Mofya-Mukuka & Kuhlgatz, 2016).
In Zambia, the government has embarked on policy initiatives to promote agricultural diversification and reduce the dependence on maize, Zambia's most commonly grown food crop. This involves cultivating various crops such as cassava, sweet potato, groundnut, sunflower, and soya beans (Sichoongwe et al., 2014). Diverse crop cultivation can boost productivity and improve the stability of agroecosystems, whereas a lack of diversification can have a negative knock-on effect on global diet quality (Jones, 2017). Agricultural diversification can help ensure food security by improving farmer adaptability and reducing vulnerability, which is crucial, given the predicted climate changes and the heavy reliance of smallholder farmers on rain-fed crops. Farmers can thus avert risk and increase/ improve income streams by adopting diversification practices.
Despite agricultural diversification being among the key policy strategies promoted to ensure nutrient adequacy and overall household food security, the link between agricultural diversification and household dietary diversity remains inconclusive and needs to be supported by evidence-based research (Jones, 2017;Sibhatu et al., 2015).
Due to the paucity of data, it remains unclear whether agricultural diversification directly impacts dietary diversity within agricultural households, as it may be affected by other sociodemographic factors and household characteristics. While higher income levels, education, and nonfarm enterprise engagement may strongly stimulate adequate nutrient intakes, increases in the number of adolescents would substantially diminish it as the household would have more individuals to feed and given the additional nutritional requirements of adolescents (Akerele et al., 2017). The purpose of this paper is to estimate the effect of agricultural diversification on household dietary diversity across the distribution of rural households with children under the age of 5 years.

| Data
The study used secondary data from the 2015 Rural Agricultural where P ij = A ij /∑A j is the proportion of the ith agricultural enterprise value of output (crop or livestock) of total agricultural value of output by household j. If SI is near zero, it indicates that a household is highly specialized. If it is close to one, then the household is fully diversified in terms of agricultural production.

| Key outcome variables
To explore the effect of agricultural diversification on dietary diversity and nutrient adequacy across households with children under 5 years, two outcome variables were considered in the current study. These were household dietary diversity score (HDDS) and months of inadequate household food provisioning (MIHFP). We describe these in turn.
Household dietary diversity score accounted for the number of different foods or food groups consumed over a 24 h period. In the RALS data set, dietary diversity was estimated using data collected during a 24 h recall period rather than a seven-day time frame.
Although longer recall periods capture a wider variety of foods consumed by a household, a 24-h recall period is beneficial because it increases estimation accuracy by reducing the level of "noise" added.
Dietary diversity and agricultural diversification (for rural households) were compared with the age and income activities of household head.
Household dietary diversity score is one of the most commonly used dietary diversity indicators at household level (Headey & Ecker, 2013). It accounts for the number of different foods or food groups consumed over a reference period, usually ranging from a day (24 h) up to 2 weeks (Heady & Ecker, 2013;Ruel, 2003

| Empirical strategy
The empirical goal for our paper was to measure the association be- Second, we estimated equation (2) by using an approach that accounts for potential sample selection bias likely to be imposed by violating the assumption that diversification is uncorrelated with the error term.
According to Mazunda et al. (2015), the link between agricultural diversification and dietary diversity is wrought with various unobservable characteristics. Important differences in food security and nutritional outcomes will likely exist between households that choose to diversify their crops and livestock production and those that do not. However, not all observed improvements in outcomes can be directly attributed to the decision to diversify agriculture since there could be other factors-the quality of land, risk, and food preferences-contributing to this and thereby improving food security and nutritional outcomes. This complexity in the linkage poses an endogeneity concern in that any observed relationship between agriculture diversification and the outcomes of interest may be due to any of these factors. To circumvent this potential pitfall, we used the Heckman two-step procedure.
As the name implies, the procedure is in two stages. In the first stage (selection equation), probit analysis is used to identify what determines the decision by a household whether to diversify agricultural production or not. The selection equation is specified here below: The selection variable z * i is not observed but rather a sign of whether or not household i diversified agricultural production (we explain how this variable was constructed in the next section), w is a vector of factors influencing the decision to diversify, γ's are the parameter estimates, and u is the error term which is normally distributed with mean zero and unit variance.  Table 1 shows the descriptive statistics of the sampled households.

| Descriptive statistics
A total of 7467 (94.2%) farmers resided in rural areas. The population can be predominantly characterized as farmers from rural households with at least one child under 5 years of age. Most household heads are educated at least up to the primary school level.
The highest percentage of farmers were from the Eastern province (n = 2061; 26.0%), with the lowest from Lusaka (n = 446; 5.6%). The mean number of hours to the nearest urban center with at least 500,000 inhabitants was 13.40 (SD = 8.57).
A total of 78.8% (n = 6251) of households were male-headed,  In Table 4, "Households with diversified diets" were defined as households that consumed six or more food groups following the HDDS. "Educated mothers" were defined as mothers that received at least a primary school education. Households with more and less than six months of adequate food provisions were calculated ac-  Figure 1 presents the distribution of our key explanatory variable, agricultural diversity, across the sampled households using the kernel density estimate overlaid by a fitted normal density. Although the kernel density plot is less peaked than the normal density plot, the graph clearly shows that it is not skewed. Given this, the mean for the SI of agricultural diversification is used as the cutoff point to create a binary variable of whether a household diversified its agricultural activities. Households below the mean of the diversification index were assigned a value of zero (did not diversify), and those above the mean are assigned a value of one (completely diversified).

| Distribution of the agricultural diversification index
This binary variable is used as a dependent variable in the selection equation for our econometric results discussed later.

| Bivariate relationships between dietary diversity and agricultural diversity
As a prelude to our robust econometric analysis, we checked for bivariate relationships between our main explanatory variable (agricultural diversity index) and the dietary diversity indicators (HDDS and MIHFP) using two-way quadratic prediction plots with confidence intervals. Without controlling for other confounding factors, HDDS has an inverse U-shaped relationship with agricultural diversity (Figure 2), while MIHFP had a U-shaped relationship with agricultural diversity (Figure 3). As households move from complete specialization toward diversification, the graphs suggest that households generally experience improvements in dietary diversity TA B L E 4 Comparison of means of selected characteristics between households with and without diversified diets and those with adequate and inadequate food provisioning ( Figure 2) or reduction in months with inadequate food provisioning.
The key finding from these bivariate relationships is that the effect of agricultural diversity on HDDS and MIHFP is not obvious, as demonstrated by the nonmonotonic relationships. These results, however, do not provide an overall picture. The next section presents results from a multivariate framework.

| Regression results
Regression results are provided in two tables. Although our regression analysis includes fixed effects controlling for the 10 provinces in Zambia, we do not report the point estimates for these variables.
First, we highlight the key findings of the relationship between HDDS and agricultural diversity conditional on other factors (Table 5) (e.g., Kumar et al., 2015;Hossain et al., 2016;Mazunda et al., 2015).
Apart from looking at the relationship between agricultural diversity and household dietary diversity, we were also interested in unpacking how this relationship differs across the distribution of households based on the number of underfive children. To implement this, we include interaction terms between the agricultural diversification index and a categorical variable that divides the number of children under 5 years in the sample into terciles (three groups).
Specifically, the interaction terms are as follows: Regression analysis of household-level data from Indonesia, Kenya, Ethiopia, and Malawi shows that agricultural diversity was not always positively associated with dietary diversity. In fact, the association was sometimes negative when agricultural diversity was already high, which was attributed to income losses from a lack of specialization (Sibhatu et al., 2015). The study also highlighted that F I G U R E 3 Relationship between MIHFP and agricultural diversity market access and transactions had a more significant impact on dietary diversity than increased agricultural diversity.
For the second set of regression results ( Despite this a priori outcome, the results are not statistically significant, implying that the association is just random and not a confirmation of a strong link between the two variables in our data.
Furthermore, the statistically insignificant coefficients on the interaction terms suggest that there are no significant differences in the number of months with inadequate household provisioning due to changes in agricultural diversification as one compares households across the distribution of the number of children under 5 years.
Finally, besides the dummy variable minimum tillage use (1 = yes), the list of factors that have a more than random contribution to improved food security (decrease in MIHFP) is similar to those affecting HDDS.

| CON CLUS I ON S AND IMPLI C ATI ON S FOR P OLI C Y
In contrast to predicted findings, agricultural diversification in It also highlights the need for policy initiatives that improve access to income and inputs among women farmers. In addition, with the relatively long distances between farmers and urban centers, improving access to urban markets by investing in rural transport systems should also be investigated from a policy perspective.
Policy interventions that focus on increasing farm productivity and hence access to income and behavior change communication to provide information on the importance of consuming a diverse diet could improve household dietary diversity. Further investigations are warranted to confirm the feasibility of these recommendations.

ACK N OWLED G M ENTS
Special thanks to the African Economic Research Consortium (AERC) for funding this research.

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest with this publication.

S TU D I E S I N VO LV I N G H U M A N O R A N I M A L S U BJ EC T S
This article does not contain any studies with human or animal subjects performed by any of the authors. Note: t statistics in parentheses * p < .10, ** p < .05, *** p < .01.

TA B L E 6
Regression results for months of inadequate household food provisioning using alternative specifications

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available on re-