1 Background

Despite the increasing consolidation of the global food system (Howard, 2021), small-scale farmers remain the dominant type of agricultural producers across the world (FAO & IFAD, 2019). Defined as farmers with limited resources that rely primarily on household members for farm labor (HLPE, 2013), small-scale farmers typically farm two hectares of land or less (IFAD, 2016; Woodhill et al., 2020). Small-scale farms are critical to the global food supply, providing an estimated 70–80% of calories consumed across much of the Global South (IFAD, 2016; Samberg et al., 2016). However, despite their importance to global food production, small-scale farm households are a particularly vulnerable group who experience high rates of food insecurity (FAO & IFAD, 2019; HLPE, 2013; IFAD, 2016; Woodhill et al., 2020). Moreover, climate-change related shocks continue to worsen and threaten global food security, most notably among vulnerable groups such as small-scale producers (Niles & Salerno, 2018).

Given the importance of small-scale agriculture to global food production, researchers and policymakers have dedicated significant attention to enhancing the food security of small-scale producers. Two strategies – diversification of on-farm production and increased market access and engagement – are often the focus of such analyses (Jones, 2017; Sibhatu & Qaim, 2018a). Diversifying on-farm crop production is often viewed as a strategy to mitigate environmental and market risks, improve dietary quality, and generate ecosystem services (Cadotte et al., 2012; IFAD, 2016; Saenz & Thompson, 2017; Tesfaye & Tirivayi, 2020). However, evidence on how household production contributes to the food and nutrition security of the household is mixed, leading some to argue that market avenues should be the primary approach for small-scale producers to improve their wellbeing (Sibhatu & Qaim, 2018b; Sibhatu et al., 2015). Typically, this perspective encourages farmers to specialize production to a limited number of crops in order to become more efficient for market-oriented production and sale (IFAD, 2016; Samberg et al., 2016). However, emerging evidence suggests that diversifying the farm’s crop portfolio can also be an avenue for market success, giving producers a better safety net in the face of volatile market prices (Chen et al., 2018; de Roest et al., 2018). Furthermore, there is some evidence that specialized production in Uganda can replace food crops with cash crops, negatively impacting food security (Ntakyo & van den Berg, 2019). While specialized producers may suffer if the one (or few) crop(s) they produce has a low price one year, farmers with more diversified production may be more resilient, as they have multiple crops to fall back on. Using this reasoning, some argue that on-farm diversification strategies should be oriented towards market competitiveness rather than household consumption (FAO & IFAD, 2019; Sibhatu & Qaim, 2018a). Still others argue for an approach that recognizes the benefits of production diversification for both home consumption and market sale (Bellon et al., 2020; IFAD, 2016).

A growing body of recent literature has emerged seeking to provide further insights into the dynamics of how crop diversity and/or market access contribute to dietary outcomes in low-income countries (Habtemariam et al., 2021; Kissoly et al., 2018; Koppmair et al., 2017; Mondal et al., 2021; Ochieng et al., 2020). However, perhaps in part due to the complex interactions between on-farm crop diversity and market access (the ability to sell and/or purchase foods at market), no consistent pattern has appeared in terms of the most effective pathways to enhancing food security. Sibhatu and Qaim (2018a) find that higher diversity in subsistence crops was associated with lower cash income and reduced dietary diversity across Kenya, Indonesia, and Uganda, although most of their data came from households that already had adequate access to markets. However, Tobin et al. (2019) find higher levels of crop diversity to be associated with positive dietary outcomes at the village level across eleven sub-Saharan African countries, though their crop diversity estimates rely on relatively coarse geospatial data for local crop production patterns. Likewise, Jones et al. (2014) provide nuanced findings through their study in Malawi, where diversity in farm production has a significant positive association with dietary diversity. However, the strength of the association varies widely by demographic factors including household size, education, and age and gender of the household head (Jones et al., 2014). Other recent studies suggest that higher levels of on-farm crop diversity in Ghana are positively associated with both income and diversity of foods consumed (Bellon et al., 2020; Ecker, 2018). Taken together, the findings from previous research do not converge, indicating that on-farm crop diversification may both positively and negatively relate to dietary diversity and household incomes depending on the context.

Past findings regarding the contributions of market access to food and nutrition security among small-scale farmers are similarly mixed. Increased market access has the potential to increase household food security through both sales and purchases. Sibhatu and Qaim, (2017) report that households in Ethiopia consume 58% of their calories from food they grow for subsistence, while 42% of their calories come from food they purchase. This split indicates that both self-produced foods and purchased foods are important to food availability for small-scale farm households, a pattern that has also been seen in Uganda (Nabuuma et al., 2021; Wichern et al., 2017). Several other studies explicitly analyze associations between subsistence production and consumption of market-sourced foods, but they also yield differing results. For example, Frelat et al. (2016) examine food availability across sub-Saharan Africa, concluding that enhancing market access for impoverished households more effectively improves food outcomes than focusing only on improving agricultural production. They find that households generally require less land to achieve food security when they have better access to markets, although they note that households often sell crops before reaching sufficient food security themselves. In other words, market engagement may be positively associated with food security in some cases, but in others, crop sales may be an indication of coping with food insecurity. On the other hand, Usman and Callo-Concha (2021) find no significant association between market access and food security or between market access and household dietary diversity, although they do find that market access influences how households procure their food.

The interactions between market engagement, crop diversity, and dietary diversity or food security are further influenced by farm characteristics and farmer demographics. Many studies investigate the influence of gender on agricultural outcomes, noting that women in agriculture tend to experience disadvantages which may limit their access to diverse crops or their ability to engage in markets (Pincus et al., 2018; Quisumbing et al., 2014; Reynolds et al., 2020; Tavenner et al., 2019). Another factor which may moderate the relationships between market engagement, crop diversity, and dietary diversity or food security is education. Several studies include education as a control variable, typically determining that higher levels of education tend to associate with better outcomes for small-scale farmer household livelihoods (Jones et al., 2014; Koppmair et al., 2017; Rajendran et al., 2017; Reynolds et al., 2020).

The mixed results from previous research suggest that complex interactions exist between and among market factors, crop diversity, and food security or dietary diversity outcomes, thereby requiring continued and further empirical investigation. To date, analyses that address the interactions between crop diversity and market factors, and their influence on both dietary diversity and food security at the household level, have rarely focused on Uganda, a country heavily reliant on small-scale agriculture that is also rapidly urbanizing (Linard et al., 2012; The Annual Labour Force Survey 2018/19, 2018). According to the Annual Labor Force Survey 2018/19 (2018), 36.0% of the working age population works in subsistence agriculture, with 63.5% working in the agriculture, forestry, and fishing industry. Despite having such a large population working in agriculture, Uganda still experiences high rates of food insecurity: 69.2% from 2018 through 2020, ranking Uganda 89th out of 113 countries in overall national food security (FAOSTAT, 2020). Ugandan small-scale farmers often experience a cycle of poverty, as falling prices for cash crops such as coffee leave farmers with little choice but to intensify production, which further drives prices down and can become devastating with a bad harvest (Gresser & Tickell, 2002; IFAD, 2016; Papaioannou, 2016). Uganda has an overall absolute poverty rate of 21.4%, with the Karamoja region reaching 60.2%—reflecting that not all regions feel the effects of poverty equally (UBOS, 2017). For impoverished small-scale farmers in Uganda, lack of irrigation means that the agricultural production is particularly vulnerable to weather events and will require policy responses to support farmer livelihoods (Tesfaye & Tirivayi, 2020). Given the conditions of Uganda, it provides an interesting case for further examination of the associations that exist among on-farm crop diversity, market engagement, and dietary diversity. The limited evidence that does exist for Uganda is partial in its consideration of these various factors, missing either a crop diversity index, nuanced market data, and/or specific measures of dietary diversity (Nabuuma et al., 2021; Ntakyo & van den Berg, 2019; Wichern et al., 2017).

This study draws on the 2015–16 farm household survey data from the Uganda National Panel Survey (UNPS) as a part of the Living Standards Measurement Study – Integrated Surveys on Agriculture (LSMS-ISA) collected by the Ugandan Bureau of Statistics with support from the World Bank to assess how on-farm crop diversity and market engagement interact to shape food security and dietary diversity outcomes among small-scale producers in Uganda. Using Structural Equation Modeling (SEM), the specific objectives of the study are to examine:

  1. i)

    how levels of on-farm crop diversity co-vary with measures of household market engagement,

  2. ii)

    how varying levels of crop diversity and market engagement relate to household dietary diversity and food security, and

  3. iii)

    the influence of farm household demographics such as farm size, gender of the household head, education, and off-farm income on market engagement and on-farm crop diversity, and, in turn, household dietary diversity and food security.

2 Conceptual framework and definitions

Given that it is unclear the degree to which crop diversity benefits households through home consumption and/or market engagement, we test how these two factors relate to dietary diversity and food security among Ugandan agricultural households. The conceptual framework outlined in Fig. 1 postulates that dietary diversity and food security are a function of crop diversity and market engagement. We further hypothesize that crop diversity and market engagement might covary to account for the possibility that market opportunities exist for households that are diversifying crop production (Sibhatu & Qaim, 2018a).

Fig. 1
figure 1

Structural Equation Model for HDDS and Food Secure Months

Full size image

2.1 Dietary diversity and food security

This study considers two outcome variables: dietary diversity (measured by Household Dietary Diversity Score (HDDS) and food security (measured by Food Secure Months). Dietary diversity is typically measured as a recall count of the number of food groups eaten in a household, usually within a 24-h or 7-day period (Ruel, 2003). Dietary diversity recall is considered an indicator of dietary quality and therefore dietary health and relates to the availability of both energy (e.g., calories) and nutrients (Niles et al., 2021). HDDS as utilized by the FAO includes 12 food groups and typically indicates household economic access to foods (Kennedy et al., 2011). The most common measure of individual dietary diversity is the minimum dietary diversity for women of childbearing age (WDDS or MDD-W) metric; this includes 9 or 10 food groups and assesses diet quality and micronutrient adequacy (FAO & FHI 360, 2016; Fongar et al., 2019; Kennedy et al., 2011).

In order to use a measurement that can be an indicator of household dietary quality and micronutrient adequacy, the HDDS we use has the same 10 food groups as the MDD-W (FAO & FHI 360, 2016; Fongar et al., 2019). We add an eleventh food group for “Other foods,” to capture the categories “Oils and Fats” and “Sweets” from the FAO HDDS, and to capture the “other foods” item listed in the UNPS food-recall survey question (Kennedy et al., 2011). The score indicates how many food groups each household consumed over a period of 7 days. It is calculated by using data from the UNPS in which a respondent reports whether their household did or did not consume foods from a list of 64 items over the past 7 days. The 64 food items were sorted into 11 food groups and household responses were used to calculate how many food groups that household had eaten during the 7-day period. The MDD-W was the base for our HDDS-11 because it provides insight into micronutrient intake (Fongar et al., 2019; Kennedy et al., 2011). The food groups in our HDDS-11 are as follows:

  1. 1.

    Grain, white roots and tubers, and plantain

  2. 2.

    Pulses (beans, peas, and lentils)

  3. 3.

    Nuts and seeds

  4. 4.

    Dairy

  5. 5.

    Meat, poultry, and fish

  6. 6.

    Eggs

  7. 7.

    Dark green vegetables

  8. 8.

    Other vitamin A-rich fruits and vegetables (passion fruits, mangoes, oranges, tomatoes, pumpkins, carrots)

  9. 9.

    Other vegetables (maize (cobs), onions, cabbages, green pepper, eggplants, other vegetables)

  10. 10.

    Other fruits (sweet bananas, avocado, watermelon, pineapple, pawpaw, apples, other fruits)

  11. 11.

    Other food (cooking oil, margarine, butter, etc., sugar, other juice, other foods)

The other outcome variable of interest in this study, food security, is often measured by reported hunger (Adjognon et al., 2021; Nwosu et al., 2022; Williams et al., 2016). The FAO metric for reported hunger is based upon respondents’ self-reported answers to questions about their experiences accessing food (Ballard et al., 2013). Reported hunger data are collected for the UNPS by asking households which, if any, months in the past year they did not have enough food to feed the household. For ease of interpretation, we compute the variable ‘Food Secure Months’ as 12 minus the number of hungry months reported (UNPS BID, 2020). This approach, through which higher values correspond to greater food security, is an accepted measure in the literature known as months of adequate household food provisioning (MAHFP) (Bilinsky & Swindale, 2010; Leah et al., 2013; Swindale & Bilinsky, 2006). The reported food security score we use for Food Secure Months ranges from 1 to 12, and is based upon how many months a household reported having had sufficient food to feed the household (UNPS BID, 2020). This does not assign a cut off for food insecure versus food secure, instead measuring relative food insecurity as households with lower scores are less are food secure than households with higher scores.

2.2 Crop diversity

Crop diversity is a dimension of agrobiodiversity, which refers to the diversity of plants, animals and other organisms which exist and interact within an agroecological landscape (Wood et al., 2015). Agrobiodiversity indices, crop species counts, animal/livestock species counts, cumulative species counts, production diversity scores, and functional diversity scores are all common methods used to capture on-farm diversity (Bellon et al., 2020; Ecker, 2018; Habtemariam et al., 2021; Kissoly et al., 2018; Muthini et al., 2020a, b; Sibhatu & Qaim, 2018a; Singh et al., 2020). Excluding animals and livestock, crop diversity is measured by various metrics ranging from richness (simple counts of the number of crops grown) to the Simpson’s Diversity Index (SDI) (accounting for both number of crops and relative area allocated to each) (Jones et al., 2014). Because SDI includes the area allocated to crops, it becomes difficult to compute when livestock are included (Sibhatu & Qaim, 2018a).

Because the SDI accounts for species richness and abundance, this study uses SDI to represent on-farm crop diversity. Using SDI, we measure crop diversity at the household level. The SDI is developed using the number of crops grown and the area of each crop. SDI is indicated by a number between 0 and 1 that tells how many crops there are for a household and how evenly distributed those crops are: 0 represents total monoculture and 1 represents perfectly even and infinite diversity.

2.3 Market engagement

This study defines market engagement as a combination of market sales and market purchases. Across many studies and contexts, market access has been shown to enhance livelihoods and lead to improved food security and dietary diversity outcomes through the increased income generated from sales of crops grown under a farm production strategy of either specialization or diversification (Fraval et al., 2018; Sibhatu & Qaim, 2018a; UBOS, 2017). Better access to markets may also allow households to source more diverse foods through purchasing. Several studies use different measures of markets to create indices for concepts like market access and market participation. Market access is typically measured as the geographic distance to a market hub, average distance to a market for specific crops, or distance to a main road; it also may be measured as travel time to a center of population of a certain number of inhabitants (Habtemariam et al., 2021; Muthini et al., 2020a, b; Singh et al., 2020; Wichern et al., 2018). Market participation is more typically computed as a farm commercialization index that is either the total value of farm products sold, or the proportion of farm production sold (Muthini et al., 2020a, b; Ogutu & Qaim, 2019). Other studies calculate market participation as the percent of a farmer’s crops they sold at market (Singh et al., 2020).

In this study, we measure market engagement as both total farm sales as reported in the UNPS and proportion of food that is eaten by the household that is bought outside the household. Together, these variables account for both the degree to which households sell their products for income as well as the degree to which households rely on markets to access food, thereby providing a more robust depiction of market use than the more limited focus of market sales or commercialization that is common in studies measuring market participation (Muthini et al., 2020a, b; Ogutu & Qaim, 2019). All farm sales data are based on self-reported commercialization of food crops produced based on the UNPS and are reported in Ugandan shillings. All market purchase data are based on self-reported purchasing habits as collected by the UNPS household survey. The proportion is calculated by dividing the value of food that was purchased outside the households by the total value of food that was consumed. Including market purchases as a part of market engagement is not typical in the literature but is included in this study to account for the possibility that the different ways that households engage in markets may influence their food security and dietary diversity status. Using both market sales and market purchases provides a better sense of where households are sourcing their food from (markets vs. own production).

2.4 Demographic variables

Demographic and farm characteristics including female-headed household, education level of the household head, travel time to market, region, and livestock ownership are included as controls (Fig. 1). These control variables were chosen based on demographics available from the UNPS and those analyzed in previous studies (Ecker, 2018; Jones et al., 2014; Koppmair et al., 2017; Mondal et al., 2021; Reynolds et al., 2020).

3 Methods

3.1 Data

Data are drawn from the 2015–2016 Uganda National Panel Survey (UNPS) as a part of the Living Standards Measurements Study – Integrated Surveys on Agriculture (LSMS-ISA) in Uganda. The data were collected by the Ugandan Bureau of Statistics in collaboration with the World Bank via a series of household surveys conducted in two visits over a twelve-month period in a nationally representative sample of households (UNPS BID, 2020). The household samples descend from an original sample of 3,123 households in 2009/10—the first year of the survey. The agriculture questionnaire from the 2015–2016 UNPS surveys 5,581 agricultural households. Non-agricultural households were not included in our analyses, given that the target population of this study was small-scale farm households. The continuous variables Farm Sales, Off-farm Income, and Farm Size were winsorized at the 99th percentile to remove outliers. Descriptive statistics for the variables used for analysis are summarized in Table 1.

Table 1 Description of Variables Used for Analysis
Full size table

3.2 Data analysis

All statistical analyses were conducted in Stata 16.0, beginning with bivariate analyses. To explore possible associations between on-farm crop diversity (SDI) and market engagement (farm sales and market purchases) (objective i) we use bivariate linear regressions. We then use ordinal logistic regressions for bivariate analyses involving the ordinal outcome variables HDDS-11 and Food Secure Months (objective ii). To examine the degree to which food sources (market vs. farm) relate to market engagement, on-farm crop diversity, and dietary outcomes including dietary diversity and food security, we run bivariate tests in the form of ordinal logistic regressions between SDI, HDDS-11, and Food Secure Months as outcome variables.

To further examine the possible associations between crop diversity, market access, and food security or dietary diversity (objective iv), we used Structural Equation Modeling (SEM), as summarized in Fig. 1. SEM estimates coefficients for the association between variables which can be both exogenous and endogenous. SEM also allows estimation of both direct and indirect effects of variables on one another.

Our model assumes an association between Crop Diversity (measured by SDI) and Market Engagement (measured by farm sales and the proportion of food purchased to food consumed). The model then considers several exogenous variables that might affect both SDI, Market Engagement, and either HDDS or Food Secure Months. We use Livestock ownership as an instrumental variable, with the assumption that it affects Market Engagement and HDDS but does not affect SDI because animal forage crops are not included in the calculation for SDI. The model predicts the relationships between each of the endogenous variables with each other and between the exogenous variables and the endogenous variables.

The SEM predicts HDDS and Food Secure Months as outcomes, as shown in Fig. 1, and estimates the association between SDI, Market Engagement, and either HDDS-11 or Food Secure Months, including demographic control variables. Market Engagement and SDI are endogenous variables as they are both being tested as predictors of HDDS-11 and are outcome variables themselves. The SEM technique allows us to include variables that are both predictors and outcomes, showing both direct and indirect effects of the variables on one another (Bliege Bird & Bird, 2021).

4 Results

4.1 Associations between crop diversity and market participation

Figure 2 shows the relationship between the Farm Sales and SDI and between the proportion of Market Purchases and SDI with a Lowess regression line. We see that the line is upward sloping for Farm Sales vs. SDI. Households that maintain lower levels of crop diversity tend to have somewhat lower sales. On the other end, households that maintain higher levels of crop diversity tend to have higher farm sales. Households that fall into the middle (maintaining closer to 0.4 SDI), also have lower levels of crop diversity. We also see a general trend for Market Purchases vs. SDI. For lower levels of crop diversity, households tend to purchase a higher proportion of their food in markets. For higher levels of crop diversity, households tend to purchase a lower proportion of their food at markets, suggesting more reliance on home production.

Fig. 2
figure 2

Proportion of Crop Production Sold vs. Simpson’s Diversity

Full size image

4.2 Crop diversity, market participation, and food security

The results of bivariate tests examining patterns in HHDS, Food Secure Months, SDI, Farm Sales, and Market Purchases are presented in Table 2. We find a significant positive association between SDI and HDDS-11. There is also a significant positive association between SDI and Food Secure Months, meaning an increase in SDI is associated with more Food Secure Months. There is also a significant positive association between total Farm Sales and SDI (coef. = 0.007), but a negative association between Market Purchases and SDI (coef. = -0.231) suggesting a potential tradeoff between market reliance and on-farm crop diversity.

Table 2 Bivariate Tests
Full size table

4.3 Structural equation models for market engagement, crop diversity, dietary diversity, and food security

The SEMs for HDDS and Food Secure Months with all variables and controls are shown in Table 3. Because there was no covariance in initial models, we ran the full SEMs without covariance terms. We use SDI as a predictor of Farm Sales and Market Purchases. Because agriculture is an operation where the planting must be done months in advance of any sales, it is unlikely that farm sales will have any influence on what has already been planted. Likewise, it is unlikely that any food purchases in the market will affect which crops the household planted in the previous months. We theorize that it is more likely that the crop diversity the household has grown will influence the sales of those same crops when it comes time to sell them. It is also more likely that the crop diversity a household has maintained will help them to determine what food they need to purchase outside the home.

Table 3 SEM for SDI, Farm Sales, and Food Security and Dietary Diversity and SDI, Market Purchases, and Food Security and Dietary Diversity
Full size table

First, we ran an SEM test with HDDS, SDI, and Farm Sales as endogenous variables and FHH, Education, Off-farm income, Farm Size, and Region as exogenous variables, with Livestock as an instrumental variable unrelated to SDI. Then we ran a SEM test with Food Secure Months, SDI, and Farm Sales as endogenous variables and FHH, education, Off-farm Income, Farm Size, and Region as exogenous variables, with Livestock as an instrumental unrelated to SDI. The former shows that Farm Sales (coef. = 0.035) and SDI (coef. = 0.856) are both significantly positively associated with HDDS. FHH is significantly positively associated with HDDS (coef. = 0.180). Education is significantly positively associated with HDDS (coef. = 0.579). Farm Sales is not significantly related to Food Secure Months. SDI is significantly positively associated with Farm Sales (coef. = 2.277) after accounting for HDDS and Food Secure Months and the control variables. After accounting for HDDS and Food Secure Months, FHH is significantly negatively associated with Farm Sales (coef. = -0.879), while education is significantly positively associated (coef. = 0.405).

Two more SEMs are shown in Table 3. The third model ran an SEM test with HDDS, SDI, and Market Purchases as endogenous variables and FHH, Education, Off-farm Income, Farm Size, and Region as exogenous variables, with Livestock as an instrumental variable unrelated to SDI. The fourth model ran an SEM test with Food Secure Months, SDI, and Market Purchases as endogenous variables and FHH, Education, Off-farm Income, Farm Size, and Region as exogenous variables, with Livestock as an instrumental variable unrelated to SDI. The former model shows that SDI (coef. = 1.154) and Market Purchases (coef. = 1.405) are significantly and positively associated with HDDS. FHH (coef. = 0.174) is positively associated with HDDS. FHH shows no significant relationship to market purchases. The fourth full SEM shows that SDI is significantly and positively associated with Food Secure Months (coef. = 0.377), meaning that increased SDI relates to more Food Secure Months. Across both models, SDI is significantly negatively associated with Market Purchases (coef. = -0.175).

5 Discussion

Overall, the results of these analyses are consistent with the literature that finds crop diversity and market engagement to be positive indicators of food security (Bellon et al., 2020; Ecker, 2018; Jones et al., 2014; Koppmair et al., 2017; Sibhatu & Qaim, 2018b; Tesfaye & Tirivayi, 2020; Tobin et al., 2019). Conducting SEM with UNPS data has allowed us to analyze the associations between on-farm crop diversity, market sales, and market purchases and the association of each with dietary diversity and Food Secure Months in Uganda. We find that although SDI and Farm Sales both have a moderating effect on HDDS-11, only SDI is significantly related to Food Secure Months. This suggests that overall, SDI is an effective mechanism to address food security concerns within Uganda.

5.1 Objective I) How levels of on-farm crop diversity co-vary with measures of household market engagement.

In the bivariate Lowess regressions between SDI and Farm Sales and between SDI and Market Purchases, we see that there are general trends in both charts. Households with lower levels of crop diversity tend to have lower farm sales and a higher proportion of market purchases and households with higher levels of crop diversity tend to have higher farm sales and lower proportion of market purchases. However, after running an initial SEM model for Market Engagement and Farm Sales with control variables included, we found that there is no covariance for SDI and Farm Sale or SDI and Market Purchases. This means that, after controlling for demographics and farm characteristics, SDI and Market Engagement do not tend to follow the same patterns. There is no significance in the way that they vary in relation to one another. Although several studies suggest that diversifying farm production will positively influence financial resilience (de Roest et al., 2018), we find that this avenue does not reflect a direct relationship in the context of Uganda; there are farm characteristics and demographic variables which explain the initial relationship seen between SDI and Market Engagement. This may indicate that diversification has the potential to be a strategy for improving general household welfare, which can relate to household characteristics (Tesfaye & Tirivayi, 2020).

In the full SEMs that include HDDS and Food Secure Months, we demonstrate that there are significant relationships between SDI and Farm Sales and SDI and Market Purchases. These relationships emerge without testing for covariance and after assuming that SDI is a predictor of Farm Sales and Market Purchases. So, although there is no significant covariance between SDI and Market Engagement, there are significant relationships between them.

5.2 Objective II) How varying levels of crop diversity and market engagement influence household dietary outcomes including dietary diversity and food security.

Seeing that SDI is strongly and significantly positively associated with HDDS-11 and Food Secure Months across all models supports evidence that maintaining high levels of crop diversity has a positive influence on dietary diversity and food security among small-scale farmers at the household level (Jones et al., 2014; Koppmair et al., 2017; Rajendran et al., 2017; Tesfaye & Tirivayi, 2020; Tobin et al., 2019). The strength of the association between crop diversity and dietary diversity that we find is stronger than what is typically reported in other studies, which may indicate that this relationship is somewhat unique to Uganda. Because the association between SDI and HDDS is strong, it means that Ugandan households who are successful in maintaining high levels of crop diversity have a higher likelihood of having a more diverse diet.

On the other hand, we see that SDI is consistently significantly negatively related to Market Purchases. There is some evidence that is may be related to cultures in which the people are eager to be self-reliant for their food security, such as the Batwa of Kanungu, Uganda, who expressed excitement at the opportunity to farm for their own food, especially when market prices for buying food were expensive (Patterson et al., 2017). The association between SDI and Market Purchases also suggests that diversification of the crop portfolio does not typically help increase a household’s market purchasing power in Uganda, perhaps because households maintaining higher levels of crop diversity have a lesser need to purchase food outside the home. Therefore, it is likely that the mechanism by which SDI increases dietary diversity and enhances food security (through fewer Food Secure Months) is not through increased food purchasing power from market sales via diversification. This leads us to postulate that SDI increases dietary diversity and food security through home consumption, or else through some other pathways not investigated in this study, such as through informal exchange. The idea that crop diversification contributes to home consumption is consistent with Bellon et al. (2020), yet contradicts findings in the same study that show diversification as beneficial for market engagement. This may indicate that market avenues in Ghana are more receptive to diversification than those in Uganda. Our results call into question the idea households may diversify their crop portfolio in order to gain access to more diverse crops via economic exchange.

However, we also find that SDI is significantly positively associated with Farm Sales, with a large coefficient. After considering this association, it appears that SDI may be a strategy through which to increase sales at markets; however, this increased income does not necessarily lead to increased purchases of food at markets. Because SDI is significantly negatively associated with Market Purchases, and significantly strongly positively associated with Farm Sales, we assume that Market Purchases and Farm Sales do not have a linear relationship. Thus, encouraging households to diversify their crop production and sell it at markets, with the intention to benefit both the household and the community, may be misled if the goal is to increase the household’s market purchases of food. Our findings are somewhat consistent with previous findings indicating the income benefits of crop diversification (Fraval et al., 2018). We find evidence that diversification can lead to positive effects on income from farm sales at market; but we do not find that markets are necessarily the direct pathway through which crop diversity increases dietary diversity, since higher crop diversity negatively impacts food purchases. The increased farm sales that relate to higher crop diversity may not be used by the household to buy food; the money may be used to cover the costs of planting, to reinvest in farming, or for other household needs. Although other studies have suggested that higher dietary diversity may be attributed to the production of cash crops, which can contribute to income and thus help households diversify their diets by increasing purchasing power, our findings instead indicate that increased crop diversity increases food security through some non-purchase-related channel, an important analytical focus for future studies (Sibhhatu & Qaim, 2018a).

Our analyses also find that Farm Sales and Market Purchases are significantly related to HDDS, but not to Food Secure Months. Although we exercise caution when comparing coefficient magnitudes, the coefficient relating Market Purchases to HDDS (1.405) is much larger than that of Farm Sales (0.035), which likely means that households spend their income on many other things in addition to food. When we think about how increases in SDI relate to increases in Farm Sales, but not to increases in Market Purchases, this makes sense. The increased income may not be going toward purchasing food. This finding is consistent with much of the literature proposing that increasing market engagement can improve diet outcomes while suggesting that purchasing habits may be more related to diets than selling habits (Rajendran et al., 2017). Nonetheless, we can see that both increased income from sale of crops, and increased purchasing of food outside the home are both associated with better dietary diversity. We also note that the coefficient relating Market Purchases to HDDS (1.405) is larger than the coefficient relating SDI to HDDS (1.154), but not by a large margin. These findings are consistent with the literature which indicate that market purchases are a more powerful indicator of food security than crop diversification, but we add that both are nearly equally important strategies (Koppmair et al., 2017; Rajendran et al., 2017). Therefore, finding some way to increase money available to the household may help to increase small-scale farmers’ dietary diversity. Although it is unclear why both indicators of market engagement (Farm Sales and Market Purchases) have no significant effect on Food Secure Months, it is consistent with Usman and Callo-Concha, (2021), who found no significant association between market access and food security in Southwestern Ethiopia. It is important to note that while Food Secure Months captures a temporal aspect of food security, dietary diversity focuses on one moment in time, meaning that the time of year data are collected may have implications to the relationships between market engagement and dietary diversity.

5.3 Objective III) The influence of farm household demographics such as farm size, gender of the household head, education, and off-farm income on market engagement, on-farm crop diversity, and diet outcomes

FHH is significantly negatively associated with SDI and Farm Sales but is not significant to Market Purchases. These findings are somewhat consistent with Reynolds et al. (2020), who found an initial negative relationship between female household heads and crop diversity; however, after controlling for more variables, including landholding, their study shows that this relationship turns significantly positive. Our findings are consistent with the initial negative association; however, we find this negative association despite controlling for farm size. This may be due to the country-specific context of Uganda. Although we find somewhat different results, we agree that special attention must be given to female-headed households, since we find that these households tend to have lower farm sales compared to general households.

Female-headed households are significantly negatively associated with Farm Sales in our full SEMs, which have added in the household, farm, and demographic characteristics. This relationship remains across both dietary diversity and food security (Food Secure Months) measures. The coefficient (coef. = -0.879) is large and equal in both SEMs for HDDS and Farm Sales and Food Secure Months and Farm Sales. This indicates that women tend to have lower farm sales even after controlling for several different variables, including farm size, which strays from the findings of Reynolds et al. (2020). However, these results are consistent with Pincus et al. (2018), who find that women earn less from agricultural sales than men do. The tendency for female-headed households to have lower farm sales and lower crop diversity is consistent with global trends that reveal a gap in access to resources for women in agriculture (Quisumbing et al., 2014). This includes access to land, seed, and technologies (Doss et al., 2015; Fisher & Carr, 2015; Jost et al., 2016).

Interestingly, we see that FHH is significantly positively associated with HDDS in the SEMs for both Farm Sales and Market purchases. Having a negative relationship with sales but a positive relationship with dietary diversity is reflective of the idea that women have more power over consumption decisions than they do over sales decisions, as presented by Tavenner et al. (2019). In order to help increase the food security of small-scale female farmer households, special attention must be paid to gender.

Education is generally a positive indicator for dietary diversity and food security outcomes, indicating that efforts to increase education across Uganda would be helpful in increasing dietary diversity and food security outcomes. This finding is consistent across most of the literature which shows that education tends to increase with positive outcome variance (Jones et al., 2014; Koppmair et al., 2017; Rajendran et al., 2017; Reynolds et al., 2020).

6 Limitations

The calculation of HDDS-11 requires further explanation. Households that have an HDDS-11 score of 1 may be overinflated; the score of 1 may include a feasible food group, such as ‘Eggs’, or it could include just the group “Other foods.” The items from the UNPS survey question that were sorted into the “Other foods” category include: cooking oil, margarine, butter, etc., sugar, other juice, and other foods. Responses that were sorted into this category may have indicated the item “other foods” with a food in mind that would have fallen into one of the other ten HDDS-11 food categories. However, that detailed information is not available from the UNPS. Therefore, households may be consuming higher levels of dietary diversity, but are only reflected with HDDS-11 = 1 based on the available data.

Because subsistence agriculture in Uganda is rainfed, food security can correspond to harvest seasons for crops; this is especially true in certain regions of Uganda such as the Karamoja sub-region, which experiences six months of rains, six months of dry season, and one harvest per year (Lunyolo et al., 2021; Muggaga et al., 2021). This seasonal aspect of food security was not captured within the model in this study. Future research should include the month the survey was given within the model to investigate the effect of changing seasons on food security and dietary diversity outcomes.

7 Conclusion

This study finds that there are positive associations between market engagement and dietary diversity, between on-farm crop diversity and dietary diversity, and between on-farm crop diversity and food security. The debates about development pathways to enhance food security and nutrition outcomes are too often pitched as a binary, by which either crop diversification or market access must be pursued. Instead, the findings from this study suggest that both strategies can be useful. There are benefits to smallholders that come from a combination of crop diversity and market engagement (Koppmair et al., 2017; Rajendran et al., 2017; Sibhatu & Qaim, 2017). Meanwhile, the question of specialization versus diversification is highly context dependent (Bellon et al., 2020; de Roest et al., 2018; Ochieng et al., 2016, 2020). We see that increased on-farm crop diversity can contribute to better dietary diversity and food security, but that does not necessarily move in tandem with market engagement: while on-farm crop diversity leads to increased market sales in Uganda, it does not lead to increased market purchases. This suggests that targeted interventions are necessary, with the recognition that some households may rely on their crop production for home consumption to diversify their diets and enhance their food security, while others may use increased income through market sales without necessarily maintaining crop diversity, while still others may use a combination of crop diversity and market strategies. Additionally, we see that market purchases have a greater effect on dietary diversity than market sales. This adds a dimension to the broader debate by indicating that, at least in Uganda, the avenue through which diversification enhances food security is not necessarily through commercialization. Should market-oriented crop diversification be pursued as suggested by Sibhatu and Qaim (2018a), efforts to stimulate market purchasing demand for diverse crops must also be undertaken. Additionally, although increased purchasing power will help to improve diet quality and food security, market sales do not contribute greatly to that power. While it may be possible to improve diet quality and food security by increasing off-farm income, it is more important to make the small-scale farmer livelihood viable, since they contribute the majority of the food supply in Latin America, sub-Saharan Africa, and South and East Asia (Samberg et al., 2016).

In sum, and in accordance with previous studies, both market engagement and crop diversity can contribute to better dietary diversity and higher food security (Frelat et al., 2016; Jones et al., 2014; Koppmair et al., 2017; Rajendran et al., 2017; Sibhatu & Qaim, 2017; Tesfaye & Tirivayi, 2020; Tobin et al., 2019). This validates past and current initiatives that aim to bolster markets and crop diversity. Development and intervention strategies that address only market engagement or only on-farm crop diversity will only have partial effectiveness and may miss out on opportunities to improve livelihoods by sacrificing one for the other. Intervention strategies that acknowledge both will be more effective than only focusing on one or the other. The context in which the solution is being implemented is of utmost importance.