The study area
Masaka District is located in southcentral Uganda in the Buganda kingdom, the largest of the traditional kingdoms of East Africa. The Bagandan people speak Luganda, Swahili, and English and are primarily engaged in smallholder agricultural activities. More than two million people live in Masaka district with a population density of 700 people per square kilometer (Masaka District Local Government, 2024). The high population density can be explained by favorable climate and proximity to Lake Victoria (Fig. 1) and Kampala, Uganda’s capital.
Masaka District experiences two rainy seasons a year, which allows for two growing seasons. Temperatures range between 10 and 30°C (Uganda Bureau of Statistics, 2022). The topography of the region is characterized by hills and ridges with an average altitude of 1,150 m above sea level. Masaka District has a total area of 1,603 km2, half of which is wetland. The dominant coffee (Coffea canephora Pierre ex Froehn.)–banana (Musa spp.) crop production system is supplemented with cassava (Manihot esculenta Crantz), sweet potatoes (Ipomoea batatas [L.] Lam), beans (Phaseolus vulgaris L.), groundnuts (Arachis hypogaea L.), and maize (Zea mays L.).
The high population density and continued growth have led to agricultural land fragmentation (Mwesigye & Barungi, 2021), cropping intensification, deforestation, and decreased soil fertility. Despite the region’s good climate and overall fertile soils, soil quality has been deteriorating at a high rate. Soil degradation is estimated to affect 50% of the Masaka District area. The situation is exacerbated by insufficient soil conservation practices and climate change, which create irregular rain patterns (Gbegbelegbe, 2018) and disrupt historic crop production cycles. As a result, the dominant banana-coffee cropping system is experiencing multiple pressures including pest and disease outbreaks that threaten farmers’ livelihoods. These factors combined make Masaka District a valuable site to study the shifting dynamics between farming systems and food security (Hashmiu et al., 2024).
Data collection
A farm household survey was administered to 150 smallholder respondents covering six sub-counties (Bukakata, Mukungwe, Buwunga, Kabonera, Kyanamukaka, and Kyesiga) and one division (Katwe-Butego) during September–October 2015. Mukungwe and Kabonera sub-counties now belong to a new district (Masaka City) that was carved out from Masaka District in 2021 (Masaka District Local Government, 2024). District-level administrative units consist of sub-counties, parishes, and villages. For respondent selection, five villages were randomly chosen per parish. Then, farmers were selected based on their status as a smallholder (determined by the size of the land farmed) and engagement in the production of two most common crops: banana and coffee. One to six farmers were identified in each village with an average of two households per village. Farmer training records kept by local extension services or by village leaders were used to aid in respondent identification. An initial list of all potential respondents (approximately 500 farmers) was created, but due to resource constraints, 150 were randomly selected. To capture cross-village variation, the number of farmers surveyed per village was limited to a maximum of six.
Interviews covered socioeconomics, agricultural production, and food security. The interviews were conducted in Luganda with the help of a local extension agent who served as an interpreter. The majority (66%) of those interviewed were women aged 30 and above. Most farming households had four to seven members. Most farm sizes were less than four acres. Seven or eight years of primary schooling was the most common educational level achieved by any household member. Only 32% of farmers surveyed had access to electricity.
Description of explanatory variables
The choice of variables was informed by the findings of previous studies, study objectives and data availability. To accurately identify dominant farm typologies and their relation to food security, we selected variables that were reflective of the characteristic smallholder farming systems and food security determinants in southcentral Uganda (Alvarez et al., 2018). Most of the farming systems in the region are characterized by intercropping of annual crops (beans, maize, cassava, sweet potatoes, and groundnuts) with banana and coffee. Because bananas serve as a dominant staple crop and coffee as a dominant cash crop (Bongers et al., 2015; Bunn et al., 2019), the study focused on these two specific crops. The significance of these crops in the agricultural development agenda in Uganda is emphasized by the disproportionate efforts at developing new banana and coffee hybrid varieties and educating farmers on their management techniques (Batte et al., 2019; Wang et al., 2015).
Food security among smallholder households in Uganda has been found to be associated with different factors that can generally be categorized into two main groups: socioeconomic and agricultural. Our analysis included variables from each group (Table 1). Specifically, the socio-economic factors were comprised of gender (Kassie at al., 2012), off-farm income (Wichern et al., 2017; d’Errico et al., 2018; Twongyirwe et al., 2019 ), land size (total amount of land to which a farmer has use rights) (Apanovich & Mazur, 2018), walking time to farmed land, and how land was acquired (inherited, received as gift, borrowed, or purchased) (Singirankabo & Ertsen, 2020; Muraoka et al., 2018). We used walking time to land as a proxy for labor requirements. Farmers who have to walk long distances often engage in agricultural practices different from those who walk smaller distances to their land (McCall, 1985). We used the “how land was acquired” variable as a proxy for land rights because the way a farmer gains access to land impacts farming practices and food security (Séogo & Zahonogo, 2023). Agricultural factors targeted the size of land dedicated to banana and coffee growing and crop yields (Apanovich & Mazur, 2018), as well as fertilizer use (organic vs. inorganic), pest management, and soil management (Pender et al., 2004, Mukuve & Fenner, 2015; Karamage et al., 2017). The use of synthetic inputs serves as a proxy for capital, because only those with access to cash can purchase inputs (Kwanmuang & Lertjunthuk, 2021; Silva et al., 2019). We hypothesized that crop management would be significant in establishing farm typologies.
The analysis also relied on two supplementary variables: number of meals in seasons of plenty and in scarcity. The season of plenty is the period right after harvest, and it usually occurs from June to August, and December to February in Uganda. The season of scarcity coincides with periods of intensive agricultural production activities, which are from March to May, and September to November. While the study acknowledges that food security is complex and consists of multiple dimensions (availability, access, utilization, and stability) and indicators, assessing the number of meals a household consumes per day across seasons has been reported as an effective proxy for food security in this part of Uganda (Apanovich & Mazur, 2018; Patterson et al., 2017). In addition, by capturing seasonal variability in meal consumption, we intended to gain a deeper understanding of the relationship between farm typologies and food security throughout the year.
Table 1
Description of variables used in establishing farm typologies
Variable | Variable No. | Type | Description | Levels /Units |
Socio-economic | | | | |
Gender | V13 | Categorical | Sex of household head | Female/Male |
OffIn | V15 | Categorical | Off-farm income | Yes/No |
TotalA | V5 | Continuous | Total land owned | Acres |
WalkT | V6 | Continuous | Walking time to plot1 | Minutes |
PlotAcq | V14 | Categorical | How land was acquired | Inherited (1)/gift(2) /borrowed(4)/purchased(6) |
Agricultural activities | | | | |
BanA | V1 | Continuous | Area planted with bananas | Acres |
BanY*** | V2 | Continuous | Banana yields | Kg/acre |
CoffA | V3 | Continuous | Area planted with coffee | Hectare |
CoffY*** | V4 | Continuous | Coffee yields | Kg/acre |
Mplenty | XX* | Categorical | Number of meals in seasons of plenty | Number |
MScarce | XX | Categorical | Number of meals in seasons of scarcity | Number |
BanOrg | V7 | Categorical | Use of organic soil amendments on bananas | Yes/No/NA** |
BanInorg | V8 | Categorical | Use of inorganic soil amendments on bananas | Yes/No/NA |
Banpest | V9 | Categorical | Use of pesticides on bananas | Yes/No/NA |
CoffOrg | V10 | Categorical | Use of organic soil amendments on coffee | Yes/No/NA |
CoffInorg | V11 | Categorical | Use of inorganic soil amendments on coffee | Yes/No/NA |
Coffpest | V12 | Categorical | Use of pesticides on coffee | Yes/No/NA |
Soilmgt | V16 | Categorical | Use of soil management practices (crop rotation, fallowing, terracing, etc.) | Yes/No |
*XX indicates supplementary variables. Supplementary variables are not used in the analysis and do not contribute to dimension variability. Therefore, they cannot be used to describe dimensions, but they can aid in visualization and post−analysis of clusters.
**NAs indicate households that did not grow the targeted crops to show that practices relating to these crops were not applicable.
Values in parenthesis indicate codes used for levels.
***The natural logarithm of yields was used as crop yields were not normally distributed (Lobell et al., 2011).
Data Analysis
Recognizing that there is no single approach to measuring food security, methods for data analysis were determined by the study objectives and the type of available information. Factor Analysis on Mixed Data (FAMD) (Kassambara, 2017) and Hierarchical Clustering on Principal Components (HCPC) (Argüelles et al., 2014) were combined to establish relevant features that differentiate profiles of the smallholder farming households in Masaka District. This method of farm typology development has been applied by multiple studies (Mądry et al. 2013; Kuivanen et al., 2016; Alvarez et al., 2018; Nyairo et al., 2020). FAMD can handle both qualitative and quantitative variables by applying Principal Component Analysis (PCA) to the quantitative data and Multiple Correspondence Analysis (MCA) to the qualitative data (Abdi et al., 2013; Greenacle, 2010). PCA is a dimensionality reduction method which helps to reveal complex relationships and separations between groups of information, that are not easily visible in the original high-dimensional data (Mishra et al., 2017). FAMD outputs dimensions, which represent variability in the data, with each dimension having an eigenvalue that corresponds to the amount of variation explained by the dimension. In principle, if the eigenvalue is greater than one, then the dimension holds more information than an individual variable. Custer analysis is a data mining process which divides samples into groups (clusters) based on information within each sample and its relationship with other samples (Tan, 2005). Samples belonging to the same cluster must show a similarity pattern among them while being as dissimilar as possible from samples associated to other clusters.
Data were imported into R version 4.3.1, normalized, pre-processed to remove any spurious entries, and formatted into supported structures (continuous variables were set to numeric while qualitative variables were set as factors). The pre-processing identified two households that had multiple entries for the variable PlotAcq (V14) which were then deleted. The FactoMineR package (Lê et al., 2008), which supports multivariate analysis on a mixture of continuous and categorical data, was used to conduct FAMD. The variables Mplenty (Number of meals in seasons of plenty) and MScarce (Number of meals in seasons of scarcity) were set as supplementary variables. Supplementary variables do not contribute to dimension variability. Therefore, they cannot be used to describe dimensions, but can aid in visualization and post-analysis of clusters. The FAMD result was then used as input in the HCPC function, outputting the appropriate number of clusters for our dataset. HCPC utilizes Ward’s method by default, which is an agglomerative clustering technique that computes Euclidean distance between entries and successively merges the most similar samples (Murtagh, 2014). Ward’s method is based on both multidimensional variance and PCA, making it the most applicable for this analysis. Chi-square tests were conducted with HCPC, giving p-values at a 0.05 significance level. The ggplot2 package was then used to visualize results. Means (for quantitative variables) and frequencies (for qualitative variables) were computed to determine proportional differences per cluster and further characterize the developed typologies.
To determine food security per cluster, the mean value of the number of meals in scarcity (MScarce) was subtracted from the mean value of the number of meals in plenty (MPlenty). If the number of meals in plenty (Mplenty) was more than the number of meals in scarcity (MScarcity) then a cluster was interpreted as being vulnerable to shifting food regimes, because they had to reduce their meal consumption during scarcity. If the number of meals was not affected during seasons of scarcity, then a cluster was assumed to be resilient and food secure.