Food Demand Characteristics inUganda: Estimation and Policy Relevance

Abstract

Uganda was highly successful in reducing poverty over the past two decades but made little progress towards household food security. This underlines the need for designing food security interventions customised for household-specific needs and behaviours. This study estimates Ugandan household demand behaviour with a focus on food consumption paying particular attention to household-specific characteristics. The results show that preferences to increase calorie-dense staple consumption, likely associated with food energy deficiency, extend far beyond the percentage of rural Ugandans officially deemed poor. Price elasticities indicate that poor rural households are largely well positioned to compensate staple price increases by substitution as long as they are not already concentrated on the cheapest foods. This flexibility applies less to urban households. The estimated demand elasticities generally vary widely between rural and urban households and depend on expenditure levels. Household-specific characteristics have significant, sometimes pronounced, influences on demand, as do seasons and regions. The results reflect highly differentiated demand behaviour, which can be utilised to improve the design and evaluation of food security interventions. JEL Classification: D12, O55, Q18

Heterogeneity in Food Demand among Rural Indian Households: The Role of Demographics

“…In particular, it allows a good to be a luxury item at one end of the income distribution but a necessity at the other end (Banks et al ). Originally proposed by Banks et al (), several studies have employed the QUAIDS model to estimate food demand, by broad category, in both developed (e.g., Abdulai [2008] for Switzerland) and developing (e.g., Boysen [] for Uganda; Meenakshi and Ray [] for India; and Fashogbon and Oni [] for Nigeria) countries. The analyses in many of the studies, such as that by Fashogbon and Oni () are augmented to account for differences in demographic factors across households using methods proposed by Poi () and Ray ().…”

“…The model accounts for differences in socio‐economic conditions across households by augmenting the demographic and household specific variables (e.g., household size) using the method proposed by Ray () and Poi (). For purposes of analysis, furthermore, like Boysen () we assume weak separability in the household's two‐stage budgeting process wherein the first stage the household makes a determination as to the percentage of the total budget will be allocated to food items and then in the second stage allocates the food budget among the different food items . As noted by Thompson (), elasticities contingent on exogenous total group expenditure in the demand system may be inappropriate when assuming a two‐stage allocation process.…”

“…The second‐stage elasticities are conditional on the first‐stage because allocation among food categories is conditional on allocation of the budget to food. Based on Carpentier and Guyomard () and Boysen (), unconditional food group demand elasticities, based on the estimated demand equations associated with total food (Equation []) and food categories, can be expressed as…”

“…To estimate the QUAIDS demand system, a nonlinear seemingly unrelated regression approach is employed . To correct for potential biases in the variance covariance matrix of demand system estimation in two‐stage models, Deis and Hill () and Boysen () propose computing nonparametric bootstrapped standard errors instead of usual asymptotic standard errors. Following this procedure, we undertook bootstrapped standard errors to obtain valid and consistent standard errors in demand system estimation .…”

“…This procedure is repeated until the iteration‐to‐iteration change of the coefficient is very small and the process is converged. As α 0 is difficult to estimate, we choose an arbitrary small value of 10, as practiced in previous such studies (see Deaton and Muellbauer , p. 316; Boysen ). We also tried with alternative values of α 0 as 5, 9, and 11 which did not change the resulting elasticities substantially but required more iterations to converge.…”