Application of a Bayesian network modelling approach to predict the cascading effects of COVID-19 restrictions on the planting activities of smallholder farmers in Uganda

Abstract

CONTEXT

There are rising concerns over the cascading effects induced by COVID-19 restrictions on the planting activities of smallholder farmers in low and middle-income countries, which may become a non-negligible threat to the long-term food security. Studies that utilize probability based models to examine the effects of COVID-19 restrictions on planting activities of smallholder farmers are scare, with no available evidence on Uganda. Yet these effects do not act in isolation, and are known to be complex, stochastic, nonlinear, and multidimensional.

OBJECTIVE

To develop a Bayesian network (BN) model based on expert knowledge, existing literature, and Uganda's High Frequency Phone Survey (HFPS) datasets on COVID-19 to bridge this gap.

METHODS

A comprehensive survey of relevant literature on the effects of COVID-19 restrictions on the planting activities of smallholder farmers was conducted based on well established guidelines. Resultantly, 23 relevant publications were obtained, and reviewed. A total of 12 variables deemed relevant to smallholder famers in Uganda were extracted, and organized into an influence diagram. The influence diagram was used to develop the BN model. A total 6313 households aggregated from Round 1, 4 and 7 of the HFPS datasets on COVID-19 was used in this study. A training portion (75%, n = 4734) was used to populate the model, and test dataset (25%, n = 1578), was used evaluate model accuracy.

RESULTS AND CONCLUSIONS

The error rate was 17.9%% implying that the model had the majority of its predictions correct (82.1%). The model's classification power, was evaluated basing on the scoring rules. The model's scoring rule results indicated that the model has a strongest predictive power with both the logarithmic loss (0.45,) and quadratic loss (0.29) scores close to zero, while a spherical payoff (0.84) approaching 1. Results reveal the maize, beans, and ground nuts, were the most grown crops during the pandemic as compared to the period before the pandemic. The sensitivity results indicate that the probability of COVID-19 restrictions to affect the planting activities of the smallholder farmers in Uganda was 30%. The variables of ‘unable to acquire seeds, and fertilizers’ affected the planting activities by 2.6 percentage points (PP), and 1.3 PP respectively. The variables ‘travel restrictions’ and reduced labour, affected the planting activities by 11 PP and 1PP respectively.

SIGNIFICANCE

These findings emphasize the importance of intervening on the highly ranked effects to enhance the resilience of local food systems, and smallholders' capacity to cope with recurring and unforeseen shocks.