Drivers of willingness to repay Parish Revolving funds (PRF) under Uganda’s Parish Development Model: a Bayesian Latent Class Logistic Regression (LCLR) approach

Abstract

This study examines the determinants of beneficiaries’ willingness to repay the Parish Revolving Fund (PRF) under Uganda’s Parish Development Model (PDM), using nationally representative data from the 2024 EPRC PDM Survey. The study applies a Bayesian Latent Class Logistic Regression (LCLR) model to account for observed and unobserved heterogeneity in repayment behaviour. Results show significant variation in repayment willingness by gender, education, region and wealth. Financial literacy, awareness of repayment terms, larger PRF amounts and shorter disbursement periods increase repayment willingness, while long delays and low household wealth reduce it. The Bayesian model identifies three borrower classes: institutionally aware borrowers (35%), economically empowered borrowers (45%) and financially constrained borrowers (20%), each with distinct traits. The model’s predictive accuracy is strong, with an area under the curve (AUC) of 0.828 and overall accuracy of 87%. Policy simulations indicate that combining better financial literacy with faster disbursement yields the highest repayment gains. The principal methodological contribution is the application of the Bayesian LCLR framework, which simultaneously estimates class-specific repayment parameters and latent class memberships with full uncertainty quantification, advancing beyond conventional homogeneous logistic models used in prior microfinance research. The study recommends the need to leverage on rural social capital for repayment enforcement, the need for borrower segmentation, focused financial training and efficient fund management to strengthen PRF repayment.