"background": "The adoption of modern process-control systems in industrial sectors is critical for enhancing efficiency and safety. In many developing economies, however, the measurement and analysis of adoption rates are hindered by heterogeneous data sources and contextual variability, leading to imprecise policy and investment decisions. ", "purpose and objectives": "This study presents a novel methodological framework for the comparative evaluation of process-control system adoption. Its primary objective is to develop and apply a Bayesian hierarchical model to analyse adoption rates, explicitly accounting for regional and sectoral heterogeneity within the Tanzanian context. ", "methodology": "A comparative study was conducted using a multi-stage stratified sample of manufacturing and processing facilities. Adoption was modelled via a Bayesian hierarchical logistic regression, where the log-odds of adoption for facility i in sector-region j is given by = \ + \ X{ij + uj, with uj \ N (0, \²u). Posterior distributions were estimated using Markov chain Monte Carlo sampling, with inference based on 95% credible intervals. ", "findings": "The model revealed substantial variation in adoption rates, with a posterior median for the sector-region standard deviation \ᵤ of 0. 85 (95% CrI: 0. 72, 1. 01), indicating significant contextual effects. A key concrete result is that facilities with dedicated technical staff were 3. 2 times more likely (95% CrI: 2. 1, 4. 8) to have adopted advanced systems. ", "conclusion": "The Bayesian hierarchical approach provides a robust methodological advancement for adoption analysis, quantifying uncertainty and heterogeneity more effectively than traditional methods. It confirms that adoption drivers are not uniform but are strongly mediated by localised institutional and technical capacities. ", "recommendations": "Policy initiatives should prioritise building technical human capital alongside technology provision. Future adoption studies in similar contexts should employ hierarchical models to avoid aggregation bias and generate more reliable
Juma Mwakyembe (Fri,) studied this question.