Journal Design Engineering Masthead
African Civil Engineering Journal | 18 September 2001

A Bayesian Hierarchical Modelling Framework for Evaluating Process-Control System Adoption in Tanzania (2000–2026)

J, u, m, a, M, w, i, n, y, i, m, v, u, a
Bayesian ModellingTechnology AdoptionIndustrial DevelopmentEngineering Methodology
A three-level Bayesian model handles sparse, longitudinal data from engineering projects.
Posterior estimates show a national adoption growth of 0.15 per annum, with high uncertainty.
Methodology allows partial pooling of information across regions to improve inference.
Provides probabilistically coherent estimates for policy and investment decisions.

Abstract

{ "background": "The adoption of modern process-control systems in engineering sectors is critical for industrial development, yet robust methodologies for evaluating adoption rates and their drivers are lacking. Existing approaches often rely on cross-sectional surveys or deterministic models, which fail to account for hierarchical data structures and inherent uncertainties in longitudinal adoption processes.", "purpose and objectives": "This article presents a novel Bayesian hierarchical modelling framework designed to quantify the adoption rates of process-control systems and to identify key technical and organisational factors influencing adoption. The objective is to provide a rigorous, probabilistic methodology that can handle sparse or incomplete data typical in such contexts.", "methodology": "A three-level hierarchical model is developed, where the adoption status of a system for a given project $i$ in region $j$ at time $t$ is modelled as $y{ijt} \\sim \\text{Bernoulli}(\\pi{ijt})$, with $\\text{logit}(\\pi{ijt}) = \\alphaj + \\beta X{ijt}$. Region-specific intercepts $\\alphaj$ are drawn from a common normal distribution, $\\alphaj \\sim N(\\mu\\alpha, \\sigma^2_\\alpha)$, allowing partial pooling of information across regions. Inference is performed using Hamiltonian Monte Carlo, with posterior distributions summarising all parameter and prediction uncertainty.", "findings": "The framework's application to a simulated dataset, constructed from expert elicitation, demonstrates its utility. A key finding is the model's ability to produce probabilistically coherent estimates, such as a posterior median for the national-level adoption growth parameter of 0.15 per annum, with a 90% credible interval of [0.08, 0.22], indicating a positive but uncertain trajectory. The hierarchical structure reveals substantial regional heterogeneity in baseline adoption probabilities.", "conclusion": "The proposed Bayesian hierarchical model provides a statistically sound and flexible methodology for evaluating technology adoption in engineering. It formally incorporates multi-level variability and quantifies uncertainty in estimates and predictions, offering a superior