Developing an Original Bayesian Framework for Evidence Synthesis in Nigerian Educational Policy Evaluation

Evidence synthesis for educational policy in Nigeria requires integrating data from heterogeneous sources including randomised trials, observational studies, administrative data, and qualitative evaluations, and a Bayesian framework adapted for the sparse evidence and measurement challenges specific to Nigerian education contexts represents an important original methodological contribution. This study developed an original Bayesian evidence synthesis framework for Nigerian educational policy evaluation. A methodological development approach employing statistical theory and empirical testing was used: systematic review of Bayesian meta-analysis and evidence synthesis methods (63 publications from 2018 to 2024), analysis of 47 Nigerian educational intervention evaluations as a synthesis test corpus, and iterative framework development validated against simulated data with known causal structures. The framework specifies four Bayesian synthesis components: hierarchical meta-analysis for effect size aggregation across heterogeneous studies, Bayesian evidence weighting that adjusts for Nigerian-specific study quality dimensions (including common cluster randomisation design violations), non-parametric priors for studies with implausible effect sizes, and predictive posterior distributions for out-of-sample policy transfer scenarios. Applied to the synthesis test corpus, the framework identified heterogeneity sources not apparent in classical random-effects meta-analysis, including a significant rural-urban effect modification that standard synthesis methods masked. Expert validation by 16 Bayesian statistics and education policy specialists confirmed the framework's methodological originality. The study recommends adoption by the Federal Ministry of Education as the standard evidence aggregation method for National Education Policy review cycles.

Keywords: Bayesian evidence synthesis, educational policy, Nigeria, meta-analysis, hierarchical modelling