Data_EDI.csv.

BackgroundThe Educational Development Index (EDI) is a critical tool for assessing and tracking the progress of education systems from local to national, and even global scales and needs to be chosen for every layer of the subnational boundaries to secure the basic human rights of the people. In reality, there are significant variations within the consecutive time breaks and the geographical boundaries that need to be examined. The authors aim to examine how the EDI relates to various spatiotemporal variables.Methods and MaterialsThis research is based on secondary data on literacy rates (EDI) from 64 districts of Bangladesh and 6 relevant variables over the period 2001 to 2021. The optimal model for the data was identified from Bayesian spatial-temporal modeling (Linear, Analysis of Variance (ANOVA), Autoregressive (AR1), and AR2) and the Markov Chain Monte Carlo (MCMC) method used to generate data about the prior and posterior realizations. To select the best model different model selection and validation criteria such as the Deviance Information Criterion (DIC), Watanabe-Akaike information criterion (WAIC), and Root Mean Square Error (RMSE) were employed in this study.ResultsThe ‘AR1’ model is a ‘temporal model’ performed better than others. Significant spatial (=0.994) and temporal (=0.347) variations were identified for the suited model. Of the factors considered for model fitting, the health index, income index, expected years of schooling, population density, and dependency ratio are found to be important components of educational development in Bangladesh.ConclusionThe variation in the spatial domain can be used to identify the districts to improve the educational index controlling responsible factors by the policymakers.背景：教育发展指数（Educational Development Index, EDI）是评估与追踪地方、国家乃至全球层面教育系统进展的关键工具，需在各级次国家行政边界层面应用该指数，以保障民众的基本人权。现实中，连续时段间隔与地理边界间均存在显著差异，亟待开展相关审视与研究。本研究旨在探讨教育发展指数与各类时空变量之间的关联。 方法与材料：本研究基于2001年至2021年间孟加拉国64个地区的识字率（教育发展指数EDI）及6项相关变量的二手数据。研究通过贝叶斯时空建模（包含线性模型、方差分析（Analysis of Variance, ANOVA）、自回归（Autoregressive, AR1）与AR2模型）筛选适配数据的最优模型，并采用马尔可夫链蒙特卡洛（Markov Chain Monte Carlo, MCMC）方法生成先验与后验分布的相关数据。本研究选用偏差信息准则（Deviance Information Criterion, DIC）、渡边-赤池信息准则（Watanabe-Akaike information criterion, WAIC）以及均方根误差（Root Mean Square Error, RMSE）等多种模型选择与验证标准，以遴选出最优模型。 结果：AR1模型作为时序模型，整体表现优于其余模型。适配模型展现出显著的空间变异（系数为0.994）与时序变异（系数为0.347）。在模型拟合所考量的各项因素中，健康指数、收入指数、预期受教育年限、人口密度与抚养比被证实为孟加拉国教育发展的重要影响因子。 结论：空间维度的差异可用于识别亟需提升教育指数的地区，政策制定者可通过管控相关影响因子，推动当地教育发展水平提升。