行政记录整合的贝叶斯分层记录链接模型及应用

Bayesian Hierarchical Record Linkage Model and Its Application in Administrative Records Integration

记录链接的技术问题与统计理论密切相关,尤其是在建立记录链接分类规则时需要构建统计模型,识别关键变量以完成数据匹配。在贝叶斯框架下构建分层模型整合行政记录,通过多元回归可以实现匹配错误率的估计,而且一对一限制下的记录链接允许通过模块反映记录信息的来源变化,基于MCMC模拟的后验分布计算方便,有助于提高数据整合效率。

Technical issues about record linkage require a combination of statistical theory to resolve, especially in the establishment of record linkage classification rules, we need to build a statistical model to identify critical variables in order to complete the data match. Build hierarchical model to integrate administrative records under the Bayesian framework, can estimate matching error rate by multiple regression, and allows block to reflect the change of information sources under restriction of one to one integration, the posterior distribution calculated based MCMC simulation is convenient, help to improve the efficiency of data matching.