Underrated Bootstrap Correction for Linkage-Induced Estimation Biasntitled Item

This paper studies estimation bias arising from imperfect record linkage and proposes an iterated bootstrap correction that accounts for the feedback between linkage error and parameter estimation. The method repeatedly re-estimates the bias using bootstrap samples that preserve the estimated linkage structure, yielding higher-order bias reduction relative to conventional bootstrap corrections. We establish consistency and higher-order bias reduction under general regularity conditions and show that the procedure delivers asymptotically valid inference even when linkage probabilities are unknown and estimated from the data. Monte Carlo experiments demonstrate substantial finite-sample improvements in bias, root mean squared error, and coverage accuracy across linear and nonlinear models. The results provide a practical and general approach to bias correction for econometric analysis using imperfectly linked data. We also introduce a statistical test to determine when additional iterations yield negligible bias reduction relative to increased variance. Simulations using hormone data and a real-world application linking Australian Bureau of Statistics labour mobility surveys demonstrate substantial bias reduction and enhanced estimation accuracy.本文针对不完全记录链接（imperfect record linkage）所引发的估计偏误展开研究，提出一种可刻画链接误差与参数估计间反馈效应的迭代自助法校正（iterated bootstrap correction）方案。该方法借助保留已估计链接结构的自助抽样样本，反复对偏误进行重新估计，相较传统自助法校正，可实现更高阶的偏误缩减。本文在一般正则性条件下证明了该方法的相合性与高阶偏误缩减特性，并表明即便链接概率（linkage probabilities）未知且需从数据中估计得到，该流程仍可生成渐近有效推断。蒙特卡洛（Monte Carlo）实验结果显示，在线性与非线性模型中，该方法在偏误、均方根误差（root mean squared error）与覆盖精度（coverage accuracy）方面均实现了显著的有限样本性能提升。研究结果为使用不完全链接数据开展计量经济学分析提供了一种实用且通用的偏误校正方案。本文还引入了一种统计检验方法，用于判断当额外迭代次数带来的偏误缩减幅度相较于其引发的方差增大可忽略不计时的临界点。我们通过激素数据（hormone data）开展的仿真实验，以及将澳大利亚统计局（Australian Bureau of Statistics）劳动力流动调查数据进行链接的现实应用案例，均验证了该方法可显著缩减偏误并提升估计精度。