Chen Qixuan, Gelman Andrew, Tracy Melissa, Norris Fran H, Galea Sandro
Department of Biostatistics, Mailman School of Public Health, Columbia University, New York, NY, U.S.A.
Department of Statistics, Columbia University, New York, NY, U.S.A.
Stat Med. 2015 Dec 10;34(28):3637-47. doi: 10.1002/sim.6618. Epub 2015 Aug 2.
We review weighting adjustment methods for panel attrition and suggest approaches for incorporating design variables, such as strata, clusters, and baseline sample weights. Design information can typically be included in attrition analysis using multilevel models or decision tree methods such as the chi-square automatic interaction detection algorithm. We use simulation to show that these weighting approaches can effectively reduce bias in the survey estimates that would occur from omitting the effect of design factors on attrition while keeping the resulted weights stable. We provide a step-by-step illustration on creating weighting adjustments for panel attrition in the Galveston Bay Recovery Study, a survey of residents in a community following a disaster, and provide suggestions to analysts in decision-making about weighting approaches.
我们回顾了针对面板损耗的权重调整方法，并提出了纳入设计变量（如分层、聚类和基线样本权重）的方法。通常可以使用多级模型或决策树方法（如卡方自动交互检测算法）将设计信息纳入损耗分析。我们通过模拟表明，这些加权方法可以有效减少因忽略设计因素对损耗的影响而导致的调查估计偏差，同时保持所得权重的稳定性。我们在加尔维斯顿湾恢复研究（一项针对灾难后社区居民的调查）中，提供了为面板损耗创建权重调整的分步说明，并为分析师在权重方法决策方面提供了建议。
