Diehr P, Grembowski D
Department of Biostatistics, School of Public Health and Community Medicine, University of Washington, Seattle 98195.
Am J Public Health. 1990 Nov;80(11):1343-8. doi: 10.2105/ajph.80.11.1343.
All small area analyses need to compare the observed variability in rates to that expected by chance alone, but the expected variability is usually not known. This paper uses patient-level data for five dental procedures to simulate the distributions of the summary statistics that are usually generated in such studies. These statistics are found to vary greatly even under the "null hypothesis" that all dentists are using procedures at the same rates. The simulated dentist rates are compared to observed rates obtained in a different study. These findings illustrate problems that can occur in small area analysis studies, and emphasize the importance of using statistical techniques that are appropriate for the data that are to be analyzed. Investigators should make every effort to obtain patient-level data, or at least to understand the underlying distribution of the number of procedures per patient, to avoid mistaking significant deviations from an incorrect model as evidence for significant variation among small areas.
所有小区域分析都需要将观察到的发病率变异性与仅由随机因素预期的变异性进行比较，但通常不知道预期的变异性。本文使用五种牙科手术的患者层面数据来模拟此类研究中通常生成的汇总统计数据的分布。即使在“零假设”（即所有牙医使用手术的比率相同）下，这些统计数据也被发现有很大差异。将模拟的牙医比率与另一项研究中获得的观察比率进行比较。这些发现说明了小区域分析研究中可能出现的问题，并强调了使用适合待分析数据的统计技术的重要性。研究人员应尽一切努力获取患者层面的数据，或者至少了解每位患者手术数量的潜在分布，以避免将与错误模型的显著偏差误认为是小区域之间显著差异的证据。
