Missing data imputation in quality-of-life assessment - Imputation for WHOQOL-BREF
Missing data imputation in quality-of-life assessment - Imputation for WHOQOL-BREF
复制标题
DOI：
10.2165/00019053-200624090-00008
复制
发表时间：
2006-01-01
影响因子：
4.4
通讯作者：
Lin, Ting Hsiang
中科院分区：
文献类型：
--
作者：
Lin, Ting Hsiang
引言：本研究调查了WHO生活质量简明问卷（WHOQOL-BREF）中缺失数据的插补效果。方法：采用实证分析和模拟研究的方法，考察缺失数据率和填补项目数对填补精度的影响。在实证分析中，进行了项目级和领域级插补，并使用不同数量的数据对缺失值进行插补。在模拟研究中，随机抽取2%、5%和10%的数据集，并用缺失值替换。为每种情况生成了20个数据集。结果：在实证研究中，用于插补的项目数对插补的准确性只有很小的影响。此外，在模拟研究中，插补的准确率并没有显着变化的缺失数据的比例增加。然而，计算中使用的项目数在一定程度上确实造成了估算的缺失值。极端的反应有最差的计算和最低的准确率。结论：建议尽可能多的项目包括在同一域内的插补。然而，使用不同领域的项目进行估算并不是特别有帮助。研究人员在解释极端反应的估算值时应格外谨慎。
Introduction: This study investigated the effects of imputing missing data in the WHO Quality of Life Abbreviated Questionnaire (WHOQOL-BREF). The imputation results from both the item and domain levels were compared and the impact of the missing data rate and the number of items included for imputation were examined.Methods: An empirical analysis and a simulation study were used to examine the effects of missing data rates and the number of items used for imputation on the accuracy for imputation. In the empirical analysis, both item-level and domain-level imputations were performed, and the missing values were imputed using different amounts of data. In the simulation study, sets of 2%, 5% and 10% of the data were drawn randomly and replaced with missing values. Twenty datasets were generated for each situation. The data were imputed and the accuracy of the imputation was reported.Results: In the empirical study, the number of items used for imputation had only a small impact on the accuracy of imputation. Furthermore, in the simulation study, the accuracy rates of imputation did not significantly change as the proportions of missing data increased. However, the number of items used in the computation did contribute to some extent to the missing values imputed. Extreme responses had the worst computations and the lowest accuracy rates.Conclusion: It is recommended that as many items as possible be included for imputation within the same domain. However, it is not particularly helpful to use items from different domains for imputation. Researchers should exercise extra caution in interpreting the imputed values of extreme responses.