消費者調查缺失資料修補所形成之分配形變

Distributional Distortion from Missing Data Replacement in Consumer Survey

缺失資料會造成統計推估及解釋結果時的偏差。本研究以六種不同方式測試修補缺失資料時偏態及峰態的改變，結果顯示當以完整資料的平均數或中位數修補缺失資料時，資料的分配會因此而產生形變。Two-sample Kolmogorov-Smirnov tests檢定結果顯示在修補缺失資料時，以逐步迴歸（ Stepwise regression）、複迴歸（ Multiple regression）、及集群分析法（Clustering method）的期望值進行資料修補時，相對較可維持資料分配的特性。 Missing data can cause biasedness in estimation and interpretations of analytical results. This study utilizes consumer survey data to examine skewness and kurtosis distortion of missing data replacements using six different approaches to provide further insights into the features of incomplete observation restorations in consumer survey data. Results of this study indicate that replacing incomplete observations with mean or median values of complete observations can distort the distributional properties of the variables. Based on the results of two-sample Kolmogorov-Smirnov tests, replacing missing data with expected values of stepwise regressions, of multiple regressions, and of cluster means are considered preferred procedures.

Missing data can cause biasedness in estimation and interpretations of analytical results. This study utilizes consumer survey data to examine skewness and kurtosis distortion of missing data replacements using six different approaches to provide further insights into the features of incomplete observation restorations in consumer survey data. Results of this study indicate that replacing incomplete observations with mean or median values of complete observations can distort the distributional properties of the variables. Based on the results of two-sample Kolmogorov-Smirnov tests, replacing missing data with expected values of stepwise regressions, of multiple regressions, and of cluster means are considered preferred procedures.