葛特曼量表之拒答處理:簡易、多重與最鄰近插補法的比較
Treatments of Guttman-type Scale Refusals: Comparisons among Simple, Multiple and Nearest Neighbor Imputation Methods
作者:廖培珊(Pei-shan LIAO)、江振東(Jeng-tung CHIANG、林定香(Ting-hsiang LIN)、李隆安(Lung-an LI)、翁宏明(Hung-ming WONG)、左宗光 | 首次發表於 2020-06-03 | 第 47 期 September 2011
DOI:https://dx.doi.org/10.6786/TJS.201109_(47).0004
研究紀要(Research Notes)
論文資訊 | Article information
摘要 Abstract
當社會科學研究者依理論或經驗,將調查訪問資料的拒答或不知道等答案歸到既有的回答類別中,以減少資料流失造成的估計偏誤時,這些插補方式的適當性通常缺乏進一步的檢證或統計模型予以支持,因此本研究以葛特曼量表中之「拒答」為例,比較不同的插補方式,包括簡易插補、多重插補以及最鄰近插補法,並考慮不同拒答率的條件下,調查資料品質的改善程度有無差異。本文對大型學術調查資料中,四道態度題目的「拒答」進行插補,將其中符合葛特曼量表的資料視為「黃金標準」,探討各類插補法的正確率表現。
研究結果發現,簡易插補法的正確率可以公式推導而得,且以實證資料為例,各類簡易插補法的正確率皆約三成左右。各種插補法的正確率表現,以最鄰近插補法最佳,其次為多重插補法的模式一。然而若考量效率,當研究者受限於沒有太多的共變項或有預測力較強的輔助變項來進行插補,而且資料符合葛特曼量表之特性時,簡易插補法的表現未 必較複雜的插補法遜色。

關鍵詞:簡易插補、多重插補、最鄰近插補、葛特曼量表、拒答
It is a common practice to treat refusals as a missing value and exclude them from data analysis. To avoid biased results obtained from complete cases, imputation and reclassification of refusals into other response categories are frequently used. The appropriateness and effectiveness of different methods, however, remain unclear. This study attempts to compare results among different imputation methods using refusals in a Guttman-type scale as an example. The results indicate that formula for estimating accuracy of single imputation can be derived from the observed frequency of the response patterns that correspond to Guttman-scale types. In addition, refusal rates did not have much impact on the accuracy of simple imputation due to the fixed refusal patterns simulated from the gold standard. On the other hand, the nearest-neighbor method achieves the highest accuracy among the imputation methods examined. Discussions on the imputation results and imputation for further research are provided.

Keywords: Simple imputation, multiple imputations, nearest neighbor imputation, Guttman scale, refusal