評定式問項上極端反應的意涵:實例分析

羅國英

東吳大學社會工作學系專任教授

 

        人們在評定式問項上的極端反應可能有三種意涵:(1)問項所欲測的意見或態度;(2)反應效應;(3)答題者的答題習慣(或性格)。其中後兩者意涵通常被視為調查或測量上的偏差。在文獻中,反應效應比較常被注意,因此本文以極端反應中所反應的答題者性格為分析及討論的重點。所作分析以三個問題為核心:(1)極端反應數量是否反應一種穩定的個別差異或特質?(2)如果問題一的答案是肯定的,這種特質和個人的其他特質(人口學變項、某些態度變項及答題習慣等)有什麼關聯;(3)極端反應性格在什麼情況下和個別問項的反應比較有關?作者選擇了數套現成的調查資料進行分析,希望能涵蓋較廣泛的調查議題。研究結果相當支持「極端反應可反應答題性格」的觀點,同時也顯示這種答題性格可能影響某些調查研究的結果甚至結論。由於推論上仍有相當多的困難及限制,這些困難或限制也是本文討論的重點。


The Analysis of Extreme Responses on Rating Scales:

An Example

KUO-YING LO

Professor, Department of Social Work, Soochow University

 

    Extreme responses on rating scales may have reflected the impact of three important determinants of the responses: (1) the opinion or attitude the scale is aimed to measure; (2) the response effects caused by the context or item characteristics; and (3) the habit or personality trait in answering questions. The last two components are often treated as "response errors." The response effects have been discussed more frequently in the literature. In the present paper, therefore, the discussion of extreme response will place the emphasis more on its characteristic.

    The following three questions are focused on: First, can the number of extreme responses indicate a kind of consistent individual difference? Second, if the answer to the first question is positive , what other personal traits are exhibited by the subjects who make more extreme responses? Finally, whether the extreme trait has different impacts on item response under different circumstances? If this is true, when or under what conditions will the extreme trait be more related to item response?

    The author selected several data to do the empirical analysis. The criterion for selection was that the single data set included as many different survey issues as possible. The results support the viewpoint that there exists individual differences in making extreme responses. Besides, such a personal trait may also have important impacts on the result or conclusion of survey research. The methodological difficulties and the limitations for inference are also discussed.