Mode Effects on Response Design: An Example of Qualify-of-Life Measures
過去研究指出數量多的選項設計在面對面訪談中較為可行，然而對於選項標示語或不平衡式量尺設計則尚無深入討論。本研究以態度量表之選項設計為主體，來分析電話訪問與面對面訪談中，不同的選項設計之結果是否有差異。資料來源包含以態度量表進行隨機分派實驗設計之2010 年及2011 年電訪資料，以及2010 年社會變遷基本調查之面訪資料。
在樣本結構部分，面訪與電訪的次樣本進行內部比較時均無差異；然而比較相同選項設計的面訪與電訪資料時，則發現個人月收入及工作狀況有明顯差別。以IRT 之等級反應模型的分析發現，不同設計在面訪的表現均較電訪要佳；若整體考量題目與選項設計結合後的表現時，則以四點平衡量尺的鑑別力最佳。此外，標示語的設計效果會因資料蒐集方法的不同而有所差異，強語氣的標示語使面訪受訪者往非極端選項靠攏的情形比電訪更為明顯。進一步以人口變項對「個人生活品質」之IRT 分數進行迴歸分析時，亦可看出方法效應。
Questionnaire design often differs among various survey modes in order to obtain better data quality. Previous studies have indicated that a higher number of response categories may perform better in face-to-face surveys than in telephone surveys. Few have examined the influence of response labels and unbalanced scale design on survey answers. This study aims to compare the responses between face-to-face and telephone surveys with respect to response design of attitudinal scales. Data used for this study are drawn from the 2010 Taiwan Social Change Survey and telephone surveys collected in 2010 and 2011 using a randomized experimental design to collect data from representative samples aged 18 years or older. By measuring the construct of individual quality of life (QoL), four types of response scales with different combinations of response number and response labels were examined to distinguish both the intensity and direction of responses.
Individual characteristics between the two modes indicated that the sub-samples were different in terms of monthly income and working status. Results of the graded response model suggested that the response design used in this study performed better in face-to-face surveys than in telephone surveys. Also, a balanced scale with four response categories and less intense labels had the best discriminating ability among the four designs. When compared to respondents to telephone surveys, respondents to face-to-face surveys were less likely to provide strongly-labeled answers, which can be seen a tendency of social desirability due to more interpersonal interactions. Findings of linear regression on the IRT scores of QoL, however, were in favor of telephone surveys, with higher values of chi-square. An intertwining effect of response design and data collection mode was also found. A discussion of the results is provided.