於2004與2005年進行的「台灣基因體意向面訪調查」應用了歐美發展的一套測量一般民衆的基因科技知識水準的題組。此題組在兩年問卷中最主要的差異是題目次序與選項設計。這差異可能造成基因科技知識水準的測量在跨年調查之間呈現信度的不穩定，以至能無法比較兩年民衆的知識水準。本研究首先比較兩年基因科技的知識題組的回答分佈，檢視是否存在著題目的次序效應。Meta analysis結果顯示了次序效應中存在者提示效應與緩衝提示效應，年齡與次序效應有關，但教育程度與次序效應的關連性則不如理論預期。其次，以潛藏類別模型(Latent Class Models)當中的LC-cluster方法，利用兩年均相同的知識題組，比較受訪者對兩種選項設計問卷的回答型態是否有明顯的差異，發現年齡與教育爲解釋回答型態差異的顯著變項。本研究結果可作爲降低測量誤差之實證研究參考，建議未來在知識題組的選項中設計「不確定」選項，以明確區辨受訪者的理解狀態。

Two waves of Taiwan Genomic Survey, collected in 2004 and 2005, contained a battery of items related to knowledge of biotechnology. The main differences between two versions of knowledge questions are question order and response options Prior to comparing general knowledge level toward biotechnology between two time-points, it is in need to examine whether different question order and response options would cause unstable reliability in the knowledge scale across surrey. In this article, we first examined order effects by comparing the distribution of response to each knowledge item. Meta analysis revealed a significant carry-over effect and the buffer of carry-over effect on few items in the scale. Such kinds of question-order effects were associated with age group. The effect of education on question-order effect was, however, not consistent with theoretical hypothesis. We then examined whether there is significant difference in response style between the two versions of response options by using Latent Class Models-LC-cluster. Age and education were two variables significantly differentiate respnse style. The present findings indicate the importance of reducing measurement errors and suggest that "not sure" response in knowledge scale is to distinguish the respondent's knowledge status.