在知識量表中，知識分數通常以回答正確的總次數表示，且選答『不知道』均被視爲回答不正確。在教育測驗或政治知識測量上對於是否鼓勵『不知道』回答雖有不同的理論觀點，其中之一是鼓勵『不知道』會引起猜題順答行爲。本文以基因科學知識題組爲例承襲前人研究，除了分析猜題行爲也探討猜題行爲對知識分數的影響，且此影響程度是否因考慮受訪者認知相關特徵與訪員經驗特徵後有所改變。分析資料來自2003年執行的『基因意向面訪調查』。此調查採分層三唱段等機率法抽取18到64歲的台灣民衆，最後完訪人數爲1090。結果發現，台灣民衆在包含10題的基因科學知識題組中，不論樣本是否加權，回答正確的平均題數約在6.15左右（標準差在2.10左右）；相較於其他國家民衆的基因知識水準是不相上下。利用兩階層線性迴歸模式(HLM)進行單獨猜題效應以及分到包含受訪者與訪員效應的分析結果顯示，不控制其他因素時，完全不猜題者的知識分數明顯的比完全猜題者低，但也部分猜題者高。控制受訪者與訪員層面因素後，無論是部分猜題者或完全猜題者的分數均較完全不猜題者低，尤其是部分猜題與完全不猜題者的比較有更明顯的差異。儘管如此，猜題順答性誤差因考慮其他影響因素而消失，但猜題率介於0.01到0.99之間的這群人到底是否猜題率越高知識分數越高則有待驗證。調查設計屬不鼓勵但接受『不知道』回答，也許這是造成本研究發現猜題順答性誤差降低的主因。本研究建議未來繼續更嚴密且深入的實驗研究。

Knowledge level has been conventionally measured by the number of correct answers, where "don't know (DK)" response was considered as incorrect answer. Different perspectives of educational measurement and political science have not reached agreement on the encouragement of the DK answers. Some of the previous studies have suggested that the encouragement of DK answers led to guessing response sets. This paper explores guessing response sets and their effect on the respondents' knowledge levels by using genetics knowledge scale as an example. In particular, the changes of guessing effect were examined by considering the characteristics of respondent and interviewer. Data were drawn from Taiwan Genomic Survey collected in 2003. A group of adults aged 18-64 were selected using three-stage stratified sampling scheme. A total of 1090 interviews were completed. The results showed that regardless of sample weighting, the average number of correct answers among the ten-item genetics knowledge scale was 6.15 (with a standard deviation of 2.10), which was similar to those found in the United States, Canada, Japan, and European countries. Without controlling for other factors, the knowledge scores of the respondents who did not guess at all were significantly lower than their counterparts who fully guessed, but were significantly higher than those who partially guessed. After controlling for the characteristics of respondent and interviewer, the knowledge scores of the respondents who partially guessed and fully guessed were lower than those who did not guess. Guessing effect turned insignificant after controlling for other factors. For the respondents with a guessing rate between 0.01 and 0.99, whether their genetics knowledge increases with guessing rate requires further investigation. One of the reasons why guessing effect turned insignificant may be the discouragement of DK in the original survey design. It remains suggested to encurage more rigid experimental studies in the future.