同居是一項重要的研究議題，但有關同居題項設計對受訪者回應行為的影響效果，研究卻極為稀少。這項研究採用家庭動態調查2014、2016年的面訪追蹤調查資料，探討同居題項變動前後受訪者應答的差異。另外，輔以調查蒐集到的周邊資料，本文也分析訪員特質、第三人在場與否對受訪者回應的影響效果。研究結果發現，在同居題項改以兩階段方式詢問後，受訪者回答同居的機率高出7倍左右。另外，訪員特質並不影響受訪者對同居題項的應答結果，顯示訪員效應的問題並不嚴重。第三人在場效應的結果則顯示，在場者的身份關係著同居的應答結果：如果受訪者的直系親屬在場，受訪者回答同居的機率顯著較沒有他人在場之下為低；非直系親屬在場的影響效果則相反。除學理上的貢獻外，本文的分析結果可回饋到調查實務，用以改善同居題項的設計方式。

In a society which lacks a registration system for cohabitation, whether an individual is cohabiting with someone relies on self-reported information. The quality of the self-reported information matters for estimates of proportions of cohabitation. Such information is also important for research in the fields of demography and family studies on the topics of, for example, reproductive behavior and gender equality. Cohabitation is often asked as one of the options for marital status, while some see it a type of living arrangement. In some societies, however, cohabitation is considered to be a sensitive issue because it is not a legal marital status, and thus respondents might under-report cohabitation when asked relevant questions. In Taiwan, self-reported survey data are the main source of information for habitation. This response option is often combined with being married due to its sensitivity, or viewed as a distinct option in the marital status question. It is important to understand whether the design of cohabitation question affects respondents’ self-reported behavior. Using panel data from the Taiwan Panel Study of Family Dynamics (PSFD), this study examines the effects of changes in the cohabitation question on the respondents’ answers. The PSFD is a longitudinal survey project initiated in 1999. Up till the survey conducted in 2014, the question on marital status contained seven answer options: unmarried, cohabiting, married, separated, divorced, widowed, and other. Starting with the 2016 PSFD survey, the seven-option marital status question has been revised into a two-question version. The question on marital status is asked first with six mutually exclusive options: unmarried, married, separated, divorced, widowed, and others. Respondents who select the options legally compatible with cohabitation (including “unmarried,” “divorced,” and “widowed”) are then asked whether they are cohabiting with someone. The difference in the probability of reporting cohabitation between the 2014 and 2016 in-person surveys is analyzed based on the sample who are legally compatible with the cohabitation status. For this analytical purpose, the sample of the 2014 survey is confined to the respondents who chose “unmarried,” “divorced,” “widowed,” or “cohabiting” from the question on marital status. As to the sample of the 2016 survey, it is limited to the respondents who selected “unmarried,” “divorced,” or “widowed” when answering the first item. Using the merged data, the random-effects logit model is applied to analyze the respondents’ likelihood of reporting cohabitation. In addition to analyzing the effects of changes in question design, we also explore whether interviewer characteristics and third-person presence matter for the respondents’ likelihood of answering cohabitation. Interviewer characteristics include the interviewers’ gender, age, and interviewing experience. The dummies on third-person presence contain three categories: presence of direct family member(s), presence of non-direct family member(s), and no third-person presence. The findings indicate that the probability of reporting cohabitation is about seven times higher in the 2016 survey with the revised question on marital status than that in the 2014 survey with the original design. With respect to interviewer effects, none of the interviewer variables is significant. The results on third-person presence effects reveal that the presence of direct family member(s) is associated with a lower probability of reporting cohabitation than that of no third-person presence. The presence of non-direct family member(s) shows an opposite effect. These findings suggest that who the third person is matters for the respondents’ self-reporting of cohabitation. This study not only contributes to our understanding of the effects of question design for cohabitation on the respondents’ answers, but also provides implications for improving the design of the cohabitation question.