主計處自1978年起按月辦理的「人力資源調查」，是採樣本輪換的方式進行調查。就相鄰兩年同一月份的人力資源調查而言，約有半數樣本是重複的。而在同年相鄰月份的人力資源調查中，亦有近半數樣本是重複的。因此，可就同年相鄰月份的資料進行合併，亦可就相鄰年份相同月份的資料進行合併，形成短期的追蹤調查資料。

雖然人力資源調查的跨時合併資料已廣被國內學界使用，但迄今仍未見任何研究就該合併資料的特性、限制提出討論。在這篇論文中，以1979-2001年5月份人力資源調查資料為資料來源，就相鄰兩年的資料合併，建構相鄰兩年的合併資料，以分析樣本流失與否的決定因素。logit模型的分析結果顯示，樣本的性別、年齡、教育程度、家戶大小、為戶長與否、由本人回答問卷與否、從業身份、居住地區、受訪年份等因素，均對樣本流失機率有顯著的影響效果。

The Manpower Utilization Survey (MUS) conducted by Directorate General of Budget Accounting and Statistics (DGBAS) since year 1978 is a rotating monthly survey. Because of this sampling frame, about a half of the sample in any two consecutive months could be matched. Further, in any given month, about a half of the sample in a given year could be matched with those in the following year. Thus, matching person records across MUS files could result in short panels.

Though the matched MUS files have been widely applied in empirical analysis, the limitation of the files has not been discussed yet. By utilizing the 1979-2001 March MUS files, this paper analyzes the determinants of attrition from the MUS. A simple logit model is used to model the dependence of match probabilities on a broad set of household and person attributes. The results show that most explanatory variables have significant effect on the match probabilities.