在複雜抽樣調查資料廣為應用的今日，研究者仍舊慣常地忽略抽樣設計的特性對統計推論的影響，包括分層、分群及加權等因素；而此將導致偏差的標準誤估計及提高型I誤差發生的機率。本文企圖彰顯不同的抽樣設計在變異數估計上的差異，以及此差異對信賴區間與檢定統計量等相關估計及推論上的影響；同時，也介紹SAS及STATA統計軟體在複雜抽樣調查資料上的應用。

Since the analysis of complex sample survey data is widespread nowadays, the effect of sample design features on statistical inference such as stratification, clustering, and weighting should be seriously considered accordingly. However, most analysts routinely ignore this issue, resulting in biased estimates of standard errors, as well as increased possibility of Type I errors. This paper attempts to illustrate the impact of sample designs on variance estimation, and thereby confidence intervals and test statistics. Also, appropriate software applications are briefly introduced.

複雜抽樣調查；自由度；設計變項；設計效果；變異數估計

complex sample survey； degree of freedom； design feature； design effect； variance estimation