『固定樣本連續訪談法』（panel studies）樣本流失問題之探討

Abstract

在固定樣本連續訪談法（panel studies）中，受訪者非隨機性的流失（mortality），對研究的效度會產生負面的影響。本文作者利用美國全國選舉研究（American National Election Studies；ANES）在西元1990到1992年進行的固定樣本連續訪問的資料進行分析，發現了接受續訪的樣本（panel sample）與流失的樣本（mortality sample），在政治興趣、政治知識以及政治參與上有著顯著的差異。因此，當我們的研究目的是在推論母體在政治興趣、政治知識以及政治參與的分佈時，「樣本流失」的問題會對研究的「外在效度」（external validity）產生顯著的影響。當影響受訪者是否參與後續訪問的因素，與我們研究模型中的變數相互關聯時， 「樣本流失」也會對研究的「內在效度」（internal validity）造成傷害。本文作者檢視「樣本流失」對投票參與模型中，解釋變數的參數估計的影響時發現：「樣本流失」使我們低估了受訪者年齡以及對報上選舉消息注意程度這兩個變數，對其投票與否的影響力。當我們要解釋國會選舉中現任者的優勢時，「樣本流失」使我們高估了民主黨議員現任者優勢的重要性，也低估了共和黨現任議員的重要性。由於受訪者對調查研究的興趣以及態度，會影響他們參與訪談的意願，因此，作者認為，調查研究工作者應該努力讓受訪者覺得接受訪問是一個愉快經驗，減少調查研究的拒訪率，使研究結果更為精確。

Mortality is a problem when it exits differential loss of respondents during our panel studies. By examining the 1990 - 1992 NES panel studies, I demonstrate that there were significant differences between the panel sample and the mortality sample on their political interest, political knowledge, and political participation. Mortality undermines the external validity when our research interest is to describe the political interest, political knowledge, and political participation of the mass public.When the selection process correlates with some variables in our model, mortality also undermines the internal validity of our research. I demonstrate that we will underestimate the importance of the respondent`s age and his /her interest on newspaper on explaining his /her voting participation when we have the mortality problem in our panel studies. When we want to explain the importance of incumbent advantage, mortality makes us overestimate the importance of Democrat`s incumbency and underestimate the importance of Republican incumbency.Participating in panel studies is a function of the respondent`s interest and attitudes toward surveys. We shall always keep this in mind since the respondent`s cooperation is the key point for the success of survey research. As the nonresponse rate increases in most surveys, social scientists have the obligation to make our interview more pleasant for our respondents.