Sugisawa H, Kishino H, Sugihara Y, Shibata H
Department of Health Sociology, Tokyo Metropolitan Institute of Gerontology.
Nihon Koshu Eisei Zasshi. 2000 Apr;47(4):337-49.
The attrition of respondents in panel studies of the elderly can create bias in data analysis. The purposes of this study are two fold; 1) to examine characteristics of dropouts, from a particular panel lost except due to natural attrition (by death) by comparison with continuing participants in that wave, and 2) to assess representativeness of those actually continuing in a particular panel by comparison with those eligible for inclusion in that wave. Only those who died were excluded from the group of respondents at the baseline survey because they constituted natural attrition in this longitudinal survey.
At baseline (1987), 2,200 individuals age 60+ from 3,288 national representative sample were interviewed. Non-response status to three contacts (1990, 1993, 1996) with the panel was examined in relation to variables included in the baseline interview. A number of characteristics of demographic background, health, life-style, and social relations obtained in the baseline survey (1987) were compared between those re-interviewed in a particular panel and subjects lost through unnatural attrition until that wave. To study the influence of unnatural attrition on variable distributions and each related factors of two health indicators (self-rated health and depressive symptoms), baseline responses were compared between those re-interviewed in a particular panel and all who were eligible to respond in that wave.
Dropouts in longitudinal research were found to appear nonrandomly. While distributions of age and health indicators in those re-interviewed were influenced by respondent attrition, related factors of health indicators may be free of bias that can be created by it.
在针对老年人的面板研究中，受访者的损耗可能会在数据分析中产生偏差。本研究有两个目的：1）通过与该轮次中继续参与的受访者进行比较，考察特定面板中除自然损耗（因死亡）外的退出者的特征；2）通过与该轮次中有资格纳入的受访者进行比较，评估特定面板中实际继续参与的受访者的代表性。在基线调查中，只有那些死亡的受访者被排除在受访者群体之外，因为他们构成了这项纵向调查中的自然损耗。
在基线（1987年）时，对来自3288个全国代表性样本的2200名60岁及以上的个体进行了访谈。考察了与该面板的三次联系（1990年、1993年、1996年）中的无回应状态与基线访谈中包含的变量之间的关系。比较了在特定面板中接受重新访谈的受访者与在该轮次中因非自然损耗而流失的受试者在基线调查（1987年）中获得的一些人口背景、健康、生活方式和社会关系特征。为了研究非自然损耗对两个健康指标（自评健康和抑郁症状）的变量分布及各相关因素的影响，比较了在特定面板中接受重新访谈的受访者与该轮次中所有有资格回应的受访者的基线回答。
1）各轮次中流失的退出者比该轮次中留下来的人年龄显著更大，社会参与水平更低。在第三轮或后续轮次之前，流失的退出者与继续参与的受访者在健康和生活方式变量上存在显著差异。2）特定面板中继续参与的受访者可能比该轮次中有资格回应的受访者更年轻，身体、心理或社会方面更健康。在特定面板中接受重新访谈的受访者与该轮次中有资格回应的受访者在两个健康指标的各相关因素上几乎相同。
纵向研究中的退出者呈现出非随机出现的情况。虽然接受重新访谈者的年龄和健康指标分布受到受访者损耗的影响，但健康指标的相关因素可能不受其可能产生的偏差影响。
