Norris F H
Urban Studies Center, University of Louisville.
J Gerontol. 1987 Nov;42(6):597-605. doi: 10.1093/geronj/42.6.597.
Using data from a five-wave panel study of older adults, this study examined the potential for attrition to affect relationships among study variables. There were few differences between those respondents who completed all five waves and those who did not in either their initial matrices of correlations or in the equations that best predicted their Wave 2 health and depression. Relationships among variables also did not differ between respondents and persistent nonrespondents, between respondents and four specific attrition types, or between the original sample and various multiwave samples. In discussing the findings, the limited effects that attrition has on relationships between variables are contrasted to the stronger effects that attrition appears to have on mean levels of those variables, showing that the former does not necessarily follow from the latter. The findings support recent conclusions in the literature that the threat posed by panel loss is not as great as typically assumed.
本研究利用一项针对老年人的五波面板研究数据，考察了样本流失对研究变量之间关系的潜在影响。在最初的相关矩阵或能最佳预测其第二波健康和抑郁状况的方程方面，完成了所有五波调查的受访者与未完成的受访者之间几乎没有差异。变量之间的关系在受访者与持续无回应者之间、受访者与四种特定样本流失类型之间，或原始样本与各种多波样本之间也没有差异。在讨论研究结果时，样本流失对变量间关系的有限影响与样本流失似乎对这些变量平均水平的更强影响形成对比，表明前者不一定由后者推导而来。这些发现支持了文献中最近的结论，即面板数据缺失所带来的威胁并不像通常认为的那么大。
