Effects of attrition on relationships between variables in surveys of older adults

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.