2.3 Challenge 3: Recognizing the limitations of survey data
Survey research and self-reported health are common elements of human health and aging studies addressing the impact of social inequalities (Black et al., 2017). Surveys have proven to be a reliable general tool for quickly producing empirical data from large representative samples, but these data also have significant disadvantages. For example, response rates are hard to control, and the data produced can lack many important details about the topic under investigation (Kelley et al., 2003). When used longitudinally (e.g., cohort studies), survey research suffers from other limitations due to the potential for significant changes in individual’s responsiveness over time, as well as in the consistency of respondents in their replies to public health-relevant questions when asked again at a later time (ethnic origin: Johnson 1974; patterns of abuse: Abramsky et al., 2022; Loxton et al., 2019; smoking: Kaestle, 2015; substance use: Broman et al., 2022; suicidal attempts: Hart et al., 2013). In general, there may be systematic differences in the ability of individuals to self-evaluate their own health across time (Black et al., 2017; Vuolo et al., 2014). Associations between self-reported health and objective health, as measured by biological risk factors, may also differ across socially stratified groups (Dowd and Zajacova, 2010; Layes et al., 2012). Conversely, interviewer effects also contribute significantly to biases in survey research, as respondents may be predisposed to provide socially desirable responses to the interviewer (Davis et al., 2010; Salazar, 1990). Interviewers also contribute to misclassification of race and ethnic groups (Massey 1980; Williams and Chiquita, 1995); biased assumptions of health among old persons (Thorslund and Wärneryd, 1990); biases in age, race, and gender perceptions (as reviewed by Davis et al., 2010); and can be affected by their level of experience interviewing people (Salazar, 1990). Taken together, these data limitations may contribute to the census miscount of socially stratified groups that ultimately affects reported rates of health conditions (Williams and Chiquita, 1995). Thus, both survey and interviewer data can result in biases in our ability to evaluate the accuracy of models quantifying the effects of the social environment on human health and consequent life courses.