Cross-linked survey analysis is an approach for separating cause and effect in survey research

Abstract

ObjectivesWe developed a new research approach, called cross-linked survey analysis, to explore how an acute exposure might lead to changes in survey responses. The goal was to identify associations between exposures and outcomes while reducing some ambiguities related to interpreting cause and effect in survey responses from a population-based community questionnaire. Study Design and SettingCross-linked survey analysis differs from a cross-sectional, longitudinal, and panel survey analysis by individualizing the timeline to the unique history of each respondent. Cross-linked survey analysis, unlike a repeated-measures self-matching design, does not track changes in a repeated survey question given to the same respondent at multiple time points. ResultsPilot data from three analyses (n = 1,177 respondents) illustrate how a cross-linked survey analysis can control for population shifts, temporal trends, and reverse causality. Accompanying graphs provide an intuitive display to readers, summarize results, and show differences in response distributions. Population-based individual-level linkages also reduce selection bias and increase statistical power compared with a single-center cross-sectional survey. Cross-linked survey analysis has limitations related to unmeasured confounding, pragmatics, survivor bias, statistical models, and the underlying artifacts in survey responses. ConclusionWe suggest that a cross-linked survey analysis may help in epidemiology science using survey data.