Ecologic Inference Problems in the Analysis of Surveillance Data

Publisher Website
Google Scholar
Abstract
Surveillance data (e.g., vital statistics and registry data) are often collected at great expense and therefore should be exploited fully. They are useful in preliminary searches for disease risk factors, and they often describe outcomes across a much broader spectrum of exposures than is found in typical cohort or case-control studies. However, surveillance data on risk factors and outcomes are usually unlinked on the individual level and thus allow only ecologic (group-level) associations to be examined. This chapter reviews the theory needed to understand the limitations of surveillance data. It emphasizes the nonseparability of contextual (group-level) effects from individual-level effects that is due to the multilevel causal structure of ecologic (aggregate) data. Contrary to common misperceptions, this problem afflicts ecologic studies in which the sole objective is to estimate contextual effects, as well as studies in which the objective is to estimate individual effects. Multilevel effects also severely complicate causal interpretations of model coefficients.

This publication has 0 references indexed in Scilit: