Abstract
Mediation analysis is a newer statistical tool that is becoming more prominent in nutrition research. Its use provides insight into the relationship among variables in a potential causal chain. For intervention studies, it can define the influence of different programmatic components and, in doing so, allows investigators to identify and refine a program's critical aspects. We present an overview of mediation analysis, compare mediators with other variables (confounders, moderators, and covariates), and illustrate how mediation analysis permits interpretation of the change process. A framework is outlined for the critical appraisal of articles purporting to use mediation analysis. The framework's utility is demonstrated by searching the nutrition literature and identifying articles citing mediation cross referenced with the terms "nutrition," "diet," "food," and "obesity." Seventy-two articles were identified that involved human subjects and behavior outcomes, and almost half mentioned mediation without tests to define its presence. Tabulation of the 40 articles appropriately assessing mediation demonstrates an increase in these techniques' appearance and the breadth of nutrition topics addressed. Mediation analysis is an important new statistical tool. Familiarity with its methodology and a framework for assessing articles will allow readers to critically appraise the literature and make informed independent evaluations of works using these techniques.