Causal diagrams and the cross-sectional study

Abstract

Causal diagrams and the cross-sectional study Eyal Shahar,1 Doron J Shahar2 1Division of Epidemiology and Biostatistics, Mel and Enid Zuckerman College of Public Health, 2Department of Mathematics, College of Science, University of Arizona, Tuscon, AZ, USA The cross-sectional study design is sometimes avoided by researchers or considered an undesired methodology. Possible reasons include incomplete understanding of the research design, fear of bias, and uncertainty about the measure of association. Using causal diagrams and certain premises, we compared a hypothetical cross-sectional study of the effect of a fertility drug on pregnancy with a hypothetical cohort study. A side-by-side analysis showed that both designs call for a tradeoff between information bias and variance and that neither offers immunity to sampling colliding bias (selection bias). Confounding bias does not discriminate between the two designs either. Uncertainty about the order of causation (ambiguous temporality) depends on the nature of the postulated cause and the measurement method. We conclude that a cross-sectional study is not inherently inferior to a cohort study. Rather than devaluing the cross-sectional design, threats of bias should be evaluated in the context of a concrete study, the causal question at hand, and a theoretical causal structure. Keywords: cross-sectional study, causal diagrams, colliding bias, information bias