Fig. 1From: Noncollapsibility and its role in quantifying confounding bias in logistic regressionDirected acyclic graphs of the four possible scenarios into which each simulated condition can be classified. Panel A: both confounding and noncollapsibility. Panel B: confounding without noncollapsibility. Panel C: noncollapsibility without confounding. Panel D: neither confounding nor noncollapsibility. C represents three continuous covariates, X represents the dichotomous exposure and Y represents the dichotomous outcome. The dotted line in panel D between the covariates and the exposure and between the exposure and the outcome indicate there may or may not be an associationBack to article page