Fig. 2From: Noncollapsibility and its role in quantifying confounding bias in logistic regressionTrue confounding bias (\({{\varvec{\beta}}}_{1}-{{\varvec{\beta}}}_{1}^{\boldsymbol{*}}\)) as a function of the confounder-outcome effect collapsed over all sample sizes. Panel A: each line represents a positive confounder-exposure effect. Panel B: each line represents a negative confounder-exposure effectBack to article page