Of particular import to this study, is collider bias originating from stratification on retreatment variables forming an embedded M or bowtie structural design. That is, rather than assume an M structural design which suggests that "X" is a collider but not a confounder, the authors adopt what they consider to be a more reasonable position and that is "X" is both a collider and confounder. Accordingly, in this study they examined the extent to which confounder induced bias exceeds collider induced bias. To inform this tradeoff, they quantified the bias from two simple linear model estimators which are asymptotically equivalent to stratification and matching on these variables (alone or with the propensity score) (e.g. Pearl, 2009). More specifically, they examined this tradeoff by quantifying the net bias induced from adjusting for "X" versus the net bias from ignoring it. As a result, stratifying on "X" removes confounding bias but induces collider bias whereas ignoring "X" alleviates collider bias but invokes confounding bias. For that reason, this study quantified the threshold by which collider and confounder bias due to a hypothesized confounder (e.g. "X") is equal. The intention is to provide pragmatic guidance as to the consequences of and the decision to stratify on covariates hypothesized to be confounders and/or colliders. (Contains 7 figures.)