MMP.15:8 - Common Anti-Patterns and How to Avoid Them

Anti-pattern	Consequence	Repair
Change a predictor value and call the prediction an effect	Conditions on naturally selected cases rather than specifying an intervention	Define the intervention target and derive its relation to the available law
Adjust for every recorded variable	Can block part of the requested effect or open a collider path	Justify the selected variables for that target
Treat no adjustment set as nonidentifiability	Misses mediated or other identifying constructions	Seek a justified broader derivation or a real obstruction
Pool experimental, selected and target records as one population	Changes the probability law behind the expression	Retain source conditions and derive the transport or selection correction
Fill unsupported cells by regression and claim data-only identification	Hides extrapolation assumptions	State the additional response restriction or return a bound
Treat a prior, fitted mechanism or simulation as fresh causal evidence	Conceals disagreement among compatible causal models	Show which assumption selected the answer and what the records distinguish