C.16.MR:8 - Common Anti-Patterns and How to Avoid Them

Failure	Effect on use	Repair
Read the display as the sought value after the procedure changes the subject	The answer can concern the measured arrangement instead of the target conditions.	Model the interaction and state which quantity is sought.
Copy a correction without its measurement relation	Its direction, magnitude or range can be wrong for the current arrangement.	Recover where the effect enters and derive its contribution.
Fill an unknown influence with zero	A numerical answer hides unresolved alternatives.	Keep the influence unknown and use available bounds or a justified distribution.
Treat expected response as an error-free observed proportion	The inferred population property loses sampling uncertainty.	Keep the response model and sampling inference distinct.
Repeat measurements to remove structural ambiguity	The same indication can still fit different sought values.	Change the relation or observation that distinguishes those values.
Model every conceivable influence before answering	The work expands without changing the result.	Test the consequence of the influence for the intended use and stop with an adequate result.