MMP.13:8 - Common Anti-Patterns and How to Avoid Them

Failure	Repair
A confidence interval is read as posterior probability for the observed interval.	State which object varies under the probability law and which claim the construction supports.
A likelihood curve or penalized optimum is treated as a posterior distribution.	Supply and normalize the prior-likelihood construction before using posterior probabilities.
A derived result is computed only from parameter means or separate standard errors.	Transform the joint inference; retain nonlinear effects and covariance.
More readings are assumed to remove a shared uncertainty.	Derive the variance under the actual dependence, as in :5.2.
Smaller Monte Carlo error is reported as less subject uncertainty.	Qualify the computed summary separately from the distribution it summarizes.
The prior or fitted model is changed without recalculating the receiving conclusion.	Carry the revised assumption through the target and uncertainty calculation.