A sensible strategy for avoiding this pitfall might entail the formulation of alternative IRT models which correspond to practically significant model violations, the selection of model tests which are most powerful against these models, and an estimation of the necessary sample size for testing against these models with sufficient power.
1
—
—