C30. Significance is not adequacy

Topic: Paired comparisons, and how far a fit is from the data · more questions on this topic
‹ C29

Does a significant predictor establish that a GLM fits adequately? Name one separate diagnostic check.

Worked answer (click to show)

Definitions/notation. A coefficient significance test addresses a specified null hypothesis about that coefficient within the assumed model; adequacy concerns whether the model describes the data sufficiently well. Significance does not establish adequacy: for example, inspect Pearson residuals (y_i-\hat\mu_i)/\sqrt{\hat\phi V(\hat\mu_i)} against fitted values for unexplained patterns or changing spread.