Is it necessary to exclude all nonsignificant parameters to choose the best model?

Question

I'm running LMM models and could I ask if I can just report the model after comparing random intercept with random intercept and slope model without excluding nonsignificant factor?

score 5 · Accepted Answer · answered Mar 21 '14 at 16:45

5

On the contrary, it is necessary to include all the insignificant predictors to get the parameter estimates and inferences "right". This has been discussed at length on this site. There are many issues but to mention one briefly, discarding "insignificant" variables greatly biases standard errors towards zero.

answered Mar 21 '14 at 16:45

Frank Harrell

91,879
6
178
397

Ok so I will report all parameters and won't exclude them at the final model. Thank you Frank Harrell. – user3288202 Mar 21 '14 at 20:39

Is it necessary to exclude all nonsignificant parameters to choose the best model?

1 Answers1

Linked