0
$\begingroup$

I am making a full factorial repeated measure test in SPSS, and the Test of Within-Subjects Effects/multivariate tests are all non-significant.

In this case would estimated marginal means pairwise comparison make sense since I would use a model that is not significant?

$\endgroup$
0

1 Answer 1

0
$\begingroup$

No. You should take at 'face value' the ANOVA result that the factor does not contain significant differences. Although not common, it could happen that an ad hoc "test" would "find" a bogus significant difference between two levels of a non-significant factor.

Depending on circumstances, this might be due to using a different measure of differences than in the main ANOVA or it might be a chance 'false discovery' due to multiple testing of the same data.

Properly done, you should not even look for such differences, but if you do look and think you have found something, you should ignore your mistake.

$\endgroup$
4
  • 1
    $\begingroup$ Ok basically nothing be said about the data since no factors are significant? What if I fit a submodel and something is significant. Would this be bad practice as the full factorial is not signf? $\endgroup$ Commented Nov 24, 2021 at 17:38
  • $\begingroup$ Not sure what you mean by fitting a submodel. If it is truly a sub-model, why was significance not found initially? Even if the main model is significant (say barely at 5%), all ad hoc tests should be use a method that protects against false discovery (perhaps requiring significance at 2% or 1% level). // If you do 20 independent ad hoc tests at 5%, chances are one will reject. $\endgroup$
    – BruceET
    Commented Nov 24, 2021 at 17:58
  • $\begingroup$ for example the model where nothing is significant has 12 parameters while the submodel has 10 of the 12 parameters but shows something is significant $\endgroup$ Commented Nov 24, 2021 at 18:19
  • $\begingroup$ On what basis did you choose those 10, in particular, out of 12. There are ${12\choose 10} = 66$ possible subsets of 10, why that particular one? $\endgroup$
    – BruceET
    Commented Nov 25, 2021 at 2:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.