Skip to content

[R-meta] Removing the intercept in rma.mv after non-significant QM/F test

4 messages · Michael Dewey, Wolfgang Viechtbauer, T D

T D
#
Dear all,

I have a question regarding the rma.mv function.

Some QM tests (test of moderation) are not significant. However, I continued to investigate both levels of the moderator regardless by removing the intercept (i.e with placing a -1 at the end)
How sensible is this and what kind of conclusions can I draw when both levels (after removing the intercept) produce significant results. I understand I can?t draw any overall conclusions for the a priori hypotheses, considering the test of moderation was not significant - but generally, I wonder, whether it is ?ok? to report this - or whether it is simply wrong to continue with any analysis if the QM test is not significant.

Many thanks for clarifying.

Kind regards,
Tina Dudenh?ffer
#
Dear Tina

When you leave the intercept in you are testing whether the two levels 
differ. When you take it out you are separately testing for each whether 
its coefficient differs from zero. Those are clearly not the same.

Michel
On 09/12/2021 12:32, T D wrote:

  
    
  
#
Indeed!

@Tina: Also see this write-up:

https://www.metafor-project.org/doku.php/tips:models_with_or_without_intercept

Best,
Wolfgang
T D
#
Hi Michael,

Yes, that?s clear - I was rather wondering how to report this but Wolfgang forwarded a link.

All clear now.

Many thanks,
Tina