Skip to content

[R-meta] Can we call moderators in subgroup analysis predictors?

7 messages · Michael Dewey, Dylan Johnson, Wolfgang Viechtbauer +1 more

#
Hello All,

A quick question. In classic meta-analysis, moderators help to form groups
of studies and conduct meta analysis separately in each group.

In meta-regression, moderators are directly predictors of effect size in a
regression analysis.

But, can we still call moderators in classic meta-analysis predictors of
effect size?

Thank you,
Simon
#
For clarification, I'm only concerned with categorical moderators.

Thanks again,
Simon
On Fri, Dec 18, 2020 at 3:33 PM Simon Harmel <sim.harmel at gmail.com> wrote:

            

  
  
#
Dear Simon

My immediate thought is, "Yes, why not?" but do you have some deep doubt 
you want to share with us?

Michael
On 18/12/2020 22:51, Simon Harmel wrote:

  
    
  
#
Thank you Michael. I had the same reaction. So, this was from a reviewer
who mentioned:

"You say moderators will eventually serve as predictors in a meta-analysis,
are you referring to meta-regression here? Moderators are not used as
predictors in a classic meta-analysis which simply pools all the effects
into a weighted effect size."

 My response was:

". . . In both cases, categorical moderators/predictors act as on-and-off
switches. I always prefer the meta-regression as it reveals the partial
effect of one moderator controlling for other moderators. But again, in my
context it is fine to generally use the term predictor. I think some may
even find the term moderator confusing because in general it may denote
some kind of interactional effect".

Just wanted to make sure I'm not too far off.

Thanks,
Simon

On Sat, Dec 19, 2020 at 4:32 AM Michael Dewey <lists at dewey.myzen.co.uk>
wrote:

  
  
#
Hello Simon,

The reviewer in this instance is correct. You are pooling the estimate on the grounds of an exposure -> outcome relationship. If you are include sex, for example, the moderator tells you how the exposure and outcome relation changes by sex. This doesn't tell you whether sex itself is a predictor of the outcome.

Does that make sense?

Best,
Dylan
On Dec. 19, 2020 6:39 p.m., Simon Harmel <sim.harmel at gmail.com> wrote:
EXTERNAL EMAIL:

Thank you Michael. I had the same reaction. So, this was from a reviewer
who mentioned:

"You say moderators will eventually serve as predictors in a meta-analysis,
are you referring to meta-regression here? Moderators are not used as
predictors in a classic meta-analysis which simply pools all the effects
into a weighted effect size."

 My response was:

". . . In both cases, categorical moderators/predictors act as on-and-off
switches. I always prefer the meta-regression as it reveals the partial
effect of one moderator controlling for other moderators. But again, in my
context it is fine to generally use the term predictor. I think some may
even find the term moderator confusing because in general it may denote
some kind of interactional effect".

Just wanted to make sure I'm not too far off.

Thanks,
Simon

On Sat, Dec 19, 2020 at 4:32 AM Michael Dewey <lists at dewey.myzen.co.uk>
wrote:
_______________________________________________
R-sig-meta-analysis mailing list
R-sig-meta-analysis at r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-meta-analysis
#
I think the reviewer uses the term "meta-analysis" exclusively when pooling a bunch of effects into a single weighted average. So, to him/her, it does not make sense to say that there are predictors/moderators/covariates (whatever you want to call them) in a meta-analysis. If there are predictors in the model, then the reviewer wants this to be called "meta-regression" and not "meta-analysis with predictors". I don't think the comment has anything to do with whether you use the term predictor or moderator.

Best,
Wolfgang
#
Hi Wolfgang,

Thank you very much. To be clear, you're approving of my response, correct?

Best,
Simon


On Sun, Dec 20, 2020 at 6:05 AM Viechtbauer, Wolfgang (SP) <
wolfgang.viechtbauer at maastrichtuniversity.nl> wrote: