[R-meta] How to Get PIs for Moderator Analyses

[Viechtbauer, Wolfgang (NP)]

Dear Tori,

The profile() function is not for getting PIs. You are misunderstanding what Michael was suggesting -- he was suggesting to use profile() to check if the variance components are identifiable by examining the profile likelihood plots. And predict() works just fine for getting PIs. Like I wrote, you have to use the 'newmods' argument in predict() to specify for which combination of moderator values you want to obtain a predicted effects (and the corresponding CI and PI).

Best,
Wolfgang

>-----Original Message-----
>From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces using r-project.org] On
>Behalf Of Tori Peña via R-sig-meta-analysis
>Sent: Thursday, 22 June, 2023 19:37
>To: R Special Interest Group for Meta-Analysis
>Cc: Tori Peña
>Subject: Re: [R-meta] How to Get PIs for Moderator Analyses
>
>Thanks, all!
>
>Unfortunately, the profile() function did not yield PIs for our moderator
>analyses.  Are there other ways to get the PIs for rma.mv models since
>predict() and profile() are not working?
>
>On Thu, Jun 15, 2023 at 12:18 PM Michael Dewey via R-sig-meta-analysis <
>r-sig-meta-analysis using r-project.org> wrote:
>
>> Dear Tori
>>
>> Just to clarify the reasoning behind my comments I was concerned that
>> your data may not be adequate to fit the model since you seemed to have
>> very small random variances and little effect from the moderators.
>>
>> Michael
>>
>> On 14/06/2023 10:22, Viechtbauer, Wolfgang (NP) via R-sig-meta-analysis
>> wrote:
>> > Michael means using the profile() function.
>> >
>> > This aside, if you use predict(<model>), then it will compute
>> predicted/fitted values based on the model matrix (and corresponding
>> CIs/PIs). If you want predicted values for particular combinations of
>> moderator values, then you have to specify the 'newmods' argument.
>> >
>> > Best,
>> > Wolfgang
>> >
>> >> -----Original Message-----
>> >> From: R-sig-meta-analysis [mailto:
>> r-sig-meta-analysis-bounces using r-project.org] On
>> >> Behalf Of Tori Peña via R-sig-meta-analysis
>> >> Sent: Tuesday, 13 June, 2023 19:51
>> >> To: Michael Dewey
>> >> Cc: Tori Peña; R Special Interest Group for Meta-Analysis
>> >> Subject: Re: [R-meta] How to Get PIs for Moderator Analyses
>> >>
>> >> Hi Michael,
>> >>
>> >> Thanks for your response!  What do you mean by profiling the variance?
>> >> Yes, some of my moderators had virtually no effect in our models.  I
>> have
>> >> the issue with the PI being displayed for each data point in the model
>> >> instead of the PI for the model across all the moderators,
>> unfortunately.
>> >>
>> >> On Tue, Jun 6, 2023 at 12:20 PM Michael Dewey <lists using dewey.myzen.co.uk>
>> >> wrote:
>> >>
>> >>> Dear Tori
>> >>>
>> >>> It might help to know exactly why you think the ouput is incorrect but
>> I
>> >>> notice that (a) your random effects seem to have very small variances
>> >>> (have you tried profiling them?) (b) your moderator does not seem to
>> >>> have much effect either.
>> >>>
>> >>> Michael
>> >>>
>> >>> On 06/06/2023 16:59, Tori Peña via R-sig-meta-analysis wrote:
>> >>>> Hello Listserv -
>> >>>>
>> >>>> I wanted to know if folks know of a way to get prediction intervals
>> for
>> >>>> moderator analyses!  I tried the predict() function but the output
>> looks
>> >>>> incorrect.  I attached a screenshot below of the code plus a part of
>> the
>> >>>> output.
>> >>>>
>> >>>> Thanks in advance!
>> >>>>
>> >>>> --
>> >>>> *Tori Peña, Ph.D. *(she/her/ella)
>> >>>> Cognitive Psychology
>> >>>> Dept. of Psychology
>> >>>> Stony Brook University
>> >>>> Stony Brook, NY 11790-2500
>> >>>> Stony Brook University logo