[R-meta] Subgroup analyses under the multi-level (3 level) meta-analysis model

[Viechtbauer, Wolfgang (SP)]

Dear Saya,

Please keep the mailing list in cc.

One level is the reference level (namely 'Corr') and you get three contrasts with this reference level. Please see:

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

Best,
Wolfgang

>-----Original Message-----
>From: Saya Fujita [mailto:sayakaf.p using gmail.com]
>Sent: Sunday, 14 March, 2021 17:43
>To: Viechtbauer, Wolfgang (SP)
>Subject: Re: Subgroup analyses under the multi-level (3 level) meta-analysis model
>
>Dear Professor Viechtbauer,
>
>Thank you very much for your prompt response!
>
>I ran the code using my original variables "Metrics" as factors.
>I can see it has 4 levels, but my regression results only show 3 factors.
>I am wondering why?
>
>The Metrics has 4 levels (Coupling, Flow, Graph, Corr) as seen below:
>
>> str(multi.data$Metrics)
> Factor w/ 4 levels "Corr","Coupling",..: 1 1 1 1 1 1 1 1 1 1 ...
>
>Here's my codes:
>
>subEEG.Metrics<-rma.mv(y,
>                   v,
>                   random = ~ 1 | PaperID/AnalysisID,
>                   tdist = TRUE,
>                   data = multi.data,
>                   method = "REML",
>                   mods = ~ Metrics)
>
>And the results are:
>
>Model Results:
>
>                 estimate      se     tval    pval    ci.lb   ci.ub
>intrcpt            0.8795  0.2458   3.5788  0.0005   0.3933  1.3658  ***
>MetricsCoupling   -0.1432  0.2982  -0.4804  0.6318  -0.7332  0.4468
>MetricsFlow        0.1238  0.4307   0.2876  0.7742  -0.7283  0.9760
>MetricsGraph      -0.0084  0.4257  -0.0196  0.9844  -0.8506  0.8339
>
>If you could kindly let me know whether I am doing something wrong, I would be so
>appretiative!
>
>Thank you very much again!
>
>Warmest regards,
>Saya
>
>On 14/03/2021, 15:52, "Viechtbauer, Wolfgang (SP)"
><wolfgang.viechtbauer using maastrichtuniversity.nl> wrote:
>
>    Dear Saya,
>
>    You do not need to create these 'dummy variables' yourself. Just declare the
>variable that distinguishes the various subgroups as a factor and include it as a
>predictor via 'mods'. You also might want to read this:
>
>    https://www.metafor-project.org/doku.php/tips:models_with_or_without_intercept
>
>    It's not focused on rma.mv() (i.e., it uses a simpler model), but the same
>principles apply.
>
>    Best,
>    Wolfgang
>
>    >-----Original Message-----
>    >From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces using r-project.org]
>On
>    >Behalf Of Saya Fujita
>    >Sent: Sunday, 14 March, 2021 16:24
>    >To: R-sig-meta-analysis using r-project.org
>    >Subject: [R-meta] Subgroup analyses under the multi-level (3 level) meta-
>analysis
>    >model
>    >
>    >Dear All,
>    >
>    >I am conducting a 3-level multi-analysis, following this Bookdown.
>    >https://bookdown.org/MathiasHarrer/Doing_Meta_Analysis_in_R/mlma.html
>    >
>    >I would like to run some sub-group analyses under this 3 level structure, and
>I
>    >can see in this bookdown (chapter 12.1) that if I use rma.mv function in
>metafor,
>    >we need to create two variables/columns for each subgroup level. In the
>example in
>    >this book down, the number of sub-groups was two: peer-review, dissertation,
>so it
>    >makes sense that they coded 1 (yes) or 0 (no).
>    >
>    >In my data I have more than two subgroups, and I wonder how I should input
>the
>    >data. I thought, if I have 4 subgroups for instance, I’d create 4 columns and
>code
>    >0 or 1 in each column, but that would group 3 different levels as 0 in one
>column,
>    >and specifying one variable/column as a moderator in rma.mv using
>    >the mods parameter would compare one subgroup with the other three subgroups
>–
>    >which is not what I want to do. Hope this makes sense!
>    >I wonder if anyone here has done a subgroup analysis for multiple subgroup
>    >comparisons, maintaining the 3 level structure of a multi-level meta-
>analysis?
>    >Could I ask how you did the analysis?
>    >
>    >Thank you very much in advance, and I apologise if it’s a very simple/naïve
>    >question!
>    >
>    >Many thanks,
>    >Saya
>    >
>    >*********
>    >Sayaka Kidby (Fujita)| PhD candidate, Marie Skłodowska-Curie Early Career
>    >Researcher [she/her]
>    >Psychology | Lancaster University
>    >s.kidby using lancaster.ac.uk
>    >www.lancaster.ac.uk