[R-meta] variable importance in the context of meta-analysis

[Viechtbauer, Wolfgang (SP)]

Dear Diego,

Yes, this makes sense. The Akaike weight of a model is the estimated probability that the model is the best model (in the sense of how 'best' is defined under such an information-theoretic approach) in the candidate set. So, by adding up the weights for all models that contain the predictor of interest, you get, roughly speaking, the probability that the predictor is relevant.

As references, you could take a look at the book "Model Selection and Multimodel Inference" by Burnham and Anderson or the somewhat shorter/condensed "Model Based Inference in the Life Sciences: A Primer on Evidence" by Anderson.

Best,
Wolfgang

>-----Original Message-----
>From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces using r-project.org]
>On Behalf Of Diego Grados Bedoya
>Sent: Friday, 18 September, 2020 8:40
>To: r-sig-meta-analysis using r-project.org
>Subject: [R-meta] variable importance in the context of meta-analysis
>
>Dear all,
>
>I was exploring model specification alternatives in the context of
>meta-analysis and I discovered the glmulti r package. The main function of
>this package fits all possible models that could be constructed using
>combinations of the moderators/predictors. However, there is something that
>triggered my attention: the definition of the variable importance. The
>relative importance value for a particular predictor is equal to the sum of
>the Akaike weights for the models in which the predictor appears. Is this
>conceptually and/or mathematically acceptable? I could not find
>references related to it.
>
>Thanks in advance,
>
>Kind regards,
>
>Diego