[R-meta] Single group mean - metafor

[Viechtbauer, Wolfgang (NP)]

Hi Manu,

Just a brief response, as I am not too familiar with this specific application (determining critical thresholds). Your general thinking sounds reasonable to me if those means you intend to synthesize are indeed estimates of such a threshold.

When dealing with complex dependency structures, it would be wise to compare the model-based results with those from using cluster-robust inference methods, although the latter gets tricky when including crossed random effects than can span the entire dataset (like phylogenetic or spatial correlations). But if these end up being negligible and one is left with a strictly hierarchical dependency structure, using robust(..., cluster=<>, clubSandwich=TRUE) with the highest level as the cluster variable will get you those results.

Best,
Wolfgang

> -----Original Message-----
> From: R-sig-meta-analysis <r-sig-meta-analysis-bounces using r-project.org> On Behalf
> Of Manuela Santos Santana via R-sig-meta-analysis
> Sent: Thursday, April 24, 2025 20:57
> To: r-sig-meta-analysis using r-project.org
> Cc: Manuela Santos Santana <manuelasantana.oc using gmail.com>
> Subject: [R-meta] Single group mean - metafor
>
> Dear all,
>
> I am seeking input on a conceptual approach I’m considering for determining
> critical thresholds for a biological response (a biomarker) using the
> metafor package.
>
> I have data from a monitoring program and aim to establish thresholds by
> estimating a central tendency of the biomarker while accounting for its
> variability across time, sampling efforts, and locations. The idea is to
> use a single-group meta-analytic model to estimate a mean value, and then
> use the confidence interval of that mean as a data-driven threshold range.
>
> Specifically, I am considering using rma.mv() to model the mean response,
> while including random effects to account for:
>
>    - site-level variability,
>    - temporal and sampling effort heterogeneity, and
>    - potentially, phylogenetic correlations among species.
>
> Would this be considered a sound statistical approach for defining such
> thresholds? Are there any caveats or alternative modeling strategies I
> should consider?
>
> I appreciate any insights or references you can offer.
>
> Best,
>
> Manu