Ecology and evolution meta-analysis

Question frameworks

Eco-evo reviews can use PECO, PICOC, or PO depending on whether the review is exposure-focused, intervention-focused, context-dependent, or descriptive. Comparison axes can represent exposure, habitat, taxon, management, temperature, dose, disturbance, or other ecological conditions rather than clinical treatments.

Effect measures

• Log response ratio (lnRR) for proportional mean change.
• Standardised mean difference (Hedges' g) where studies measure the same construct on different scales.
• Fisher z-transformed correlation for association questions.
• Log coefficient of variation ratio (lnCVR) for differences in relative variability.
• Log variability ratio (lnVR) for differences in variability (SD ratio).
• A generic inverse-variance option for protocol-defined effect sizes that carry their own variance.

These are computed from arm-level means, standard deviations, and sample sizes (or from correlations), or accepted pre-computed when a study reports the effect directly. Analysis runs on the transformed scale, and results are back-transformed for reporting: response and variability ratios return to the ratio scale (null of 1), correlations return via the inverse Fisher transform, and the null stays 0 on the difference and correlation scales. Summaries state the scale, the back-transform, and the null anchor so estimates are not misread.

Dependent effects and moderators

One ecology study can contribute multiple effects because it reports several species, sites, outcomes, timepoints, treatments, or shared controls. The default synthesis is a standard pairwise random-effects model. When a study contributes multiple dependent effects — multiple outcomes per study, shared controls, or repeated measures — the analysis uses a multilevel random-effects model (metafor rma.mv) that nests effects within studies and is estimated by REML with t-based inference, so dependent effects are not treated as independent observations.

When sampling covariances among correlated effects are supplied, the model uses them. When they are not available, the effects are modelled as independent and that assumption is surfaced as a limitation on the result, because the pooled uncertainty may then be understated.

Recorded moderators (for example habitat, taxon, or stratum) can be fitted as an additive meta-regression with an omnibus test of the moderators. The regression is additive only, with no interaction terms; if it cannot be fitted the run falls back to the unadjusted model. Diagnostics include between- and within-study heterogeneity (variance components), prediction intervals, publication-bias and sensitivity checks, and model convergence status.

Reviewers should still document dependency groups, moderator choices, covariance assumptions, and sensitivity analyses. Axelium can surface missing or assumed covariance information, but reviewers remain responsible for deciding whether assumptions are defensible.

Certainty and reporting

Ecology certainty is narrative by default. Report search scope, screening, data extraction, dependency assumptions, moderators, heterogeneity, prediction intervals, and sensitivity analyses. Do not present ecology outputs as clinical GRADE, CINeMA, or ranked treatment recommendations.

Reporting follows ecology and evolution standards — PRISMA-EcoEvo and ROSES — and systematic maps are supported for reviews that describe the evidence base rather than pool it.

Current limitations

• Broad availability may depend on your workspace and review configuration.
• Phylogenetic meta-analysis is not supported.
• Meta-regression is additive only; interaction terms between moderators are not fitted.
• Partitioned I² is not reported for multilevel models; a total heterogeneity and variance-component summary with prediction intervals is used instead.
• Covariance values are not automatically inferred when the evidence does not support them; when they are not supplied, dependent effects are treated as independent and this is surfaced on the result.
• Clinical certainty and ranking frameworks do not apply to ecology profiles.