REVIEW 3 major objections 4 minor 7 references
Is there a robust effect of mainland mutualism rates on species richness of oceanic islands?
T0 review · 3 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash
Pith's one-line read A reported effect of mainland mutualists on island species richness does not survive re-analysis.
desk verdict A convincing re-analysis showing the mutualism effect vanishes with better interpolation and nonlinear terms, though the authors should quantify residual concurvity to rule out an identifiability artifact. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing object is the covariate 'mutualism filter strength' — the proportion of mutualist species in the mainland source pool. The original result relied on a generalized additive model that predicted this value from latitude alone, which made the covariate a deterministic function of latitude and therefore nonlinearly confounded with it. The re-analysis replaces that with a random forest trained on latitude and longitude, decoupling the covariate from latitude, and re-estimates the species-deficit regression as a generalized additive model with smooth splines for all non-focal predictors. The mechanism that produced the original effect is thus a combination of measurement error and confounding in the interpolated covariate plus unaccounted nonlinearity in the other predictors.
What would settle it
Use direct island-level observations of mutualist proportions (instead of mainland-derived interpolations) to test whether island species deficit increases with mainland mutualist proportion after spline adjustment for area, distance, and latitude; a clear positive relation would overturn the paper's conclusion.
Extended reading notes
Core claim
On the authors' analysis, the mutualism filter effect reported in the original study is not robust. When mainland mutualism strength is interpolated with a random forest that uses both latitude and longitude, the covariate no longer perfectly tracks absolute latitude; and when the main model is refit as a generalized additive model with smooth splines for area, distance, and latitude, the mutualism coefficient becomes non-significant regardless of which interpolation is used. The paper further reports that the proportion of mutualists observed on oceanic islands does not differ systematically from the proportion predicted at the corresponding mainland locations, which the authors read as evidence against a strong mutualist establishment disadvantage. Their overall conclusion is that the available data do not provide convincing statistical support for a mutualism effect on island species richness.
Load-bearing premise
The conclusion depends on the assumption that the random forest interpolation and the spline-based nonlinear model correctly isolate the true effect of latitude, so that the mutualism effect disappears because it truly was an artifact—not because the new model is misspecified in a different way.
Editorial extensions
If this is right
- The original mutualism-based explanation of a weaker latitudinal diversity gradient among oceanic islands should be treated as unsupported by current data.
- Future island biogeography analyses should test for nonlinear effects of area, distance, and absolute latitude before attributing residual signal to biological covariates.
- Interpolated environmental predictors that depend on a single correlated variable can create apparent effects in downstream regressions unless the confounding is explicitly modeled.
- The absence of a difference between island and mainland mutualist proportions suggests that establishment limitations specific to mutualists are not easily detectable at the whole-island richness scale.
Reading between the lines
- The same confound-and-nonlinearity critique may apply to other macroecological studies that interpolate coarse spatial covariates from latitude only; reanalyses with multivariate interpolation and spline adjustment could reveal which reported gradients are robust.
- A testable extension would be to repeat the analysis using direct measurements of mutualist richness on islands, where such data exist, to bypass interpolation uncertainty entirely.
- If the null result holds, the weaker latitudinal gradient of oceanic islands might instead be explained by area, distance, and their nonlinear interactions, or by other unmeasured island attributes.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript re-analyzes the data of Delavaux et al. (2024) and challenges the claim that the proportion of mutualists on the mainland drives the species deficit of oceanic islands. The authors replace the original latitude-only GAM interpolation of mutualism strength with a random forest using latitude and longitude, refit the island richness regression as a GAM with smooths for area, distance, and absolute latitude, and compare predicted mainland and observed island mutualism proportions. They report that the mutualism effect becomes non-significant and small under these changes, and that island mutualism proportions do not differ systematically from mainland values. They conclude there is no robust statistical evidence for the original effect.
Significance. If the re-analysis holds, it is a valuable correction to a high-profile result in island biogeography and to the claimed link between mutualism rates and the latitudinal diversity gradient. The paper's strengths are its transparency (code and data provided), its multiple sensitivity checks (Fig. S3), and its attempt to test the mechanism with an independent island-mainland comparison. Its principal limitations are the lack of explicit identifiability diagnostics for the key GAM and the informal nature of the island-mainland comparison. These are fixable and do not invalidate the overall approach, but they currently leave the central negative conclusion less certain than the text suggests.
major comments (3)
- [Accounting for nonlinear effects; Fig. 2B, Fig. S3] The paper's central negative result depends on the assumption that the random forest-predicted mutualism covariate is identifiable from the smooth of absolute latitude in the final GAM. The random forest still uses latitude as a predictor, so the mutualism covariate may retain a nonlinear dependence on latitude even after adding longitude. The sensitivity analysis in Fig. S3 varies the latitude specification, but it does not diagnose the final model actually used. Please report concurvity diagnostics for the GAM in Fig. 2B (e.g., the concurvity() output from mgcv for the mutualism term against the latitude smooth), and if concurvity is high, either fit an alternative model with explicit orthogonalization or soften the claim that the effect is 'small'.
- [Island-mainland comparison, Fig. 2C] The conclusion that mutualists appear at the same proportion on islands and on the corresponding mainland is presented as visual evidence ('we did not find such an effect'), but no statistical test, effect size, or confidence interval is reported. Because island communities are small and the mainland values are spatial predictions with uncertainty, a formal paired test or mixed-effects model with confidence intervals is needed to support the claim of no establishment disadvantage. Please also state explicitly how the 'corresponding mainland' value was defined for each island.
- [Fig. 1D, Fig. 2B] The paper describes the mutualism effect as 'non-significant and small' but does not quote estimated coefficients, standard errors, or confidence intervals. Under collinearity or concurvity, the point estimate and its variance are both unstable, so 'small' and 'non-significant' are not sufficient. Report the standardized coefficient for the mutualism term (with its confidence interval) for the final GAM under both the original and the random forest-predicted covariates.
minor comments (4)
- [General] The manuscript text contains numerous typographical artifacts or OCR-like errors (e.g., 'isola3on', 'beTer', 'misift', 'Har3g' in the author line); please ensure the final PDF is properly rendered.
- [Figure captions] The figure captions should state how the confidence intervals in Figs. 1B/1D and 2B were computed (for example, from the GAM coefficient covariance matrix) and whether the model used for Fig. 2B is identical to the one behind Fig. S3.
- [Fig. 2C caption] The distinction between 'predicted' and 'observed' is important in Fig. 2C; the caption should clarify that the mainland values are spatial predictions from the random forest while the island values are directly observed proportions, and should refer to the statistical test requested in the major comments.
- [References] The Delavaux et al. paper is reference 2 in the main text and appears again in the supplementary reference list; please use a consistent numbering scheme across the main text and supplementary file.
Circularity Check
No substantial circularity: this is a re-analysis of an external dataset, and its conclusions are not wired into its inputs by construction.
full rationale
The paper is a re-analysis of data and models from Delavaux et al. (2024). It does not claim to derive a first-principles prediction; instead, it tests whether the original mutualism effect survives under alternative modeling choices. The random forest interpolation of mutualism strength from latitude and longitude is a covariate-reconstruction step, not a fit to the island species-deficit response. The GAM re-fit uses smooth terms for area, distance, and latitude while keeping mutualism strength linear, and the reported non-significance is an empirical outcome rather than an algebraic consequence of the model specification. The island-versus-mainland mutualism comparison in Fig. 2C is an independent observational check. The only self-citation is Hartig (2022) for the DHARMa residual-diagnostics package, which is non-load-bearing software and does not supply any premise of the argument. The skeptical concern about residual concurvity between the latitude spline and the interpolated mutualism covariate is a legitimate statistical identifiability risk, but it is a question of model validity and does not make the derivation circular. No fitted parameter is renamed as a prediction, no uniqueness theorem is imported, and no known result is repackaged under new coordinates. Therefore the circularity score is 0.
Assumptions & free parameters
assumptions (3)
- domain assumption The random forest model using latitude and longitude provides unbiased predictions of mainland mutualism filter strength.
- domain assumption The generalized additive model with smooth terms adequately captures the true non-linear relationships between area, distance, latitude, and species deficit.
- domain assumption Comparing observed island mutualism ratios to predicted mainland ratios at corresponding locations is a valid test of establishment disadvantage.
Cite this review
Pith. "Pith review of Is there a robust effect of mainland mutualism rates on species richness of oceanic islands?." pith.science (2026). https://pith.science/paper/H3SFKZTN
@misc{pith2026241115105,
author = {Pith},
title = {Pith review of: Is there a robust effect of mainland mutualism rates on species richness of oceanic islands?},
year = {2026},
howpublished = {\url{https://pith.science/paper/H3SFKZTN}},
note = {Machine review of arXiv:2411.15105}
}
read the original abstract
In island biogeography, it is widely accepted that species richness on island depends on the area and isolation of the island as well as the species pool on the mainland. Delavaux et al. (2024) suggest that species richness on oceanic islands also depends on the proportion of mutualists on the mainland, based on the idea that mutualists require specific interaction partners for their survival and thus have lower chances of establishment after successful immigration. As the proportion of mutualists increases towards the tropics, this effect could explain a weaker latitudinal diversity gradient (LDG) for oceanic islands. However, after re-analyzing their data, we have doubts if these conclusions are supported by the available data.
Figures
Reference graph
Works this paper leans on
-
[1]
MacArthur, R. H., & Wilson, E. O. (2001). The theory of island biogeography. Princeton University Press
work page 2001
-
[3]
DeZo, M., Visser, M.D., Wright, S.J. and Pacala, S.W. (2019), Bias in the detec3on of nega3ve density dependence in plant communi3es. Ecol LeZ, 22: 1923-1939. hZps://doi.org/10.1111/ele.13372
-
[4]
Freckleton, R. P . (2011). Dealing with collinearity in behavioural and ecological data: model averaging and the problems of measurement error. Behavioral Ecology and Sociobiology, 65, 91-101. 5
work page 2011
-
[5]
Random measurement error and regression dilu3on bias BMJ 2010; 340:c2289 doi:10.1136/bmj.c2289
Hutcheon J A, Chiolero A, Hanley J A. Random measurement error and regression dilu3on bias BMJ 2010; 340:c2289 doi:10.1136/bmj.c2289
-
[6]
Wood, S.N. (2011) Fast stable restricted maximum likelihood and marginal likelihood es3ma3on of semiparametric generalized linear models. Journal of the Royal Sta3s3cal Society (B) 73(1):3-36 6 Supplementary Informa4on S1 Figure S1: Residuals of the model by Delavaux et al.1. Most predictors show residuals paTerns, indica3ng nonlineari3es that could poten...
work page 2011
-
[7]
Delavaux, C.S., Crowther, T.W., Bever, J.D. et al. Mutualisms weaken the la3tudinal diversity gradient among oceanic islands. Nature 627, 335–339 (2024). hZps://doi.org/10.1038/s41586-024-07110-y
-
[8]
DHARMa: Residual Diagnos3cs for Hierarchical (Mul3-Level / Mixed) Regression Models
Har3g F (2022). DHARMa: Residual Diagnos3cs for Hierarchical (Mul3-Level / Mixed) Regression Models. R package version 0.4.6, hZps://CRAN.R-project.org/package=DHARMa
work page 2022
Reviewed August 12, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.