{"id":"dbe645ae-d324-4599-827a-e9ee960dc328","arxiv_id":"1908.07960","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Directional ecological selection is predicted and partly observed to lower species diversity more than stabilizing selection at equal strength, via a null-comparison framework using diversity, community-weighted mean, and community-weighted variance.","lead":"This paper proposes a way to tell whether ecological selection favors average or extreme traits by comparing species diversity and trait spread to null models. A simulation predicts directional selection reduces diversity more than stabilizing selection, and a plant mesocosm test partly supports this, with one habitat needing a different explanation.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The empirical attribution to seed-mass selection is not identifiable: trait-shuffling nulls cannot exclude selection on an unmeasured trait correlated with seed mass, a limitation the authors acknowledge but do not resolve.","rationale":"The reader's weakest assumption correctly identifies the unresolved identification problem in the empirical analysis: the trait-shuffling null cannot distinguish selection on seed mass from selection on a correlated unmeasured trait. This is the most load-bearing concern because the empirical test is a central part of the paper's claim to demonstrate the framework's utility; if the attribution to seed mass is not identifiable, the empirical demonstration is substantially weakened, even though the simulation prediction itself remains internally consistent. I considered whether the comparability of selection strength across stabilizing and directional regimes in the simulation might be a more fundamental issue, but because the paper's practical prediction is framed in terms of comparable CWV reductions, and because robustness checks were provided, the empirical identification problem is more decisive. The authors explicitly flag the wrong-trait limitation in Appendix S1, but they do not test whether their trait-shuffling null resolves it, and they do not address the correlated-trait case. The less productive habitat's CWV being higher than null also indicates that the framework does not apply when niche partitioning is present, but this is acknowledged and discussed. The recommended verdict remains CONDITIONAL, since the simulation prediction is well supported and the empirical test is suggestive but not uniquely diagnostic.","tokens_in":13073,"tokens_out":10225,"duration_ms":101014,"concrete_test":"Re-run the Appendix S1 wrong-trait simulation with the selected trait correlated with the measured trait at several levels (e.g., r = 0, 0.3, 0.6, 0.9), and apply the same trait-shuffling null used in the empirical analysis (randomizing seed mass among observed species within observed abundances). If CWV falls below the 2.5th percentile of the null in a substantial fraction of replicate simulations for r > 0, the empirical CWV test cannot uniquely identify seed mass as the selected trait; if it does not, the correlation confound is less severe. Optionally, use the interspecific trait correlation matrix from DeMalach et al. 2019 to choose realistic correlation values.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim requires that an observed CWV lower than the null expectation indicates selection specifically on the measured trait (seed mass). The trait-shuffling null in Methods randomizes seed mass among species within the observed abundance distribution, which controls for abundance and dominance effects but does not control for correlation between seed mass and unmeasured traits. If selection acts on an unmeasured trait correlated with seed mass, the surviving species will be similar in that unmeasured trait and, due to the correlation, also similar in seed mass; CWV of seed mass will be lower than the shuffled null and CWM will deviate, producing exactly the pattern reported for the productive habitat. The authors concede in Appendix S1 (wrong-trait scenario) that a decrease in CWV can be driven by selection on an uncorrelated trait, and the Methods paragraph states: 'a decrease in CWV could be driven by a selection acting on a different uncorrelated trait.' That simulation, however, permutes fitness to be uncorrelated with the measured trait and does not apply the trait-shuffling null, so it neither resolves the uncorrelated-trait case nor addresses the more damaging correlated-trait case. The empirical attribution to seed mass is therefore not unique; it is compatible with selection on any trait sufficiently correlated with seed mass. Additionally, the less productive habitat shows CWV higher than the null, which lies outside the framework's decision tree and is explained post hoc as niche partitioning, further weakening the empirical test as confirmation of the framework.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a framework for inferring the type (stabilizing vs. directional) and strength of ecological selection by jointly comparing species diversity, community-weighted mean (CWM), and community-weighted variance (CWV) of a trait against null expectations. A spatially implicit metacommunity simulation shows that, at equal selection strength, directional selection reduces species diversity more than stabilizing selection while both reduce CWV, and that CWM deviation indicates directionality. The framework is applied to a mesocosm experiment of annual plants in two soil-depth habitats: species diversity was lower than the drift null in both habitats and lower in the productive habitat, CWM of seed mass was higher and CWV lower than the trait-shuffling null in the productive habitat (interpreted as directional selection for large seeds), while in the less productive habitat CWV was higher than the null (interpreted as niche partitioning). The authors conclude that simultaneous analysis of diversity and traits helps identify selection processes.","tokens_in":13310,"tokens_out":2852,"duration_ms":29152,"significance":"If the framework holds, it offers a practical way to infer ecological selection attributes from commonly measured community-level data. The simulation result that directional selection reduces diversity more than stabilizing selection at equal strength is a useful, nontrivial prediction that is robust to alternative trait distributions, fitness functions, and temporal scales, as shown by extensive supplementary analyses. The empirical case study demonstrates the framework's application to real data and highlights a potential link between productivity, seed mass selection, and diversity. However, the empirical attribution to seed mass is compromised by an identifiability problem that the authors acknowledge but do not resolve, and the less-productive-habitat result falls outside the proposed decision framework. The theoretical contribution is solid; the empirical inference needs substantial reframing or additional analysis to support the paper's central claim.","major_comments":[{"comment":"The empirical attribution of the observed CWM/CWV pattern to selection on seed mass is not identifiable from the trait-shuffling null used. The null randomizes seed mass among species within the observed abundance distribution, which controls for abundance and dominance but does not control for correlations between seed mass and unmeasured traits. If selection acts on an unmeasured trait correlated with seed mass, surviving species will be similar in that unmeasured trait and, because of the correlation, also similar in seed mass, generating exactly the observed decrease in CWV and shift in CWM. The authors acknowledge in the Methods that 'a decrease in CWV could be driven by a selection acting on a different uncorrelated trait,' and Appendix S1 Fig. S10 simulates a wrong-trait scenario, but that simulation permutes fitness to be uncorrelated with the measured trait and does not apply the trait-shuffling null; it neither addresses the correlated-trait case nor provides a way to rule it out empirically. Consequently, the claim that the productive habitat experienced directional selection specifically on seed mass is not uniquely supported by the data; it is equally compatible with selection on any trait sufficiently correlated with seed mass.","section":"Methods, 'Mesocosm experiment' paragraph; Appendix S1, Fig. S10"},{"comment":"The finding in the less productive habitat that CWV is higher than the null expectation lies outside the proposed decision framework. The framework in Fig. 2 only specifies interpretations for CWV lower than or not different from the null expectation; a higher-than-expected CWV is not part of the scheme. The authors interpret this result as evidence for niche partitioning and limiting similarity in seed mass, but this interpretation is post hoc and is not derived from the simulation model, which assumes frequency-independent selection only. No formal test or model is presented to distinguish niche partitioning from other processes that could inflate CWV relative to the trait-shuffling null (e.g., environmental heterogeneity within a treatment or sampling artifacts). This gap weakens the claim that the framework can jointly explain diversity and trait patterns across both habitats.","section":"Results, 'Mesocosm experiment'; Discussion, 'Interpretation of the mesocosm experiment'; Fig. 2 decision scheme"},{"comment":"The logical chain from observed patterns to selection type relies on a sequential decision rule (diversity lower than null → selection; CWV lower than null → trait-specific selection; CWM deviation → directional vs. stabilizing). However, the two null models are constructed differently: diversity is compared to a dynamic drift simulation with no trait-based selection, while CWM/CWV are compared to trait-shuffling nulls that condition on the observed abundances. Because the observed abundances are themselves a product of selection, the trait-shuffling null does not represent the 'no selection on this trait' expectation under the same community-level constraints. This mismatch means that the two steps of the decision rule are not testing the same null hypothesis, and the inference of selection type/strength is not as direct as Fig. 2 implies. The authors should clarify what biological question each null answers and address whether the sequential rule is valid when abundances are selected.","section":"Fig. 2 and Methods, 'Mesocosm experiment'"}],"minor_comments":[{"comment":"The sentence 'In both habitats, selection has taken place, as indicated by species diversity being lower than the null expectation (Fig. 2)' cites Fig. 2, which is the conceptual scheme, but the relevant empirical results are in Fig. 4; the citation should be corrected.","section":"Discussion, first paragraph"},{"comment":"The species list contains an apparent typo: 'Papaveraceaer umbonatum' should likely be 'Papaver umbonatum' or similar; the family abbreviation is also inconsistent.","section":"Appendix S2, Table S1"},{"comment":"The description of the trait-shuffling null would benefit from explicitly stating whether the randomization is constrained to preserve the species pool's trait distribution and whether the observed abundances are held fixed; the current text is ambiguous about how the null accounts for the nine replicate communities.","section":"Methods, 'Simulation model'"},{"comment":"Several supplementary figures (e.g., S4-S6, S9) are described as showing results 'qualitatively similar' to the main simulation, but no quantitative comparison or effect-size metric is provided; a brief statement of the range of deviations would strengthen the robustness claim.","section":"Figure 3 and Appendix S1"}],"recommendation":"major_revision","confidential_remarks":"The theoretical core of the paper (simulation prediction about diversity under directional vs. stabilizing selection) is solid and well tested. The main weakness is that the empirical test does not deliver what the title promises: the attribution to seed mass is not identifiable from the trait-shuffling null, and the less-productive-habitat result is outside the framework. I would encourage the authors to either (a) reframe the empirical section as an illustrative application that is consistent with, but does not uniquely establish, seed-mass selection, or (b) add a formal sensitivity analysis or additional null model that addresses correlated unmeasured traits. If the authors can do that, the paper would be a valuable contribution; as it stands, the empirical claim overreaches."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper's real contribution is the simulation-based prediction that, at equal selection strength, directional selection reduces species diversity more than stabilizing selection, and the proposed decision framework that jointly reads species diversity, CWM, and CWV against nulls. That part is solid. The simulation is simple, clearly specified, and robust to alternative fitness functions and trait distributions, including a uniform trait distribution where the difference shrinks but does not vanish. The framework is a nice practical step: it gives ecologists a low-cost way to generate hypotheses about selection type and strength from community-level data.\n\nThe soft spots are in the empirical application. The trait-shuffling null used for CWM and CWV conditions on the observed abundance distribution, so it cannot distinguish selection on seed mass from selection on an unmeasured trait correlated with seed mass. The appendix's wrong-trait scenario only randomizes fitness with respect to the measured trait, which handles the uncorrelated case but not the correlated one. The authors do acknowledge the possibility of correlated traits in the discussion and point to ordination techniques, but the case study itself does not resolve it. So the productive-habitat result is consistent with directional selection on seed mass, but it is not uniquely attributable to seed mass. The less-productive habitat result, where CWV is higher than null, falls outside the decision tree and is explained post hoc as niche partitioning. That is honest but weakens the empirical test as confirmation. Minor issues: simulation code is only promised, and the simulation results are presented without error bars.\n\nNone of this sinks the central claim, which is the simulation prediction. The paper does what it says: it proposes a framework, verifies its internal logic, and gives a candid empirical demonstration with limitations stated. The citation pattern is appropriate, including self-citations to the prior mesocosm analyses. The authors are not overselling; they explicitly flag the uncorrelated-trait confound and call for future extensions.\n\nThis paper deserves a serious referee. A good referee should push for a clearer treatment of the correlated-trait confound in the empirical inference, and a more cautious interpretation of the less-productive habitat pattern. With revisions, it would be a useful contribution to community ecology methods.","headline":"Useful framework with a clean simulation result; the empirical test is less clean than the title suggests, but the limitations are acknowledged.","tokens_in":623,"tokens_out":2482,"would_cite":true,"duration_ms":49878,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Trait statistics plus diversity identify the type of selection.","keywords":["ecological selection","species diversity","community-weighted mean","community-weighted variance","null models","seed mass","stabilizing selection","directional selection"],"falsifier":"Measure an additional trait correlated with seed mass in the productive habitat and build the same trait-shuffling null for it; if that trait's CWV also lies below its null while seed mass CWV does not, the case that seed mass itself was the selected trait would collapse.","tokens_in":12843,"feed_emoji":"🌱","tokens_out":9524,"duration_ms":80749,"temperature":0.7,"pith_summary":"The paper argues that ecologists can infer not only the strength but also the type of ecological selection acting on a community by reading species diversity and trait distributions together against null expectations. Its central prediction, confirmed in a simulation model, is that at equal selection strength directional selection reduces species diversity more than stabilizing selection, even though both lower the community-weighted variance (CWV) of the selected trait. Applying this logic to a five-year mesocosm experiment of annual plants, the authors attribute the lower diversity of the productive habitat to strong directional selection for large seed mass, while the less productive habitat shows elevated CWV, which they interpret as niche partitioning among seed sizes rather than stabilizing selection. The payoff would be a way to identify the mechanism of community assembly from patterns that are already routinely measured.","feed_headline":"Trait statistics plus diversity identify the type of selection","feed_subtitle":"Species diversity alone hides the kind of filtering; adding trait mean and variance exposes it.","key_machinery":"The load-bearing piece is a pair of null comparisons plus a fitness model. In the model, the Lorentzian function $\\omega_i = 1/(1+\\theta(\\delta_{\\mathrm{best}}-\\delta_i)^2)$ maps each species' trait value $\\delta_i$ to its ecological fitness, where $\\theta$ sets selection strength and $\\delta_{\\mathrm{best}}$ is the optimal trait value; the framework then reads CWM as a marker of the optimum's location (selection type), CWV as a marker of how steeply fitness declines with distance from the optimum (selection strength), and species diversity as the total filtering intensity. The empirical nulls are constructed by shuffling trait values across species within the observed abundance distribution (for CWM and CWV) and by a pure-drift simulation (for diversity).","core_discovery":"The central claim is that selection type and strength can be recovered by comparing three community statistics to null expectations built from the species pool: species diversity, which falls under any selection; the community-weighted mean trait value (CWM), which shifts away from the pool mean only under directional selection; and the community-weighted variance (CWV), which falls with selection strength under either type. The simulation shows that because a normal species pool contains more species near the intermediate optimum, stabilizing selection preserves more diversity at a given strength, whereas directional selection, favoring only species at one extreme, drives more species extinct. In the mesocosm data, both habitats have diversity below the drift null, but only the productive habitat shows elevated CWM and reduced CWV, supporting directional selection for large seeds; the less productive habitat's elevated CWV points to coexistence of different seed sizes rather than seed-mass filtering.","pith_inferences":["The same logic could be applied to multiple traits at once: if CWM shifts only along one trait axis, the analysis localizes the trait under directional selection even when the true selective environment is unknown.","The diversity gap between selection types suggests a testable experimental prediction for any system with a manipulable trait–fitness map: rotate trait values among species and the gap should shrink or disappear when pool trait geometry is destroyed.","A CWV-above-null result in other empirical systems could be used as a distributional signal for competition–colonization trade-offs, since such trade-offs maintain high trait variance in low-stress habitats without invoking stabilizing selection on the trait itself."],"forward_implications":["Whenever a community's diversity falls below the drift null, selection has acted; if CWV of a trait also falls below its null, that trait is implicated, and a CWM shift signs the direction of selection.","At equal selection strength, directional selection should leave a community less diverse than stabilizing selection, so across-habitat diversity gaps become a rough gauge of selection directionality.","Along environmental gradients, species diversity and CWV should track each other when selection strength varies, but decouple when the gradient mostly flips stabilizing to directional selection.","A CWV above the trait-shuffling null can flag niche partitioning on the measured trait, as inferred for the less productive habitat.","The framework is cleanest in closed experimental systems, since blocking dispersal removes the confounding between reproductive output and colonizing ability."],"supporting_citations":[{"why":"Defines ecological selection, its strength, and the stabilizing/directional classification that the framework formalizes.","marker":"Vellend 2016"},{"why":"Supplies the trait-based assembly perspective and the fitness-through-traits premise the simulation assumes.","marker":"Shipley 2010"},{"why":"Previous model of trait selection that the authors extend by adding species-diversity predictions.","marker":"Loranger et al. 2018"},{"why":"Established the expected CWM/CWV signatures for stabilizing versus directional trait selection.","marker":"Rolhauser and Pucheta 2017"},{"why":"Resource-competition theory predicting large-seed advantage under high productivity, motivating the directional-selection hypothesis.","marker":"DeMalach and Kadmon 2018"},{"why":"Shows seed mass is the main abundance predictor in this mesocosm system, justifying the choice of seed mass as the focal trait.","marker":"DeMalach et al. 2019"},{"why":"Provides the mesocosm experiment and its diversity patterns that the framework is applied to.","marker":"Ron et al. 2018"},{"why":"Supplies the limiting-similarity concept used to interpret the elevated CWV in the less productive habitat.","marker":"Macarthur and Levins 1967"}],"fun_headline_variants":["Trait stats unmask selection hidden in diversity","Combining community stats exposes selection type","Directional selection trims diversity: new framework","Why diversity drops more under directional selection"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The empirical attribution to seed mass assumes that shuffling seed masses among species within the observed community yields the right null, so it cannot distinguish selection on seed mass from selection on any unmeasured trait correlated with seed mass.","fun_headline_variants_meta":{"raw":{"variants":["Trait stats unmask selection hidden in diversity","Combining community stats exposes selection type","Directional selection trims diversity: new framework","Why diversity drops more under directional selection"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000491,"raw_usage":{"total_tokens":2433,"prompt_tokens":982,"completion_tokens":1451,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":598,"completion_tokens_details":{"reasoning_tokens":1396}},"tokens_in":598,"tokens_out":1451,"duration_ms":14247,"temperature":1.0,"reasoning_tokens":1396,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:52:48.353601+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure an additional trait correlated with seed mass in the productive habitat and build the same trait-shuffling null for it; if that trait's CWV also lies below its null while seed mass CWV does not, the case that seed mass itself was the selected trait would collapse.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines ecological selection, its strength, and the stabilizing/directional classification that the framework formalizes."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the trait-based assembly perspective and the fitness-through-traits premise the simulation assumes."},{"cited_title":"Munoz, B","cited_arxiv_id":null,"evidence_quote":"Previous model of trait selection that the authors extend by adding species-diversity predictions."},{"cited_title":"G., and E","cited_arxiv_id":null,"evidence_quote":"Established the expected CWM/CWV signatures for stabilizing versus directional trait selection."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Resource-competition theory predicting large-seed advantage under high productivity, motivating the directional-selection hypothesis."},{"cited_title":"Ron, and R","cited_arxiv_id":null,"evidence_quote":"Shows seed mass is the main abundance predictor in this mesocosm system, justifying the choice of seed mass as the focal trait."},{"cited_title":"Fragman-Sapir, and R","cited_arxiv_id":null,"evidence_quote":"Provides the mesocosm experiment and its diversity patterns that the framework is applied to."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the limiting-similarity concept used to interpret the elevated CWV in the less productive habitat."}],"review_version":1}