{"id":"2caf1ffa-df54-4b37-a76b-fe0f45ae2e43","arxiv_id":"2607.07588","paper_version":2,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"A unified U-statistic framework tests equality of smooth parameters across k populations via Wald and ANOVA statistics, with fixed-d asymptotics, weighted bootstrap, and normal limits when d grows slower than n.","lead":"The paper gives a single nonparametric toolkit for testing whether many populations share the same U-statistic-based parameters, from variances and Gini indices to high-dimensional means and covariances. Practitioners get Wald, ANOVA, bootstrap, and increasing-dimension normal tests with jackknife covariances and clear finite-sample guidance.","discovery_kind":"unification","skeptic_critique":{"model":"grok-4.5","headline":"No significant objection identified","rationale":"The paper's strongest claim is carefully scoped: under independence, finite second moments, positive first-order kernel variances, comparable sample sizes, continuous differentiability of f, and (for the high-d case) the additional moment and eigenvalue conditions, the ATS admits a weighted-χ^{2} limit for fixed d and a normal limit after trace centering/scaling when d\to∞ with d/n\to0. The proofs follow classical lines and the delicate steps are flagged (Remark 2, Remark 3). The reader's identification of the d/n\to0 + (16)–(17) package as the weakest link is accurate, but that package is an explicit hypothesis, not a tacit assumption that can be violated while still claiming the theorem. Simulations and practical guidelines further document the finite-sample boundary of the regime. Consequently the mathematical claims as stated stand, and the ACCEPT verdict with high confidence needs no adjustment.","tokens_in":32412,"tokens_out":470,"duration_ms":6161,"concrete_test":"Independently re-derive the order of the remainder R in Step 1–2 of the proof of Theorem 2 (after the Hoeffding + Taylor expansion) under only the paper's moment and eigenvalue assumptions; confirm that ||HR||^{2}/σ_n = O_P(√d/n) still vanishes when d/n\to0. If the order fails, the normal approximation would require stronger conditions than claimed.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The reader's weakest_assumption correctly flags the d/n\to0 regime, uniform fourth-moment bounds (16), and eigenvalue sandwich (17) as the most delicate conditions for the ATS-ID normal limit (Theorem 2 / Proposition 5). These are stated explicitly, used at the precise places in the proof (vanishing of remainder terms, Lindeberg condition, and consistency of the feasible traces), and the paper does not claim validity outside them. Within the stated regime the argument is standard (Hoeffding decomposition + Delta method + martingale CLT + jackknife consistency) and internally consistent. No hidden gap that would invalidate the central claim under the paper's own hypotheses was found.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The paper develops a unified nonparametric framework for testing equality of parameters θ_i = f(η_i) across k independent populations, where each η_i is a vector of expectations of symmetric kernels and is estimated by U-statistics. Two quadratic-form statistics are studied: a Wald-type statistic T_n whose null limit is χ^{2}_{(k-1)d} under fixed d (Proposition 1), and an ANOVA-type statistic Q_n = n θ̂ᵀ H θ̂ whose null limit is a weighted sum of χ^{2}_1 variables for fixed d (Proposition 3). For the latter, a weighted bootstrap calibration is proved consistent (Theorem 1), and under d \to ∞ with d/n \to 0, uniform fourth-moment bounds on projected influence functions, and eigenvalue sandwich conditions on HΣH, the centered and scaled ATS converges to N(0,1) (Theorem 2 / Proposition 5). Covariance matrices are estimated by the jackknife (Lemma 1, extending Arvesen). Local-power and consistency results are given, finite-sample behavior is examined in extensive simulations (univariate variance/Gini and multivariate mean-covariance settings), and the methods are illustrated on CPS1988 wage data.","tokens_in":32557,"tokens_out":1109,"duration_ms":14831,"significance":"If the asymptotics hold as stated, the paper supplies a single, assumption-light procedure that recovers many classical multi-sample tests (variances, correlations, Gini indices, coefficients of variation, etc.) and extends them to arbitrary k ≥ 2 and to moderately growing dimension. The jackknife covariance estimator, the weighted-bootstrap validity proof, and the martingale-CLT argument for the increasing-dimension normal limit are concrete technical contributions that go beyond ad-hoc special cases. The simulation design and the practical guidelines in §4.3 (kd/n thresholds, effective-dimension requirements for ATS-ID, relative computational cost) make the methods usable. The work is therefore a useful consolidation and moderate extension of the U-statistic multi-sample literature.","major_comments":[{"comment":"The increasing-dimension theory (Theorem 2, Proposition 5, Remark 2) is restricted to fixed k and d/n \to 0. The paper correctly flags that joint growth of k and d would require additional tracking of n/n_i factors and moment constants, yet the simulations vary k up to 10 while reporting only d/n. Because the effective dimension of the quadratic form is of order kd, the finite-sample guidance in §4.3 (kd/n ≲ 0.03, kd ≳ 600) should be stated as the primary regime indicator already in the statement of Theorem 2 / Remark 2, not only in the discussion of simulations; otherwise readers may misapply ATS-ID when k is large and d moderate.","section":null},{"comment":"Condition (16) requires a uniform fourth-moment bound on the projected influence functions e_rᵀ D_i h_1(X_i1) that is independent of d. For parameters whose dimension grows by stacking many distinct functionals (e.g., all pairwise correlations or a full covariance matrix), this bound is not automatic and can fail under heavy tails. The manuscript should either supply a verifiable sufficient condition on the original kernels that implies (16) for the leading examples (means + covariances, correlation matrices), or explicitly list the examples for which (16) is known to hold.","section":null}],"minor_comments":[{"comment":"In §2.4 the claim “with this particular choice of C we have H = C” is true for the chosen projection, but the surrounding text first introduces a general C and then specialises; a one-sentence clarification that H coincides with C only for this Kronecker choice would avoid confusion.","section":null},{"comment":"Table 1 (CPU times) reports averages over the three weight distributions for WBS; a short note on the Monte-Carlo size used for the bootstrap p-values would help readers reproduce the timings.","section":null},{"comment":"Several typographical inconsistencies appear (e.g., “ANOV A-type”, “bpImhof”, “estimates as follows”). A careful copy-edit pass is needed.","section":null},{"comment":"The real-data section (§4.2) refers to Tables S9 etc. in the SM; a one-sentence summary of the log-wage versus raw-wage contrast already in the main text would make the application self-contained.","section":null}],"recommendation":"minor_revision","confidential_remarks":"The manuscript is solid and well within the scope of a methods journal. The self-citations to the authors’ related k\to∞ work are appropriate background and do not inflate novelty claims. I see no integrity or scope issues; minor revision is sufficient."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"This is a clean methodological paper that does what it claims. The real addition is a single multi-sample template (jackknife covariances + WTS/ATS) that covers smooth functions of U-statistics, plus a weighted bootstrap for the ATS and a normal limit for the ATS when d grows but d/n\to0. That package unifies a lot of existing two- and multi-sample tests (Gini, correlations, CVs, correlation matrices, etc.) without inventing a new functional class.\n\nWhat works: the asymptotics are standard tools applied carefully—Hoeffding, Delta method, Arvesen-style jackknife consistency, continuous mapping, martingale CLT for the high-d case. Local-power and consistency results are there. Simulations are extensive (univariate variance/Gini, multivariate means+covariances, real CPS data) and the practical guidelines section is actually useful: WTS gets liberal as k or d grows, ATS/WBS stay closer to nominal when kd/n is small, ATS-ID is the computational winner once kd is large enough. The d/n\to0 regime is stated explicitly and used where the proofs need it; they do not overclaim.\n\nSoft spots are real but proportional. The high-d theory keeps k fixed and needs d/n\to0 plus uniform fourth-moment and eigenvalue sandwich conditions; if d grows like n or the projected covariance is badly conditioned the normal approximation can fail. No code is shipped. Self-citations to their own k\to∞ work are background, not circular. None of this sinks the central claims under the stated hypotheses.\n\nThis is for people who need multi-group nonparametric tests on common functionals and want one framework instead of a pile of ad-hoc papers. It deserves a serious referee. I would send it out.","headline":"Solid unification of multi-sample U-statistic equality tests with usable fixed-d and moderate high-d procedures; theory is standard but carefully done, simulations informative.","tokens_in":33177,"tokens_out":451,"would_cite":true,"duration_ms":83631,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["62G10","62H15","62E20"],"pacs":[],"model":"grok-4.5","headline":"A single U-statistic framework tests equality of many parameters across populations, with valid asymptotics even as dimension grows slower than sample size.","keywords":["non-parametric testing","multivariate inference","U-statistics","increasing dimension","jackknife covariance","ANOVA-type statistic","Wald-type statistic","weighted bootstrap"],"falsifier":"Generate independent samples from k populations that truly share the same parameter vector of growing dimension d, with d/n \to 0, compute the centered and scaled ANOVA-type statistic with jackknife covariances, and check whether its empirical distribution is standard normal and its type-I error stays near the nominal level; systematic departure would falsify the increasing-dimension claim.","tokens_in":33344,"feed_emoji":"📊","tokens_out":788,"duration_ms":9183,"temperature":0.7,"pith_summary":"Researchers often need to decide whether several populations share the same variance, correlation, Gini index, or other functional of the data. Existing tests are usually hand-crafted for one parameter and two groups. This paper shows that any parameter that can be written as a smooth function of expectations of symmetric kernels can be handled by the same two quadratic-form statistics, once the jackknife supplies consistent covariance estimates. For fixed dimension the Wald statistic is asymptotically chi-squared and the ANOVA-type statistic is a weighted sum of chi-squares; a weighted bootstrap also approximates the latter. When the parameter dimension d grows with total sample size n but d/n tends to zero, the ANOVA-type statistic, after centering and scaling by traces, becomes standard normal under the null. The result therefore supplies a single, distribution-free procedure that covers classical comparisons and still works when many coordinates are tested at once.","feed_headline":"One U-statistic test covers variances, Gini, correlations","feed_subtitle":"Asymptotically exact multi-sample comparisons even when parameter dimension grows slower than n","key_machinery":"The ANOVA-type quadratic form Q_n = n θ̂ᵀ H θ̂, where H is the orthogonal projection that encodes the equality contrast C = P_k ⊗ I_d. Its null behaviour is controlled by the eigenvalues (or traces) of the projected asymptotic covariance of the stacked U-statistic estimators.","core_discovery":"Under independence of the samples, finite second moments of the kernels, positive first-order variance components, comparable sample sizes, and continuous differentiability of the map f, the ANOVA-type statistic Q_n = n θ̂ᵀ H θ̂ converges under the null to a weighted sum of chi-squares when d is fixed, and, after centering by tr(Σ_n H) and scaling by the square root of 2 tr((Σ_n H)^{2}), converges to N(0,1) when d \to ∞ with d/n \to 0. The same limiting normal law remains valid when the unknown covariance is replaced by its jackknife estimator.","pith_inferences":[],"forward_implications":[],"fun_headline_variants":["Unified U-stat tests for equal params across samples","ANOVA U-stats compare variances Gini correlations multi-group","Tests of equal estimable parameters as dimension grows","Wald ANOVA stats for multi-sample U-parameter equality","Jackknife bootstrap for U-stat multi-pop equality tests"],"cache_read_input_tokens":16512,"weakest_assumption_plain":"The normal approximation for growing dimension requires that the parameter dimension stays much smaller than the total sample size and that the projected influence functions have uniformly bounded fourth moments and non-degenerate eigenvalues; if dimension grows as fast as sample size or the covariance becomes extremely sparse, the normal limit can fail.","fun_headline_variants_meta":{"raw":{"variants":["Unified U-stat tests for equal params across samples","ANOVA U-stats compare variances Gini correlations multi-group","Tests of equal estimable parameters as dimension grows","Wald ANOVA stats for multi-sample U-parameter equality","Jackknife bootstrap for U-stat multi-pop equality tests"]},"model":"grok-4.5","effort":"low","cost_usd":0.003528,"raw_usage":{"total_tokens":1172,"prompt_tokens":780,"num_sources_used":0,"completion_tokens":63,"cost_in_usd_ticks":35280000,"prompt_tokens_details":{"text_tokens":780,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":329,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":780,"tokens_out":63,"duration_ms":4596,"temperature":1.0,"reasoning_tokens":329,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-10T18:35:25.181052+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Generate independent samples from k populations that truly share the same parameter vector of growing dimension d, with d/n \to 0, compute the centered and scaled ANOVA-type statistic with jackknife covariances, and check whether its empirical distribution is standard normal and its type-I error stays near the nominal level; systematic departure would falsify the increasing-dimension claim.","supporting_citations":[],"review_version":2}