betaFeedback

WorksMathematics

Uniform inference for UU-statistics of exchangeable arrays under degeneracy

Abstract

Many UU-statistics of row-column exchangeable arrays become degenerate at natural null hypotheses: their convergence rates change and their limiting laws may or may not remain Gaussian. We establish conditions under which inference based on the classical unbiased variance estimator is uniformly valid across these changes, without knowledge of the degeneracy order. It is known that UU-statistics can be decomposed into orthogonal projections indexed by bipartite graphs. Under uniform fourth-moment assumptions, we establish that the standardized statistic is asymptotically Gaussian and the classical unbiased variance estimator is ratio-consistent, as long as a certain measure of concentration for projections indexed by disconnected graphs tends to zero. Under compactness of the normalized projections, both properties hold if and only if the proportion of variance contributed by disconnected graphs tends to zero. The necessity of this condition follows from a general limit theorem for these UU-statistics, in which each graph contributes to a Wiener chaos of order equal to its number of connected components. We note that the row-column jackknife variance estimator, although nonnegative, does not generally give a studentized statistic with a standard normal limit in degenerate regimes. We illustrate our theory with an application to estimation and testing based on a distance from product form, exhibiting Gaussian and non-Gaussian limits at various rates depending on the model. We also propose a conservative inference procedure that does not require a Gaussian limit, valid under additional moment and projection conditions.

Introduction

A UU-statistic on a two-way array (Yij)i≥1,j≥1(Y_{ij})_{i\geq1,j\geq1} is a statistic of the form

UN=(mp)−1(nq)−1∑WHW,HW:=h(YW),(1)U_N = \binom{m}{p}^{-1}\binom{n}{q}^{-1}\sum_W H_W,\qquad H_W := h(Y_W), \tag*{(1)}

where the sum is over the pairs W=(W1,W2)W=(W_1,W_2) of a row set W1⊆[m]W_1\subseteq[m] and a column set W2⊆[n]W_2\subseteq[n], with ∣W1∣=p|W_1|=p and ∣W2∣=q|W_2|=q, YWY_W is the subarray induced by the indices of WW, and the kernel hh is a function of a p×qp\times q subarray, symmetric under separate permutations of the rows and columns. We assume that the array is row-column exchangeable (RCE), that means, its law is invariant under separate permutations of its rows and its columns, and dissociated, that means, subarrays sharing neither rows nor columns are independent. The study of RCE matrices is motivated by bipartite network analysis (for example, recommender systems [40], scientific authorship networks [31], ecological pollination networks [11]), where the RCE property implies vertex-exchangeability.

UU-statistics are a large class of statistics on these arrays, yet, for a wide range of null models, UU-statistics are degenerate in that their first-order Hoeffding projection is zero [26]. In this case, first-order variance estimators are no longer suitable to describe the actual variance of the UU-statistics, the rate of convergence might be different, and the limiting distribution might be Gaussian or non-Gaussian depending on the situation [25]. Thus, a test that is asymptotically valid at each fixed law does not necessarily remain valid when the law approaches such a null hypothesis. This paper characterizes the region under which Gaussian inference, that is, tests and confidence intervals based on Gaussian critical values, is uniformly valid, and shows that there is a boundary beyond which it fails.

Contributions

Main result: Uniform inference and its boundary. We establish conditions under which Gaussian confidence intervals based on the classical unbiased variance estimator is uniformly valid across degeneracy regimes. More precisely, under uniform fourth-moment bounds and a condition on the graph projections, Theorem 1 states that the standardized U-statistic is asymptotically Gaussian, and the classic unbiased estimator of variance is ratio-consistent. Assuming compactness of the normalized projections, a necessary-and-sufficient condition is derived: the two properties are satisfied if and only if the fraction of variance due to disconnected graphs is asymptotically vanishing (Thm. 3 and Cor. 4).

Limit distributions underlying the boundary. To determine when the standardized statistic can have a Gaussian limit, we establish a general limit theorem under square integrability: each principal support graph contributes to a Wiener chaos of order equal to its number of connected components (Thm. 6). This is the RCE analogue of Janson and Nowicki [20] and resolves the multi-component fixed-law case left open in Le Minh [25]. Under compactness, a nonvanishing contribution from disconnected graphs then prevents the standardized statistic from being asymptotically Gaussian (Thm. 3).

Jackknife bias and correction. Even when the statistic is asymptotically Gaussian, the row-column jackknife does not generally yield valid Gaussian inference because it is asymptotically biased. We show that its expectation reweights each variance contribution by a factor tending to its number of vertices, irrespective of its edges. Then, we establish an identity showing that the classical unbiased variance estimator can be interpreted as a bias-corrected jackknife that removes these excess weights exactly.

Conservative inference without a Gaussian limit. At a null hypothesis with a known structure, the conditions for Gaussian inference can be verified analytically, but they do not hold in general and cannot be verified from a single observed array. We therefore also construct a conservative test using independent diagonal blocks. Under uniform fourth-moment bounds, this test remains valid even when these conditions fail, although its confidence intervals are wider. Under additional moment and projection conditions, a statistic computed from the data determines whether to use this conservative test or the Gaussian test. For any fixed α∈(0,1)\alpha\in(0, 1), the resulting procedure has a rejection probability asymptotically at most α\alpha, uniformly over the null class, and selects the Gaussian test when the quantity used for the diagnostic tends to zero sufficiently quickly.

Applications across degeneracy regimes. We apply the theory to estimation and testing based on a measure of departure from the product-form model, where the expected array entries can be factorized into a row effect and a column effect. Assuming conditionally Poisson entries, we use a ratio statistic that is degenerate under the product-form null hypothesis but has a Gaussian limit. Inference near the null hypothesis requires both the uniform theorem and a nonlinear remainder negligible compared with the standard deviation. We also consider two variations of this problem: a departure from product form that remains degenerate but has a non-Gaussian limit, and a statistic allowing Gaussian inference without the Poisson assumption, at a different rate. Table 1 summarizes these regimes: neither the convergence rate nor the degeneracy order determines whether the limit is Gaussian. With our theoretical results, Gaussian inference based on the unbiased variance estimator applies in all Gaussian regimes using the corresponding kernel or linearization. In simulation studies, we analyze the coverage and rejection probabilities of the Gaussian test and the conservative procedures across degeneracy regimes. In an application to UN General Assembly roll calls, we assess whether additive effects can account for the departures from product form detected by our test, by fitting the classical Rasch model ([34, 7]).

RegimePrincipal support graphsLimitSD order
Nondegenerate departure∘ / □\circ\ / \ \squareGaussianN−1/2N^{-1/2}
Product form∘□\circ\squareGaussianN−1N^{-1}
Two-group model∘□ / ∘∘ / □□ / ∘□\circ\square\ / \ \circ\circ\ / \ \square\square\ / \ \circ\squareNon-GaussianN−1N^{-1}
Product form, general cells (Ψ\Psi)∘□∘□\begin{matrix}\circ& \square\\ \circ& \square\end{matrix}GaussianN−2N^{-2}

Table 1. Product-form regimes at fixed laws with m,n≍Nm,n \asymp N, under the conditions of § 6. Circles denote rows and squares denote columns.

Relation to earlier work

Our starting point is the theory of RCE UU-statistics developed by Le Minh [24], which studied nondegenerate array UU-statistics, Le Minh et al. [26], which developed a coarse decomposition and first-order variance estimation, and Le Minh [25], which supplied the bipartite graph decomposition and Gaussian fixed-law limits. We address uniform inference over changing laws, including changes in degeneracy, and establish the general fixed-law limit theorem to determine its boundary.

On i.i.d. observations, degenerate UU-statistics have a non-Gaussian limiting law which lives in a higher Wiener chaos [35]. For generalized UU-statistics, Janson and Nowicki [20] showed that their Hoeffding decomposition can be indexed by graphs, and that, under degeneracy, connected principal support graphs can still produce Gaussian limits. They also obtained Wick-product limits for disconnected principal support graphs. In jointly exchangeable arrays, similar limits arise for subgraph counts [22, 2].

Unbiased variance estimators have been studied for UU-statistics of i.i.d. observations [36, 38] and in multi-reader multi-case inference [14, 15]. Fuchs et al. [13] and Xu et al. (2024) stated consistency results that cover nondegenerate regimes. Our results establish the ratio-consistency of the unbiased variance estimator for general kernels on two-way arrays, even under degeneracy.

Davezies et al. [8] developed empirical-process results for RCE arrays, notably with a chaos limiting law for degenerate dyadic processes. For dyadic density estimation, Graham et al. [16] established Gaussian inference across degenerate and nondegenerate cases. Cattaneo et al. [4] constructed uniform confidence bands under degeneracy. Here, we study RCE UU-statistics with general kernels, and establish conditions for the standardized statistic to be asymptotically Gaussian and the variance estimator to be ratio-consistent, uniformly over changing laws, which includes potential changes in degeneracy regimes.

For two-way means, Menzel [29] established uniform bootstrap guarantees and a conservative inference procedure. Chiang et al. [6], Davezies et al. [9] and Hounyo and Lin [18] developed similar results for means, GMMs, and quantile regression. However, general array kernels introduce higher-order connected and disconnected graph contributions that do not arise in the two-way mean. Chatterjee et al. [5] developed a multiplier bootstrap and joint confidence sets for network moments in jointly exchangeable graphs, allowing both Gaussian and non-Gaussian limits.

Our analysis of the behavior of the row-column jackknife also relates to earlier findings that resampling can overcount variance components in crossed data [28, 32, 33], and to work on two-way cluster jackknives [27]. We give the exact component weights responsible for this behavior, with a graph-theoretic interpretation.

Outline

Section § 2 introduces the general framework. Section § 3 establishes uniform Gaussian inference and its boundary. Section § 4 discusses the jackknife. Section § 5 gives conservative procedures, combining them with Gaussian inference. Section § 6 develops the product-form application, and Section § 7 presents simulations and the UN roll-call illustration. The supplementary material contains the proofs and additional numerical results.

Notation and convention

→d\xrightarrow{d} denotes convergence in distribution, and OPO_P, oPo_P are stochastic orders. When a statement is made uniformly over a class of laws, both are understood along every sequence of laws in the class. For sequences, aN≍bNa_N \asymp b_N means that aN/bNa_N/b_N is bounded away from 00 and ∞\infty. We write Φ\Phi for the standard normal distribution function and zu=Φ−1(u)z_u = \Phi^{-1}(u) for its uu-quantile. For graphs, ⊔\sqcup denotes the disjoint union, Aut⁡(G)\operatorname{Aut}(G) the automorphism group of GG preserving the row and column vertex sets, and ∑Gdisc⁡\sum_{G}^{\operatorname{disc}} and ∑Gconn⁡\sum_G^{\operatorname{conn}} are sums over disconnected and connected classes (definition in § 2.2). By convention, we set sup⁡∅:=0\sup\varnothing:= 0. We call a test with rejection event RNR_N uniformly valid at level α\alpha on a null class P0,N\mathcal{P}_{0,N} if lim sup⁡Nsup⁡P∈P0,NPP(RN)≤α\limsup_N \sup_{P\in\mathcal{P}_{0,N}} \mathbb{P}_P(R_N) \leq\alpha. Analogously, a confidence interval INI_N for θN,P\theta_{N,P} is uniformly valid at level 1−α1-\alpha over PN\mathcal{P}_N if lim inf⁡Ninf⁡P∈PNPP{θN,P∈IN}≥1−α\liminf_N \inf_{P\in\mathcal{P}_N} \mathbb{P}_P\{\theta_{N,P} \in I_N\} \geq1-\alpha.

Framework and preliminaries

Row-column exchangeable arrays with drifting laws

Let Y=(Yij)i≤m, j≤nY = (Y_{ij})_{i\le m,\,j\le n} be the upper-left block of a dissociated RCE array. By the Aldous–Hoover–Kallenberg (AHK) representation [1, 17, 21],

Yij=a.s.ϕ(ξi,ηj,ζij),Y_{ij} \mathrel{\overset{\mathrm{a.s.}}{=}} \phi(\xi_i,\eta_j,\zeta_{ij}),

with (ξi)i≥1(\xi_i)_{i\ge1}, (ηj)j≥1(\eta_j)_{j\ge1}, (ζij)i≥1,j≥1(\zeta_{ij})_{i\ge1,j\ge1} independent families of i.i.d. U[0,1]\mathcal{U}[0,1] variables (the AHK variables). Given the row and column variables, the cells are independent. When E∣Y11∣<∞\mathbb{E}|Y_{11}| < \infty, we call w(x,y)=E[Yij∣ξi=x,ηj=y]w(x,y) = \mathbb{E}[Y_{ij} \mid\xi_i=x,\eta_j=y] the mean graphon of the array. Throughout, m=mNm=m_N, n=nNn=n_N with mN+nN=Nm_N+n_N=N and mN/N→ρ∈(0,1)m_N/N\to\rho\in(0,1), so m,n≍Nm,n\asymp N. We also define LN=min⁡(m,n)L_N=\min(m,n).

We allow the law of the array to change with NN by replacing ϕ\phi with ϕN\phi_N, while the AHK variables remain independent and uniformly distributed. Classes of laws can also be fixed or depend on NN. In notation, we omit the indices NN or PP indicating dependence when there is no ambiguity.

The graph-indexed Hoeffding decomposition

Write θN=E[UN]\theta_N=\mathbb{E}[U_N] and σN2=Var⁡(UN)\sigma_N^2=\operatorname{Var}(U_N), where UNU_N is defined in (1). Since hh is symmetric, the same statistic is obtained by averaging over ordered tuples of distinct rows and columns. We use this when counting graph copies, with (x)k=x(x−1)⋯(x−k+1)(x)_k=x(x-1)\cdots(x-k+1).

Graphs. A bipartite graph GG has row vertices V1(G)V_1(G), column vertices V2(G)V_2(G) and edges E(G)⊆V1(G)×V2(G)E(G)\subseteq V_1(G)\times V_2(G). Isolated vertices are allowed. We write r(G)r(G) and c(G)c(G) for its numbers of row and column vertices, d(G)=r(G)+c(G)d(G)=r(G)+c(G) for its total number of vertices, and b(G)b(G) for its number of connected components. A subgraph of GG, written F⊆GF\subseteq G, is a bipartite graph FF with V1(F)⊆V1(G)V_1(F)\subseteq V_1(G), V2(F)⊆V2(G)V_2(F)\subseteq V_2(G) and E(F)⊆E(G)E(F)\subseteq E(G). It is proper, written F⊊GF\subsetneq G, when F≠GF\ne G.

The AHK variables associated with GG are

References

  1. [1]Aldous, D. J. (1981). Representations for partially exchangeable arrays of random variables. Journal of Multivariate Analysis, 11(4):581–598.DOI
  2. [2]Bhattacharya, B. B., Chatterjee, A., and Janson, S. (2023). Fluctuations of subgraph counts in graphon based random graphs. Combinatorics, Probability and Computing, 32:428–464.arxiv.org/abs/2104.07259
  3. [3]Billingsley, P. (1999). Convergence of Probability Measures. Wiley, New York, 2 edition.
  4. [4]Cattaneo, M. D., Feng, Y., and Underwood, W. G. (2024). Uniform inference for kernel density estimators with dyadic data. Journal of the American Statistical Association, 119(548):2695–2708.arxiv.org/abs/2201.05967
  5. [5]Chatterjee, A., Dan, S., and Bhattacharya, B. B. (2026). Higher-order graphon theory: fluctuations, degeneracies, and inference. The Annals of Statistics, 54(4):2136–2163.arxiv.org/abs/2404.13822
  6. [6]Chiang, H. D., Matsushita, Y., and Otsu, T. (2025). Multiway empirical likelihood. Journal of Econometrics, 249:105861.
  7. [7]Clinton, J., Jackman, S., and Rivers, D. (2004). The statistical analysis of roll call data. American Political Science Review, 98(2):355–370.DOI
  8. [8]Davezies, L., D’Haultfœuille, X., and Guyonvarch, Y. (2021). Empirical process results for exchangeable arrays. The Annals of Statistics, 49(2):845–862.DOI
  9. [9]Davezies, L., D’Haultfœuille, X., and Guyonvarch, Y. (2025). Analytic inference with two-way clustering. arXiv preprint arXiv:2506.20749.arxiv.org/abs/2506.20749
  10. [10]Döbler, C. and Peccati, G. (2017). Quantitative de Jong theorems in any dimension. Electronic Journal of Probability, 22:1–35.arxiv.org/abs/1603.00804
  11. [11]Dormann, C. F., Fründ, J., Blüthgen, N., and Gruber, B. (2009). Indices, graphs and null models: analyzing bipartite ecological networks. The Open Ecology Journal, 2(1).DOI
  12. [12]Efron, B. and Stein, C. (1981). The jackknife estimate of variance. The Annals of Statistics, 9(3):586–596.DOI
  13. [13]Fuchs, M., Hornung, R., Boulesteix, A.-L., and De Bin, R. (2020). On the asymptotic behaviour of the variance estimator of a U-statistic. Journal of Statistical Planning and Inference, 209:101–111.DOI
  14. [14]Gallas, B. D. (2006). One-shot estimate of MRMC variance: AUC. Academic Radiology, 13(3):353–362.pubmed.ncbi.nlm.nih.gov/16488848
  15. [15]Gallas, B. D., Bandos, A., Samuelson, F. W., and Wagner, R. F. (2009). A framework for random-effects ROC analysis: biases with the bootstrap and other variance estimators. Communications in Statistics – Theory and Methods, 38(15):2586–2603.DOI
  16. [16]Graham, B. S., Niu, F., and Powell, J. L. (2024). Kernel density estimation for undirected dyadic data. Journal of Econometrics, 240(2):105336.
  17. [17]Hoover, D. N. (1979). Relations on probability spaces and arrays of random variables. Preprint, Institute for Advanced Study, Princeton, NJ, 2:275.
  18. [18]Hounyo, U. and Lin, J. (2026). A two-way cluster-robust variance estimator for quantile regression. arXiv preprint arXiv:2602.16376.
  19. [19]Janson, S. (1997). Gaussian Hilbert Spaces, volume 129 of Cambridge Tracts in Mathematics. Cambridge University Press, Cambridge.
  20. [20]Janson, S. and Nowicki, K. (1991). The asymptotic distributions of generalized U-statistics with applications to random graphs. Probability Theory and Related Fields, 90:341–375.DOI
  21. [21]Kallenberg, O. (2005). Probabilistic Symmetries and Invariance Principles, volume 9. Springer.
  22. [22]Kaur, G. and Röllin, A. (2021). Higher-order fluctuations in dense random graph models. Electronic Journal of Probability, 26.arxiv.org/abs/2006.15805
  23. [23]Le Minh, T. (2023a). U-statistics of row-column exchangeable matrices: application to ecological network analysis. PhD thesis, Université Paris-Saclay. NNT 2023UPASM027, tel-04321993.DOI
  24. [24]Le Minh, T. (2023b). U-statistics on bipartite exchangeable networks. ESAIM: Probability and Statistics, 27:576–620.
  25. [25]Le Minh, T. (2026). Characterization of the asymptotic behaviour of U-statistics on row-column exchangeable matrices. Information and Inference: A Journal of the IMA, 15:iaag001.DOI
  26. [26]Le Minh, T., Donnet, S., Massol, F., and Robin, S. (2025). Hoeffding-type decomposition for U-statistics on bipartite networks. Electronic Journal of Statistics, 19:2829–2875.arxiv.org/abs/2308.14518
  27. [27]MacKinnon, J. G., Nielsen, M. Ø., and Webb, M. D. (2024). Jackknife inference with two-way clustering. arXiv preprint arXiv:2406.08880.arxiv.org/abs/2406.08880
  28. [28]McCullagh, P. (2000). Resampling and exchangeable arrays. Bernoulli, 6(2):285–301.DOI
  29. [29]Menzel, K. (2021). Bootstrap with cluster-dependence in two or more dimensions. Econometrica, 89(5):2143–2188.DOI
  30. [30]Michalska-Smith, M. J. and Allesina, S. (2019). Telling ecological networks apart by their structure: a computational challenge. PLoS Computational Biology, 15(6):e1007076.
  31. [31]Newman, M. E. J. (2001). The structure of scientific collaboration networks. Proceedings of the National Academy of Sciences, 98(2):404–409.
  32. [32]Owen, A. B. (2007). The pigeonhole bootstrap. The Annals of Applied Statistics, 1(2):386–411.DOI
  33. [33]Owen, A. B. and Eckles, D. (2012). Bootstrapping data arrays of arbitrary order. The Annals of Applied Statistics, 6(3):895–927.
  34. [34]Rasch, G. (1960). Probabilistic Models for Some Intelligence and Attainment Tests. Danmarks Paedagogiske Institut, Copenhagen.
  35. [35]Rubin, H. and Vitale, R. A. (1980). Asymptotic distribution of symmetric statistics. The Annals of Statistics, 8(1):165–170.DOI
  36. [36]Schucany, W. R. and Bankson, D. M. (1989). Small sample variance estimators for U-statistics. Australian Journal of Statistics, 31:417–426.DOI
  37. [37]Voeten, E. (2013). Data and analyses of voting in the United Nations General Assembly. In Reinalda, B., editor, Routledge Handbook of International Organization. Routledge.
  38. [38]Wang, Q. and Lindsay, B. G. (2014). Variance estimation of a general U-statistic with application to cross-validation. Statistica Sinica, 24(3):1117–1141.DOI
  39. [40]Zhou, T., Ren, J., Medo, M., and Zhang, Y.-C. (2007). Bipartite network projection and personal recommendation. Physical Review E, 76(4):046115.pubmed.ncbi.nlm.nih.gov/17995068

Paper details

vlnc:W_3dsnnun4
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY-SA 4.0
Usage

views · 2

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

ChatGPT, Claude

View statement
Looking for related papers…
Contents