Pith. sign in

REVIEW 1 major objections 4 minor 60 references

Consistent Infill Estimability of the Regression Slope Between Gaussian Random Fields Under Spatial Confounding

T0 review · 1 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read In spatial regression with an unmeasured confounder, the slope is consistently estimable unless the exposure is more than d/2 degrees smoother than the confounder; when estimable, local differencing estimators recover it without knowing…

desk verdict Sufficient theory and the new multivariate equivalence result are solid, but the paper's central sharp 'only if' for correlated confounders is only proved for independent X and W. read the letter →

arxiv 2506.09267 v3 pith:4IJJYGMK submitted 2025-06-10 math.ST stat.TH

Abhirup Datta , Michael L. Stein This is my paper · ORCID
classification math.STstat.TH MSC 62M3060G6062F1262G20
keywords spatialconfoundingGaussianrandomfieldsinfillasymptoticsconsistentestimabilityprincipalirregulartermMatérncovariancelocaldifferencingspectralequivalence
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper asks a basic question that the spatial confounding literature has only answered empirically or estimator-by-estimator: if the full processes $Y$ and $X$ were observed on a fixed spatial domain, with an unmeasured process $W$ entering $Y = X\beta + W$ and correlated with $X$, when does any estimator of the slope $\beta$ exist? The answer given is a smoothness comparison: $\beta$ is consistently estimable under infill asymptotics when the exposure is not too smooth relative to the confounder and the cross-covariance is smoother than the exposure covariance. The claim matters because it turns a widely debated bias problem into a checkable condition on local behavior of covariances, and because the proof supplies explicit non-parametric estimators—discrete differences or Laplacians of suitable order—that need no covariance parameters. For the Matérn, power exponential, generalized Cauchy, and coregionalization families the paper shows the condition is sharp, with a $d/2$ smoothness gap in $d$ dimensions; beyond that gap no estimator can be consistent, even in principle. A by-product is a new spectral criterion for equivalence of multivariate Gaussian random fields whose components have different smoothness.

What carries the argument

The load-bearing object is the principal irregular term of a covariance function: the leading non-analytic term $c\,t^{\alpha}$ in the near-zero expansion, whose exponent $\alpha$ measures the smoothness of the process. Under the estimability conditions, the slope $\beta$ appears as the ratio of the principal irregular coefficients of the cross-covariance and of the covariance of $X$. The constructive estimators use local differencing or discrete Laplacians of order high enough to annihilate the analytic part of the covariance, leaving the principal irregular term to dominate; the OLS ratio of differenced $Y$ to differenced $X$ then converges to $\beta$ through variance calculations built on stationarity. For the impossibility direction, the machinery is spectral equivalence in multivariate Paley-Wiener spaces: comparing the spectral density matrices of $(X,Y)$ under two values of $\beta$, the paper proves a sufficient condition for equivalence of multivariate Gaussian measures with component fields of different smoothness, which yields the $\int f_X/f_W<\infty$ criterion.

What would settle it

Simulate a bivariate Matérn field with $\nu_X>\nu_W+d/2$ and nonzero cross-correlation between $X$ and $W$, observe $(X,Y)$ on a regular grid with growing density, and compute the likelihood ratio or the maximum likelihood estimate of $\beta$ for two different slope values; if the likelihood separates or the estimator concentrates as the grid densifies, the claimed non-estimability regime fails for correlated confounding.

Watch

Extended reading notes

Core claim

The paper's central claim is that for $Y(s)=X(s)\beta+W(s)$ with $(X,W)$ a bivariate Gaussian random field on a fixed domain in $\mathbb{R}^d$, the slope $\beta$ is consistently estimable from the paths of $(X,Y)$ exactly when the exposure is not too smooth relative to the confounder. Writing $\alpha_{11}$, $\alpha_{22}$, and $\alpha_{12}$ for the exponents of the principal irregular terms of the covariance of $X$, of $W$, and of the cross-covariance, the sufficient conditions are $\alpha_{11}<\alpha_{12}$ and $\alpha_{11}<\alpha_{22}+d$; in that regime $\beta$ equals the limiting ratio of the principal irregular coefficients and is recovered by regressing $p$-th order differences ($d=1$) or $m$-th order discrete Laplacians ($d>1$) of $Y$ on those of $X$. The necessary condition is spectral: when $X$ and $W$ are independent and $\int f_X/f_W<\infty$, the measures on $(X,Y)$ for two different $\beta$ values are equivalent, so no consistent estimator exists even with complete observation. For Matérn, power exponential, generalized Cauchy, and coregionalization covariances the sufficient and necessary conditions coincide except at a boundary, giving a sharp characterization: $\beta$ is consistently estimable if $\nu_X<\nu_W+d/2$ and $\nu_{XW}>\nu_X$, and not estimable if $\nu_X>\nu_W+d/2$.

Load-bearing premise

The impossibility half of the sharp Matérn characterization is proved only for the special case where $X$ and $W$ are independent; the paper relies on the unstated premise that adding cross-correlation between $X$ and $W$ cannot make two different slopes distinguishable when they are already indistinguishable under independence.

Editorial extensions

If this is right

  • If the sharp characterization holds, consistent estimation of $\beta$ requires no knowledge of the covariance family or its parameters: only the order of differencing or Laplacian chosen from the smoothness of $X$.
  • Under spatial confounding, $\beta$ is consistently estimable even when the exposure is up to $d/2$ degrees smoother than the confounder, not only when it is rougher.
  • For Matérn, power exponential, generalized Cauchy, and coregionalization processes, $\nu_X>\nu_W+d/2$ is an impossibility boundary: no estimator based on the observed $(X,Y)$ paths on the fixed domain can be consistent.
  • Measurement error does not change the estimability boundary; with noise, a local-averaging-then-differencing estimator is consistent under the same smoothness conditions.
  • With multivariate exposures, the condition must hold for every nonzero linear combination of the exposures, which restricts how collinear or smooth the exposure components may be relative to the confounder.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the paper's unstated premise survives—that cross-correlation cannot restore identifiability where independence already prevents it—the results imply a practical diagnostic: estimate the local smoothness exponents of $X$ and $Y$ from data, and when $\alpha_{11}\ge\alpha_{22}+d$ abandon slope estimation on these variables alone and look for instruments or additional covariates.
  • The new multivariate spectral equivalence theorem is a transferable tool; it should apply to identifiability of cross-covariance parameters and, as the authors conjecture, to Gaussian fields on manifolds, where the dimension gap $d/2$ reappears through eigenvalue growth.
  • The boundary $\nu_X=\nu_W+d/2$ is where the interesting rates live: the paper shows consistency survives on the boundary in one dimension only at a logarithmic rate and only for integer smoothness, so higher-dimensional boundary cases likely need estimators beyond plain differencing or Laplacians.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

1 major / 4 minor

Summary. The paper studies infill (fixed-domain) consistent estimability of the slope β in the spatial regression Y(s) = X(s)β + W(s), where X and W are correlated Gaussian random fields and W is unobserved. The central claim is an 'if and only if' characterization: β is consistently estimable if and only if the exposure is not too smooth relative to the confounder, roughly α_X < α_W + d, together with a condition that the cross-covariance is smoother than the covariance of X (α_XW > α_X). The sufficient direction is established via explicit local-differencing or discrete-Laplacian OLS estimators, whose consistency is proved without parametric covariance knowledge. The necessary direction is approached through a new multivariate Paley-Wiener equivalence theorem, yielding the spectral condition ∫ f_X/f_W < ∞ as implying non-estimability. For Matern, power exponential, generalized Cauchy, and coregionalization families, the paper claims the sufficient and necessary conditions coincide (except at a boundary), giving a sharp characterization. The paper also treats multivariate exposures, measurement error, some non-stationary and non-Gaussian processes, irregular designs, and provides simulations that match the predicted rates.

Significance. If the characterization is correct, this is a substantial and long-needed contribution to the spatial confounding literature. The paper moves beyond estimator-specific analyses and gives a fundamental information-theoretic boundary for when the slope can be recovered at all, even if the entire bivariate field were observed. The main strengths are: (i) explicit, parameter-free estimators (differencing and Laplacian OLS) that do not require fitting covariance parameters; (ii) a genuinely new sufficient condition for equivalence of multivariate Gaussian measures with components of unequal smoothness (Theorem 4), which is of independent interest; (iii) detailed proofs of the variance rates, with simulations that empirically confirm the predicted convergence rates; and (iv) a broad set of extensions (multivariate X, measurement error, non-stationarity, heavy tails, irregular designs). The central weakness is that the necessity direction is proved only for the independent submodel X ⊥ W, whereas the headline claim concerns the spatially confounded setting where X and W are correlated.

major comments (1)
  1. [§S3.3, proof of Corollary 1(b); §4.3, Theorem 5] The necessity direction for the correlated case is not established. Corollary 1(b) states that for a non-degenerate bivariate Matern field with ν_X > ν_W + d/2, β is not consistently estimable by any estimator. The proof in Section S3.3 explicitly considers only the two parameter choices (β=0, ρ_XW=0) and (β=1, ρ_XW=0), and then invokes Theorem 5, which is stated in the main text for 'independent stationary GRFs' X and W. Thus the supplied argument proves non-estimability only on the diagonal submodel where X and W are independent. The same gap appears in Corollary 3(b) (power exponential), Corollary 4(b) (generalized Cauchy, d=1), Corollaries 7(b) and 8(b), and the second (non-estimability) part of Theorem 6. This is load-bearing because the paper's abstract and Section 1 advertise an if-and-only-if characterization for spatial confounding, which is precisely the setting where W is correlated with X. If adding cross-correlation can make different β values distinguishable in the ν_X > ν_W + d/2 regime (e.g., through a sufficiently smooth cross-covariance), then part (b) is false as stated; if it cannot, the result may be true but the proof as written does not support it. The manuscript should either extend the equivalence argument to the correlated case (for example, by proving a bivariate spectral condition involving the cross-spectrum, or by a coupling argument showing that equivalence under independence implies equivalence under any valid cross-correlation) or explicitly restrict the non-estimability claims to independent X and W, which would substantially weaken the central claim.
minor comments (4)
  1. [§4.3, after Theorem 5] The text says 'generalized Cauchy or powered exponential'; 'powered exponential' should be 'power exponential' for consistency with Section 5.2.
  2. [§4.2, proof of Theorem 4] In the case analysis, the case where F^{(0)}(ω) ≥ F^{(1)}(ω) pointwise is not explicitly treated; it follows by symmetry but should be stated to avoid a gap in the reader's eyes.
  3. [§5.1, Corollary 2 and the following paragraph] The boundary case ν_X = ν_W + d/2 is only partially resolved (d=1, integer ν_X) and the general case is left as a conjecture. This is fine, but the wording 'sharp characterization' in the abstract and Section 5 should be tempered to 'sharp except at the boundary' whenever referencing the boundary.
  4. [§6, Theorem 6 and Figure 2] The local-averaging-and-differencing estimator in Theorem 6 depends on the choice of the averaging block size (indexed by ρ in the proof), but the statement does not tell the reader how to choose it in practice; a short remark would help.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: beta is derived as a ratio of principal irregular coefficients and estimated by direct differencing; necessity rests on external measure-equivalence arguments.

full rationale

The paper's sufficient-side derivation is self-contained and non-circular. From Y = X beta + W with alpha_12 > alpha_11, the leading irregular term of Cov(Y(s+h), X(s)) is beta c_11 h^alpha_11, so beta is literally the limiting ratio of the leading irregular coefficients; the OLS-of-differences estimator converges to that ratio through explicit variance calculations, not through fitted covariance parameters. The necessity side (Theorem 5 and its corollaries) is an external impossibility argument: it uses Paley-Wiener spaces and spectral equivalence of Gaussian measures, with Matern spectral bounds from [48] and [3], and does not quote the paper's own estimator as evidence. Self-citations by the authors (e.g., [14] and Stein's book [38]) are contextual or standard references and are not load-bearing in the proof chain. The concern raised about Corollary 1(b) — that the proof shows non-estimability only for rho_XW = 0 even though the theorem is stated for general bivariate Matern fields — is a possible proof gap or correctness issue about the correlated case, not a circularity: nothing in that step reduces the conclusion to an input assumption by construction. No fitted quantity is renamed as a prediction, and no uniqueness result is imported solely from the authors' prior work.

Assumptions & free parameters 2 free parameters · 6 assumptions · 0 invented entities

The paper introduces no new particles, forces, or latent entities. Its axiomatic burden is the assumed form of the covariance functions (a single principal irregular term) and the spectral comparability with Paley-Wiener functions. The free parameters are the differencing order and the averaging block size, both chosen from the smoothness exponent rather than fitted to the target beta.

free parameters (2)
  • differencing order p (or Laplacian order m) = p integer with alpha_11 < 2p in d=1; m integer with alpha_11 < 4m in d>1
    Consistency of the proposed estimator requires choosing p or m larger than half the smoothness exponent alpha_11. In practice this is estimated from data, so the estimator is not fully adaptive in the theory.
  • averaging block exponent rho in Theorem 6 = rho > max(0, alpha_11 - 1), and rho > max(0, alpha_11) when X also has noise
    The local-averaging-and-differencing estimator for measurement error requires the averaging neighborhood to grow at a rate tied to the unknown alpha_11; this is a hand-chosen quantity depending on model smoothness.
assumptions (6)
  • domain assumption Gaussianity and stationarity of (X, W) in the main theorems
    The measure-equivalence and differencing proofs rely on Gaussian calculus such as Isserlis's theorem and on stationarity; extensions relax some of these but not the core characterization.
  • domain assumption Covariance decomposition Assumption 1/2: K(t) = A(t) + B(t) with A analytic and B(t) = c t^alpha + o(t^alpha), alpha not an even integer
    This restricts the local behavior to a single leading irregular term. It covers Matern, power exponential, generalized Cauchy, and coregionalization families, but excludes more exotic local behavior.
  • domain assumption Spectral density regularity condition (16): f_X and f_W are comparable to squares of Paley-Wiener functions
    The necessary condition in Theorem 5 requires this comparability. It fails for analytic processes such as squared exponential, which the paper explicitly excludes.
  • domain assumption Fixed-domain infill asymptotics and regular lattices
    All consistency statements are under n growing within a fixed bounded domain. Irregular designs are treated only in supplementary Theorem S1 under quasi-uniform spacing assumptions.
  • standard math Existence of phi_X and phi_W satisfying (S21) via Zastavnyi and Lemma A.4 of Bachoc et al.
    External results are used to verify the spectral comparability condition for Matern covariance functions; the paper relies on these without reproving them.
  • domain assumption Non-degeneracy and alpha_12 > alpha_11 for the sufficient direction
    The paper assumes the cross-covariance between X and W is smoother than the covariance of X; otherwise the proposed differencing estimators need not converge to beta.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Consistent Infill Estimability of the Regression Slope Between Gaussian Random Fields Under Spatial Confounding." pith.science (2026). https://pith.science/paper/4IJJYGMK

@misc{pith2026250609267,
  author       = {Pith},
  title        = {Pith review of: Consistent Infill Estimability of the Regression Slope Between Gaussian Random Fields Under Spatial Confounding},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4IJJYGMK}},
  note         = {Machine review of arXiv:2506.09267}
}
read the original abstract

The problem of estimating the slope parameter in regression between two spatial processes under confounding by an unmeasured spatial process has received widespread attention in the recent statistical literature. Yet, a fundamental question remains unresolved: when is this slope consistently estimable under spatial confounding, with existing insights being largely empirical or estimator-specific. We characterize conditions for consistent estimability of the regression slope between Gaussian random fields (GRFs), the common stochastic model for spatial processes, under spatial confounding. Under fixed-domain (infill) asymptotics, we give sufficient conditions for consistent estimability in terms of the smoothness or local behavior of the exposure and confounder processes. When estimability holds, we provide consistent estimators of the slope using local differencing (taking discrete differences or Laplacians of the processes of suitable order). Using functional analysis results on Paley-Wiener spaces, we then provide an easy-to-verify necessary condition for consistent estimability of the slope in terms of the relative spectral tail decays of the confounder and exposure. As a by-product, we establish a novel and general spectral condition on the equivalence of measures on the paths of multivariate GRFs with component fields of varying smoothnesses. We show that for many covariance classes like the Mat\'ern, power-exponential, generalized Cauchy, and coregionalization families, the necessary and sufficient conditions become identical, thereby providing a sharp characterization of consistent estimability of the slope for these processes. The results are extended to multivariate slopes, to accommodate measurement error, to popular classes of non-stationary Gaussian random fields and some non-Gaussian random fields, and for irregular designs.

Figures

Figures reproduced from arXiv: 2506.09267 by the authors.

Figure 1
Figure 1. Region of consistent estimability of the slope [PITH_FULL_IMAGE:figures/full_fig_p016_1.png] view at source ↗
Figure 2
Figure 2. Local-averaging-and-differencing based estimation of [PITH_FULL_IMAGE:figures/full_fig_p020_2.png] view at source ↗
Figure 3
Figure 3. Estimates of β for regression between GRF Y = Xβ + W and X when both the exposure X and the unmeasured confounder W have Mat´ern covariances with smoothnesses νX and νW respectively. where there is no confounding, i.e., ρ = 0. So the OLS estimators should be unbiased but not consistent, as β is not consistently estimable according to Corollary 1. We see this corroborated in the results. For both choices of νX, all t… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Comparison of empirical standard deviations of [PITH_FULL_IMAGE:figures/full_fig_p027_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

60 extracted references · 53 canonical work pages

  1. [1]

    On the consistent separation of scale and variance for gaussian random fields.The Annals of Statistics, 38(2):870–893, 2010

    Ethan Anderes. On the consistent separation of scale and variance for gaussian random fields.The Annals of Statistics, 38(2):870–893, 2010

    work page 2010

  2. [2]

    Cross-covariance functions for multivari- ate random fields based on latent dimensions.Biometrika, 97(1):15–30, 2010

    Tatiyana V Apanasovich and Marc G Genton. Cross-covariance functions for multivari- ate random fields based on latent dimensions.Biometrika, 97(1):15–30, 2010

    work page 2010

  3. [3]

    Asymptotically equivalent prediction in multivariate geostatistics.Bernoulli, 28(4): 2518–2545, 2022

    Fran¸ cois Bachoc, Emilio Porcu, Moreno Bevilacqua, Reinhard Furrer, and Tarik Faouzi. Asymptotically equivalent prediction in multivariate geostatistics.Bernoulli, 28(4): 2518–2545, 2022

    work page 2022

  4. [4]

    N. H. Bingham. A tauberian theorem for integral transforms of hankel type.Journal of 28 the London Mathematical Society, s2-5(3):493–503, 1972. URLhttps://londmathsoc. onlinelibrary.wiley.com/doi/abs/10.1112/jlms/s2-5.3.493

    work page doi:10.1112/jlms/s2-5.3.493 1972

  5. [5]

    Spatial self-confounding: Smoothness-related estimation bias in spatial regression models.Biometrika, page (In press), 2025

    David Bolin and Jonas Wallin. Spatial self-confounding: Smoothness-related estimation bias in spatial regression models.Biometrika, page (In press), 2025

    work page 2025

  6. [6]

    Spatial correlation in ecological analysis.International Journal of Epidemiology, 22(6):1193–1202, 1993

    David G Clayton, L Bernardinelli, and C Montomoli. Spatial correlation in ecological analysis.International Journal of Epidemiology, 22(6):1193–1202, 1993

    work page 1993

  7. [7]

    Cambridge University Press, 2014

    Giuseppe Da Prato and Jerzy Zabczyk.Stochastic equations in infinite dimensions. Cambridge University Press, 2014

    work page 2014

  8. [8]

    Accounting for unobservable heterogeneity in cross section using spatial first differences

    Hannah Druckenmiller and Solomon Hsiang. Accounting for unobservable heterogeneity in cross section using spatial first differences. Technical report, National Bureau of Economic Research, 2018

    work page 2018

Show all 60 references
  1. [9]

    Wood, and Nicole H

    Emiko Dupont, Simon N. Wood, and Nicole H. Augustin. Spatial+: A novel approach to spatial confounding.Biometrics, 78(4):1279–1290, March 2022. URLhttps://doi. org/10.1111/biom.13656

    2022 doi

  2. [10]

    L´ evy flights: exact results and asymptotics beyond all orders.Journal of Mathematical Physics, 43(5):2670–2689, 2002

    Timothy M Garoni and Norman E Frankel. L´ evy flights: exact results and asymptotics beyond all orders.Journal of Mathematical Physics, 43(5):2670–2689, 2002

    2002

  3. [11]

    d-dimensional L´ evy flights: Exact and asymptotic.Journal of Mathematical Physics, 43(10):5090–5107, 2002

    TM Garoni and NE Frankel. d-dimensional L´ evy flights: Exact and asymptotic.Journal of Mathematical Physics, 43(10):5090–5107, 2002

    2002

  4. [12]

    Proper multivariate conditional autoregressive models for spatial data analysis.Biostatistics, 4(1):11–15, 2003

    Alan E Gelfand and Penelope Vounatsou. Proper multivariate conditional autoregressive models for spatial data analysis.Biostatistics, 4(1):11–15, 2003

    2003

  5. [13]

    A causal infer- ence framework for spatial confounding.arXiv preprint arXiv:2112.14946, 2021

    Brian Gilbert, Abhirup Datta, Joan A Casey, and Elizabeth L Ogburn. A causal infer- ence framework for spatial confounding.arXiv preprint arXiv:2112.14946, 2021

    2021 arXiv

  6. [14]

    Consistency of common spatial estimators under spatial confounding.Biometrika, 112(2), December 2025

    Brian Gilbert, Elizabeth L Ogburn, and Abhirup Datta. Consistency of common spatial estimators under spatial confounding.Biometrika, 112(2), December 2025. ISSN 1464-

    2025

  7. [15]

    Stochastic models that separate fractal di- mension and the Hurst effect.SIAM Review, 46(2):269–282, 2004

    Tilmann Gneiting and Martin Schlather. Stochastic models that separate fractal di- mension and the Hurst effect.SIAM Review, 46(2):269–282, 2004

    2004

  8. [16]

    Mat´ ern cross-covariance functions for multivariate random fields.Journal of the American Statistical Association, 105(491):1167–1177, 2010

    Tilmann Gneiting, William Kleiber, and Martin Schlather. Mat´ ern cross-covariance functions for multivariate random fields.Journal of the American Statistical Association, 105(491):1167–1177, 2010

    2010

  9. [17]

    Spectral adjustment for spatial confounding.Biometrika, 110(3):699–719, 12 2022

    Yawen Guan, Garritt L Page, Brian J Reich, Massimo Ventrucci, and Shu Yang. Spectral adjustment for spatial confounding.Biometrika, 110(3):699–719, 12 2022. ISSN 1464-

    2022

  10. [18]

    Hanks, Erin M

    Ephraim M. Hanks, Erin M. Schliep, Mevin B. Hooten, and Jennifer A. Hoeting. Re- stricted spatial regression in practice: geostatistical models, confounding, and robust- ness under model misspecification.Environmetrics, 26(4):243–254, February 2015. URL https://doi.org/10.1002...

    2015 doi

  11. [19]

    URLhttps://doi.org/10.1093/biomet/asac069

    doi

  12. [20]

    Springer Science & Business Media, 2012

    Ildar Abdulovich Ibragimov and Yurii Antol’evich Rozanov.Gaussian random processes, volume 9. Springer Science & Business Media, 2012

    2012

  13. [21]

    Hodges and Brian J

    James S. Hodges and Brian J. Reich. Adding spatially-correlated errors can mess up the fixed effect you love.The American Statistician, 64(4):325–334, 2010. URLhttps: //doi.org/10.1198/tast.2010.10052

    2010 arXiv

  14. [22]

    Gaussian processes and kernel methods: A review on connections and equivalences

    Motonobu Kanagawa, Philipp Hennig, Dino Sejdinovic, and Bharath K Sriperumbudur. Gaussian processes and kernel methods: A review on connections and equivalences. arXiv preprint arXiv:1807.02582, 2018

    2018 arXiv

  15. [23]

    Exponential bases, Paley–Wiener spaces and applica- tions.Journal of Functional Analysis, 268(2):363–375, 2015

    Alex Iosevich and Azita Mayeli. Exponential bases, Paley–Wiener spaces and applica- tions.Journal of Functional Analysis, 268(2):363–375, 2015. 29

    2015

  16. [24]

    Restricted spatial regression methods: Implications for inference.Journal of the American Statistical Association, 117(537):482–494, 2022

    Kori Khan and Catherine A Calder. Restricted spatial regression methods: Implications for inference.Journal of the American Statistical Association, 117(537):482–494, 2022

    2022

  17. [25]

    Re-thinking spatial confounding in spatial linear mixed models.Statistical Science, 2025

    Kori Khan and Candae Berrett. Re-thinking spatial confounding in spatial linear mixed models.Statistical Science, 2025

    2025

  18. [26]

    Inference for gaussian processes with mat´ ern covariogram on compact riemannian manifolds.Journal of Machine Learning Research, 24(101):1–26, 2023

    Didong Li, Wenpin Tang, and Sudipto Banerjee. Inference for gaussian processes with mat´ ern covariogram on compact riemannian manifolds.Journal of Machine Learning Research, 24(101):1–26, 2023

    2023

  19. [27]

    Nonstationary modeling for multivariate spatial processes.Journal of Multivariate Analysis, 112:76–91, 2012

    William Kleiber and Douglas Nychka. Nonstationary modeling for multivariate spatial processes.Journal of Multivariate Analysis, 112:76–91, 2012

    2012

  20. [28]

    On the effects of spatial confounding in hierarchical models.International Statistical Review, 89(2): 302–322, 2021

    Widemberg S Nobre, Alexandra M Schmidt, and Jo˜ ao BM Pereira. On the effects of spatial confounding in hierarchical models.International Statistical Review, 89(2): 302–322, 2021

    2021

  21. [29]

    Gaussian fields and Gaussian sheets with generalized Cauchy covariance structure.Stochastic Processes and Their Applications, 119(4):1325–1356, 2009

    SC Lim and Lee P Teo. Gaussian fields and Gaussian sheets with generalized Cauchy covariance structure.Stochastic Processes and Their Applications, 119(4):1325–1356, 2009

    2009

  22. [30]

    Paciorek

    Christopher J. Paciorek. The importance of scale for spatial-confounding bias and precision of spatial regression estimators.Statistical Science, 25(1):107 – 125, 2010. URLhttps://doi.org/10.1214/10-STS326

    2010 doi

  23. [31]

    Univariate stable distributions.Springer Series in Operations Research and Financial Engineering, 10:978–3, 2020

    John P Nolan. Univariate stable distributions.Springer Series in Operations Research and Financial Engineering, 10:978–3, 2020

    2020

  24. [32]

    Page, Yajun Liu, Zhuoqiong He, and Donchu Sun

    Garritt L. Page, Yajun Liu, Zhuoqiong He, and Donchu Sun. Estimation and prediction in the presence of spatial confounding for spatial linear models.Scandinavian Journal of Statistics, 44(3):780–797, April 2017. URLhttps://doi.org/10.1111/sjos.12275

    2017 doi

  25. [33]

    Spatial modelling using a new class of nonstationary covariance functions.Environmetrics, 17(5):483–506, 2006

    Christopher J Paciorek and Mark J Schervish. Spatial modelling using a new class of nonstationary covariance functions.Environmetrics, 17(5):483–506, 2006

    2006

  26. [34]

    Effects of residual smoothing on the posterior of the fixed effects in disease-mapping models.Biometrics, 62(4):1197–1206, 2006

    Brian J Reich, James S Hodges, and Vesna Zadnik. Effects of residual smoothing on the posterior of the fixed effects in disease-mapping models.Biometrics, 62(4):1197–1206, 2006

    2006

  27. [35]

    Adjusting for unmea- sured spatial confounding with distance adjusted propensity score matching.Biostatis- tics, 20(2):256–272, January 2018

    Georgia Papadogeorgou, Christine Choirat, and Corwin M Zigler. Adjusting for unmea- sured spatial confounding with distance adjusted propensity score matching.Biostatis- tics, 20(2):256–272, January 2018. ISSN 1465-4644. URLhttps://doi.org/10.1093/ biostatistics/kxx074

    2018

  28. [36]

    Nonparametric estimation of nonstationary spatial covariance structure.Journal of the American Statistical Association, 87(417):108–119, 1992

    Paul D Sampson and Peter Guttorp. Nonparametric estimation of nonstationary spatial covariance structure.Journal of the American Statistical Association, 87(417):108–119, 1992

    1992

  29. [37]

    Equivalence of Gaussian measures of multivariate ran- dom fields.Stochastic Environmental Research and Risk Assessment, 29:325–334, 2015

    MD Ruiz-Medina and E Porcu. Equivalence of Gaussian measures of multivariate ran- dom fields.Stochastic Environmental Research and Risk Assessment, 29:325–334, 2015

    2015

  30. [38]

    Springer Science & Business Media, 1999

    Michael L Stein.Interpolation of Spatial Data: Some Theory for Kriging. Springer Science & Business Media, 1999. 30

    1999

  31. [39]

    On absolute con- tinuity of measures corresponding to homogeneous Gaussian fields.Theory of Probability & Its Applications, 18(1):27–40, 1973

    Anatolii Volodimirovich Skorokhod and Mikhaılo Iosipovich Yadrenko. On absolute con- tinuity of measures corresponding to homogeneous Gaussian fields.Theory of Probability & Its Applications, 18(1):27–40, 1973

    1973

  32. [40]

    Structural equation models for dealing with spatial confounding.The American Statistician, 72(3):239–252, March 2018

    Hauke Thaden and Thomas Kneib. Structural equation models for dealing with spatial confounding.The American Statistician, 72(3):239–252, March 2018. URLhttps: //doi.org/10.1080/00031305.2017.1305290

    2018 arXiv

  33. [41]

    Wenpin Tang, Lu Zhang, and Sudipto Banerjee. On identifiability and consistency of the nugget in Gaussian spatial process models.Journal of the Royal Statistical Society Series B: Statistical Methodology, 83(5):1044–1070, 2021

    2021

  34. [42]

    Wakefield

    J. Wakefield. Disease mapping and spatial regression with count data.Biostatistics, 8 (2):158–183, June 2006. URLhttps://doi.org/10.1093/biostatistics/kxl008

    2006 doi

  35. [43]

    Springer Science & Business Media, 2003

    Hans Wackernagel.Multivariate geostatistics: An introduction with applications. Springer Science & Business Media, 2003

    2003

  36. [44]

    An instrumental variables framework to unite spatial confounding methods.arXiv preprint arXiv:2411.10381, 2024

    Sophie M Woodward, Mauricio Tec, and Francesca Dominici. An instrumental variables framework to unite spatial confounding methods.arXiv preprint arXiv:2411.10381, 2024

    2024

  37. [45]

    On prediction properties of kriging: Uniform error bounds and robustness.Journal of the American Statistical Association, 115(530): 920–930, 2020

    Wenjia Wang, Rui Tuo, and CF Jeff Wu. On prediction properties of kriging: Uniform error bounds and robustness.Journal of the American Statistical Association, 115(530): 920–930, 2020

    2020

  38. [46]

    PhD thesis, University of Maryland, College Park, 2022

    Nathan Yu.Parametric estimation in spatial regression models. PhD thesis, University of Maryland, College Park, 2022

    2022

  39. [47]

    Spatial confounding in multivariate areal data analysis.arXiv preprint arXiv:2505.07232, 2025

    Kyle Lin Wu and Sudipto Banerjee. Spatial confounding in multivariate areal data analysis.arXiv preprint arXiv:2505.07232, 2025

    2025

  40. [48]

    On some properties of Buhmann functions.Ukrainian Mathematical Journal, 58(8), 2006

    Viktor Zastavnyi. On some properties of Buhmann functions.Ukrainian Mathematical Journal, 58(8), 2006

    2006

  41. [49]

    Deep compositional spatial models.Journal of the American Statistical Association, 117(540): 1787–1808, 2022

    Andrew Zammit-Mangion, Tin Lok James Ng, Quan Vu, and Maurizio Filippone. Deep compositional spatial models.Journal of the American Statistical Association, 117(540): 1787–1808, 2022

    2022

  42. [50]

    Parameter estimation for fractional brownian surfaces.Statistica Sinica, pages 863–883, 2002

    Zhengyuan Zhu and Michael L Stein. Parameter estimation for fractional brownian surfaces.Statistica Sinica, pages 863–883, 2002

    2002

  43. [51]

    Inconsistent estimation and asymptotically equal interpolations in model- based geostatistics.Journal of the American Statistical Association, 99(465):250–261, 2004

    Hao Zhang. Inconsistent estimation and asymptotically equal interpolations in model- based geostatistics.Journal of the American Statistical Association, 99(465):250–261, 2004

    2004

  44. [52]

    American Mathematical Society, 1986

    Vladimir M Zolotarev.One-dimensional stable distributions, volume 65. American Mathematical Society, 1986. 31 Supplementary materials S1 Some considerations for estimation ofβin practice The focus of this manuscript is primarily on characterizing conditions for the consistent ...

    1986

  45. [53]

    On deconfounding spatial confounding in linear models.The American Statistician, 76(2):159–167, 2022

    Dale L Zimmerman and Jay M Ver Hoef. On deconfounding spatial confounding in linear models.The American Statistician, 76(2):159–167, 2022

    2022

  46. [55]

    1 nhα11−2 n−1X j=0 {∇(1) h Z1(hj)}2 # =    O(n−1) if 2α 11 <3 O(n−1 logn) if 2α 11 = 3 O(n2α11−4) if 3<2α 11 <4, (S9) 40 and Var

    forα <2; similar results hold for largerαby considering higher order differences). If αY Y is estimated to be less thanα XX −d, then we know from our theory that one cannot consistently estimateβ. If that is not the case, then we can use the OLS estimator based on pth order di...

    2000

  47. [56]

    For the caseβ= 0, choose aβ 0 ̸= 0, and defineY ∗ =Xβ 0 +W, and eY=Y+ Y ∗ =Xβ 0 + 2W

    Thus (Z 1, Z2) = (X, Y) satisfies all condi- tions of Theorem 1, andβis consistently estimable with the consistent estimator given by 42 OLS (p) n (Z1, Z2) =OLS (p) n (X, Y). For the caseβ= 0, choose aβ 0 ̸= 0, and defineY ∗ =Xβ 0 +W, and eY=Y+ Y ∗ =Xβ 0 + 2W. Asβ 0 ̸= 0, from...

  48. [57]

    ∂4K(∥u∥) ∂u2 g∂u2 g′ + ∆h,g∆h,g′ [K(∥u∥) +r(u)]− ∂4K(∥u∥) ∂u2 g∂u2 g′ !# . 54 LetC ∗(1)(u) = dX g,g′=1 ∂4K(∥u∥) ∂u2 g∂u2 g′ and r(1)(u) = dX g,g′=1

    for a detailed discussion on how (S21) is established. As sup ω∈Rd (1 +∥ω∥) −2νX −d/fX(ω) and sup ω∈Rd (1 +∥ω∥) −2νW −d/fW (ω) are uniformly bounded 49 away from 0 and∞, the condition (16) of Theorem 5 is satisfied. Also, asν X > νW , we have sup ω∈Rd fX(ω)/fW (ω)<∞. So (17) w...

  49. [58]

    We prove the same result below using our assumptions

    used their Condition 1 to show thatq(u k,n, uj,n)→q(g k, gj). We prove the same result below using our assumptions. Note that|q(u, v)| ≤ ∥u∥F (0)∥v∥F (0)∥B∥2,F (0). We then have |q(uk,n, uj,n)−q(g k, gj)| ≤ |q(uk,n −g k, uj,n)|+|q(g k, uj,n −g j)| →0 asn→ ∞. Here the limit fol...

  50. [59]

    We consider the case whereα 11 andα ∗ 22 are both less than two, so that we can work with first differences

    By the statement of the theorem,K ∗ 22 andK ∗ 23 satisfy Assumption 1 with exponent α∗ 22 <min{α 33 + 1, α∗ 23}. We consider the case whereα 11 andα ∗ 22 are both less than two, so that we can work with first differences. More generally, we would work withp th order difference...

  51. [60]

    So, eα13 ≥min{α ∗ 12, α13}> α 11

    AsK ∗ 12(0)−K ∗ 12(h) =c 12hα12 − c12 c11 c11hα11+smaller order terms, andα 12 =α 11 cancelling out the leading order term, we haveα ∗ 12 > α11. So, eα13 ≥min{α ∗ 12, α13}> α 11. Also,eα 33 ≥min{α ∗ 22, α33}. Nowα 12 =α 11 impliesα 22 =α 11 (since ifα 22 > α11, thenα 12 ≥ α11 ...

    2000

  52. [3510]

    URLhttps://doi.org/10.1093/biomet/asae070

    doi

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.