Robust Nonparametric Testing Approaches for Spatial Regression

Hyoeun Kim; Jaewoo Park; Jorge Mateu; Kanghyun Wi; Tom\'a\v{s} Mrkvi\v{c}ka

arxiv: 2604.27569 · v3 · submitted 2026-04-30 · 📊 stat.ME

Robust Nonparametric Testing Approaches for Spatial Regression

Kanghyun Wi , Hyoeun Kim , Tom\'a\v{s} Mrkvi\v{c}ka , Jorge Mateu , Jaewoo Park This is my paper

Pith reviewed 2026-05-12 02:24 UTC · model grok-4.3

classification 📊 stat.ME

keywords spatial regressionnonparametric testingrandom shift testMonte Carlo methodsasymptotic exactnesssample covarianceincreasing domainrobust inference

0 comments

The pith

Random shift Monte Carlo tests deliver asymptotically exact inference for covariate significance in spatial regression without parametric assumptions on dependence or errors.

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

Spatial data analysis often produces invalid conclusions because parametric tests require correct specification of the correlation structure, mean function, and error distribution, all of which are hard to verify. This paper develops a nonparametric alternative that measures association between a covariate and residuals after removing nuisance effects, then uses random shifts of the covariate field to approximate the null distribution via Monte Carlo. The method needs no closed-form expression for the test statistic and works across different models. The authors prove that the procedure becomes exact in the limit when the sample covariance serves as the statistic and the observed region grows larger. This matters because it lets analysts obtain reliable p-values from real spatial datasets where confirming every modeling assumption is unrealistic.

Core claim

The paper establishes that the random shift test is asymptotically exact for the sample covariance test statistic in the increasing-domain asymptotic regime. This means that as the spatial region expands, the Monte Carlo p-values obtained by shifting the covariate converge in probability to the true p-values under the null of no partial association, without any parametric model for the spatial process or error distribution.

What carries the argument

The random shift Monte Carlo procedure, which holds residuals fixed after regressing out nuisance covariates and generates null replicates by randomly translating the spatial locations of the covariate of interest.

If this is right

Analysts obtain correct type I error control when testing covariate effects even if the true spatial dependence is complex or unknown.
The procedure achieves power comparable to correctly specified parametric tests while avoiding their inflated error rates under misspecification.
Valid tests become available for spatial regression models whose test statistics lack closed-form null distributions.
Inference remains feasible when the spatial process cannot be parameterized in a simple way.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same shifting idea could be adapted to test associations in other dependent-data settings such as time series or point patterns on networks.
Practical use would improve by examining how the number of shifts affects the precision of the p-value approximation for moderate-sized domains.
Different dependence measures beyond sample covariance might be substituted while preserving the asymptotic exactness result.

Load-bearing premise

The spatial sampling region expands without bound and randomly shifting the covariate accurately reproduces the null distribution of the sample covariance statistic without needing any further parametric structure on the process.

What would settle it

A simulation in which, under the true null of no covariate effect and for successively larger spatial domains, the empirical rejection rate of the random shift test stays far from the nominal significance level.

Figures

Figures reproduced from arXiv: 2604.27569 by Hyoeun Kim, Jaewoo Park, Jorge Mateu, Kanghyun Wi, Tom\'a\v{s} Mrkvi\v{c}ka.

**Figure 1.** Figure 1: Simulated spatial random fields under different error structures. From left to right, view at source ↗

**Figure 2.** Figure 2: Empirical rejection rates under the linear model (left) and nonlinear model (right). view at source ↗

**Figure 3.** Figure 3: Empirical rejection rates under the linear model (left) and the nonlinear model view at source ↗

**Figure 4.** Figure 4: Empirical rejection rates under the linear model (left) and the nonlinear model view at source ↗

**Figure 5.** Figure 5: (a) Scatter plots of the covariates versus the response variable. (b) Spatial maps view at source ↗

**Figure 6.** Figure 6: Empirical rejection rates under the linear model simulation (left) and nonlinear view at source ↗

**Figure 7.** Figure 7: Empirical rejection rates under the linear (left) and nonlinear (right) simulation view at source ↗

**Figure 8.** Figure 8: Empirical rejection rates under the linear (left) and nonlinear (right) simulation view at source ↗

**Figure 9.** Figure 9: Empirical rejection rates under the linear (left) and nonlinear (right) simulation view at source ↗

**Figure 10.** Figure 10: Empirical rejection rates under the linear model simulation (left) and nonlinear view at source ↗

**Figure 11.** Figure 11: Empirical rejection rates under the linear model (left) and the nonlinear model view at source ↗

**Figure 12.** Figure 12: Scatter plots of the covariates. References Banerjee, S., Carlin, P. B., and Gelfand, E. A. (2014). Hierarchical Modeling and Analysis for Spatial Data, 2nd Edition. Chapman and Hall/CRC. Cressie, N. (2015). Statistics for spatial data. John Wiley & Sons. Dale, M. R. and Fortin, M.-J. (2002). Spatial autocorrelation and statistical tests in ecology. Ecoscience ´ , 9(2):162–167. Davison, A. C. and Hinkley,… view at source ↗

read the original abstract

Reliable inference for spatial regression remains challenging because it requires the correct specification of the spatial dependence structure, the mean trend, and the error distribution. Existing parametric testing methods rely on restrictive assumptions that are difficult to verify in practice and can lead to inaccurate conclusions under misspecification. To address this, we develop a robust nonparametric Monte Carlo testing framework for spatial regression based on random shifts. We construct test statistics that measure the dependence between residuals, obtained after removing the effects of nuisance covariates, and the covariate of interest. This allows us to assess the significance of the covariate in the sense of partial correlation. The proposed framework enables robust inference across various models without requiring parametric assumptions or even a closed-form distribution of the test statistics. Furthermore, we establish the asymptotic exactness of the random shift test in the increasing-domain setting when the sample covariance is used as the test statistic. Through extensive numerical experiments, we demonstrate that our method maintains the nominal significance level while achieving competitive power, whereas parametric methods can exhibit inflated type I error rates, even when they are correctly specified.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper gives a random-shift Monte Carlo test for partial correlation in spatial regression with an asymptotic exactness result under increasing domains, but the nonparametric positioning rests on unstated weak-dependence conditions.

read the letter

The main takeaway is that this work adapts random-shift Monte Carlo testing to residuals from a spatial regression so you can test whether one covariate still matters after adjusting for others, and they prove the test is asymptotically exact when the domain grows and the statistic is ordinary sample covariance. That combination is not in the earlier literature they cite. They also run simulations showing the test keeps nominal level where standard parametric approaches can go wrong even when correctly specified. That part is useful for applied spatial work where the covariance structure is hard to get right. The approach is straightforward to implement once you have the residuals, and it avoids needing a closed-form null distribution. The soft spot is the asymptotic claim. The abstract says the result holds without parametric assumptions on the process, yet random shifts only reproduce the null distribution of the covariance if the field becomes roughly independent of its shifted copy at large distances. That requires some form of spatial mixing or ergodicity whose rate must interact with domain growth. The paper needs to state those conditions explicitly in the theorem; otherwise the “fully nonparametric” framing overstates what is shown. If the proof already lists the minimal mixing rate, then the concern is minor and just needs clearer wording. This paper is for spatial statisticians and applied researchers in ecology or public health who need a test that does not force a parametric covariance model. A reader already working with Monte Carlo spatial tests or nonparametric regression will see the targeted extension and can judge whether the conditions fit their data. It deserves a serious referee because the core idea is practical, the simulations address a real failure mode of existing methods, and the theory is a concrete step even if the dependence conditions need tightening in revision.

Referee Report

1 major / 1 minor

Summary. The paper develops a robust nonparametric Monte Carlo testing framework for spatial regression based on random shifts. Test statistics are constructed as measures of dependence (specifically sample covariance) between residuals obtained after removing nuisance covariates and the covariate of interest, allowing assessment of significance in the sense of partial correlation. The central theoretical result is the establishment of asymptotic exactness for the random-shift test under increasing-domain asymptotics when using the sample covariance statistic. Simulations across various models demonstrate that the method maintains nominal significance levels with competitive power, while parametric alternatives can exhibit inflated type I error rates even under correct specification.

Significance. If the asymptotic exactness holds under minimal conditions, the work provides a useful nonparametric alternative for inference in spatial regression settings where correct specification of dependence, trends, and error distributions is difficult. The residualization approach for handling nuisance covariates and the emphasis on Monte Carlo testing without closed-form distributions are practical strengths. The simulation evidence of level control and power is supportive, but the overall significance depends on whether the theoretical guarantee is fully nonparametric or implicitly relies on unstated dependence conditions.

major comments (1)

[Abstract and theoretical development] Abstract and main theoretical result on asymptotic exactness: the claim of asymptotic exactness 'without requiring parametric assumptions' for the random-shift test in the increasing-domain setting is load-bearing for the paper's positioning as robust and nonparametric. For shifted copies to asymptotically reproduce the null distribution of the sample covariance between residuals and the target covariate, the underlying spatial random field must become approximately independent of its shifted version, which generally requires weak dependence (e.g., mixing or ergodicity conditions whose rate interacts with domain growth). If the proof invokes such conditions without stating them explicitly or reconciling them with the 'fully nonparametric' claim, the result is narrower than advertised. Please identify the precise assumptions in the proof (e.g., any mixing rates or domain-growth conditions

minor comments (1)

[Numerical experiments] The simulation section would benefit from additional detail on the specific spatial correlation functions, domain sizes, and shift mechanisms used, to facilitate reproducibility and to confirm that the reported nominal level control is not sensitive to particular dependence strengths.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful reading and constructive feedback on the manuscript. We address the major comment below.

read point-by-point responses

Referee: [Abstract and theoretical development] Abstract and main theoretical result on asymptotic exactness: the claim of asymptotic exactness 'without requiring parametric assumptions' for the random-shift test in the increasing-domain setting is load-bearing for the paper's positioning as robust and nonparametric. For shifted copies to asymptotically reproduce the null distribution of the sample covariance between residuals and the target covariate, the underlying spatial random field must become approximately independent of its shifted version, which generally requires weak dependence (e.g., mixing or ergodicity conditions whose rate interacts with domain growth). If the proof invokes such conditions without stating them explicitly or reconciling them with the 'fully nonparametric' claim, the result is narrower than advertised. Please identify the precise assumptions in the proof (e.g., any

Authors: We appreciate the referee highlighting the need for explicit assumptions. The proof of asymptotic exactness (Theorem 3.1) is established under increasing-domain asymptotics together with a weak dependence condition on the spatial random field: specifically, the field is assumed to be alpha-mixing with mixing coefficients decaying at a rate that ensures the covariance between the original and sufficiently shifted copies vanishes asymptotically (see Assumption 2.2). This is a standard nonparametric regularity condition that does not impose a parametric form on the dependence structure (e.g., no requirement that the covariance belongs to a specific family such as exponential or Matérn). The random-shift Monte Carlo procedure then reproduces the null distribution of the sample covariance because the shifted residuals become asymptotically independent of the target covariate under these conditions. We agree that the abstract and introduction would benefit from greater precision to avoid any implication that the result holds with literally no regularity conditions whatsoever. We will revise the abstract to qualify the claim as holding 'under mild weak dependence conditions' and expand the discussion in Section 3 to state Assumption 2.2 explicitly, reconcile it with the nonparametric positioning, and note that the mixing rate interacts with domain growth in the usual way for spatial ergodic theorems. revision: yes

Circularity Check

0 steps flagged

No circularity: asymptotic exactness derived from independent mixing/ergodicity arguments under increasing-domain regime

full rationale

The central result establishes asymptotic exactness of the random-shift Monte Carlo test for the sample-covariance statistic between residuals and the target covariate. The test statistic is explicitly constructed from the data (residuals after nuisance regression), while the reference distribution is generated by random shifts whose validity rests on domain-expansion arguments that are external to the fitted values. No equation reduces the prediction to a quantity fitted from the same data under the null, no self-citation supplies a uniqueness theorem that forces the method, and the nonparametric positioning is not achieved by renaming a known result. The derivation chain therefore remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The framework relies on standard spatial mixing or weak dependence conditions implicit in increasing-domain asymptotics and on the validity of random shifts for generating the null distribution.

axioms (1)

domain assumption Increasing-domain asymptotic regime for the spatial sampling design
Invoked to establish asymptotic exactness of the random-shift test.

pith-pipeline@v0.9.0 · 5498 in / 1057 out tokens · 44768 ms · 2026-05-12T02:24:40.532995+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

we establish the asymptotic exactness of the random shift test in the increasing-domain setting when the sample covariance is used as the test statistic
IndisputableMonolith/Foundation/DimensionForcing.lean alexander_duality_circle_linking unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Assumption 2. The random fields Φ(s) and Ψ(s) are strongly mixing with coefficient α(a;b)

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

6 extracted references · 6 canonical work pages

[1]

B., and Gelfand, E

Banerjee, S., Carlin, P. B., and Gelfand, E. A. (2014).Hierarchical Modeling and Analysis for Spatial Data, 2nd Edition. Chapman and Hall/CRC. Cressie, N. (2015).Statistics for spatial data. John Wiley & Sons. Dale, M. R. and Fortin, M.-J. (2002). Spatial autocorrelation and statistical tests in ecology. ´Ecoscience, 9(2):162–167. Davison, A. C. and Hinkl...

work page 2014
[2]

45 DiCiccio, C

Cambridge university press. 45 DiCiccio, C. J. and Romano, J. P. (2017). Robust permutation tests for correlation and regression coefficients.Journal of the American Statistical Association, 112(519):1211–

work page 2017
[3]

J., Tawn, J

Diggle, P. J., Tawn, J. A., and Moyeed, R. A. (1998). Model-based geostatistics.Journal of the Royal Statistical Society Series C: Applied Statistics, 47(3):299–350. Dvoˇ r´ ak, J. and Mrkviˇ cka, T. (2024). Nonparametric testing of the covariate significance for spatial point patterns under the presence of nuisance covariates.Journal of Computational and...

work page 1998
[4]

and Lane, D

Freedman, D. and Lane, D. (1983). A nonstochastic interpretation of reported significance levels.Journal of Business and Economic Statistics, 1(4):292 –

work page 1983
[5]

and Tibshirani, R

Hastie, T. and Tibshirani, R. (1986). Generalized additive models.Statistical science, 1(3):297–310. Hayfield, T. and Racine, J. S. (2008). Nonparametric econometrics: The np package.Journal of Statistical Software, 27(5):1–32. Helwig, N. E. (2019). Robust nonparametric tests of general linear model coefficients: A com- parison of permutation methods and ...

work page 1986
[6]

Robinson, P. (2011). Asymptotic theory for nonparametric regression with spatial data. Journal of Econometrics, 165(1):5–19. Moment Restriction-Based Econometric Methods. 47 Sherman, M. and Carlstein, E. (1994). Nonparametric estimation of the moments of a general statistic computed from spatial data.Journal of the American Statistical Association, 89(426...

work page 2011

[1] [1]

B., and Gelfand, E

Banerjee, S., Carlin, P. B., and Gelfand, E. A. (2014).Hierarchical Modeling and Analysis for Spatial Data, 2nd Edition. Chapman and Hall/CRC. Cressie, N. (2015).Statistics for spatial data. John Wiley & Sons. Dale, M. R. and Fortin, M.-J. (2002). Spatial autocorrelation and statistical tests in ecology. ´Ecoscience, 9(2):162–167. Davison, A. C. and Hinkl...

work page 2014

[2] [2]

45 DiCiccio, C

Cambridge university press. 45 DiCiccio, C. J. and Romano, J. P. (2017). Robust permutation tests for correlation and regression coefficients.Journal of the American Statistical Association, 112(519):1211–

work page 2017

[3] [3]

J., Tawn, J

Diggle, P. J., Tawn, J. A., and Moyeed, R. A. (1998). Model-based geostatistics.Journal of the Royal Statistical Society Series C: Applied Statistics, 47(3):299–350. Dvoˇ r´ ak, J. and Mrkviˇ cka, T. (2024). Nonparametric testing of the covariate significance for spatial point patterns under the presence of nuisance covariates.Journal of Computational and...

work page 1998

[4] [4]

and Lane, D

Freedman, D. and Lane, D. (1983). A nonstochastic interpretation of reported significance levels.Journal of Business and Economic Statistics, 1(4):292 –

work page 1983

[5] [5]

and Tibshirani, R

Hastie, T. and Tibshirani, R. (1986). Generalized additive models.Statistical science, 1(3):297–310. Hayfield, T. and Racine, J. S. (2008). Nonparametric econometrics: The np package.Journal of Statistical Software, 27(5):1–32. Helwig, N. E. (2019). Robust nonparametric tests of general linear model coefficients: A com- parison of permutation methods and ...

work page 1986

[6] [6]

Robinson, P. (2011). Asymptotic theory for nonparametric regression with spatial data. Journal of Econometrics, 165(1):5–19. Moment Restriction-Based Econometric Methods. 47 Sherman, M. and Carlstein, E. (1994). Nonparametric estimation of the moments of a general statistic computed from spatial data.Journal of the American Statistical Association, 89(426...

work page 2011