FLAT: Fused Lasso Regression with Adaptive Minimum Spanning Tree with Applications on Thermohaline Circulation

Cuiwen Che; Wei Zhong; Yifan Chen; Zhaoyu Xing

REVIEW 3 major objections 5 minor 30 references

FLAT: Fused Lasso Regression with Adaptive Minimum Spanning Tree with Applications on Thermohaline Circulation

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read FLAT is a spatial regression method that uses an adaptive minimum spanning tree to simultaneously estimate smoothly varying coefficients and detect sharp boundaries, outperforming GWR and SCC in simulations and revealing new…

desk verdict Adaptive-MST fused lasso is a worthwhile idea, but the paper's core Lasso reparameterization is dimensionally wrong, so the algorithm isn't justified. read the letter →

arxiv 2507.09800 v2 pith:PDCFZV6Y submitted 2025-07-13 stat.ME

Cuiwen Che , Yifan Chen , Zhaoyu Xing , Wei Zhong This is my paper · ORCID

classification stat.ME MSC 62J0762M3062H30

keywords spatialheterogeneityfusedlassominimumspanningtreeadaptiveregularizationspatiallyvaryingcoefficientsthermohalinecirculationtemperature-salinityrelationshipDBSCAN

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 introduces FLAT, a spatial regression method that estimates spatially varying coefficients while simultaneously detecting sharp boundaries between regions with different behavior. The method builds an adaptive minimum spanning tree whose edges connect locations that are close both in space and in an initial estimate of their coefficients, then applies a fused-lasso penalty along those edges together with a Lasso penalty that selects features. Because the fused penalty can be reparameterized as a standard Lasso, the whole model is solved by efficient coordinate descent. Simulations on data with smooth change inside clusters and sharp jumps between them show FLAT estimating coefficients more accurately and recovering clusters more reliably than geographically weighted regression (GWR) and spatially clustered coefficients (SCC). Applied to public oceanographic data, FLAT identifies an inverse temperature-salinity surface near the equator and delineates Atlantic water masses, offering a data-driven way to find thermohaline fronts.

What carries the argument

The central mechanism is the adaptive minimum spanning tree, a tree connecting all n spatial locations with n-1 edges, built so that edges tend to join locations that are close in space and have similar first-stage coefficient estimates. The fused penalty term $\|\pi H \beta_k\|_1$ forces coefficient differences along these tree edges to be small or zero, with adaptive weights $\pi = 1/\tilde{d}^\gamma$. The key algebraic step is the reparameterization $\theta_k = \tilde{H} \beta_k$ with $\tilde{H} = [\pi H;\ (\lambda_2/\lambda_1) I]$, which is invertible, so the transformed design matrix turns the FLAT objective into a standard Lasso in $\theta$; coordinate descent on this Lasso and the inverse map $\beta = \tilde{H}^{-1}\theta$ complete the estimation, and DBSCAN converts the estimates into spatial regions.

What would settle it

Simulate a coefficient field with a true boundary that is narrow and has a small signal-to-noise ratio, then inspect the adaptive minimum spanning tree returned by the first-stage estimate and count how many of its edges connect locations on opposite sides of the boundary. If the tree contains no such boundary-crossing edge while a tree built from the true coefficients does, the FLAT cluster solution cannot separate the two regions, showing that boundary recovery depends on the first-stage estimate rather than on the tree construction alone.

Watch

Extended reading notes

Core claim

The paper claims that the FLAT estimator, obtained by minimizing a combined penalized objective with a fused penalty on the edges of an adaptive minimum spanning tree and a Lasso penalty on the coefficients, recovers both smooth spatial variation and sharp boundaries in a spatial regression. The tree is built from a first-stage estimate's coefficient dissimilarities, and the fused regularization is reparameterized through an invertible matrix into an equivalent standard Lasso problem solvable by coordinate descent, after which the estimated coefficients are clustered by DBSCAN to output discrete regions. In simulations, FLAT reports lower RMSE, MAE, and SD and higher Rand indices and silhouette coefficients than GWR and SCC. In Atlantic sea-surface and cross-section data, it finds the temperature-salinity coefficient crossing zero near the equator, an 'inverse temperature-salinity surface,' and identifies thermohaline water-mass structure.

Load-bearing premise

The load-bearing premise is that the first-stage fused-lasso estimate is accurate enough that the minimum spanning tree built from coefficient distances contains the edges that cross the true region boundaries; if that initial estimate is too noisy, the fused penalty shrinks along the wrong edges and the detected clusters become unreliable.

Editorial extensions

If this is right

Users of spatial regression gain a single procedure that both estimates smoothly varying coefficients and outputs discrete regions, without a two-step clustering design.
The Lasso reparameterization means existing Lasso solvers can fit the model, making computation practical for large spatial datasets.
The $\lambda_2$ Lasso penalty includes feature selection, so the method can zero out entire spatial coefficient fields for irrelevant predictors.
In oceanography, the method can map thermohaline fronts and water masses directly from temperature and salinity measurements, as demonstrated on Atlantic data.
Simulation results indicate FLAT can be used where GWR smooths over boundaries and SCC oversimplifies within-cluster variation, giving more accurate coefficients and cluster recovery.

Reading between the lines

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

A natural next extension is a diagnostic that checks whether the adaptive minimum spanning tree actually contains edges crossing every true boundary; when it does not, the fused penalty shrinks along the wrong edges, so boundary detection depends heavily on the quality of the first-stage estimate.
The same construction should transfer to other problems with front-like spatial transitions, such as atmospheric cold fronts, ecological ecotones, or geological fault zones, where a reasonable first-stage estimate of the coefficients is available.
Because the reparameterized objective is a standard Lasso, existing theoretical tools for Lasso and fused Lasso could be adapted to derive conditions under which FLAT recovers the true boundary set, which would strengthen the method's formal grounding.
One could extend FLAT to non-Gaussian responses, such as binary or count outcomes, by replacing the squared-error loss with a generalized linear model likelihood while keeping the same fused-tree penalty.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Colleague, here's my read of arXiv:2507.09800. The headline: the idea is new and the simulations look decent, but the paper's central algorithm is built on a dimensional error. The stress-test note is correct: tilde H in eq. (12) is (2n-1) x n, not n x n, so it can't be inverted and eq. (14) doesn't follow. That makes Algorithm 1's coordinate descent unjustified as written.

What's genuinely new: building the MST from coefficient dissimilarity (eq. 3) rather than pure spatial distance is a sensible adaptive step, and the weighted fusion penalty in (5) is a reasonable way to model both smooth variation and sharp boundaries. The simulation comparison against GWR and SCC is useful, and the reported numbers suggest FLAT does better on both estimation and cluster recovery. Code is public, which helps.

Soft spots, in order of seriousness. First, the invalid reparameterization is not a typo; it's the entire basis for the algorithm and the claimed Lasso equivalence. Without a correct algorithm, the simulation results are not backed by the math in the paper. Second, the premise that the preliminary estimate beta0 yields an MST containing boundary-crossing edges is just assumed; no theory or diagnostics. Third, the simulation tables give means but no measures of precision across the 100 replications, so we can't judge whether the improvements are real. Fourth, the real-data analysis is purely descriptive; no validation or uncertainty quantification, and the DBSCAN tuning is vague.

Is the central argument salvageable? The objective (5) is still convex and can be solved by, say, ADMM or proximal gradient without the invalid transformation. So the flaw is fixable, but the paper as written doesn't deliver. I'd send it to a serious referee with a request to verify the algorithm; if the code works and the method can be re-derived, it's a decent contribution. If not, it's a reject.

Referee Report

3 major / 5 minor

Summary. The paper proposes FLAT, a fused-lasso-type estimator for spatial regression with spatially varying coefficients that exhibit both smooth variation within regions and sharp transitions at boundaries. The method first obtains a preliminary estimate from a fused lasso on a spatial proximity graph, builds a minimum spanning tree using both spatial distance and coefficient dissimilarity, and then minimizes a convex objective combining an adaptive tree-based fusion penalty and a Lasso penalty. The authors claim that this objective can be reparameterized into a standard Lasso via an invertible matrix tilde H, which justifies a coordinate-descent algorithm; afterward DBSCAN is applied to the estimated coefficients to recover discrete spatial clusters. The paper reports simulation comparisons against GWR and SCC, and an application to Atlantic temperature-salinity relationships, where it claims improved estimation and new empirical findings.

Significance. If the methodology is sound, the paper addresses a genuinely relevant problem: modeling continuous and discrete spatial heterogeneity simultaneously in a data-driven way. The adaptive minimum-spanning-tree idea is intuitive, the objective function is convex, and the authors provide code. The simulation design is reasonable in scope, and the empirical application touches an important oceanographic question. However, the central formal claim—that objective (5) can be transformed to a standard Lasso—is invalid as written, and the numerical results are therefore not supported by the derivation presented. This is a load-bearing issue that must be resolved before the paper's claims can be accepted.

major comments (3)

[Appendix, Eqs. (12)-(14); Algorithm 1] The reparameterization is dimensionally invalid. In Eq. (12), H is (n-1) x n and pi is (n-1) x (n-1), so pi H is (n-1) x n. Stacking pi H with the n x n matrix (lambda_2/lambda_1) I gives tilde H of dimension (2n-1) x n, not n x n. Consequently tilde H is not square, is not invertible, the expression beta_k = tilde H^{-1} theta_k is undefined, the reparameterized design matrix tilde X in Eq. (13) does not exist, and Eq. (14) is not equivalent to the FLAT objective (5)/(11). Because Algorithm 1's coordinate descent updates are derived from this incorrect transformation, the simulation results in Tables 1-2 and the real-data results in Table 3 and Figures 3-8 are not backed by the paper's derivation. The authors must either provide a correct equivalent formulation (for example, a direct proximal-gradient or ADMM treatment of objective (5)) or prove the transformation with correct dimensions, and then rerun all experiments under the corrected algorithm.
[Section 2, Eqs. (2)-(3)] The adaptive minimum spanning tree E is built from the preliminary estimate beta_0 obtained at a fixed tuning parameter lambda in Eq. (2). The success of the fused penalty in Eq. (5) depends critically on E containing edges that cross true coefficient boundaries: if the preliminary estimate is too noisy, or if lambda is poorly chosen, the tree may place edges within homogeneous regions or fail to connect across boundaries, and the subsequent fusion will shrink along the wrong edges. The paper provides no theoretical condition on beta_0 or lambda under which the MST captures the boundary structure, and no sensitivity analysis. I would like to see either a concentration or support-recovery result for the preliminary estimate, or at least a simulation study that varies the first-stage lambda and reports boundary-recovery rates.
[Section 3, Tables 1-2] The simulation section does not describe how the tuning parameters lambda_1, lambda_2, and the DBSCAN hyperparameters epsilon and minPts are selected. If these are chosen using the true cluster labels or with a separate validation procedure, the comparison with GWR and SCC may be unfair, and the reported clustering indices may be optimistic. Please state the tuning mechanism explicitly for both the estimation and the clustering stages, and report the variability of the performance measures across the 100 replications (for example, standard errors or interquartile ranges).

minor comments (5)

[Abstract and Introduction] There are multiple typographical and grammatical errors, including 'hyper-plain' instead of 'hyper-plane', 'we constructs', and 'disinclines' in the first sentence. The manuscript would benefit from a careful language edit.
[Eq. (12)] In the sentence defining tilde H, the text reads 'where I denotes the .' and is incomplete; the identity matrix should be specified fully, and the dimensions of I should be stated.
[Tables 2 and 3] The scaling notation for the Calinski-Harabasz index is inconsistent: Table 2 says values are scaled by ×10^4, while Table 3 reports CHI(×10^3). Please use a single consistent scaling and clarify that CHI is otherwise unit-free and depends on sample size.
[Algorithm 1] In Algorithm 1, the reparameterization in line 4 appears after the construction of H in line 3, but the text says the reparameterization is 'defined in (12)' and then 'Compute adaptive weights pi' again; the ordering and duplication are confusing. Also, the initialization of the hyperparameter sequences epsilon and minPts is not described.
[Section 2, Eq. (3)] The adaptive weight pi(s_i, s_j) = 1/d_tilde(s_i, s_j)^gamma is undefined when d_tilde = 0, which can occur if the preliminary estimate is constant or if duplicate locations are present. This should be addressed, perhaps by a small additive constant in the denominator.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: FLAT is an adaptive two-step fused-lasso procedure and its simulation claims are benchmarked against independently generated truth; the appendix reparameterization error is a correctness issue, not circularity.

full rationale

The central estimator is not circular by construction. The preliminary fused-lasso estimate beta0 in equation (2) and the minimum spanning tree in equation (3) are standard adaptive-lasso ingredients; the FLAT objective (5) is then fit to the data, but the final estimator is evaluated against independently generated true coefficient surfaces in simulation, and the MST does not encode the true cluster labels. No fitted parameter is renamed as a prediction, and no load-bearing claim rests only on the authors' own citations. The self-citations (Xing et al. 2024 for cluster metrics, Xing et al. 2025 for soft-thresholding updates) concern standard, externally available tools and are not load-bearing. The principal formal flaw is in the appendix, equations (12)-(14): tilde H is defined as [pi H; (lambda2/lambda1) I], which has (2n-1) rows and n columns, not n x n as claimed, so tilde H^{-1} and the reparameterized design matrix tilde X in (13) are not defined as written; this invalidates the displayed equivalence to a standard lasso in (14) and the coordinate descent in Algorithm 1. That is a derivation and correctness error, not a reduction of the result to its own inputs. Similarly, the unproved assumption that the MST built from beta0 contains boundary-crossing edges is a statistical risk, not circular reasoning. Overall circularity is minimal.

Assumptions & free parameters 5 free parameters · 3 assumptions · 0 invented entities

The method rests on a two-stage estimation with several tuning parameters, none of which have an explicit specification. The adaptive MST premise and the invalid reparameterization are the main auxiliary assumptions beyond standard Lasso theory.

free parameters (5)

lambda (first-stage) = not reported
Controls the initial fused lasso in eq. (2) that produces beta0 for the adaptive weights; selection rule is not described.
lambda_1 = not reported
Weights the fused penalty on MST edges in eq. (5); tuning procedure not specified in the main text.
lambda_2 = not reported
Weights the L1 sparsity penalty in eq. (5); tuning procedure not specified.
gamma = 1
Exponent in the adaptive weight pi = 1/dtilde^gamma; set to 1 following Zou (2006) for adaptive lasso, but no justification that this transfers to the fused MST setting.
DBSCAN epsilon and minPts = selected by internal clustering criteria over a grid
Used in the clustering step to detect discrete regions; selection over a grid is data-dependent and could inflate cluster quality metrics.

assumptions (3)

domain assumption Spatial regression model (1) with independent errors is the correct generating model.
The method and all simulations assume y(si) = beta(si)'x(si) + epsilon(si) with i.i.d. errors; if this is misspecified, the coefficient estimates and detected boundaries are not interpretable.
ad hoc to paper The minimum spanning tree built from the preliminary estimates captures the true boundary structure.
No consistency result is given for beta0 or for the tree topology; the entire method depends on this heuristic.
ad hoc to paper The reparameterization matrix tilde H in eq. (12) is square and invertible.
The appendix states tilde H is in R^{n x n}, but the stacked matrix [pi H; (lambda2/lambda1) I] is (2n-1) x n; the Lasso equivalence (14) relies on this invalid claim.

how reviews work

0 comments

Cite this review

Pith. "Pith review of FLAT: Fused Lasso Regression with Adaptive Minimum Spanning Tree with Applications on Thermohaline Circulation." pith.science (2026). https://pith.science/paper/PDCFZV6Y

@misc{pith2026250709800,
  author       = {Pith},
  title        = {Pith review of: FLAT: Fused Lasso Regression with Adaptive Minimum Spanning Tree with Applications on Thermohaline Circulation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PDCFZV6Y}},
  note         = {Machine review of arXiv:2507.09800}
}

read the original abstract

This article introduces a new methodology model both discrete and continuous spatial heterogeneity simultaneously with an application in detection of hyper-plain in thermohaline circulation. To enable the data-driven detection of spatial boundaries with heterogeneity, we constructs an adaptive minimum spanning tree guided by both spatial proximity and coefficient dissimilarity, and combines both a spatial fused regularization and LASSO-type regularization to estimate the spatial coefficients under the framework of spatial regression. Numerical simulations demonstrate the effectiveness of proposed method in both estimation and heterogeneity detection. The usefulness of the approach is further illustrated via an analysis of oceanic data that provides new empirical finds about Atlantic with detected surfaces in temperature-salinity relationship.

Figures

Figures reproduced from arXiv: 2507.09800 by the authors.

**Figure 1.** True spatial parameters βk and estimated parameters βˆ k for k = 1, 2, 3 by GWR, SCC and FLAT: In each subgraph, the coordinate axes correspond to two spatial dimensions and the points represent the spatial locations si accordingly. The dataset is generated from DGP5 and the values of parameters and estimations are indicated by color [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗

**Figure 2.** Map of the North Atlantic and its coastal [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗

**Figure 5.** Identified T-S coefficient contours with [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗

Figures from the paper (3 more)

**Figure 6.** Figure 6: Vertical cross-section of the estimated T-S slope coefficients using FLAT [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]

**Figure 7.** Figure 7: Vertical SDQ of the T-S coefficients along the ocean transect [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]

**Figure 8.** Figure 8: Identified T-S coefficient contours with clustering in the vertical cross-section [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

30 extracted references · 25 canonical work pages

[1]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...
[2]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...
[3]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...
[4]

, year 2010

author Anselin, L. , year 2010 . title Thirty years of spatial econometrics . journal Papers in Regional Science volume 89 , pages 3--26

2010
[5]

, year 2013

author Anselin, L. , year 2013 . title Spatial econometrics: methods and models . volume volume 4 . publisher Springer Science & Business Media

work page 2013
[6]

, author Gatrell, A.C

author Bailey, T.C. , author Gatrell, A.C. , et al., year 1995 . title Interactive spatial data analysis . volume volume 413 . publisher Longman Scientific & Technical Essex

work page 1995
[7]

, author Fotheringham, A.S

author Brunsdon, C. , author Fotheringham, A.S. , author Charlton, M.E. , year 1996 . title Geographically weighted regression: a method for exploring spatial nonstationarity . journal Geographical Analysis volume 28 , pages 281--298

work page 1996
[8]

, author Zhang, Q

author Chen, Y. , author Zhang, Q. , author Ma, S. , author Fang, K. , year 2024 . title Heterogeneity-aware clustered distributed learning for multi-source data analysis . journal Journal of Machine Learning Research volume 25 , pages 1--60

work page 2024

Show all 30 references

[9]

, author Pisias, N.G

author Clark, P.U. , author Pisias, N.G. , author Stocker, T.F. , author Weaver, A.J. , year 2002 . title The role of the thermohaline circulation in abrupt climate change . journal Nature volume 415 , pages 863--869

2002
[10]

, author Church, R.L

author Duque, J.C. , author Church, R.L. , author Middleton, R.S. , year 2011 . title The p-regions problem . journal Geographical Analysis volume 43 , pages 104--126

2011
[11]

, author Kriegel, H.P

author Ester, M. , author Kriegel, H.P. , author Sander, J. , author Xu, X. , year 1996 . title A density-based algorithm for discovering clusters in large spatial databases with noise , pp. pages 226--231

1996
[12]

, author Paparella, F

author Ferrari, R. , author Paparella, F. , author Rudnick, D. , author Young, W. , year 2001 . title The temperature-salinity relationship of the mixed layer . journal From Stirring to Mixing in a Stratified Ocean , pages 95--104

2001
[13]

, author Armi, L

author Flament, P. , author Armi, L. , author Washburn, L. , year 1985 . title The evolving structure of an upwelling filament . journal Journal of Geophysical Research: Oceans volume 90 , pages 11765--11778

1985
[14]

, author Yang, W

author Fotheringham, A.S. , author Yang, W. , author Kang, W. , year 2017 . title Multiscale geographically weighted regression (mgwr) . journal Annals of the American Association of Geographers volume 107 , pages 1247--1265

2017
[15]

, author Souza, P.C

author Lam, C. , author Souza, P.C. , year 2020 . title Estimation and selection of spatial weight matrix in a spatial lag model . journal Journal of Business & Economic Statistics volume 38 , pages 693--710

2020
[16]

, author Pace, R.K

author LeSage, J. , author Pace, R.K. , year 2009 . title Introduction to spatial econometrics . publisher Chapman and Hall/CRC

2009
[17]

, author Sang, H

author Li, F. , author Sang, H. , year 2019 . title Spatial homogeneity pursuit of regression coefficients for large datasets . journal Journal of the American Statistical Association volume 114 , pages 1050--1062

2019
[18]

, author Yang, Y

author Li, J. , author Yang, Y. , year 2021 . title Thermohaline interleaving induced by horizontal temperature and salinity gradients from above . journal Journal of Fluid Mechanics volume 927 , pages A12

2021
[19]

, author Sang, H

author Luo, Z.T. , author Sang, H. , author Mallick, B. , year 2021 . title A bayesian contiguous partitioning method for learning clustered latent variables . journal Journal of Machine Learning Research volume 22 , pages 1--52

2021
[20]

, author Han, J

author Miller, H.J. , author Han, J. , year 2009 . title Geographic data mining and knowledge discovery . publisher CRC press

2009
[21]

, year 2002

author Raudenbush, S.W. , year 2002 . title Hierarchical linear models: Applications and data analysis methods . journal Advanced Quantitative Techniques in the Social Sciences Series/SAGE

2002
[22]

, year 1969

author Reid, J. , year 1969 . title Sea-surface temperature, salinity, and density of pacific ocean in summer and in winter . journal Deep-Sea Research , pages 215

1969
[23]

, year 1975

author Roden, G.I. , year 1975 . title On north pacific temperature, salinity, sound velocity and density fronts and their relation to the wind and energy flux fields . journal Journal of Physical Oceanography volume 5 , pages 557--571

1975
[24]

, author Luyten, J.R

author Rudnick, D.L. , author Luyten, J.R. , year 1996 . title Intensive surveys of the azores front: 1. tracers and dynamics . journal Journal of Geophysical Research: Oceans volume 101 , pages 923--939

1996
[25]

, year 2011

author Talley, L.D. , year 2011 . title Descriptive physical oceanography: an introduction . publisher Academic press

2011
[26]

, year 2013

author Talley, L.D. , year 2013 . title Closure of the global overturning circulation through the indian, pacific, and southern oceans: Schematics and transports . journal Oceanography volume 26 , pages 80--97

2013
[27]

, author Nissen, J

author Van Dusen, B. , author Nissen, J. , year 2019 . title Modernizing use of regression models in physics education research: A review of hierarchical linear modeling . journal Physical Review Physics Education Research volume 15 , pages 020108

2019
[28]

, author Gleditsch, K.S

author Ward, M.D. , author Gleditsch, K.S. , year 2018 . title Spatial regression models . volume volume 155 . publisher Sage Publications

2018
[29]

, author Liu, J

author Zhang, X. , author Liu, J. , author Zhu, Z. , year 2024 . title Learning coefficient heterogeneity over networks: a distributed spanning-tree-based fused-lasso regression . journal Journal of the American Statistical Association volume 119 , pages 485--497

2024
[30]

, year 2006

author Zou, H. , year 2006 . title The adaptive lasso and its oracle properties . journal Journal of the American Statistical Association volume 101 , pages 1418--1429

2006

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...

[2] [2]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...

[3] [3]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...

[4] [4]

, year 2010

author Anselin, L. , year 2010 . title Thirty years of spatial econometrics . journal Papers in Regional Science volume 89 , pages 3--26

2010

[5] [5]

, year 2013

author Anselin, L. , year 2013 . title Spatial econometrics: methods and models . volume volume 4 . publisher Springer Science & Business Media

work page 2013

[6] [6]

, author Gatrell, A.C

author Bailey, T.C. , author Gatrell, A.C. , et al., year 1995 . title Interactive spatial data analysis . volume volume 413 . publisher Longman Scientific & Technical Essex

work page 1995

[7] [7]

, author Fotheringham, A.S

author Brunsdon, C. , author Fotheringham, A.S. , author Charlton, M.E. , year 1996 . title Geographically weighted regression: a method for exploring spatial nonstationarity . journal Geographical Analysis volume 28 , pages 281--298

work page 1996

[8] [8]

, author Zhang, Q

author Chen, Y. , author Zhang, Q. , author Ma, S. , author Fang, K. , year 2024 . title Heterogeneity-aware clustered distributed learning for multi-source data analysis . journal Journal of Machine Learning Research volume 25 , pages 1--60

work page 2024