Pith. sign in

REVIEW 2 major objections 5 minor 41 references

Network Weighted Functional Regression: a method for modeling dependencies between functional data in a network

T0 review · 2 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Network-Weighted Functional Regression (NWFR) weights each node's functional regression by graph geodesic distances, and its conformal wrapper claims distribution-free, finite-sample prediction bands for the whole response curve at a new…

desk verdict The coverage guarantee advertised for the D2 pointwise band is invalid—the conformal set is an L2 ball—and the paper's own simulations show it; the D∞ version works, and the estimation part is sound. read the letter →

arxiv 2501.18221 v3 pith:HAU4CUWO submitted 2025-01-30 stat.ME

Elvira Romano , Antonio Irpino , Claire Miller This is my paper · ORCID
classification stat.ME MSC 62R1062M3062G1562F40
keywords functionaldataanalysisnetworkconformalpredictionregressiongeographicallyweightedintervalscommunitydetectionsensornetworks
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

Functional data that live on a network—sensor readings, brain signals, social measurements—violate the independence assumption of classical functional regression because nearby nodes tend to behave alike. This paper introduces the Network-Weighted Functional Regression (NWFR) model, which replaces the global least-squares fit with a node-wise weighted fit: the influence of another node on the fit at a given node is a Gaussian kernel of the shortest-path distance between them on the graph. The authors report that on simulated stochastic-block networks and on indoor micro-climate sensor data, NWFR raises explained variability from roughly 20% for the unweighted model to above 86% in nearly all scenarios, and it outperforms the spatial GWFR model. To quantify uncertainty, the paper develops a split-conformal prediction procedure in which the vertices are first stratified into communities (via Louvain) to recover approximate exchangeability, and a calibration-set quantile $k_S$ is used to form a prediction band $\hat{Y}_{\nu_j}(t) \pm k_S S(t)$ claimed to contain the full response curve at a new vertex with probability at least $1-\alpha$. The purpose of the paper is to establish that explicitly modeling the network structure yields both better point forecasts and honest, distribution-free prediction intervals for functional responses on networks.

What carries the argument

The central object is the node-specific diagonal weight matrix $W_{\nu_i}$, whose diagonal entries are $w(i,j) = \exp\left\{-\frac{1}{2}(d_{ij}/\theta)^2\right\}$, with $d_{ij}$ the shortest-path (geodesic) distance between vertices $\nu_i$ and $\nu_j$ on the weighted graph and $\theta$ a bandwidth chosen by cross-validation. Given B-spline bases for the covariates and the response, the fitted coefficient block at node $\nu_i$ has the closed form $B_{\nu_i} = (X^T W_{\nu_i} X)^{-1} X^T W_{\nu_i} Y$, a network-weighted version of the functional concurrent regression estimate. The conformal machinery has three parts: a community partition of the vertices (via Louvain) from which at most one vertex per community is drawn into the calibration set; a modulation function $S(t)$, the pointwise root-mean-square calibration error; and the nonconformity scores $D_h = \left(\int_T \left( \frac{\hat{Y}_{\nu_i}(t)-Y_{\nu_i}(t)}{S(t)} \right)^h dt\right)^{1/h}$, with $h=2$ and $h=\infty$ compared. A permutation test on the integrated variance of the coefficient functions across nodes tests whether the network structure has a significant effect on each covariate's coefficient surface.

What would settle it

Simulate networks where within-community pairs are generated with clear hub-dependent variances, use the paper's Louvain split-conformal procedure with the $D_2$ score, and record the empirical fraction of complete curves contained in $\hat{Y}_{\nu_j}(t) \pm k_S S(t)$ on a large test set; if that fraction is substantially below $1-\alpha$ while the fraction of curves inside the $L^2$ ball of radius $k_S$ is at least $1-\alpha$, the pointwise band is not the conformal set and the stated guarantee fails.

Watch

Extended reading notes

Core claim

The paper claims that functional regression on network-indexed data can be made both more accurate and more reliably quantified by (i) replacing the global least-squares fit with a node-wise weighted fit in which the weight between two nodes is a Gaussian kernel of their geodesic distance on the graph, and (ii) surrounding the fitted curves with conformal prediction bands built from an exchangeability-stabilizing partition of the vertices into communities. The authors state that for a new vertex $\nu_j$, with probability at least $1-\alpha$, the whole curve $Y_{\nu_j}$ lies inside the band $\hat{Y}_{\nu_j}(t) \pm k_S S(t)$, where $k_S$ is the $(1-\alpha)$-quantile of the nonconformity scores and $S(t)$ is a data-driven modulation function. Two scores are considered: the $L^2$-based $D_2$ score, which averages squared errors over the domain, and the supremum-based $D_\infty$ score, which focuses on the worst point. In their simulations and in the Intel sensor micro-climate case study, the conformal NWFR bands typically exceed 95% global coverage when $D_\infty$ is used, while $D_2$ gives narrower bands with lower coverage, and the paper interprets this as the standard validity-efficiency trade-off.

Load-bearing premise

The paper assumes that grouping the network's nodes into communities and taking at most one node per community for calibration makes the observations interchangeable in the way conformal prediction requires, and that the reported pointwise band is exactly the prediction set; neither is proved, and the grouping is estimated from the same data.

Editorial extensions

If this is right

  • For network-indexed functional data, NWFR supplies node-specific coefficient surfaces and predictions that explicitly encode graph proximity; a new node's response curve can be predicted from its covariates once its geodesic distances to the training nodes are known.
  • If the conformal coverage claim holds, practitioners get finite-sample marginal coverage without distributional assumptions, so uncertainty bands remain meaningful for non-Gaussian error processes on networks.
  • The comparison of $D_2$ and $D_\infty$ scores makes the validity-efficiency trade-off explicit: supremum-based scores give higher global coverage with wider bands, while $L^2$-based scores give narrower bands at the cost of coverage.
  • The permutation test on coefficient variance provides a direct check of whether the network topology materially changes the regression coefficients for a given covariate.
  • The bandwidth $\theta$ of the Gaussian geodesic kernel controls the neighborhood scale; cross-validation selects it, and the model reduces to the classical concurrent functional regression when all weights are equal.

Reading between the lines

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

  • The paper's coverage guarantee is stated for the pointwise band $\hat{Y}_{\nu_j}(t) \pm k_S S(t)$ together with the $L^2$ score $D_2$, but the conformal set associated with an $L^2$ norm is an $L^2$ ball, not a pointwise envelope; a separate argument would be needed to transfer the coverage guarantee to the reported band.
  • The Louvain partition is estimated from the same data that are later used for calibration, so the 'at most one vertex per community' rule conditions on a data-dependent grouping; treating the partition as fixed rather than estimated could understate the coverage risk.
  • A natural stress test is to study sensitivity to the kernel bandwidth $\theta$: network regions with heterogeneous edge densities may require locally varying bandwidths, and the single global $\theta$ could limit the model's flexibility.
  • The method assumes the new vertex's graph connections (and hence its geodesic distances) are known at prediction time; extending the framework to nodes that join the graph later, or to entirely unseen components, would require estimating distances or a different weight construction.
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.

Referee Report

2 major / 5 minor

Summary. The paper proposes Network-Weighted Functional Regression (NWFR), an extension of geographically weighted functional regression to data observed on a graph. The model estimates node-specific functional coefficient surfaces via weighted least squares, with weights determined by a Gaussian kernel of the geodesic distance on the network. The paper further proposes a split-conformal prediction procedure for functional responses at new vertices, using either an L2-based nonconformity score D2 or a sup-norm score D∞, and claims distribution-free finite-sample prediction bands with guaranteed coverage. The method is evaluated on twelve simulated scenarios with network-structured functional data and on the Intel Berkeley indoor sensor dataset, comparing NWFR against classical functional regression and geographically weighted functional regression. The paper reports improvements in point prediction accuracy and discusses a coverage-efficiency trade-off between D2 and D∞.

Significance. If the stated guarantees held, NWFR would be a useful addition to the literature on second-generation functional data, combining network-aware functional regression with distribution-free uncertainty quantification. The algebraic derivation of the coefficient estimator in Theorem 3.1 (Eq. 7 from Eq. 6) is correct for a fixed weight matrix, and the simulation study is extensive (12 scenarios, 100 replicates). The paper also usefully reports several evaluation metrics (CovG, CovL, ABW, interval score) and is explicit about the validity-efficiency trade-off. However, the central claim of guaranteed coverage for the proposed D2-based pointwise band is invalid, and the paper's own Table 4 empirically contradicts it. The D∞ variant does achieve approximately nominal coverage, but that variant is a direct extension of existing functional conformal methods and is not the paper's proposed nonconformity score.

major comments (2)
  1. [Section 4.1, Step 5] The prediction set defined as C(X_{ν_j}) = {y(t) : Ŷ_{ν_j}(t) − k_S S(t) ≤ y(t) ≤ Ŷ_{ν_j}(t) + k_S S(t) for all t} is not the conformal set associated with the D2 nonconformity score. For D2(Ŷ, Y) = sqrt(∫_T ((Ŷ(t) − Y(t))/S(t))^2 dt), the conformal set is {Y : D2(Ŷ, Y) ≤ k_S}, which is an L2 ball in the metric weighted by 1/S(t), not a pointwise band. Membership in the pointwise band is neither necessary nor sufficient for membership in that ball: a curve can leave the band at a single time point yet have small L2 score, and a curve can remain inside the pointwise envelope while accumulating enough L2 mass to have D2 > k_S. Consequently, even if the exchangeability assumption in Section 4.1 is granted, the finite-sample coverage guarantee applies to the L2 ball, not to the band the paper constructs and evaluates. Table 4 confirms this: for D2 at α = 0.05, global coverage CovG ranges from 80.5% to 85.8% across the twelve scenarios, far below 95%, whereas D∞, whose conformal set is exactly the pointwise band, gives approximately 95–96% coverage. This is an internal inconsistency in the paper's central claim.
  2. [Section 4.1, paragraph on partition] The validity of the entire conformal procedure rests on the assertion that there exists a partition G_1,...,G_K of the vertex set such that within each group the pairs (X_{ν_i}, Y_{ν_i}) are exchangeable, and that drawing at most one vertex from each group into the calibration set 'recovers the exchangeability needed for valid conformal prediction.' This is stated without proof or a precise description of the sampling scheme. Moreover, the Louvain partition used in Step 1 is estimated from the same data, which can itself induce dependence among the sampled units; no result is given showing that the estimated partition satisfies the required exchangeability condition with high probability or in expectation. The citation to Lunde et al. [27] provides intuition about network-aware sampling but does not establish the assertion for the proposed split-conformal algorithm with an estimated community structure. Since the finite-sample coverage guarantee in Step 5 depends on this assumption, it is a load-bearing gap.
minor comments (5)
  1. [Abstract and Introduction] There are several typos: 'asses' should be 'assess', 'uncertainity' should be 'uncertainty', and 'on on appropriately modelling' should be 'on appropriately modelling'.
  2. [Section 5, parameter CBC] The description of the Connectivity Between Communities parameter is contradictory: 'Communities inter link probability is 0.2 (CaseHigh), respectively the0.5 (Case High)' should presumably read '0.2 (Case Low) and 0.5 (Case High)'.
  3. [Section 6, table cross-references] The text refers to 'Table 6' for the model goodness-of-fit metrics, but the table is labeled Table 5; the p-values table is Table 6, not Table 7 as stated in the text.
  4. [Section 4.1, modulation function] The definition of S(t) contains a duplicated phrase: 'Formally, the modulation function can be defined as: as:' should be 'Formally, the modulation function can be defined as:'.
  5. [Section 4.1, nonconformity scores] The sup-norm score D∞ is used in the tables but is never formally defined; the paper should state explicitly that D∞ is the limit h→∞ of D_h, i.e., sup_{t∈T} |Ŷ(t) − Y(t)| / S(t).

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the NWFR estimator and conformal band radius derive from the data and calibration scores, not from a self-referential loop.

full rationale

The derivation chain is self-contained. The NWFR coefficients are the weighted least-squares solution in Eq. (7) of the estimating equation Eq. (6), with network weights in Eq. (5) computed from geodesic distances. The conformal radius k_S is the (1-alpha)-quantile of calibration nonconformity scores R_i = D_h(hatY_nu_i, Y_nu_i), and the band is pointwise +- k_S S(t) with S(t) from calibration residuals; this is the standard split-conformal procedure, not a fitted parameter relabeled as a prediction. The only self-citation, [12], is used in Section 4.2 to name evaluation indices (CovG, CovL, ABW, interval score) and is not load-bearing for the coverage claim. The substantive weaknesses are soundness issues, not circularity: Section 4.1 asserts without proof that stratified calibration recovers exchangeability, and the D2 score yields an L2-ball conformal set rather than the reported pointwise band. These are not reductions of outputs to inputs, so no circular step exists.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The model introduces no new entities. Its free parameters are tuning choices (kernel bandwidth, basis dimension, nonconformity norm, alpha). The key unproven input is the exchangeability of community-stratified observations, on which the coverage guarantee rests.

free parameters (4)
  • bandwidth theta in the Gaussian kernel (Eq. 5) = not reported; selected by cross-validation
    Controls how quickly weights decay with graph geodesic distance; all local regressions depend on it.
  • number of basis functions K/Kp (B-spline) = K=21 in simulation; 'optimal' by cross-validation in real data
    Determines the dimensions of X and the smoothness of estimated coefficient functions.
  • nonconformity exponent h = h=2 proposed; h=infinity compared
    The choice of L2 vs supremum norm determines whether the reported pointwise band equals the conformal prediction set, which is the source of the coverage mismatch.
  • miscoverage level alpha = 0.05
    All simulations and the real-data evaluation use alpha=0.05 to build prediction bands.
assumptions (3)
  • domain assumption The graph geodesic distance d(i,j) (Eq. 1), defined as the minimum sum of edge weights, is the appropriate measure of dependence between nodes.
    All weights in Eq. 5 are a monotone function of this distance; if dependence is governed by other graph features, the model is misspecified.
  • ad hoc to paper There exists a partition G_1,...,G_K of vertices such that within each group the functional pairs are exchangeable, and a new vertex is exchangeable with the calibration set after sampling at most one vertex per group.
    Section 4.1 states this to justify the conformal guarantee; it is not proved and the Louvain partition is data-dependent.
  • standard math The weighted Gram matrix X^T W_nu_i X is invertible for each node nu_i.
    Theorem 3.1 solves B_nu_i = (X^T W_nu_i X)^{-1} X^T W_nu_i Y; in high-dimensional basis settings this may require regularization, which is not discussed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Network Weighted Functional Regression: a method for modeling dependencies between functional data in a network." pith.science (2026). https://pith.science/paper/HAU4CUWO

@misc{pith2026250118221,
  author       = {Pith},
  title        = {Pith review of: Network Weighted Functional Regression: a method for modeling dependencies between functional data in a network},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HAU4CUWO}},
  note         = {Machine review of arXiv:2501.18221}
}
read the original abstract

In this paper, we propose a Network-Weighted Functional Regression (NWFR) model, an extension of Spatially Weighted Functional Regression (SWFR) to functional data defined on network-structured settings. To asses predictive uncertainity, we develop a functional conformal prediction procedure that yields a distribution free prediction intervals with guaranteed coverage. Through extensive evaluation on both simulated and real-world datasets, we demonstrate that the explicit modeling of network structure yields substantive improvements in point-prediction accuracy and markedly enhances the validity and precision of the resulting prediction intervals.

Figures

Figures reproduced from arXiv: 2501.18221 by the authors.

Figure 1
Figure 1. The environmental variables: Relative Humidity (a), Temperature (b) and [PITH_FULL_IMAGE:figures/full_fig_p016_1.png] view at source ↗
Figure 2
Figure 2. The regression coefficient functions β1(t, s, ν24) (up) and β2(t, s, ν24) (down) re￾spectively for the NWFR (left) and GWFR (right). where: V is the set of sensors; E ⊆ V × V are the links between (all the) sensors2 ; FV : V → L2(T) 3 are the functional attributes HUM, LUX and T EMP observed in each node3 ; FE : E → R + are weights associated with each edge defined as the prob￾ability of one sensor of receiving the … view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 35 canonical work pages

  1. [12]

    Diana, E

    A. Diana, E. Romano, and A. Irpino. Distribution free prediction for geograph- ically weighted functional regression models.Spatial Statistics, 57:100765, 2023. ISSN 2211-6753. doi: 10.1016/j.spasta.2023.100765

    work page arXiv 2023

  2. [27]

    Lunde, E

    R. Lunde, E. Levina, and J. Zhu. Conformal prediction for network-assisted re- gression, 2023. URL https://arxiv.org/abs/2302.10095

    arXiv 2023

  3. [1]

    Ajroldi, J

    N. Ajroldi, J. Diquigiovanni, M. Fontana, and S. Vantini. Conformal prediction bands for two-dimensional functional time series.Computational Statistics & Data Analysis, 187:107821, 2023. doi: 10.1016/j.csda.2023.107821

    work page arXiv 2023

  4. [2]

    Andersson, A

    J. Andersson, A. Boerstra, D. Clements-Croome, K. Fitzner, and S. Hanssen. Indoor climates and productivity in offices. REHVA Guidebook no. 6, 2006

    work page 2006

  5. [3]

    Aneiros-Pérez and P

    G. Aneiros-Pérez and P. Vieu. Semi-functional partial linear regression.Statist. Probab. Lett., 76(11):1102–1110, 2006

    work page 2006

  6. [4]

    Arnone, L

    E. Arnone, L. Azzimonti, F. Nobile, and L. Sangalli. Modeling spatially dependent functional data via regression with differential regularization.J. Multivar. Anal., 170:275–295, 2019

    work page 2019

  7. [5]

    Conformalprediction beyond exchangeability

    R.F.Barber, E.J.Candès, A.Ramdas, andR.J.Tibshirani. Conformalprediction beyond exchangeability. The Annals of Statistics, 51(2):816 – 845, 2023. doi: 10.1214/23-AOS2276. URL https://doi.org/10.1214/23-AOS2276

    doi:10.1214/23-aos2276 2023

  8. [6]

    M.-P. D. Bellefon and J.-M. Floch. Geographically weighted regression. InHand- book of Spatial Analysis: Theory and Application with R. Insee M, pages 231–254. Insee Méthodes, October 2018

    work page 2018

Show all 41 references
  1. [7]

    V. D. Blondel, J.-L. Guillaume, R. Lambiotte, and E. Lefebvre. Fast unfolding of communities in large networks.Journal of Statistical Mechanics: Theory and Ex- periment, 2008(10):P10008, oct 2008. doi: 10.1088/1742-5468/2008/10/P10008. URL https://dx.doi.org/10.1088/1742-5468/...

    2008 doi

  2. [8]

    Brunsdon, S

    C. Brunsdon, S. Fotheringham, and M. Charlton. Geographically weighted regres- sion. Journal of the Royal Statistical Society: Series D (The Statistician), 47(3): 431–443, 1998. doi: https://doi.org/10.1111/1467-9884.00145

    1998

  3. [9]

    Cardot, F

    H. Cardot, F. Ferraty, and P. Sarda. Functional linear model.Statist. Probab. Lett., 45(1):11–22, 1999

    1999

  4. [10]

    Chernozhukov, K

    V. Chernozhukov, K. Wüthrich, and Y. Zhu. Distributional conformal prediction. Proceedings of the National Academy of Sciences, 118(48):e2107794118, 2021. doi: 10.1073/pnas.2107794118. 20

    2021 doi

  5. [11]

    Chiou, H.-G

    J.-M. Chiou, H.-G. Muller, and J.-L. Wang. Functional response models.Statistica Sinica, 14(3):675–694, 2004

    2004

  6. [13]

    Diquigiovanni, M

    J. Diquigiovanni, M. Fontana, and S. Vantini. Conformal prediction bands for multivariate functional data.Journal of Multivariate Data Analysis, 2022

    2022

  7. [14]

    L. Fang. Impact of temperature and humidity on the perception of indoor air quality.Indoor Air, 8 (2):80–90, 1998. doi: 10.1111/j.1600-0668.1998.t01-2-00003. x

    1998

  8. [15]

    J. J. Faraway. Regression analysis for a functional response.Technometrics, 39 (3):254–261, 1997

    1997

  9. [16]

    Febrero-Bande and W

    M. Febrero-Bande and W. González-Manteiga. Generalized additive models for functional data. Test, 22(2):278–292, 2013. doi: 10.1007/s11749-012-0308-0

    2013 doi

  10. [17]

    Febrero-Bande, P

    M. Febrero-Bande, P. Galeano, and W. González-Manteiga. Measures of influence for the functional linear model with scalar response.J. Multivariate Anal., 101 (2):327–339, 2010

    2010

  11. [18]

    Fontanella, S

    L. Fontanella, S. Fontanella, R. Ignaccolo, L. Ippoliti, and P. Valentini. G-lasso network analysis for functional data. InFunctional and High-Dimensional Statis- tics and Related Fields. IWFOS 2020. Springer, Cham, 2020

    2020

  12. [19]

    Gneiting and A

    T. Gneiting and A. E. Raftery. Strictly proper scoring rules, prediction, and estimation. Journal of the American Statistical Association, 102 (477):359–378, 2007

    2007

  13. [20]

    P. W. Holland, K. Blackmond Laskey, and S. Leinhardt. Stochastic blockmodels: First steps. Social Networks, 5(2):109–137, 1983. ISSN 0378-8733. doi: https:// doi.org/10.1016/0378-8733(83)90021-7. URL https://www.sciencedirect. com/science/article/pii/0378873383900217

    1983

  14. [21]

    Hollands, E

    J. Hollands, E. Sesto, and A. Korjenic. Thermal comfort in a greened office build- ing: Investigation and evaluation through measurement and survey.Sustainability (Switzerland), 2022

    2022

  15. [22]

    Y. Kato, D. M. Tax, and M. Loog. A review of nonconformity measures for con- formal prediction in regression. In H. Papadopoulos, K. A. Nguyen, H. Boström, and L. Carlsson, editors,Proceedings of the Twelfth Symposium on Conformal and Probabilistic Prediction with Applications...

    2023

  16. [23]

    Kenny, P

    G. Kenny, P. Webb, M. Ducharme, F. Reardon, and O. Jay. Calorimetric mea- surement of post-exercise net heat loss and residual body heat storage.Medicine & Science in Sports & Exercise, 40 (9):1629–1636, 2008. 21

    2008

  17. [24]

    Koner and A.-M

    S. Koner and A.-M. Staicu. Second-generation functional data.Annual Review of Statistics and Its Application, 10(Volume 10, 2023):547–572, 2023. ISSN 2326- 831X. doi: https://doi.org/10.1146/annurev-statistics-032921-033726

    2023 doi

  18. [25]

    J. Lei, M. G’Sell, A. Rinaldo, R. J. Tibshirani, and L. Wasserman. Distribution- free predictive inference for regression.Journal of the American Statistical Asso- ciation, 113(523):1094–1111, 2018. doi: 10.1080/01621459.2017.1307118

    2018

  19. [26]

    R. Lunde. On the validity of conformal prediction for network data under non- uniform sampling, 2023. URLhttps://arxiv.org/abs/2306.07252

    2023 arXiv

  20. [28]

    H. Mao, R. Martin, and B. J. R. and. Valid model-free spatial prediction.Jour- nal of the American Statistical Association, 119(546):904–914, 2024. doi: 10. 1080/01621459.2022.2147531. URL https://doi.org/10.1080/01621459. 2022.2147531

    2024

  21. [29]

    J. Morris. Functional regression.Annu. Rev. Stat. Appl., 2:321–359, 2015

    2015

  22. [30]

    T. L. J. Ng and T. B. Murphy. Weighted stochastic block model. Statisti- cal Methods & Applications, 30(5):1365–1398, December 2021. doi: 10.1007/ s10260-021-00590-

    2021

  23. [31]

    K. Parsons. Human Thermal Environments. The Effects of Hot, Moderate, and Cold Environments on Human Health. Comfort and Performance. Taylor & Fran- cis, 2003

    2003

  24. [32]

    Ramsay and B

    J. Ramsay and B. Silverman. Functional Data Analysis. Springer, New York, 2005

    2005

  25. [33]

    Romano, J

    E. Romano, J. Mateu, and O. Butzbach. Heteroskedastic geographically weighted regression model for functional data. Spatial Statistics, 38:100444, 2020. ISSN 2211-6753. doi: https://doi.org/10.1016/j.spasta.2020.100444

    2020

  26. [34]

    L. M. Sangalli, J. O. Ramsay, and T. O. Ramsay. Spatial spline regression models. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 75: 681–703, 2013

    2013

  27. [35]

    Sawka, R

    M. Sawka, R. Francesconi, A. Young, and K. Pandolf. Influence of hydration level and body fluids on exercise performance in the heat.Medicine & Science in Sports & Exercise, 252 (9):1165–1169, 1984

    1984

  28. [36]

    D. D. Stefano, V. Fuccella, M. Vitale, and S. Zaccarin. Quality issues in co- authorship data of a national scientific community.NETWORK SCIENCE, 11: 98–112, 2023. ISSN 2050-1242

    2023

  29. [37]

    Vimalanathan and T

    K. Vimalanathan and T. Babu. The effect of indoor office environment on the work performance, health and well-being of office workers.Journal of En- vironmental Health Science and Engineering, 12 (1):113, 2014. doi: 10.1186/ s40201-014-0113-7. 22

    2014

  30. [38]

    V. Vovk, A. Gammerman, and G. Shafer. Algorithmic Learning in a Random World. Springer, New York, 2005

    2005

  31. [39]

    M. Ward, K. Stovel, and A. Sacks. Network analysis and political science.Annual Review of Political Science, 14:245–264, 2011

    2011

  32. [40]

    Yamanishi and Y

    Y. Yamanishi and Y. Tanaka. Geographically weighted functional multiple regres- sion analysis: A numerical investigation.Journal of Japanese Society of Compu- tational Statistics, 15:307–317, 2003

    2003

  33. [41]

    Functionallinearregressionanalysisforlon- gitudinal data

    F.Yao, H.-G.Müller, andJ.-L.Wang. Functionallinearregressionanalysisforlon- gitudinal data. The Annals of Statistics, 33(6):2873–2903, 2005. ISSN 00905364. URL http://www.jstor.org/stable/25463398. 23

    2005

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.