pith. sign in

arxiv: 2605.14796 · v1 · pith:TLY6OQBXnew · submitted 2026-05-14 · 📊 stat.ME

A Class of Higher-Order INAR Random Fields for Poisson Counts and Beyond

Christian H. Wei{\ss} , Angelika Silbernagel This is my paper

Pith reviewed 2026-06-30 20:14 UTC · model grok-4.3

classification 📊 stat.ME
keywords INAR modelscount random fieldsdiscrete self-decomposable distributionsPoisson marginalnegative binomial marginalautoregressive dependenceconditional probabilitiesparameter estimation
0
0 comments X

The pith

Combined INAR models allow any discrete self-decomposable marginal distribution while retaining classical autoregressive dependence for count random fields.

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

Existing integer-valued autoregressive models for count random fields face difficulties specifying the stationary marginal distribution and computing the conditional probabilities required for likelihood inference. The paper proposes a new class of combined INAR models that integrate the standard autoregressive thinning with an independent innovation process. This construction keeps the autoregressive dependence structure but permits the marginal distribution to be chosen freely from the broad class of discrete self-decomposable distributions, including Poisson and negative binomial. Stochastic properties are derived in closed form, estimation methods are developed, and the approach is illustrated on agricultural count data.

Core claim

The CINAR construction combines the autoregressive thinning mechanism with an independent innovation process such that the stationary marginal remains any chosen discrete self-decomposable distribution and the conditional probability mass function retains a simple explicit form, removing the characterization barriers of earlier INAR random-field models.

What carries the argument

The CINAR construction, which merges autoregressive thinning with an independent innovation process to enforce a target discrete self-decomposable marginal.

If this is right

  • Conditional probabilities admit an explicit formula usable directly in likelihood or composite-likelihood estimation.
  • Random fields can be equipped with Poisson, negative-binomial, or other discrete self-decomposable marginals without altering the autoregressive dependence structure.
  • Higher-order and multivariate extensions remain feasible under the same preservation property.
  • Parameter estimation routines developed for the model class apply uniformly across the allowable marginal families.

Where Pith is reading between the lines

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

  • The same preservation property may simplify specification of spatial dependence in lattice count data beyond the agricultural example.
  • Overdispersion can be controlled directly through the marginal choice rather than through the dependence parameters.
  • The construction suggests a route to non-stationary versions in which the marginal law stays fixed while the autoregressive coefficients vary with location or time.

Load-bearing premise

Merging the autoregressive thinning operator with an independent innovation process preserves both the chosen discrete self-decomposable marginal and the simple conditional probability expression.

What would settle it

A Monte Carlo simulation of a CINAR process with a Poisson marginal in which the long-run empirical distribution deviates from Poisson or the computed conditional probabilities fail to match the derived closed-form expression.

Figures

Figures reproduced from arXiv: 2605.14796 by Angelika Silbernagel, Christian H. Wei{\ss}.

Figure 1
Figure 1. Figure 1: (a) Process’ “past” corresponding to “time” (s, t). Spatial lags (k, l) and Quadrants I–IV, where (b) equation (7) or (c) equation (8) holds. Graphs adapted from Silbernagel and Weiß (2026a, [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Boxplots of simulated estimates for Poi-CINAR(1, 1) random field (true parameter values at horizontal lines), with increasing sample size (n1, n2) from left to right. and also its standard deviations are substantially lower than those of YW and CLS. This discrepancy in performance can also be seen in [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Boxplots of simulated estimates for NB-CINAR(1, 1) random field (true parameter values at horizontal lines), with increasing sample size (n1, n2) from left to right. Boxplots with white background correspond to model misspecification. 5. Data Example: Yields from an Agricultural Experiment We consider the data set published in the Appendix of Iyer (1942), which is also offered by the R package “agridat” th… view at source ↗
Figure 4
Figure 4. Figure 4: Boxplots of simulated estimates for Poi-CINAR(2, 2) random field (true parameter values at horizontal lines), with increasing sample size (n1, n2) from left to right. Boxplots with white background correspond to model misspecification. horizontal and vertical direction) appears to be another reasonable candidate model; see [PITH_FULL_IMAGE:figures/full_fig_p013_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Wheat yields data from Section 5: (a) gray-scale plot of data, where counts range between 0 (white) and 56 (black), (b) sample PMF (black) compared to Poi(33.8425)-PMF (gray), and (c) sample ACF for spatial lags (k, l) with −2 ≤ k, l ≤ 2. nℓ expresses the number of summands in the conditional log-likelihood function. From the results in [PITH_FULL_IMAGE:figures/full_fig_p014_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Simplified CINAR(1, 1) model for wheat yields data according to [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Simplified CINAR(2, 2) model for wheat yields data according to [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗
read the original abstract

Existing integer-valued autoregressive (INAR) models for count random fields suffer from difficulties in characterizing the stationary marginal distribution and in computing conditional probabilities (as required for likelihood inference). To overcome these drawbacks, the novel class of combined INAR (CINAR) models is proposed, which both exhibits the classical autoregressive dependence structure and allows to specify the marginal distribution within the wide class of discrete self-decomposable distributions. In particular, CINAR random fields can be equipped with a Poisson or negative-binomial marginal distribution. The CINAR's key stochastic properties are derived (including a simple expression for conditional probabilities), and special cases as well as possible extensions are discussed. Approaches for parameter estimation are developed and investigated, and the practical relevance of the novel CINAR family is demonstrated by an agricultural data application.

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.

Referee Report

0 major / 3 minor

Summary. The manuscript proposes a novel class of combined integer-valued autoregressive (CINAR) random fields. By combining the autoregressive thinning operator with an independent innovation process drawn from a discrete self-decomposable distribution, the models retain the classical autoregressive dependence structure while permitting the stationary marginal to be any member of the discrete self-decomposable class (in particular Poisson or negative binomial). The paper derives the key stochastic properties, including an explicit simple form for the conditional probabilities, discusses special cases and extensions, develops estimation procedures, and illustrates the approach on agricultural count data.

Significance. If the preservation of both the target marginal and the simple conditional-probability representation holds after the combination step, the construction removes two long-standing obstacles in INAR modeling of random fields and supplies a flexible, likelihood-friendly framework for spatial count data. The explicit derivation of stochastic properties, the development of estimation methods, and the real-data demonstration constitute concrete strengths.

minor comments (3)
  1. [Introduction] The abstract states that CINAR random fields 'exhibit the classical autoregressive dependence structure'; an explicit statement of the precise dependence order (first-order versus higher-order) and the lattice dimension should appear in the first paragraph of the introduction to match the title.
  2. [Model definition and properties] The claim that the conditional probability expression remains simple for arbitrary discrete self-decomposable innovations is central; placing the explicit formula (rather than only a verbal description) immediately after the model definition would improve readability.
  3. [Estimation] In the estimation section, the simulation study should report coverage probabilities or interval lengths for the marginal parameters in addition to point-estimate bias, to confirm that the marginal specification is recovered reliably.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive summary, significance assessment, and recommendation of minor revision. The report does not list any specific major comments.

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper's central construction combines the standard INAR thinning operator with an independent innovation term to extend the model to random fields while preserving any discrete self-decomposable marginal. This rests on the external, pre-existing definition and representation properties of discrete self-decomposability rather than on any equation or parameter fitted inside the present work. No derivation step equates a claimed prediction to a fitted input by construction, and no load-bearing uniqueness claim is imported solely via self-citation. The derivation chain therefore remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the existence and closure properties of the discrete self-decomposable class under the proposed combination operation; no free parameters, new physical entities, or ad-hoc axioms are stated in the abstract.

axioms (1)
  • domain assumption Discrete self-decomposable distributions remain closed under the CINAR combination of autoregressive thinning and independent innovation.
    Invoked when the abstract states that CINAR models can be equipped with any member of this class while preserving the marginal.

pith-pipeline@v0.9.1-grok · 5665 in / 1311 out tokens · 29782 ms · 2026-06-30T20:14:13.899693+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

21 extracted references · 20 canonical work pages

  1. [1]

    Abdulhamid A

    doi: 10.1007/s00184-023-00934-0. Abdulhamid A. Alzaid and Mohamed A. Al-Osh. An integer-valued pth-order autoregressive structure (INAR(p)) process.Journal of Applied Probability, 27(2):314–324,

    work page doi:10.1007/s00184-023-00934-0

  2. [2]

    Sabyasachi Basu and Gregory C Reinsel

    doi: 10.2307/3214650. Sabyasachi Basu and Gregory C Reinsel. Properties of the spatial unilateral first-order ARMA model.Advances in Applied Probability, 25(3): 631–648,

    work page doi:10.2307/3214650

  3. [3]

    15 Table 2:Wheat yields data from Section 5: CML estimates (standard errors in parentheses) and information criteria for CINAR(2,2) candidate models

    doi: 10.2307/1427527. 15 Table 2:Wheat yields data from Section 5: CML estimates (standard errors in parentheses) and information criteria for CINAR(2,2) candidate models. Modelµ ε θ02 θ01 θ10 θ11 θ12 θ20 θ21 θ22 AIC BIC full 9.420 0.233 0.100 0.313 0.002 0.016 0.057 0.000 0.000 11868.0 11918.4 (0.542) (0.025) (0.023) (0.026) (0.024) (0.020) (0.023) (0.01...

    work page doi:10.2307/1427527

  4. [4]

    Patrick Billingsley.Probability and Measure

    doi: 10.1111/j.2517-6161.1974.tb00999.x. Patrick Billingsley.Probability and Measure. John Wiley & Sons, Ltd, 3rd ed edition,

    work page doi:10.1111/j.2517-6161.1974.tb00999.x 1974

  5. [5]

    Jin-Guan Du and Yuan Li

    doi: 10.2436/20.8080.02.110. Jin-Guan Du and Yuan Li. The integer-valued autoregressive (INAR(p)) model.Journal of Time Series Analysis, 12(2):129–142,

    work page doi:10.2436/20.8080.02.110

  6. [6]

    Alireza Ghodsi

    doi: 10.1111/j.1467-9892.1991.tb00073.x. Alireza Ghodsi. Conditional maximum likelihood estimation of the first-order spatial integer-valued autoregressive (SINAR(1, 1)) model.Journal of the Iranian Statistical Society, 14(2):15–36,

    work page doi:10.1111/j.1467-9892.1991.tb00073.x 1991

  7. [7]

    Alireza Ghodsi and Hassan S

    doi: 10.7508/jirss.2015.02.002. Alireza Ghodsi and Hassan S. Bakouch. Spatial INAR(1,1) model based on mixing Pegram and binomial thinning operators with fitting striga counts.Communications in Statistics - Theory and Methods, 54(21):6988–6996,

    work page doi:10.7508/jirss.2015.02.002 2015

  8. [8]

    Alireza Ghodsi, Mahendran Shitan, and Hassan S Bakouch

    doi: 10.1080/03610926.2025.2465647. Alireza Ghodsi, Mahendran Shitan, and Hassan S Bakouch. A first-order spatial integer-valued autoregressive SINAR(1,

    work page doi:10.1080/03610926.2025.2465647 2025

  9. [9]

    Alireza Ghodsi, Hassan S

    doi: 10.1080/03610926.2011.560739. Alireza Ghodsi, Hassan S. Bakouch, and Mahendran Shitan. First-order spatial dependent count integer-valued autoregressive (Sp-DCINAR(1,1)) process.Communications in Statistics - Simulation and Computation, 53(12):6050–6060,

    work page doi:10.1080/03610926.2011.560739 2011

  10. [10]

    Stephanie Glaser

    doi: 10.1080/03610918.2023.2234681. Stephanie Glaser. A review of spatial econometric models for count data. Hohenheim Discussion Papers in Business, Economics and Social Sciences 19-2017, Universit ¨at Hohenheim, Fakult ¨at Wirtschafts- und Sozialwissenschaften, Stuttgart,

    work page doi:10.1080/03610918.2023.2234681 2023

  11. [11]

    doi: 10.2307/622202. P. V . Krishna Iyer. Studies with wheat uniformity trial data. I. Size and shape of experimental plots and the relative efficiency of different layouts. The Indian Journal of Agricultural Science, 12(1):240–262,

    work page doi:10.2307/622202

  12. [12]

    Robert C

    URLhttps://archive.org/stream/in.ernet.dli.2015.7638/ 2015.7638.The-Indian-Journal-Of-Agricultural-Science-Vol-xii-1942#page/n267/mode/2up. Robert C. Jung and Stephanie Glaser. Modelling and diagnostics of spatially autocorrelated counts.Econometrics, 10(3),

    arXiv 2015

  13. [13]

    Gilberto Pereira Sassi and Carolina Costa Mota Para ´ıba

    doi: 10.1111/stan.12298. Gilberto Pereira Sassi and Carolina Costa Mota Para ´ıba. Conditional least squares estimation for the SINAR(1,

    work page doi:10.1111/stan.12298

  14. [14]

    Angelika Silbernagel and Christian H Weiß

    doi: 10.1080/03610918.2020.1871489. Angelika Silbernagel and Christian H Weiß. The autocorrelation structure of integer-valued autoregressive random fields.Statistics&Probability Letters, 233:110681, 2026a. doi: 10.1016/j.spl.2026.110681. Angelika Silbernagel and Christian H Weiß. The integer-valued moving-average random field.Preprint, 2026b. URLhttps://...

    work page doi:10.1080/03610918.2020.1871489 2020

  15. [15]

    Christian H Weiß

    doi: 10.1080/03610918.2022.2083164. Christian H Weiß. Thinning operations for modeling time series of counts–a survey.AStA Advances in Statistical Analysis, 92(3):319–341, 2008a. doi: 10.1007/s10182-008-0072-3. Christian H Weiß. The combined INAR(p) models for time series of counts.Statistics&Probability Letters, 78(13):1817–1822, 2008b. doi: 10.1016/j.sp...

    work page doi:10.1080/03610918.2022.2083164 2022

  16. [16]

    Christian H Weiß.An Introduction to Discrete-Valued Time Series

    doi: 10.1080/03610920802233937. Christian H Weiß.An Introduction to Discrete-Valued Time Series. John Wiley & Sons, Ltd, 1st ed edition,

    work page doi:10.1080/03610920802233937

  17. [17]

    doi: 10.1007/s11749-025-00994-6. P. Whittle. On stationary processes in the plane.Biometrika, 41(3–4):434–449,

    work page doi:10.1007/s11749-025-00994-6

  18. [18]

    Kai Yang, Mingming Jia, and Xiaogang Dong

    doi: 10.1093/biomet/41.3-4.434. Kai Yang, Mingming Jia, and Xiaogang Dong. An integer-valued spatial autoregressive model with application to covid-19 counts.Journal of Applied Statistics, 0(0):1–33,

    work page doi:10.1093/biomet/41.3-4.434

  19. [19]

    Rong Zhu and Harry Joe

    doi: 10.1080/02664763.2025.2565593. Rong Zhu and Harry Joe. A new type of discrete self-decomposability and its application to continuous-time Markov processes for modeling count data time series.Stochastic Models, 19(2):235–254,

    work page doi:10.1080/02664763.2025.2565593 2025

  20. [20]

    Rong Zhu and Harry Joe

    doi: 10.1081/STM-120020388. Rong Zhu and Harry Joe. Modelling count data time series with Markov processes based on binomial thinning.Journal of Time Series Analysis, 27(5):725–738,

    work page doi:10.1081/stm-120020388

  21. [21]

    Appendix A

    doi: 10.1111/j.1467-9892.2006.00485.x. Appendix A. Derivations Appendix A.1. Proof of Proposition 2.1.1 LetZ(s,t) denote all Bernoulli counting series involved inX s,t at “time” (s,t). From the model recursion in (4), we see thatX s,t can be understood as a function of (Ds,t, εs,t,Z(s,t)). Hence, we may write σ(Xs,t,X s−1,t,X s,t−1, . . .)⊆σ Ds,t, εs,t,Z(...

    work page doi:10.1111/j.1467-9892.2006.00485.x 2006