pith. sign in

arxiv: 2604.09895 · v1 · submitted 2026-04-10 · 📊 stat.AP · physics.data-an

Blume-Capel model: Estimation of a three stable state network for -bf 1, bf 0 and bf +1 data

Lourens Waldorp , Jonas Dalege , Maarten Marsman , Adam Finnemann , Irene Ferri , Han L. J. van der Maas This is my paper

Pith reviewed 2026-05-10 16:02 UTC · model grok-4.3

classification 📊 stat.AP physics.data-an
keywords Blume-Capel modelIsing modelpseudo-likelihoodlasso regularizationnetwork estimationthree-state dataparameter recoveryconfidence intervals
0
0 comments X

The pith

The Blume-Capel model enables accurate estimation of three-state networks from data valued at -1, 0, and +1 using pseudo-likelihood and lasso.

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

The paper proposes extending the Ising model to the Blume-Capel model to handle responses that include a neutral option alongside negative and positive values. This allows each node in the network to occupy one of three stable states. The model is shown to be a member of the exponential family and to be identified except for the inverse temperature parameter. Estimation proceeds by maximizing the pseudo-likelihood regularized with lasso, which recovers the true parameters accurately even when the network contains few nodes. The desparsified lasso combined with sandwich variance estimation and shrinkage then supplies confidence intervals that achieve good coverage of the true values, as shown in simulations and an application to voting preference data.

Core claim

An extension of the Ising model called the Blume-Capel model is proposed for the inverse problem of estimating parameters from -1, 0, +1 data. The model belongs to the exponential family and is identified apart from the inverse temperature. Pseudo-likelihood estimation combined with the lasso produces accurate parameter recovery for the Blume-Capel model even in small networks. In addition, confidence intervals with good coverage properties are obtained by using the desparsified lasso together with sandwich and shrinkage techniques.

What carries the argument

The Blume-Capel model, which generalizes the Ising model to three states and permits a neutral response while supporting three stable states in the network.

If this is right

  • Parameters can be recovered with high accuracy in small networks by using pseudo-likelihood and lasso regularization.
  • Confidence intervals with reliable coverage are available through the combination of desparsified lasso, sandwich estimators, and shrinkage.
  • The magnetization properties of the model can be characterized using both mean-field theory and direct simulations.
  • The estimation procedure applies directly to empirical data sets such as those collected from online voting advice platforms.

Where Pith is reading between the lines

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

  • The three-state structure opens the possibility of modeling centrist or undecided positions in social or psychological networks without forcing a binary choice.
  • Similar estimation strategies might be adapted to other three-valued data types encountered in biology or economics.
  • Direct comparisons with alternative three-state models could reveal when the Blume-Capel assumptions provide better recovery than simpler multinomial approaches.

Load-bearing premise

The data are generated from the Blume-Capel distribution or a distribution sufficiently close to it.

What would settle it

Repeated simulations from the Blume-Capel model on small networks where the lasso estimates show large bias or the confidence intervals exhibit coverage far from the nominal level would disprove the accuracy and coverage claims.

Figures

Figures reproduced from arXiv: 2604.09895 by Adam Finnemann, Han L. J. van der Maas, Irene Ferri, Jonas Dalege, Lourens Waldorp, Maarten Marsman.

Figure 1
Figure 1. Figure 1: Average proportions of a 5 node BC model (in equilibrium) observed 100 [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: In (a) is the free energy (equation (A.2)), showing three stable fixed points. The settings are β = 2, α 2 = 2, τ = 0, σ = 1 and d = 5. The red circles are repelling fixed points and the blue circles are attracting fixed points. In (b) is the mean field approximation µ (see equation (4)) for the same settings; the intersection with the diagonal (red) line indicates fixed points (circles), with the same int… view at source ↗
Figure 3
Figure 3. Figure 3: Marginal polytope of the BC model with two nodes and one edge, which [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Results for random networks of 10, 20 and 30 nodes with edge probability [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Results for networks of size 10 (circles), 20 (triangles) and 30 (squares), with [PITH_FULL_IMAGE:figures/full_fig_p015_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: In (a) is the network estimate using the BC model with the lasso algorithm, [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: In (a) are the desparsified estimates and the corresponding confidence inter [PITH_FULL_IMAGE:figures/full_fig_p017_7.png] view at source ↗
read the original abstract

An extension of the Ising model is proposed as a viable alternative for data with values $-1$, $0$ and $+1$ in the inverse problem, i.e., estimation of the parameters. This model is called the Blume-Capel (BC) model, adapted from physics for small networks. The advantage of the BC model is not only the fact that it is possible to have a neutral (centrist) position on the response scale, but also that this model allows for three stable states. We illustrate magnetisation properties of the BC model using simulations and mean field results. For estimation of the BC parameters, we show that the BC model is part of the exponential family of distributions and show that the model is identified, except for the (inverse) temperature. We then show that combining pseudo-likelihood with lasso yields accurate parameter recovery for the BC model, even in small networks. Moreover, confidence intervals with good coverage properties can be obtained using the desparsified lasso together with sandwich and shrinkage techniques. We apply the methods to data obtained from the online platform \textit{Stemwijzer}, intended to aid people in deciding for whom to vote.

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

2 major / 2 minor

Summary. The manuscript proposes the Blume-Capel model as an extension of the Ising model suitable for inverse problems with ternary data taking values -1, 0, and +1. It illustrates magnetization properties via mean-field theory and simulations, establishes that the model belongs to the exponential family and is identified except for the inverse temperature, and claims that pseudo-likelihood combined with lasso achieves accurate parameter recovery even for small networks, with desparsified lasso plus sandwich and shrinkage methods yielding confidence intervals with good coverage. The approach is applied to Stemwijzer voting data.

Significance. If the central claims on recovery accuracy and coverage hold, the work supplies a statistically grounded framework for modeling three-state network interactions in small systems, with direct relevance to applications in political science and social network analysis. The integration of pseudo-likelihood, lasso regularization, and desparsified inference addresses high-dimensional estimation challenges in a principled way.

major comments (2)
  1. [§4 and abstract] §4 (Estimation procedure) and abstract: the claim that 'combining pseudo-likelihood with lasso yields accurate parameter recovery for the BC model, even in small networks' and that desparsified lasso gives 'good coverage properties' is load-bearing for the paper's contribution, yet the supporting evidence consists of mean-field analysis plus simulations whose finite-N regimes are not shown to match the small, sparse graphs arising in the Stemwijzer application (N typically < 30).
  2. [§3] §3 (Magnetisation properties): the mean-field magnetisation results are derived under the infinite-N limit; no explicit finite-size correction or scaling analysis is provided for the exact small-N, sparse-graph settings used in the estimation experiments and real-data application, leaving open whether stronger dependence or selection bias in small networks degrades recovery error and sandwich-based coverage.
minor comments (2)
  1. [§2] Notation for the inverse temperature parameter is introduced but its non-identifiability is stated without an explicit statement of the resulting invariance in the likelihood or pseudo-likelihood.
  2. [§6] The Stemwijzer application section would benefit from a table reporting the estimated parameters, their desparsified-lasso confidence intervals, and a brief comparison to an Ising-model baseline on the same data.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments on our manuscript. We address each major comment below and outline the revisions we will make to strengthen the alignment between our theoretical claims, simulations, and the small-network application.

read point-by-point responses

  1. Referee: [§4 and abstract] §4 (Estimation procedure) and abstract: the claim that 'combining pseudo-likelihood with lasso yields accurate parameter recovery for the BC model, even in small networks' and that desparsified lasso gives 'good coverage properties' is load-bearing for the paper's contribution, yet the supporting evidence consists of mean-field analysis plus simulations whose finite-N regimes are not shown to match the small, sparse graphs arising in the Stemwijzer application (N typically < 30).

    Authors: We agree that a closer match between the simulation regimes and the Stemwijzer data characteristics (N typically below 30 and sparse) would better support the central claims. Our current simulations already include small network sizes (N ranging from 10 to 50), but we will add a new set of targeted experiments in the revised Section 4 specifically for N=15, 20, and 25 with edge densities calibrated to the empirical Stemwijzer graphs. These will report parameter recovery error and coverage rates for the desparsified lasso with sandwich and shrinkage estimators, directly addressing whether performance holds in the relevant finite-N, sparse regime. revision: yes

  2. Referee: [§3] §3 (Magnetisation properties): the mean-field magnetisation results are derived under the infinite-N limit; no explicit finite-size correction or scaling analysis is provided for the exact small-N, sparse-graph settings used in the estimation experiments and real-data application, leaving open whether stronger dependence or selection bias in small networks degrades recovery error and sandwich-based coverage.

    Authors: The mean-field analysis in Section 3 is presented to illustrate the existence of three stable states and qualitative magnetization behavior in the thermodynamic limit, serving as theoretical motivation rather than a direct basis for the estimation results. The pseudo-likelihood estimation and desparsified inference procedures are exact for finite N and do not rely on mean-field approximations. Our existing finite-N simulations already validate recovery and coverage for small networks. To address the concern about potential finite-size effects or selection bias, we will insert a brief discussion in the revised manuscript (near the end of Section 3 or in Section 4) that references the small-N simulation outcomes and notes that no substantial degradation is observed empirically, while acknowledging that a full scaling analysis lies beyond the current scope. revision: yes

Circularity Check

0 steps flagged

No significant circularity; standard estimation validated by independent simulation

full rationale

The paper extends the Ising model to the Blume-Capel form for ternary data and applies standard pseudo-likelihood plus lasso estimation, with accuracy claims supported by separate mean-field analysis and simulations rather than any reduction by construction. Model membership in the exponential family and identification (except temperature) are shown directly from the distribution definition without self-referential loops. No load-bearing self-citations, uniqueness theorems imported from the authors, or fitted inputs renamed as predictions appear in the derivation chain. This is the expected non-circular outcome for a methodological statistics paper whose central results rest on external validation techniques.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claims rest on the Blume-Capel model belonging to the exponential family, being identifiable except for temperature, and on the suitability of pseudo-likelihood lasso for small networks; these are stated without derivation in the abstract.

free parameters (1)
  • inverse temperature
    Explicitly stated as the only non-identified parameter in the abstract.
axioms (1)
  • domain assumption The Blume-Capel model is part of the exponential family of distributions
    Invoked to justify estimation and identification results.

pith-pipeline@v0.9.0 · 5538 in / 1196 out tokens · 65979 ms · 2026-05-10T16:02:14.108105+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

61 extracted references · 61 canonical work pages

  1. [1]

    Ak nc , \"U . (2016). Crystal field dilution in s-1 blume capel model: Hysteresis behaviors. Physics Letters A , 380(14-15):1352--1357

    work page 2016

  2. [2]

    and Swendsen, R

    Albert, J. and Swendsen, R. H. (2014). The inverse ising problem. Physics Procedia , 57:99--103

    work page 2014

  3. [3]

    and Cox, D

    Barndorff-Nielsen, O. and Cox, D. (1994). Inference and asymptotics . Chapman and Hall

    work page 1994

  4. [4]

    Barrat, A., Barthelemy, M., and Vespignani, A. (2008). Dynamical processes on complex networks . Cambridge University Press

    work page 2008

  5. [5]

    Besag, J. (1974). Spatial interaction and the statistical analysis of lattice systems. Journal of the Royal Statistical Society. Series B (Methodological) , 36(2):192--236

    work page 1974

  6. [6]

    Blume, M. (1966). Theory of the first-order magnetic phase change in uo. Physical Review , 141(2):517

    work page 1966

  7. [7]

    J., and Griffiths, R

    Blume, M., Emery, V. J., and Griffiths, R. B. (1971). Ising model for the transition and phase separation in he 3-he 4 mixtures. Physical review A , 4(3):1071

    work page 1971

  8. [8]

    Bouabci, M. B. and Carneiro, C. (2000). Random-cluster representation for the blume--capel model. Journal of Statistical Physics , 100:805--827

    work page 2000

  9. [9]

    B \"u hlmann, P. (2013). Statistical significance in high-dimensional linear models. Bernoulli , 19(4):1212--1242

    work page 2013

  10. [10]

    B \"u hlmann, P., Kalisch, M., and Meier, L. (2014). High-dimensional statistics with a view toward applications in biology. Annual Review of Statistics and Its Application , 1:255--278

    work page 2014

  11. [11]

    and van de Geer, S

    B \"u hlmann, P. and van de Geer, S. (2011). Statistics for High-Dimensional Data: Methods, Theory and Applications . Springer

    work page 2011

  12. [12]

    Capel, H. W. (1966). On the possibility of first-order phase transitions in ising systems of triplet ions with zero-field splitting. Physica , 32(5):966--988

    work page 1966

  13. [13]

    and Hjort, N

    Claeskens, G. and Hjort, N. (2008). Model selection and model averaging . Cambridge University Press, Cambridge

    work page 2008

  14. [14]

    Cox, D. (2007). Principles of statistical inference . Cambridge University Press

    work page 2007

  15. [15]

    Dalege, J., Borsboom, D., Van Harreveld, F., Van den Berg, H., Conner, M., and Van der Maas, H. L. (2016). Toward a formalized account of attitudes: The causal attitude network (can) model. Psychological review , 123(1):2

    work page 2016

  16. [16]

    Dezeure, R., B \"u hlmann, P., Meier, L., and Meinshausen, N. (2015). High-dimensional inference: confidence intervals, p-values and r-software hdi. Statistical science , pages 533--558

    work page 2015

  17. [17]

    Epskamp, S., M. G. W. L. J. . B. D. (2015). Network psychometrics. In Irwing, P., Hughes, D., and Booth, T., editors, Handbook of Psychometrics . John Wiley & Sons, New York

    work page 2015

  18. [18]

    A., Korutcheva, E., and de la Rubia, F

    Fernandez, M. A., Korutcheva, E., and de la Rubia, F. J. (2016). A 3-states magnetic model of binary decisions in sociophysics. Physica A: Statistical Mechanics and its Applications , 462:603--618

    work page 2016

  19. [19]

    Ferri, I., P \'e rez-Vicente, C., Palassini, M., and D \' az-Guilera, A. (2022). Three-state opinion model on complex topologies. Entropy , 24(11):1627

    work page 2022

  20. [20]

    and Drton, M

    Foygel, R. and Drton, M. (2010). Extended bayesian information criteria for gaussian graphical models. Advances in neural information processing systems , 23

    work page 2010

  21. [21]

    and Grimmett, G

    Graham, B. and Grimmett, G. (2006). Random-cluster representation of the blume--capel model. Journal of statistical physics , 125:283--316

    work page 2006

  22. [22]

    Grimmett, G. (2022). The potts and random-cluster models. In Handbook of the Tutte Polynomial and Related Topics , pages 378--394. Chapman and Hall/CRC

    work page 2022

  23. [23]

    Grimmett, G. R. (2006). The random-cluster model , volume 333. Springer Science & Business Media

    work page 2006

  24. [24]

    a ggstr \

    H \"a ggstr \"o m, O. (2002). Finite Markov chains and algorithmic applications , volume 52. Cambridge University Press

    work page 2002

  25. [25]

    Haslbeck, J. M. and Waldorp, L. J. (2020). mgm: Structure estimation for time-varying mixed graphical models in high-dimensional data. Journal of Statistical Software , 93(8):1--46

    work page 2020

  26. [26]

    Hastie, T., Tibshirani, R., and Wainwright, M. (2015). Statistical learning with sparsity: the lasso and generalizations . CRC Press

    work page 2015

  27. [27]

    Hirsch, M., Smale, S., and Devaney, R. (2004). Differential equations, dynamical systems, and an introduction to chaos . Academic Press, 2nd edition

    work page 2004

  28. [28]

    Holmgren, R. (1996). A first course in discrete dynamical systems . Springer Science & Business Media

    work page 1996

  29. [29]

    and Montanari, A

    Javanmard, A. and Montanari, A. (2014). Confidence intervals and hypothesis testing for high-dimensional regression. Technical report, arXiv:1306.317

    work page 2014

  30. [30]

    and Kanner, M

    Kaufman, M. and Kanner, M. (1990). Random-field blume-capel model: Mean-field theory. Physical Review B , 42(4):2378

    work page 1990

  31. [31]

    L., et al

    Kindermann, R., Snell, J. L., et al. (1980). Markov random fields and their applications , volume 1. American Mathematical Society Providence, RI

    work page 1980

  32. [32]

    Lehmann, E. L. (1983). Theory of point estimation . Wiley and Sons, New York

    work page 1983

  33. [33]

    Maathuis, M., Drton, M., Lauritzen, S., and Wainwright, M. (2018). Handbook of graphical models . CRC Press

    work page 2018

  34. [34]

    Marsman, M., Borsboom, D., Kruis, J., Epskamp, S., van Bork, R., Waldorp, L., Maas, H. v. d., and Maris, G. (2018). An introduction to network psychometrics: Relating ising network models to item response theory models. Multivariate Behavioral Research , 53(1):15--35

    work page 2018

  35. [35]

    McCulloch, R. E. (1988). Information and the likelihood function in exponential families. The American Statistician , 42(1):73--75

    work page 1988

  36. [36]

    Nelder, J. A. and Wedderburn, R. W. M. (1972). Generalized linear models. Journal of the Royal Statistical Society. Series A (General) , 135(3):370--384

    work page 1972

  37. [37]

    Nguyen, H. (2017). Near universal consistency of the maximum pseudolikelihood estimator for discrete models. Annals of Statistics , 2(2):22--23

    work page 2017

  38. [38]

    Plascak, J., Moreira, J., et al. (1993). Mean field solution of the general spin blume-capel model. Physics Letters A , 173(4-5):360--364

    work page 1993

  39. [39]

    and Bergersen, B

    Plischke, M. and Bergersen, B. (2006). Equilibrium statistical physics . World Scientific Publishing Company, 3rd edition

    work page 2006

  40. [40]

    P \"o tscher, B. M. and Leeb, H. (2009). On the distribution of penalized maximum likelihood estimators: The lasso, scad, and thresholding. Journal of Multivariate Analysis , 100(9):2065--2082

    work page 2009

  41. [41]

    Ravikumar, P., Wainwright, M., and Lafferty, J. (2010). High-dimensional ising model selection using _1 -regularized logistic regression. The Annals of Statistics , 38(3):1287--1319

    work page 2010

  42. [42]

    P., Shaikh, A

    Romano, J. P., Shaikh, A. M., et al. (2012). On the uniform asymptotic validity of subsampling and the bootstrap. The Annals of Statistics , 40(6):2798--2822

    work page 2012

  43. [43]

    and Strimmer, K

    Sch\" a fer, J. and Strimmer, K. (2005). A shrinkage approach to large-scale covariance matrix estimation and implications for functional genomics. Statistical Applications in Genetics and Molecular Biology , 4(1):32

    work page 2005

  44. [44]

    Schmid, C. S. and Hunter, D. R. (2023). Computing pseudolikelihood estimators for exponential-family random graph models. Journal of Data Science , 21(2)

    work page 2023

  45. [45]

    S., Yang, E., and Ravikumar, P

    Suggala, A. S., Yang, E., and Ravikumar, P. (2017). Ordinal graphical models: A tale of two approaches. In International conference on machine learning , pages 3260--3269. PMLR

    work page 2017

  46. [46]

    Tibshirani, R. (1996). Regression shrinkage and selection via the lasso. Journal of the Royal Statistical Society. Series B (Methodological) , 58(1):267--288

    work page 1996

  47. [47]

    Vaart, A. v. d. (1998). Asymptotic Statistics . New York: Cambridge University Press

    work page 1998

  48. [48]

    D., Borsboom, D., Epskamp, S., Blanken, T

    van Borkulo, C. D., Borsboom, D., Epskamp, S., Blanken, T. F., Boschloo, L., Schoevers, R. A., and Waldorp, L. J. (2014). A new method for constructing networks from binary data. Scientific Reports , 4:5918

    work page 2014

  49. [49]

    van de Geer, S., B \"u hlmann, P., Ritov, Y., and Dezeure, R. (2014). On asymptotically optimal confidence regions and tests for high-dimensional models. The Annals of Statistics , 42(3):1166--1202

    work page 2014

  50. [50]

    van der Maas, H., Borsboom, D., and Waldorp, L. (2026). The statistical physics of psychological networks: Zero matters. Psychological Review , ((accepted))

    work page 2026

  51. [51]

    Wainwright, M. J. (2009). Sharp thresholds for high-dimensional and noisy sparsity recovery using-constrained quadratic programming (lasso). IEEE Transactions on Information Theory , 55(5):2183--2202

    work page 2009

  52. [52]

    Wainwright, M. J. and Jordan, M. I. (2008). Graphical models, exponential families, and variational inference. Foundations and Trends in Machine Learning , 1(1-2):1--305

    work page 2008

  53. [53]

    and Haslbeck, J

    Waldorp, L. and Haslbeck, J. (2024). Network inference with the lasso. Multivariate Behavioral Research , 12(in press):1--20

    work page 2024

  54. [54]

    and Kossakowski, J

    Waldorp, L. and Kossakowski, J. (2020). Mean field dynamics of stochastic cellular automata for random and small-world graphs. Journal of Mathematical Psychology , 97:102380

    work page 2020

  55. [55]

    Waldorp, L., Marsman, M., and Maris, G. (2019). Logistic regression and ising networks: prediction and estimation when violating lasso assumptions. Behaviormetrika , 46(1):49--72

    work page 2019

  56. [56]

    White, H. (1981). Consequences and detection of misspecified and nonlinear regression models. Journal of the American Statistical Association , 76(374):419--433

    work page 1981

  57. [57]

    Williams, D. R. (2020). Beyond lasso: A survey of nonconvex regularization in gaussian graphical models

    work page 2020

  58. [58]

    Yang, E., Allen, G., Liu, Z., and Ravikumar, P. K. (2012). Graphical models via generalized linear models. In Advances in Neural Information Processing Systems , pages 1358--1366

    work page 2012

  59. [59]

    and Smith, R

    Young, G. and Smith, R. (2005). Essentials of statistical inference . Cambridge University Press

    work page 2005

  60. [60]

    Bibitem after note You are using a bibitem after a note in a subbibitems environment; note should the last item in a subbibitems environment \@itemnote @bb@error [] @noitemnote \@tempa \@noitemnote \@noitemnote \@itemnote @bibitem\@bibitem @lbibitem\@lbibitem \@bibitem#1 \@itemnote @bibitem #1 \@tempa @noitemnote \@lbibitem[#1]#2 \@itemnote @lbibitem[#1] ...

    work page

  61. [61]

    ' 2<vFۍ/!7u*Q Ƕ ZƫN V] 3RU = v=΍

    @stdbsttrue @ctr \@lbibitem [ @ctr] \@bibitem##1 @ctr \@lbibitem[ @ctr] ##1 @bb@error\@mkbberr @filesw @natbibloaded \@auxout \@itemslabel @bibnum a-- @ctr \@auxout \@itemslabel :s @bibnum \@auxout \@itemslabel @bibnum a-- @ctr \@auxout \@itemslabel :s @bibnum @ctr @bibnum @citex\@citex \@tempcntc @citex[#1]#2 @filesw \@auxout #2 \@tempcnta @\@tempcntb @n...

    work page 1991