pith. sign in

arxiv: 1907.01248 · v1 · pith:PP5CQFQQnew · submitted 2019-07-02 · 📊 stat.CO

Integrated Nested Laplace Approximations (INLA)

Sara Martino , Andrea Riebler This is my paper

Pith reviewed 2026-05-25 10:30 UTC · model grok-4.3

classification 📊 stat.CO
keywords INLAlatent Gaussian modelsBayesian inferencedeterministic approximationsR-INLAposterior marginalsMCMC alternativenumerical integration
0
0 comments X

The pith

INLA approximates posterior distributions deterministically in latent Gaussian models using nested Laplace approximations and numerical integration.

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

The paper describes INLA as a deterministic approach to Bayesian inference for latent Gaussian models that combines analytical approximations with efficient numerical integration to compute posterior quantities. This replaces the sampling process of Markov chain Monte Carlo methods, delivering speed advantages for large or complex models while avoiding convergence and mixing problems. The R-INLA package implements the method and directly supplies posterior marginals for parameters along with model selection and predictive tools. Extensions that pair INLA with MCMC are outlined for cases outside the core model class.

Core claim

INLA is a deterministic paradigm for Bayesian inference in latent Gaussian models introduced in Rue et al. (2009) that relies on a combination of analytical approximations and efficient numerical integration schemes to achieve highly accurate deterministic approximations to posterior quantities of interest, with the primary benefit being computational speed even for large, complex models and the absence of sampling-related issues such as slow convergence.

What carries the argument

Integrated nested Laplace approximations, which apply Laplace approximations inside a numerical integration scheme to obtain posterior marginals for latent Gaussian models.

If this is right

  • Posterior marginals for all model parameters become available without sampling.
  • Model choice criteria and predictive diagnostics are obtained directly from the approximation.
  • Computation remains feasible for large and complex latent Gaussian models.
  • Deterministic execution removes dependence on convergence diagnostics and mixing checks required by MCMC.
  • Hybrid use with MCMC extends the approach to models outside the standard latent Gaussian class.

Where Pith is reading between the lines

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

  • Practitioners could apply the method to real-time updating tasks where repeated MCMC runs would be too slow.
  • The availability of an R package may shift default practice toward deterministic approximations in spatial statistics and time-series modeling.
  • Further work could test whether the same approximation structure extends usefully to nearby model families with minor modifications.

Load-bearing premise

The models must belong to the class of latent Gaussian models for which the specific analytical and numerical approximations of INLA were derived.

What would settle it

Direct comparison of INLA and MCMC posterior marginals on a standard latent Gaussian model where the two differ by more than numerical tolerance would indicate the approximations are not accurate.

Figures

Figures reproduced from arXiv: 1907.01248 by Andrea Riebler, Sara Martino.

Figure 1
Figure 1. Figure 1: a) Observed time series (dots) together with the posterior estimated mean (black line). [PITH_FULL_IMAGE:figures/full_fig_p018_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Posterior marginal for the hyperparameter: on the precision scale (left) and on the standard [PITH_FULL_IMAGE:figures/full_fig_p019_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Posterior mean (solid line) together with 2 [PITH_FULL_IMAGE:figures/full_fig_p019_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Posterior marginal for the linear predictor (left) and for the linear predictor at the obser [PITH_FULL_IMAGE:figures/full_fig_p020_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Estimate of the posterior predictive density [PITH_FULL_IMAGE:figures/full_fig_p020_5.png] view at source ↗
read the original abstract

This is a short description and basic introduction to the Integrated nested Laplace approximations (INLA) approach. INLA is a deterministic paradigm for Bayesian inference in latent Gaussian models (LGMs) introduced in Rue et al. (2009). INLA relies on a combination of analytical approximations and efficient numerical integration schemes to achieve highly accurate deterministic approximations to posterior quantities of interest. The main benefit of using INLA instead of Markov chain Monte Carlo (MCMC) techniques for LGMs is computational; INLA is fast even for large, complex models. Moreover, being a deterministic algorithm, INLA does not suffer from slow convergence and poor mixing. INLA is implemented in the R package R-INLA, which represents a user-friendly and versatile tool for doing Bayesian inference. R-INLA returns posterior marginals for all model parameters and the corresponding posterior summary information. Model choice criteria as well as predictive diagnostics are directly available. Here, we outline the theory behind INLA, present the R-INLA package and describe new developments of combining INLA with MCMC for models that are not possible to fit with R-INLA.

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 / 2 minor

Summary. This manuscript is a short descriptive introduction to the Integrated Nested Laplace Approximations (INLA) method for Bayesian inference in latent Gaussian models (LGMs), originally presented in Rue et al. (2009). It outlines the combination of analytical approximations and numerical integration used by INLA, emphasizes its computational speed and deterministic character relative to MCMC for LGMs, describes the R-INLA R package and the posterior summaries it produces, and sketches recent extensions that hybridize INLA with MCMC for models outside the strict LGM class.

Significance. If the outline is accurate, the paper supplies a concise, practitioner-oriented entry point to an established methodology and its software implementation. The computational advantages cited are supported by the existence and documented performance of the R-INLA package together with the prior literature; the hybrid MCMC extension is a natural and useful broadening of scope. No new empirical claims or derivations are advanced that would require independent validation within this document.

minor comments (2)
  1. [Abstract] The abstract states that the manuscript 'outline[s] the theory behind INLA'; the full text should include at least one or two key equations (e.g., the Laplace approximation step or the nested integration scheme) with explicit pointers to the corresponding derivations in Rue et al. (2009) so that readers can locate the original technical details.
  2. When describing the hybrid INLA-MCMC extension, the manuscript should state the precise class of models for which the hybrid is required and any remaining limitations, to avoid implying that the extension removes the LGM restriction entirely.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive summary of the manuscript and the recommendation to accept.

Circularity Check

0 steps flagged

Descriptive introduction citing prior work; no internal reductions

full rationale

This is an expository overview of the INLA method originally introduced in Rue et al. (2009). The text outlines existing theory, describes the R-INLA package, and notes a hybrid MCMC extension for non-LGM cases. No new derivations, equations, or empirical predictions are advanced within the document itself. The sole citation to the 2009 source paper is not load-bearing for any claim made here, as the paper does not attempt to re-derive or validate the core approximations. This matches the default expectation of no significant circularity for survey-style manuscripts.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no new free parameters, axioms, or invented entities are introduced beyond the established INLA framework described in the cited 2009 work.

pith-pipeline@v0.9.0 · 5715 in / 1005 out tokens · 37134 ms · 2026-05-25T10:30:22.168662+00:00 · methodology

discussion (0)

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

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Self-Supervised Laplace Approximation for Bayesian Uncertainty Quantification

    stat.ML 2026-05 unverdicted novelty 6.0

    SSLA approximates the posterior predictive distribution by refitting Bayesian models on self-predicted data, providing a sampling-free method that improves predictive calibration over classical Laplace approximations ...

Reference graph

Works this paper leans on

37 extracted references · 37 canonical work pages · cited by 1 Pith paper · 1 internal anchor

  1. [1]

    and Dawid, A

    Baio, G. and Dawid, A. P. (2011). Probabilistic sensitivity analysis in health economics.Statistical Methods in Medical Research

    work page 2011

  2. [2]

    Bakka, H., Rue, H., Fuglstad, G.-A., Riebler, A., Bolin, D., Illian, J., Krainski, E., Simpson, D., and Lindgren, F. (2018). Spatial modeling with R-INLA: A review. Wiley Interdisciplinary Reviews: Computational Statistics , 10(6):e1443

    work page 2018

  3. [3]

    Besag, J., York, J., and Molli´ e, A. (1991). Bayesian image restoration, with two applications in spatial statistics. Annals of the Institute of Statistical Mathematics , 43(1):1–20

    work page 1991

  4. [4]

    S., G´ omez-Rubio, V., and Rue, H

    Bivand, R. S., G´ omez-Rubio, V., and Rue, H. (2015). Spatial data analysis with R-INLA with some extensions. Journal of Statistical Software , 63(20):1–31

    work page 2015

  5. [5]

    S., G´ omez-Rubio, V., and Rue, H

    Bivand, R. S., G´ omez-Rubio, V., and Rue, H. (2014). Approximate Bayesian inference for spatial econometrics models. Spatial Statistics, 9:146 – 165

    work page 2014

  6. [6]

    and Lindgren, F

    Bolin, D. and Lindgren, F. (2015). Excursion and contour uncertainty regions for latent Gaussian models. Journal of the Royal Statistical Society: Series B (Statistical Methodology) , 77(1):85–106

    work page 2015

  7. [7]

    Breslow, N. E. (1984). Extra-poisson variation in log-linear models. Journal of the Royal Statis- tical Society. Series C (Applied Statistics) , 33(1):38–44

    work page 1984

  8. [8]

    Brown, P. (2015). Model-based geostatistics the easy way.Journal of Statistical Software, Articles , 63(12):1–24

    work page 2015

  9. [9]

    and Rue, H

    Ferkingstad, E. and Rue, H. (2015). Improving the INLA approach for approximate Bayesian inference for latent Gaussian models. Electron. J. Statist. , 9(2):2706–2731

    work page 2015

  10. [10]

    and Diggle, P

    Giorgi, E. and Diggle, P. (2017). Prevmap: An r package for prevalence mapping. Journal of Statistical Software, Articles , 78(8):1–29. 15

    work page 2017

  11. [11]

    G´ omez-Rubio, V., Molitor, J., and Moraga, P. (2018). Fast bayesian classification for disease mapping and the detection of disease clusters. In Cameletti, M. and Finazzi, F., editors, Quanti- tative Methods in Environmental and Climate Research , pages 1–27, Cham. Springer International Publishing

    work page 2018

  12. [12]

    and Rue, H

    G´ omez-Rubio, V. and Rue, H. (2018). Markov chain Monte Carlo with the integrated nested Laplace approximation. Statistics and Computing , 28(5):1033–1051

    work page 2018

  13. [13]

    K., Friede, T., and Held, L

    Guenhan, B. K., Friede, T., and Held, L. (2018). A design-by-treatment interaction model for network meta-analysis and meta-regression with integrated nested Laplace approximations. Research Synthesis Methods, pages 179–194

    work page 2018

  14. [14]

    and Riebler, A

    Guo, J. and Riebler, A. (2018). meta4diag: Bayesian bivariate meta-analysis of diagnostic test studies for routine practice. Journal of Statistical Software, Articles , 83(1):1–31

    work page 2018

  15. [15]

    and Palm´ ı-Perales, F

    G´ omez-Rubio, V. and Palm´ ı-Perales, F. (2019). Multivariate posterior inference for spatial models with the integrated nested Laplace approximation. Journal of the Royal Statistical Society: Series C (Applied Statistics) , 68(1):199–215

    work page 2019

  16. [16]

    Hastings, W. K. (1970). Monte Carlo sampling methods using Markov chains and their appli- cations. Biometrika, 57(1):97–109

    work page 1970

  17. [17]

    Held, L., Schr¨ odle, B., and Rue, H. (2010). Posterior and cross-validatory predictive checks: A comparison of MCMC and INLA. In Statistical Modelling and Regression Structures . Physica- Verlag HD

    work page 2010

  18. [18]

    Holand, A., Steinsland, I., Martino, S., and Jensen, H. (2013). Animal models and integrated nested Laplace approximations. G3 (Bethesda) , 8(3):1241–1251

    work page 2013

  19. [19]

    Estimating the marginal likelihood with Integrated nested Laplace approximation (INLA)

    Hubin, A. and Storvik, G. (2016). Estimating the marginal likelihood with integrated nested Laplace approximation (INLA). arXiv e-prints , page arXiv:1611.01450

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  20. [20]

    Lindgren, F., Rue, H., and Lindstr¨ om, J. (2011). An explicit link between Gaussian fields and Gaussian Markov random fields: the stochastic partial differential equation approach. Journal of the Royal Statistical Society, Series B (Statistical Methodology) , 73(4):423–498

    work page 2011

  21. [21]

    G., Simpson, D., Lindgren, F., and Rue, H

    Martins, T. G., Simpson, D., Lindgren, F., and Rue, H. (2013). Bayesian computing with INLA: new features. Computational Statistics & Data Analysis , 67:68–83

    work page 2013

  22. [22]

    D., Wakefield, J., Pantazis, A., Lutambi, A

    Mercer, L. D., Wakefield, J., Pantazis, A., Lutambi, A. M., Masanja, H., and Clark, S. (2015). Space–time smoothing of complex survey data: Small area estimation for child mortality. Ann. Appl. Stat. , 9(4):1889–1905

    work page 2015

  23. [23]

    W., Rosenbluth, M

    Metropolis, N., Rosenbluth, A. W., Rosenbluth, M. N., Teller, A. H., and Teller, E. (1953). Equation of state calculations by fast computing machines. The Journal of Chemical Physics , 21(6):1087–1092

    work page 1953

  24. [24]

    Meyer, S., Held, L., and H¨ ohle, M. (2017). Spatio-temporal analysis of epidemic phenomena using the R package surveillance. Journal of Statistical Software , 77(11):1–55

    work page 2017

  25. [25]

    and Held, L

    Riebler, A. and Held, L. (2017). Projecting the future burden of cancer: Bayesian age–period– cohort analysis with integrated nested Laplace approximations. Biometrical Journal , 59(3):531– 549

    work page 2017

  26. [26]

    and Held, L

    Rue, H. and Held, L. (2005). Gaussian Markov Random Fields: Theory And Applications (Monographs on Statistics and Applied Probability) . Chapman & Hall/CRC. 16

    work page 2005

  27. [27]

    Rue, H., Martino, S., and Chopin, N. (2009). Approximate Bayesian inference for latent Gaussian models by using integrated nested Laplace approximations.Journal of the Royal Statistical Society: Series B (Statistical Methodology) , 71(2):319–392

    work page 2009

  28. [28]

    H., Illian, J

    Rue, H., Riebler, A., Sørbye, S. H., Illian, J. B., Simpson, D. P., and Lindgren, F. K. (2017). Bayesian computing with INLA: A review. Annual Review of Statistics and Its Application , 4(1):395–421

    work page 2017

  29. [29]

    Salmon, M., Schumacher, D., and H¨ ohle, M. (2016). Monitoring count time series in R: Aber- ration detection in public health surveillance. Journal of Statistical Software , 70(10):1–35

    work page 2016

  30. [30]

    and Held, L

    Sauter, R. and Held, L. (2016). Quasi-complete separation in random effects of binary response mixed models. Journal of Statistical Computation and Simulation , 86(14):2781–2796

    work page 2016

  31. [31]

    G., and Sørbye, S

    Simpson, D., Rue, H., Riebler, A., Martins, T. G., and Sørbye, S. H. (2017). Penalising model component complexity: A principled, practical approach to constructing priors. Statist. Sci. , 32(1):1–28

    work page 2017

  32. [32]

    J., Best, N

    Spiegelhalter, D. J., Best, N. G., Carlin, B. P., and Van Der Linde, A. (2002). Bayesian mea- sures of model complexity and fit. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 64(4):583–639

    work page 2002

  33. [33]

    and Kadane, J

    Tierney, L. and Kadane, J. B. (1986). Accurate approximations for posterior moments and marginal densities. Journal of the American Statistical Association , 81(393):82–86

    work page 1986

  34. [34]

    A., Leday, G

    Van De Wiel, M. A., Leday, G. G., Pardo, L., Rue, H., Van Der Vaart, A. W., and Van Wieringen, W. N. (2012). Bayesian analysis of RNA sequencing data by estimating multiple shrinkage priors. Biostatistics, 14(1):113–128

    work page 2012

  35. [35]

    Watanabe, S. (2010). Asymptotic equivalence of Bayes cross validation and widely applicable information criterion in singular learning theory. J. Mach. Learn. Res. , 11:3571–3594

    work page 2010

  36. [36]

    and Wakefield, J

    Wilson, K. and Wakefield, J. (2019). Spatial modeling with incomplete data location information. Technical report, University of Washington. in preparation

    work page 2019

  37. [37]

    Wood, S. (2017). Generalized Additive Models: An Introduction with R . Chapman and Hall/CRC, 2 edition. 17 ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● −10 −5 0 5 10 0 10 20 30 40 50 (a) ● ● ● ● ● ● ● ● ● ●0.000 0.005 0.010 0.015 0.020 θ1 θ2 θ3 θ4 θ5 θ6 θ7 θ8 θ9 θ10 (b) 0.0 0.5 1.0 −4 −3 −2 −1 0 (c) 0.00 ...

    work page 2017