Prior distributions for variance parameters in hierarchical models (comment on article by Browne and Draper)
Open accessThe DOI matches its Crossref record; 42 of the 42 references it cites were found in a registry or an index; 1 of the 23 sources its claims cite were checked for support.
Abstract
Various noninformative prior distributions have been suggested for scale parameters in hierarchical models. We construct a new folded-noncentral-$t$ family of conditionally conjugate priors for hierarchical standard deviation parameters, and then consider noninformative and weakly informative priors in this family. We use an example to illustrate serious problems with the inverse-gamma family of "noninformative" prior distributions. We suggest instead to use a uniform prior on the hierarchical standard deviation, using the half-$t$ family when the number of groups is small and in other settings where a weakly informative prior is desired. We also illustrate the use of the half-$t$ family for hierarchical modeling of multiple variance parameters such as arise in the analysis of variance.
Overview
CoChat Summary
TL;DR
Gelman argues that inverse-gamma “noninformative” priors for hierarchical variance parameters can badly distort posterior inference, especially when the group-level variance is near zero or the number of groups is small. He recommends a uniform prior on the hierarchical standard deviation when groups are not few, and half-t/half-Cauchy weakly informative priors otherwise, developed within a folded-noncentral-t conditionally conjugate family.
Key findings
- The inverse-gamma(ε, ε) family is not reliably noninformative: as ε → 0 it yields an improper posterior, and in the 8-schools example inverse-gamma(1, 1) and inverse-gamma(0.001, 0.001) priors on σ²_α constrained posterior inferences and increased shrinkage relative to a uniform prior on σ_α.
- A uniform prior on σ_α performed well for the 8-schools data (J = 8), but with only J = 3 schools it produced an extremely long right tail and undershrinkage; a half-Cauchy prior with scale A = 25 removed much of that unrealistic upper tail.
- For multiple variance components, Gelman recommends a hierarchical half-Cauchy model with scale estimated from data; in the 5 × 5 × 2 split-plot latin square Anova with nine variance components, the Cauchy scale parameter A was estimated at 1.8 with a 95% posterior interval of [0.5, 5.1].
- The folded-noncentral-t family, obtained via a redundant multiplicative parameterization, provides conditionally conjugate priors for σ_α and includes the improper uniform density (ν = −1) and proper half-Cauchy (ν = 1) as special cases.
Methods
- Conceptual/comment paper using a simple two-level normal hierarchical model: y_ij ∼ N(μ + α_j, σ²_y), α_j ∼ N(0, σ²_α), focusing on the prior for σ_α.
- Illustrates with the 8-schools example (J = 8, 6000 Bugs simulation draws), a 3-schools subset, and a 5 × 5 × 2 split-plot latin square Anova with nine variance components.
- Uses a parameter-expanded form y_ij ∼ N(μ + ξη_j, σ²_y), η_j ∼ N(0, σ²_η), with σ_α = |ξ|σ_η, to construct and compare uniform, inverse-gamma, half-Cauchy, and hierarchical half-Cauchy priors.
Limitations
- The paper states that noninformative and weakly informative priors are inherently provisional: after fitting, one should inspect the posterior and revise the prior if it does not make sense.
- The paper notes that the uniform prior on σ_α has a positive miscalibration and yields an improper posterior for J = 1 or 2; inverse-gamma inferences are sensitive to ε; and half-Cauchy scale choices are weakly informative rather than fully objective.
- Evident from the text: the recommendations rest on illustrative examples and conceptual argument rather than a broad simulation study or systematic empirical comparison, and extensions to nonlinear, multivariate, and more complex hierarchical models are left for further work.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
0 Valid · 1 Not valid · 14 Not checked · 56 Own results and interpretations
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“Similarly, there are no always-nonnegative classical unbiased estimators of σ α or σ 2 α in the hierarchical model.”
- Also stated in: § 7.1 Prior distributions for variance parameters · plain
“Similarly, there are no alwaysnonnegative classical unbiased estimators of σ α or σ 2 α in the hierarchical model.”
§ Prior distributions for variance parameters in hierarchical models(Comment on Article by Browne and Draper) > 2.4 Calibration
What the sources say
- Could not verifycites Bickel and Blackwell (1967) matched to “A Note on Bayes Estimates” (1967) sourceExperimental — accuracy not yet measured
The introduction (p. 1) and Theorem 1.3 (p. 3) concern when an unbiased Bayes estimate must have zero Bayes risk. Searches for hierarchical models, σₐ, σ²ₐ, and always-nonnegative classical unbiased estimators found no passage addressing the claimed nonexistence result. (LLM, conf=0.77)
- Not checked — no full text of the sourcecites Meng and Zaslavsky (2002) matched to “Single observation unbiased priors” (2002) source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § 7.1 Prior distributions for variance parameters · plain
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“If J is low, the uniform prior density tends to lead to high estimates of σ α , as discussed in Section 5.2.”
§ Prior distributions for variance parameters in hierarchical models(Comment on Article by Browne and Draper) > 7.1 Prior distributions for variance parameters
hedgedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In contrast, the hierarchical half-Cauchy model performs a great deal of shrinkage, especially of the high ranges of the intervals.”
- Also stated in: § Bugs model: a half-Cauchy prior distribution on sigma.theta is induced # using a normal prior on xi and an inverse-gamma on tau.eta model { for (j in 1:J){ # J = the number of schools y[j] ~dnorm (theta[j], tau.y[j]) # data model: the likelihood theta[j] <-mu.theta+ xi*eta[j] tau.y[j] <-pow(sigma.y[j],-2) } xi ~dnorm (0, tau.xi) tau.xi <-pow(prior.scale,-2) for (j in 1:J){ eta[j] ~dnorm (0, tau.eta) # hierarchical model for theta } tau.eta ~dgamma (.5, .5)# chi^2 with 1 d.f.sigma.theta<-abs(xi)/sqrt(tau.eta)# cauchy = normal/sqrt(chi^2) mu.theta ~dnorm (0.0, 1.0E-6) # noninformative prior on mu } · plain
“The half-Cauchy model gives much sharper inferences, using the partial pooling that comes with fitting a hierarchical model.”
§ Prior distributions for variance parameters in hierarchical models(Comment on Article by Browne and Draper) > 6.2 Superpopulation and finite-population standard deviations
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § Bugs model: a half-Cauchy prior distribution on sigma.theta is induced # using a normal prior on xi and an inverse-gamma on tau.eta model { for (j in 1:J){ # J = the number of schools y[j] ~dnorm (theta[j], tau.y[j]) # data model: the likelihood theta[j] <-mu.theta+ xi*eta[j] tau.y[j] <-pow(sigma.y[j],-2) } xi ~dnorm (0, tau.xi) tau.xi <-pow(prior.scale,-2) for (j in 1:J){ eta[j] ~dnorm (0, tau.eta) # hierarchical model for theta } tau.eta ~dgamma (.5, .5)# chi^2 with 1 d.f.sigma.theta<-abs(xi)/sqrt(tau.eta)# cauchy = normal/sqrt(chi^2) mu.theta ~dnorm (0.0, 1.0E-6) # noninformative prior on mu } · plain
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This prior distribution is even more sharply peaked near zero and further distorts posterior inferences, with the problem arising because the marginal likelihood for σ α remains high near zero.”
§ Prior distributions for variance parameters in hierarchical models(Comment on Article by Browne and Draper) > 5.1 Noninformative prior distributions for the 8-schools problem
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The uniform prior distribution seems fine for the 8-school analysis”
- Also stated in: § 5.2 Weakly informative prior distribution for the 3-schools problem · plain
“the default uniform prior gave reasonable results”
- Also stated in: § 7.1 Prior distributions for variance parameters · hedged
“Figure 1 illustrates the generally robust properties of the uniform prior density on σ α .”
§ Prior distributions for variance parameters in hierarchical models(Comment on Article by Browne and Draper) > 5.2 Weakly informative prior distribution for the 3-schools problem
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § 5.2 Weakly informative prior distribution for the 3-schools problem · plain
26 more claims
Create a free account to see all 71 claims and every source verdict.
- It may state two separate claims as one. 34
- Probably not a substantial claim. 12
- The quote only partly states this claim. 1
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Because the uniform distribution tends to unduly favor presence of heterogeneity in discrimination and calibration estimates across studies,58,59”
From A framework for meta-analysis of prediction model studies with binary and time-to-event outcomes (2019) · cites this paper as [59]
Experimental — accuracy not yet measured
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Prior distributions for variance parameters in hierarchical models(Comment on Article by Browne and Draper) > 7.1 Prior distributions for variance parameters
“we see conditional conjugacy in the wider class of half-t distributions on σ α , which include the uniform and half-Cauchy densities on σ α (as well as inverse-gamma on σ 2 α ) as special cases.”
§ Prior distributions for variance parameters in hierarchical models(Comment on Article by Browne and Draper) > 3.2 Folded-noncentral-t prior distribution for σ α
“In the limit that the denominator is specified exactly, we have a folded normal distribution; conversely, specifying the numerator exactly yields the square-root-inverse-χ 2 distribution for σ α , as in Section 3.1.”
Citations
Cites
Gelman, A., Carlin, J. B., Stern, H. S., & Rubin, D. B. (2003) · doi:10.1201/9780429258480 · Cited by 27,276
Self-citationKass, R. E., & Raftery, A. E. (1995) · Journal of the American Statistical Association · doi:10.1080/01621459.1995.10476572 · Cited by 12,558
Sampling-based approaches to calculating marginal densities
Gelfand, A. E., & Smith, A. F. M. (1990) · Journal of the American Statistical Association · doi:10.1080/01621459.1990.10476213 · Cited by 6,606 · cited 1×
“Typically, enough data will be available to estimate µ and σy that one can use any reasonable noninformative prior distribution—for example, p(µ, σy) ∝ 1 or p(µ, log σy) ∝ 1.”
Introducing Multilevel Modeling
Kreft, I., & de Leeuw, J. (1998) · doi:10.4135/9781849209366 · Cited by 2,997
POSTERIOR PREDICTIVE ASSESSMENT OF MODEL FITNESS VIA REALIZED DISCREPANCIES
Gelman, A., Meng, X., & Stern, H. S. (1996) · Cited by 2,638
Self-citationBayes and Empirical Bayes Methods for Data Analysis
(2001) · Technometrics · doi:10.1198/tech.2001.s608 · Cited by 2,098
INADMISSIBILITY OF THE USUAL ESTIMATOR FOR THE MEAN OF A MULTIVARIATE NORMAL DISTRIBUTION
Stein, C. (1956) · Contribution to the Theory of Statistics · doi:10.1525/9780520313880-018 · Cited by 1,841
Bayesian Approaches to Clinical Trials and Health‐Care Evaluation
Spiegelhalter, D. J., Abrams, K. R., & Myles, J. P. (2003) · Wiley eBooks · doi:10.1002/0470092602 · Cited by 1,673
Parametric empirical Bayes inference: theory and applications
Morris, C. N. (1983) · Journal of the American Statistical Association · doi:10.1080/01621459.1983.10477920 · Cited by 1,448 · cited 1×
“On the theoretical side, hierarchical models allow a more “objective” approach to inference by estimating the parameters of prior distributions from data rather than requiring them to be specified using subjective information (see James and Stein, 1960, Efron and Morris, 1975, and Morris, 1983).”
The Selection of Prior Distributions by Formal Rules
Kass, R. E., & Wasserman, L. (1996) · Journal of the American Statistical Association · doi:10.1080/01621459.1996.10477003 · Cited by 1,308
Research works printed but not linked: 5
- Box, G. E. P. and Tiao, G. C. (1973). Bayesian Inference in Statistical Analysis. Reading, Mass.: Addison-Wesley.
- James, W. and Stein, C. (1960). "Estimation with quadratic loss." In Neyman, J. (ed.), Proceedings of the Fourth Berkeley Symposium, volume 1, 361–380. Berkeley: University of California Press.
- Johnson, N. L. and Kotz, S. (1972). Distributions in Statistics. New York: Wiley. 4 vols.
- Liu, C. (2001). "Bayesian analysis of multivariate probit models. Discussion of “The art of data augmentation” by D. A. van Dyk and X. L. Meng." Journal of Computational and Graphical Statistics, 10: 75–81.
- Raudenbush, S. W. and Bryk, A. S. (2002). Hierarchical Linear Models. Thousand Oaks, Calif.: Sage., second edition.
Cited by
Items 2,001–2,100 of 4,089
Gajic‐Veljanoski, O., Cheung, A. M., Bayoumi, A. M., & TOMLINSON, G. A. (2012) · Medical Decision Making · doi:10.1177/0272989x12453504 · Cited by 11
“Third, we decided to research the priors on between-study variance that have been suggested by the Bayesian community.(14,15,26) Having a palette of different noninformative priors supports an idea that even ignorance about the amount of between-study heterogeneity can be described in many ways.”
BackgroundMethodResultInfluentialBayesian semiparametric regression for longitudinal binary processes with missing data
Su, L., & Hogan, J. W. (2008) · Statistics in Medicine · doi:10.1002/sim.3265 · Cited by 9
On Divergence Measures Leading to Jeffreys and Other Reference Priors
Liu, R., Chakrabarti, A., Samanta, T., Ghosh, J. K., & Ghosh, M. (2014) · Bayesian Analysis · doi:10.1214/14-ba862 · Cited by 18
Bayesian multiscale modeling for aggregated disease mapping data
Aregay, M., Lawson, A. B., Faes, C., et al. (2014) · doi:10.1145/2676629.2676640 · Cited by 9
Honest Importance Sampling With Multiple Markov Chains
Tan, A., Doss, H., & Hobert, J. P. (2014) · Journal of Computational and Graphical Statistics · doi:10.1080/10618600.2014.929523 · Cited by 10
“Indeed, Gelman (2006) endorses it, but Bernardo (1996) states that “. . . the use of ‘standard’ improper power priors on the variances is a well documented case of careless prior specification . . . . ” Bernardo goes on to recommend the so-called reference prior of Berger and Bernardo (1992), which…”
BackgroundGeoadditive modeling for extreme rainfall data
Bocci, C., Caporali, E., & Petrucci, A. (2012) · AStA Advances in Statistical Analysis · doi:10.1007/s10182-012-0192-7 · Cited by 9
Small Area Estimation to Quantify Discontinuities in Repeated Sample Surveys
van den Brakel, J., Buelens, B., & Boonstra, H. J. (2015) · Journal of the Royal Statistical Society Series A (Statistics in Society) · doi:10.1111/rssa.12110 · Cited by 9
“The use of a flat prior for σ 2 v or σv instead of the more common inverse gamma prior was recommended by Gelman (2006).”
MethodTurek, D. (2015) · Journal of Probability and Statistics · doi:10.1155/2015/420483 · Cited by 11
Bayesian Model-Averaged Benchmark Dose Analysis via Reparameterized Quantal-Response Models
Fang, Q., Piegorsch, W. W., Simmons, S. J., Li, X., Chen, C., & Wang, Y. (2015) · Biometrics · doi:10.1111/biom.12340 · Cited by 10
Webb, J. A. (2012) · Aquaculture Environment Interactions · doi:10.3354/aei00048 · Cited by 9
“This figure should be set at a level higher than expected for the final estimate (Gelman 2006).”
BackgroundMethodInfluentialPirani, M., Panton, A., Purdie, D. A., & Sahu, S. K. (2016) · The Science of The Total Environment · doi:10.1016/j.scitotenv.2016.04.129 · Cited by 9
“Using a half-Cauchy, in fact, we can restrict the standard deviation parameter, σb, to be away from very large values (Gelman, 2006), that could bias the distribution against zero.”
MethodGeostatistical mixed beta regression: a Bayesian approach
Lagos-Álvarez, B. M., Fustos-Toribio, R., Figueroa-Zúñiga, J., & Mateu, J. (2016) · Stochastic Environmental Research and Risk Assessment · doi:10.1007/s00477-016-1308-5 · Cited by 10
“Specifically, we evaluated the correlation between li and yi given by ql ;y 1⁄4 Corfl ; y g, and also an R(2) measure of explained variation, or predictive power (Gelman 2006).”
MethodDetailed temperature mapping–Warming characterizes archipelago zones
Veneranta, L., Vanhatalo, J. P., & Urho, L. (2016) · Estuarine Coastal and Shelf Science · doi:10.1016/j.ecss.2016.09.011 · Cited by 14
“The length-scale 𝑙ℎ and the magnitude parameters were given a half-Student-t prior (Gelman, 2006).”
MethodMultilevel covariance regression with correlated random effects in the mean and variance structure
Quintero, A., & Lesaffre, E. M. E. H. (2017) · Biometrical Journal · doi:10.1002/bimj.201600193 · Cited by 9
“This option works well in situations with many clusters but has been shown to be troublesome for the estimation of higher order level variances, for example b in model (3), in scenarios with sparse data, see for example Gelman (2006).”
BackgroundNguyen, L. H., & Holmes, S. (2017) · BMC Bioinformatics · doi:10.1186/s12859-017-1790-x · Cited by 12
“[15, 16] as a weakly informative prior for scale parameters, and a default prior for routine applied use in regression models.”
MethodApplications of Small Area Estimation to Generalization With Subclassification by Propensity Scores
Chan, W. (2017) · Journal of Educational and Behavioral Statistics · doi:10.3102/1076998617733828 · Cited by 9
“Specifically, Gelman (2006) Applications of Small Area Estimation to Generalization discovered that using the noninformative uniform (0, B) prior distribution for sv2 led to very large values of the posterior variance estimate of sv2 while using the inverse gamma (E, E) as E!”
BackgroundMethodInfluentialModels for Small Area Estimation for Census Tracts
LOGAN, J. R., Bauer, C., Ke, J., Xu, H., & Li, F. (2019) · Geographical Analysis · doi:10.1111/gean.12215 · Cited by 11
“Gelman (2006) provided a comprehensive review of the prior specifications for variance parameters in Gaussian hierarchical models, and pointed out the challenge in choosing appropriate priors.”
BackgroundAn efficient algorithm for a class of stochastic forward and inverse Maxwell models in R3
Ganesh, M., Hawkins, S. C., & Volkov, D. (2019) · Journal of Computational Physics · doi:10.1016/j.jcp.2019.108881 · Cited by 8
Soil rock content influences the maximum seedling emergence depth of a dominant arid zone grass
Masarei, M., Astfalck, L., Guzzomi, A. L., et al. (2020) · Plant and Soil · doi:10.1007/s11104-020-04493-5 · Cited by 9
“Following recommendations provided in Gelman (2006) and Polson and Scott (2012) prior distributions on the hierarchical variance and mean parameters were set to be half-Cauchy and Normal distributions, respectively.”
Tolvanen, J., Kivelä, S. M., Doligez, B., Morinay, J., Gustafsson, L., Bijma, P., Pakanen, V., & Forsman, J. T. (2020) · Evolution · doi:10.1111/evo.14071 · Cited by 8
“V = 1000; Gelman 2006) for the other variance components.”
BackgroundMadjidian, J. A., Smith, H. G., Andersson, S., et al. (2020) · Journal of Evolutionary Biology · doi:10.1111/jeb.13684 · Cited by 9
“We specified random factors excluding interactions with stage using a compound symmetry structure with the prior of the covariance parameter following a Cauchy distribution with a standard deviation of 25 (Gelman, 2006).”
MethodGin, B. C., Sim, N., Skrondal, A., & Rabe‐Hesketh, S. (2020) · Psychometrika · doi:10.1007/s11336-020-09718-1 · Cited by 15
Paths to social licence for tracking-data analytics in university research and services
White, J. P., Dennis, S. J., Tomko, M., Bell, J., & Winter, S. (2021) · PLoS ONE · doi:10.1371/journal.pone.0251964 · Cited by 10
“Uninforma-tive priors were used for the intercept and random effect standard deviation parameters; both were Cauchy distributed with a location parameter of 0 and a scale parameter of 2.5 [34, 35].”
MethodInfluentialInference in OSNs via Lightweight Partial Crawls
Avrachenkov, K. E., Ribeiro, B., & Sreedharan, J. K. (2016) · ACM SIGMETRICS Performance Evaluation Review · doi:10.1145/2964791.2901477 · Cited by 9
Hierarchical modeling of clinical and expression quantitative trait loci
Sillanpää, M. J., & Noykova, N. (2008) · Heredity · doi:10.1038/hdy.2008.58 · Cited by 11
Thorson, J. T., & Berkson, J. (2010) · Canadian Journal of Fisheries and Aquatic Sciences · doi:10.1139/f10-040 · Cited by 10
“Random-effect variance terms generally require an informative prior to prevent convergence to zero, especially when using a small number of random-effect groups (Gelman et al. 2004; Gelman 2006).”
BackgroundRota, C. T. (2013) · MOspace Institutional Repository (University of Missouri) · doi:10.32469/10355/37807 · Cited by 11
“We assume uniform prior distributions for standard deviation hyper-parameters because the inverse gamma distribution, which is often used as a prior distribution for variance hyper-parameters, can have a strong influence on posterior distributions (Gelman 2006).”
MethodCampbell, H., Pd, V., Maxwell, L., de Jong, V. M. T., Debray, T. P. A., Jänisch, T., & Gustafson, P. (2020) · arXiv (Cornell University) · Cited by 9
Precise Estimation of In Vivo Protein Turnover RatesPreprint
O’Brien, J. J., Narayan, V., Wong, Y. L., Seitzer, P., Sandoval, C. M., Haste, N., Smith, M., Rad, R., Gaun, A., Baker, A., et al. (2020) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2020.11.10.377440 · Cited by 8
JAGS, NIMBLE, Stan: a detailed comparison among Bayesian MCMC softwarePreprint
Beraha, M., Falco, D., & Guglielmi, A. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2107.09357 · Cited by 28
An integrated population model for southern sea otters
Tinker, M. T., Carswell, L. P., Tomoleoni, J. A., Hatfield, B. B., Harris, M. D., Miller, M. A., Moriarty, M. E., Johnson, C. K., Young, C., Henkel, L. A., et al. (2021) · Antarctica A Keystone in a Changing World · doi:10.3133/ofr20211076 · Cited by 14
“…(location = 5 and scale = 1) for initial population size ( N a,1 ) and Cauchy distributions (location = 0 and scale = 2.5) for all other parameters except those constrained to be positive (σ, α, ψ), for which we used half-Cauchy priors with scale =2.5 (Gelman, 2006; Gelman and others, 2008).”
MethodInfluentialSherley, R. B., Barham, B. J., Barham, P., Campbell, K. J., Crawford, R. J. M., Grigg, J. L., Horswill, C., McInnes, A., Morris, T. L., Pichegru, L., et al. (2021) · Proceedings of the Royal Society B Biological Sciences · doi:10.1098/rspb.2021.2129 · Cited by 10
Bayesian model comparison for mortality forecasting
Wong, J. S. T., Forster, J. J., & Smith, P. W. F. (2023) · Journal of the Royal Statistical Society Series C (Applied Statistics) · doi:10.1093/jrsssc/qlad021 · Cited by 13
Fiscal multipliers in recession and expansion. An analysis for the Italian regions
Coppola, G., Destefanis, S., Serio, M. D., et al. (2024) · Structural Change and Economic Dynamics · doi:10.1016/j.strueco.2024.08.012 · Cited by 8
Parvej, M., Devashish, D., & Khan, A. A. (2024) · International Journal of Reliability Quality and Safety Engineering · doi:10.1142/s0218539324500384 · Cited by 8
Ön, Z. B., Ateş, M., & Kaiser, J. (2025) · Progress in Physical Geography Earth and Environment · doi:10.1177/03091333241310742 · Cited by 9
“Following Gelman (2006), we de fi ne the scale parameters of the t -distributions using half-Cauchy distributions that are weakly informative: We sample the location parameter for the change point distribution, μ cp , from a uniform distribution: As suggested by Kruschke (2015), the normality…”
Multi-scale approaches to cope with scale effect issues in macroscopic crash analysis
Zhai, X., Sze, N. N., Lee, J. J., et al. (2025) · Accident Analysis & Prevention · doi:10.1016/j.aap.2025.107971 · Cited by 8
Rao, Y., Zhang, L., Gao, L., Wang, S., & Yang, L. (2025) · Animals · doi:10.3390/ani15081172 · Cited by 13
“Within the Bayesian inference framework, the scaled inverse chi-squared distribution serves as the conjugate prior for the variance parameter of the normal distribution, facilitating posterior derivation and sampling, thereby improving computational efficiency [34].”
Fogarty, C. B., Fay, M. P., Flegg, J. A., Stepniewska, K., Fairhurst, R. M., & Small, D. S. (2015) · Biometrics · doi:10.1111/biom.12307 · Cited by 7
“Our priors on {γ, σ2β } and {η, σ2α } are specified so that we have proper posterior distributions, avoiding the trap described in Hobert and Casella (1996) and following the recommendations of Gelman (2006).”
Brighenti, C. R. G., Cirillo, M. Â., Costa, A. L. A., da Rosa, S. D. V. F., & Guimarães, R. M. (2019) · Scientia Agricola · doi:10.1590/1678-992x-2017-0123 · Cited by 7
“According to Gelman (2006), the Beta conjugated prior distribution P ~ Beta(a,b) is often used to model random variables representing proportions and percentages.”
MethodGhosal, S., Lau, T. S., Gaskins, J., & Kong, M. (2020) · Journal of the Royal Statistical Society Series C (Applied Statistics) · doi:10.1111/rssc.12434 · Cited by 9
Profitability of cryptocurrency Pump and Dump schemes
Tsuchiya, T. (2021) · Digital Finance · doi:10.1007/s42521-021-00034-6 · Cited by 12
Qi, H., Rizopoulos, D., & van Rosmalen, J. (2022) · Research Synthesis Methods · doi:10.1002/jrsm.1561 · Cited by 11
Stoudt, S., Pintar, A. L., & Possolo, A. M. (2021) · Journal of Research of the National Institute of Standards and Technology · doi:10.6028/jres.126.004 · Cited by 11
“The prior distribution assigned to σ was half-Cauchy, following Gelman’s recommendation [31], with median set equal to the median of the absolute deviations of the replicated determinations from their median, rescaled to be a consistent estimate of the standard deviation for Gaussian samples, as implemented in R funct…”
MethodSchmidt, A. M., Freitas, L. P., Cruz, O. G., & Carvalho, M. S. (2022) · Statistical Methods in Medical Research · doi:10.1177/09622802221102628 · Cited by 8
Cheng, Y.-C., Stein, S., Nardone, A., Liu, W., Ma, W., Cohen, G., Guarducci, C., McDonald, T. O., Jeselsohn, R. M., & Michor, F. (2023) · Cancer Research Communications · doi:10.1158/2767-9764.crc-23-0257 · Cited by 8
Lauterbach, A. S., Tober, T., Kunze, F., & Busemeyer, M. R. (2023) · PLoS ONE · doi:10.1371/journal.pone.0295229 · Cited by 14
“Following the recommendation by Gelman [65] for multilevel models with a small number of groups, priors of the half-t family are assigned on the random components.”
MethodA novel fatigue design modeling method under small-sample test data with generalized fiducial theory
Zou, Q., & Wen, J. (2024) · Applied Mathematical Modelling · doi:10.1016/j.apm.2024.01.019 · Cited by 8
Functional analysis of variance (ANOVA) for carbon flux estimates from remote sensing data
Hobbs, J., Katzfuß, M., Nguyen, H., et al. (2024) · Geoscientific model development · doi:10.5194/gmd-17-1133-2024 · Cited by 8
Gómez-Méndez, I., & Amornbunchornvej, C. (2024) · Heliyon · doi:10.1016/j.heliyon.2024.e27968 · Cited by 11
“As commented by [40], hierarchical models allow a more “objective” approach to inference by estimating the parameters of prior distributions from data rather than requiring them to be specified using subjective information.”
Bayesian hierarchical modelling of size spectra
Wesner, J. S., Pomeranz, J. P. F., Junker, J. R., & Gjoni, V. (2024) · Methods in Ecology and Evolution · doi:10.1111/2041-210x.14312 · Cited by 8
“The literature on prior choice is broad and active (Banner et al., 2020; Wesner & Pomeranz, 2021), particularly for priors on hyperparameters like 𝜎 j (Aguilar & Bürkner, 2023; Gelman, 2006).”
BackgroundAn Introduction to Statistical Issues and Methods in Metrology for Physical Science and Engineering
Vardeman, S. B., Hamada, M. S., Burr, T. L., Morris, M. D., Wendelberger, J. R., Jobe, J. M., Moore, L. M., & Wu, H. (2014) · Journal of Quality Technology · doi:10.1080/00224065.2014.11917953 · Cited by 11
“Gelman (2006) discusses this issue in detail.”
MethodZhang, J., Crist, T. O., & Hou, P. (2014) · Environmental and Ecological Statistics · doi:10.1007/s10651-013-0271-2 · Cited by 9
The value of a datum – how little data do we need for a quantitative risk analysis?
Leung, B., & Steele, R. J. (2013) · Diversity and Distributions · doi:10.1111/ddi.12062 · Cited by 8
“We also recognize that there are deep philosophical debates regarding the choice of vague priors in Bayesian analyses (Gelman, 2006), and that these choices can change outcomes (Van Dongen, 2006).”
BackgroundSpatial correlation in Bayesian logistic regression with misclassification
Bihrmann, K., Toft, N., Nielsen, S. S., & Ersbøll, A. K. (2014) · Spatial and Spatio-temporal Epidemiology · doi:10.1016/j.sste.2014.02.002 · Cited by 8
Public investment does not crowd out private supply of environmental goods on private land
Duncan, D. H., Kyle, G., Morris, W. K., & Smith, F. P. (2014) · Journal of Environmental Management · doi:10.1016/j.jenvman.2014.01.041 · Cited by 8
Learning Combinations of Multiple Feature Representations for Music Emotion Prediction
Madsen, J., Jensen, B. S., & Larsen, J. O. (2015) · doi:10.1145/2813524.2813534 · Cited by 8
Separating Biogenic and Adsorbed Pools of Silicon in Sediments Using Bayesian Inference
Siipola, V., Mäntyniemi, S., Lehtimäki, M., et al. (2012) · Silicon · doi:10.1007/s12633-012-9120-4 · Cited by 8
“A compromise between the uninformative and informative priors also exists, called a weakly informative prior [40, 43, 44].”
BackgroundYang, C., Xu, J. X., & Li, Y. (2016) · International Journal of Climatology · doi:10.1002/joc.4607 · Cited by 8
“While assigning considerable weights to values of zero and above, it still keeps very heavy tails (Gelman, 2006) so that both wiggly (small λk) and smooth (large λk) thin plate splines can be accommodated if the model needs to do so.”
BackgroundMethodInfluentialFabrizi, E., & Trivisano, C. (2016) · Scandinavian Journal of Statistics · doi:10.1111/sjos.12229 · Cited by 8
“Similarly, the uniform prior over the range .0; A/ for (Gelman, 2006) implies that p. 2/ / 1 1.0;A/, which may be seen as an approximation to a Gamma 1 2 ; (where D .4A2/ 1) truncated at A2.”
BackgroundInfluentialPredicting coin flips: using resampling and hierarchical models to help untangle the NHL’s shoot-out
Lopez, M. J., & Schuckers, M. (2017) · Journal of Sports Sciences · doi:10.1080/02640414.2016.1198046 · Cited by 4
“We follow Gelman (2006) in using an inverse χ2 for the prior distribution of our precision parameters, the τ’s.”
MethodPhenoCurve: capturing dynamic phenotype-environment relationships using phenomics data
Yang, Y., Xu, L., Feng, Z., Cruz, J. A., Savage, L. J., Kramer, D., & Chen, J. (2016) · Bioinformatics · doi:10.1093/bioinformatics/btw673 · Cited by 8
Fresnedo‐Ramírez, J., Famula, T. R., & Gradziel, T. M. (2017) · Ikushugaku zasshi · doi:10.1270/jsbbs.16027 · Cited by 8
“…⎢ ⎥ ⎣ ⎦ ∼ a ; a Gamma-distributed prior with shape and rate equal to 0.001 Page 15 2 (0.001,0.001)σ Γ ⎡ ⎤ ⎢ ⎥ ⎣ ⎦ ∼ a as shown by Damgaard (2007); and, finally, a uniformly distributed prior with minimum zero and maximum ten Page 16 2 (0,10)σ ⎡ ⎤ ⎢ ⎥ ⎣ ⎦ ∼ a unif , as recommended by Gelman (2006).”
InfluentialA hierarchical Bayesian design for randomized Phase II clinical trials with multiple groups
Yin, J., Qin, R., Sargent, D. J., et al. (2017) · Journal of Biopharmaceutical Statistics · doi:10.1080/10543406.2017.1321007 · Cited by 8
“In the future, we would like to include a uniform improper prior and folded-noncentral-t distribution as recommended by Gelman (2006) to further investigate the sensitivity on priors.”
MethodThe Other Renewable: Hydropower Upgrades and Renewable Portfolio Standards
Fleten, S., Mauritzen, J., & Ullrich, C. J. (2018) · The Energy Journal · doi:10.5547/01956574.39.2.sfle · Cited by 8
“We follow Gelman, Jakulin, Pittau, and Su [8] and Gelman [5] in using weakly informative Cauchy prior distributions on many of the parameters.”
MethodLi, X., Choudhary, P. K., Biswas, S., et al. (2018) · Statistics in Medicine · doi:10.1002/sim.7920 · Cited by 8
“…for variance parameters in hierarchical models has attracted a lot of attention, and various sensible choices have been suggested, especially when no meaningful information is available for such parameters (eg, chapter 5.7 in the work of Spiegelhalter et al, 11 and the work of Gelman 12 ).”
BackgroundMethodInfluentialBayesian hierarchical models for predicting individual performance in soccer
Egidi, L., & Gabry, J. (2018) · Journal of Quantitative Analysis in Sports · doi:10.1515/jqas-2017-0066 · Cited by 8
“See Gelman et al. (2006) and Gelman (2016) for a discussion related to the choice of these priors for the scale parameters.”
BackgroundCampaign finance and perceptions of interest group influence in Australia
Cameron, S., & Wynter, T. (2018) · Political Science · doi:10.1080/00323187.2018.1562307 · Cited by 10
“See Bryan and Jenkins (2015), Gelman (2006), Robson and Pevalin (2015, 26).”
BackgroundRobust approximate Bayesian inference
Ruli, E., Sartori, N., & Ventura, L. (2019) · Journal of Statistical Planning and Inference · doi:10.1016/j.jspi.2019.05.006 · Cited by 12
“ffects (α,γ). As in this case there is no extra-experimental information, we assume vague priors. In particular, αj ∼ N(0,100) and γj ∼ N(0,100), for j= 1,...,6. For the variance components, following Gelman (2006), we assume σ2 1 ∼ halfCauchy(7) and σ2 2 ∼ halfCauchy(7) in both models. However, we note that one of the…”
BackgroundMethodBodnar, O. (2020) · Theory of Probability and Mathematical Statistics · doi:10.1090/tpms/1095 · Cited by 7
Smith, C. L., Thomas, Z. M., Enas, N. H., Thorn, K., Lahn, M., Benhadji, K. A., & Cleverly, A. (2020) · Pharmaceutical Statistics · doi:10.1002/pst.1990 · Cited by 11
Fault Diagnosis Strategy for Wind Turbine Generator Based on the Gaussian Process Metamodel
Zhang, D., Yuan, J., Zhu, J., Ji, Q., Zhang, X., & Liu, H. (2020) · Mathematical Problems in Engineering · doi:10.1155/2020/4295093 · Cited by 9
Linsinger, T. P. J. (2020) · Analytica Chimica Acta X · doi:10.1016/j.acax.2020.100049 · Cited by 8
Bayesian inference for heterogeneity in meta-analysis
Bodnar, O., Nalule, R. M., & Possolo, A. M. (2020) · Metrologia · doi:10.1088/1681-7575/abb064 · Cited by 8
Robust discrete choice models with t-distributed kernel errors
Krueger, R., Bierlaire, M., Gasos, T., & Bansal, P. (2022) · Statistics and Computing · doi:10.1007/s11222-022-10182-3 · Cited by 9
Flexible Bayesian subgroup analysis in early and confirmatory trials
Bunn, V., Liu, R., Lin, J., et al. (2020) · Contemporary Clinical Trials · doi:10.1016/j.cct.2020.106149 · Cited by 8
Stock Assessment of the Lemon Shark off the Southeast United States
Hansell, A. C., Curtis, T. H., Carlson, J. K., Cortés, E., Fay, G., & Cadrin, S. X. (2020) · North American Journal of Fisheries Management · doi:10.1002/nafm.10518 · Cited by 10
Adaptive versus non-adaptive responses to drought in a non-native riparian tree / shrub, Tamarix spp
Bush, S. E., Guo, J., Dehn, D., et al. (2021) · Agricultural and Forest Meteorology · doi:10.1016/j.agrformet.2021.108342 · Cited by 8
“The population-level standard deviations (σa,p, σb,p,j, and σx,p) were given folded Student-t priors with a mean of zero, a standard deviation of 10, and a shape parameter of 2 (Gelman, 2006).”
MethodAlimi, K., & Chakroun, M. (2021) · Journal of the Knowledge Economy · doi:10.1007/s13132-021-00751-8 · Cited by 11
“Unlike other methods, it consists in taking into account the econometric beliefs of the parameters when transforming the likelihood function of the model (see Smets and Wouters, 2003; Gelman, 2006; Koopman et al., 2007; Adjemian and Pelgrin, 2008; Adjemian and Devulder, 2011).”
MethodProduct-form estimators: exploiting independence to scale up Monte Carlo
Kuntz, J., Crucinio, F. R., & Johansen, A. M. (2021) · Statistics and Computing · doi:10.1007/s11222-021-10069-9 · Cited by 9
“However, the estimators we discuss can be applied analogously to the many commonplace hierarchical models [e.g., see Gelman and Hill (2006), Gelman (2006), Koller and Friedman (2009), Hoffman et al. (2013), Blei et al. (2003), and the many references therein] for which such expressions are not…”
BackgroundMethodRobust Bayesian matrix decomposition with mixture of Gaussian noise
Wang, H., Zhang, C., & Zhang, S. (2021) · Neurocomputing · doi:10.1016/j.neucom.2021.04.004 · Cited by 10
Radcliffe, A. J., & Reklaitis, G. V. (2021) · Computers & Chemical Engineering · doi:10.1016/j.compchemeng.2021.107446 · Cited by 8
Shrinkage in the Bayesian analysis of the GGE model: A case study with simulation
de Oliveira, L. A., da Silva, C. P., da Silva, A. Q., Mendes, C. T. E., Nuvunga, J. J., Muniz, J. A., Filho, J. S. D. S. B., & Balestre, M. (2021) · PLoS ONE · doi:10.1371/journal.pone.0256882 · Cited by 8
“This improper prior, however, results in proper full conditional posterior distributions [52, 54], and has no inferential problem as we performed a numerical evaluation of marginal likelihoods [55].”
MethodThe Bayesian Spatial Bradley–Terry Model: Urban Deprivation Modelling in Tanzania
Seymour, R. G., Sirl, D., Preston, S., et al. (2022) · Journal of the Royal Statistical Society Series C (Applied Statistics) · doi:10.1111/rssc.12532 · Cited by 8
“1 which results in a somewhat noninformative distribution (Gelman, 2006).”
BackgroundInfluentialBayesian adaptive design for pediatric clinical trials incorporating a community of prior beliefs
Wang, Y., Travis, J. W., & Gajewski, B. J. (2022) · BMC Medical Research Methodology · doi:10.1186/s12874-022-01569-x · Cited by 10
Kernel-based system identification with manifold regularization: A Bayesian perspective
Mazzoleni, M., Chiuso, A., Scandella, M., Formentin, S., & Previdi, F. (2022) · Automatica · doi:10.1016/j.automatica.2022.110419 · Cited by 8
Using Small Area Estimation to Produce Official Statistics
Young, L. J., & Chen, L. (2022) · Stats · doi:10.3390/stats5030051 · Cited by 8
“The uniform priors for scale parameters σ 2 µ and σ 2 υ are motived by Gelman [16] and Browne and Draper [17].”
MethodBayesian hierarchical random-effects meta-analysis and design of phase I clinical trials
Lin, R., Shi, H., Yin, G., et al. (2022) · The Annals of Applied Statistics · doi:10.1214/22-aoas1600 · Cited by 8
RIMeta : An R shiny tool for estimating the reference interval from a meta‐analysis
Jiang, Z., Cao, W., Chu, H., Bazerbachi, F., & Siegel, L. K. (2023) · Research Synthesis Methods · doi:10.1002/jrsm.1626 · Cited by 11
“After scaling and centering the aggregate data, RIMeta places a N(0, 1000) prior on μ and Unif (0,10) priors on σ and τ: Gelman et al.20 recommended the use of noninformative uniform priors on standard deviation parameters when the number of studies is larger than 5.”
MethodÖn, Z. B., Macdonald, N., Ön, S. A., et al. (2023) · The Holocene · doi:10.1177/09596836231163508 · Cited by 8
“The distribution of the scale parameter of the t -distribution, σ , is selected to be half-Cauchy (equation (3)) as offered by Gelman (2006).”
MethodKwon, H., Lall, U., & Engel, V. (2011) · Water Resources Research · doi:10.1029/2010wr009552 · Cited by 8
King, M. D., Calamante, F., Clark, C. A., & Gadian, D. G. (2011) · Journal of Statistical Software · doi:10.18637/jss.v044.i02 · Cited by 8
“Some recommendations and other information regarding the specification of non-informative priors in the random effects context are given in Section 9.2 of the Classic BUGS manual (Spiegelhalter et al. 1995), while a more comprehensive discussion is provided by Gelman (2006).”
MethodInfluentialZhu, X., Tallman, R. F., Howland, K. E., et al. (2015) · Journal of Great Lakes Research · doi:10.1016/j.jglr.2015.08.013 · Cited by 8
An estimation method for latent traits and population parameters in Nominal Response Model
Azevedo, C., & de Andrade, D. F. (2010) · Brazilian Journal of Probability and Statistics · doi:10.1214/09-bjps102 · Cited by 8
Shahbaba, B., & Johnson, W. O. (2012) · Statistics in Medicine · doi:10.1002/sim.5680 · Cited by 9
Confidence About Possible Explanations
Apolloni, B., & Bassis, S. (2011) · IEEE Transactions on Systems Man and Cybernetics Part B (Cybernetics) · doi:10.1109/tsmcb.2011.2158306 · Cited by 8
A Bayesian Reliability Analysis of Neutron-Induced Errors in High Performance Computing Hardware
Storlie, C. B., Michalak, S., Quinn, H., Dubois, A. J., Wender, S. A., & DuBois, D. H. (2013) · Journal of the American Statistical Association · doi:10.1080/01621459.2013.770694 · Cited by 9
Bayesian multivariate growth curve latent class models for mixed outcomes
Leiby, B. E., Have, T. R. T., Lynch, K. G., et al. (2012) · Statistics in Medicine · doi:10.1002/sim.5596 · Cited by 8
Spatial mapping of hepatitis C prevalence in recent injecting drug users in contact with services
Harris, R., Hope, V. D., MORONGIU, A., Hickman, M., Ncube, F., & Angelis, D. D. (2011) · Epidemiology and Infection · doi:10.1017/s0950268811001634 · Cited by 10
A Bayesian Analysis of Correlated Interval-Censored Data
Zuma, K. K. (2007) · Communication in Statistics- Theory and Methods · doi:10.1080/03610920601033710 · Cited by 8
“For a comprehensive review of methods for eliciting priors, see for example Gelman (2006).”
Method
Metadata
- Authors
- Gelman, A.
- Affiliations
- Columbia University (US)
- Corresponding author
- Gelman, Andrew
- ORCID
- Gelman, Andrew
- Venue
- Bayesian Analysis
- Publisher
- Institute of Mathematical Statistics
- Published
- 2006-09-01
- Volume
- 1
- Issue
- 3
- Pages
- 515-534
- ISSN
- 1931-6690 (ISSN-L)
- 1936-0975
- OpenAlex
- W1536497620
- Semantic Scholar
- 63d40805d4635b70f6b5aa07fc2c9821480cfc1e
- Cited by
- 4,050
- References
- 34
- Citations by year
- 2026: 194 · 2025: 218 · 2024: 237 · 2023: 237 · 2022: 282 · 2021: 312 · 2020: 320 · 2019: 281 · 2018: 266 · 2017: 263 · 2016: 235 · 2015: 217 · 2014: 176 · 2013: 142 · 2012: 184
- FWCI
- 78.70
- Type
- journal article
- Language
- en
- Open access
- diamond
- doi.org
- Licence
- cc-by
- Locations
- Bayesian Analysis (publishedVersion, cc-by)
- Project Euclid (Cornell University) (submittedVersion)
- Funders
- National Science Foundation
- Topics
- Physical Sciences › Mathematics › Statistics and Probability › Statistical Methods and Bayesian Inference
- Physical Sciences › Mathematics › Statistics and Probability › Statistical Distribution Estimation and Applications
- Physical Sciences › Computer Science › Artificial Intelligence › Bayesian Methods and Mixture Models
- Keywords
- prior distributions, hierarchical models, scale parameter, non-informative priors, weakly informative priors, uniform prior, analysis of variance
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-02, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-02 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.