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 2,401–2,500 of 4,089
Climate and weather have differential effects in a high latitude passerine community
Mizel, J. D., Schmidt, J. H., & McIntyre, C. L. (2021) · Oecologia · doi:10.1007/s00442-020-04847-x · Cited by 5
“We specified vague normal priors for all regression coefficients and half-Cauchy priors for the hierarchal standard deviation parameters (Gelman 2006).”
MethodBayesian Analysis for Multiple-baseline Studies Where the Variance Differs across Cases in OpenBUGS
Baek, E., & Ferron, J. M. (2021) · Developmental Neurorehabilitation · doi:10.1080/17518423.2020.1858455 · Cited by 5
Miec: A Bayesian hierarchical model for the analysis of nearby young open clusters
Olivares, J., Bouy, H., Sarro, L. M., Moraux, E., Berihuete, Á., Galli, P. A. B., & Miret-Roig, N. (2021) · Astronomy and Astrophysics · doi:10.1051/0004-6361/202140282 · Cited by 7
Modelling Nonstationary Spatial Lag Models with Hidden Markov Random Fields
Ghiringhelli, C., Bartolucci, F., Mira, A., & Arbia, G. (2021) · Spatial Statistics · doi:10.1016/j.spasta.2021.100522 · Cited by 6
Patterns of low birth weight in greater Mexico City: A Bayesian spatio-temporal analysis
Lome-Hurtado, A., Li, G., Touza, J., et al. (2021) · Applied Geography · doi:10.1016/j.apgeog.2021.102521 · Cited by 5
A Uniform Shrinkage Prior in Spatiotemporal Poisson Models for Count Data
Lekdee, K., Yang, C., Ingsrisawang, L., & Li, Y. (2021) · Emerging topics in statistics and biostatistics · doi:10.1007/978-3-030-72437-5_4 · Cited by 5
An approach of Bayesian variable selection for ultrahigh‐dimensional multivariate regression
Dai, X., Fu, G., Reese, R., Zhao, S., & Shang, Z. (2022) · Stat · doi:10.1002/sta4.476 · Cited by 5
Sy, M. M., Ancelet, S., Henner, P., Hurtevent, P., & Simon-Cornu, M. (2015) · Journal of Environmental Radioactivity · doi:10.1016/j.jenvrad.2015.05.007 · Cited by 5
Improving rating curve accuracy by incorporating water balance closure at river bifurcations
Gensen, M. R. A., Warmink, J. J., Berends, K. D., et al. (2022) · Journal of Hydrology · doi:10.1016/j.jhydrol.2022.127958 · Cited by 5
Different routes to liking: how readers arrive at narrative evaluations
Mak, M., Faber, M., & Willems, R. M. (2022) · Cognitive Research Principles and Implications · doi:10.1186/s41235-022-00419-0 · Cited by 4
“As variance can only be positive, weakly regularizing, half-cauchy priors with a mean of 0 and a standard deviation of 1 were used for the variance of the random effects as well as the overall variance (as suggested by Gelman, 2006; McElreath, 2016).”
MethodVariable selection in linear-circular regression models
Çamlı, O., Kalaylıoğlu, Z., & SenGupta, A. (2022) · Journal of Applied Statistics · doi:10.1080/02664763.2022.2110860 · Cited by 5
“One such prior is HC hyper-prior which is becoming increasingly popular in the Bayesian literature as a robust alternative [12,23,26,28].”
BackgroundKasada, M., Nakashima, Y., Fukasawa, K., Yajima, G., Yokomizo, H., & Miyashita, T. (2022) · Population Ecology · doi:10.1002/1438-390x.12138 · Cited by 6
“We applied vague or weakly informative priors for the other parameters: normal distribution with mean 0 and standard deviation 100 for b 0 , b 1 , b 2 , and μ 1 ; half-Cauchy with mode 0 and scale 5 for σ s and σ 1 (Gelman, 2006), and uniform distribution with range ( (cid:5) 5, 15) for ln( m ).”
Applying Negative Binomial Distribution in Diagnostic Classification Models for Analyzing Count Data
Liu, R., Heo, I., Liu, H., Shi, D., & Jiang, Z. (2022) · Applied Psychological Measurement · doi:10.1177/01466216221124604 · Cited by 5
“Speci fi cally, Gelman (2006) argued that if the variance estimates are very small (e.g., close to 0), inverse gamma may be sensitive to inference problems.”
BackgroundMethodWang, G., Poulin-Costello, M., Pang, H., Zhu, J., Helms, H., Reyes‐Rivera, I., Platt, R. W., Pang, M., & Koukounari, A. (2023) · Pharmaceutical Statistics · doi:10.1002/pst.2336 · Cited by 10
“More detailed discussion about prior distributions for variance parameters in Bayesian hierarchical models can be found in Gelman (2006).”
BackgroundBayesian hierarchical models for adaptive basket trial designs
Chen, C., & Hsiao, C. (2023) · Pharmaceutical Statistics · doi:10.1002/pst.2289 · Cited by 9
Parker, P. A., Janicki, R., & Holan, S. H. (2023) · Journal of Survey Statistics and Methodology · doi:10.1093/jssam/smad022 · Cited by 5
Bayesian inference of W-boson mass
Rallapalli, A., & Desai, S. (2023) · The European Physical Journal C · doi:10.1140/epjc/s10052-023-11754-x · Cited by 5
Kuile, A. M., Ogle, K., Cavanaugh, C. T., Dudley, J. G., Markus, A., Saab, V. A., Wright, M. E., & Sanderlin, J. S. (2023) · Forest Ecology and Management · doi:10.1016/j.foreco.2023.121443 · Cited by 7
Bayesian analysis of systematic errors in the determination of the constant of gravitation
Rinaldi, S., Middleton, H., Pozzo, W. D., et al. (2023) · The European Physical Journal C · doi:10.1140/epjc/s10052-023-12078-6 · Cited by 5
“We will consider several possible choices for this distribution, following some of the prescriptions discussed in [29]:”
Poursina, D., & Brorsen, B. W. (2023) · Stochastic Environmental Research and Risk Assessment · doi:10.1007/s00477-023-02615-w · Cited by 8
“Gelman (2006) suggests using a uniform distribution for hierarchical variance parameters.”
Bayesian joint-regression analysis of unbalanced series of on-farm trials
Turbet-Delof, M., Rivière, P., Dawson, J. C., Gauffreteau, A., Goldringer, I., van Frank, G., & David, O. (2025) · Peer Community Journal · doi:10.24072/pcjournal.495 · Cited by 8
“Weakly informative prior distributions are placed on σ ε and ν (Cao et al., 2022; Gelman, 2006; Juárez and Steel, 2010): Michel Turbet Delof et al. 5 Peer Community Journal, Vol. 5 (2025), article e4 https://doi.org/10.24072/pcjournal.495 where λ ε is a known prior value of the standard deviation…”
InfluentialThe Evolution of Rodent Tail Morphology
Sheard, C., Skinner, N., & Caro, T. (2024) · The American Naturalist · doi:10.1086/729751 · Cited by 5
Abe, H., & Kawasaki, S. (2024) · International Journal of Food Microbiology · doi:10.1016/j.ijfoodmicro.2024.110618 · Cited by 5
Harnessing historic records and long-term monitoring data to evaluate amphibian extinction dynamics
West, M., Hunter, D. A., Scroggie, M. P., Johnson, G. N., Smith, S. A., Mccarthy, M. A., & Gillespie, G. R. (2024) · Biological Conservation · doi:10.1016/j.biocon.2024.110477 · Cited by 5
Song, J., Liu, J., Zhang, X., Chen, X., Shang, Y., & Gao, F. (2024) · Land · doi:10.3390/land13060833 · Cited by 6
Chen, G., & O’Malley, A. J. (2024) · Applied Network Science · doi:10.1007/s41109-024-00627-1 · Cited by 5
Bayesian multivariate spatiotemporal statistical modeling of bus and taxi ridership
Luan, H., Zhang, S., & Fu, X. (2024) · Journal of Transport Geography · doi:10.1016/j.jtrangeo.2024.104032 · Cited by 3
Considering ensemble spread improves rainfall forecast post‐processing
Wang, Q. J., Huang, Z., Robertson, D. E., et al. (2025) · Quarterly Journal of the Royal Meteorological Society · doi:10.1002/qj.4925 · Cited by 5
“…level, 𝜎 g ( t ) receives a simple inverse prior: Parameters 𝛽 and 𝜎 𝜇 h receive half-Cauchy priors, which is common for hierarchical variance parameters (Gelman, 2006): 𝜇 g ( t ) receives a normal distribution prior, which we developed to prevent runaways in the augmentation of censored…”
Quintero, D., Mishra, V., Limaye, A., Abel, N. V., Ross, J. B., & Rashid, A. (2025) · European Journal of Agronomy · doi:10.1016/j.eja.2025.127524 · Cited by 5
A Multiplicative Model for Gauge R & R Studies
Burr, T. L., & Hamada, M. S. (2014) · Quality and Reliability Engineering International · doi:10.1002/qre.1638 · Cited by 5
Young, J., Mucsi, I., Rollet‐Kurhajec, K. C., et al. (2015) · HIV Medicine · doi:10.1111/hiv.12305 · Cited by 5
Use of Bayesian Methods for Multivariate Bioequivalence Measures
de Souza, R. M., Achcar, J. A., & Martínez, E. Z. (2009) · Journal of Biopharmaceutical Statistics · doi:10.1080/10543400802513676 · Cited by 5
“In Tables 2 and 3 we also have posterior summaries obtained assuming uniform U 0 100 prior distributions for 2 and 2 (see, for example, Gelman, 2006).”
MethodUsing Item Response Theory to Model Multiple Phenotypes and Their Joint Heritability in Family Data
Fragoso, T. D. M., Giolo, S. R., Pereira, A. C., de Andrade, M., & Soler, J. M. P. (2014) · Genetic Epidemiology · doi:10.1002/gepi.21784 · Cited by 5
“) ∼ G amma(1, 1) to avoid degenerate variance estimates [Gelman, 2006; Wang et al., 1994].”
BackgroundBayesian hierarchical models for food frequency assessment
Song, H., Lawson, A. B., & Nitcheva, D. K. (2010) · Canadian Journal of Statistics · doi:10.1002/cjs.10052 · Cited by 5
Paddock, S. M. (2007) · Journal of the Royal Statistical Society Series C (Applied Statistics) · doi:10.1111/j.1467-9876.2007.00578.x · Cited by 5
“…problems that are inherent with specifying vague gamma prior distributions for variance components in a hierarchical model when the number of higher level units is small (Gelman, 2006), the inverse variance of the programme random effects, σ−1φ , is given a tighter prior, a gamma(1,1) distribution.”
BackgroundConvergence analysis of block Gibbs samplers for Bayesian linear mixed models with $p>N$
Abrahamsen, T., & Hobert, J. P. (2016) · Bernoulli · doi:10.3150/15-bej749 · Cited by 5
Arcuti, S., Pollice, A., Ribecco, N., & D’Onghia, G. (2015) · Biometrical Journal · doi:10.1002/bimj.201400123 · Cited by 5
“Distributions, belonging to the family of conditionally conjugate folded noncentral-t priors, are considered for random effect standard deviations (Gelman, 2006; Lunn et al., 2013).”
BackgroundRobust Bayesian hierarchical model using normal/independent distributions
Chen, G., & Luo, S. (2015) · Biometrical Journal · doi:10.1002/bimj.201400255 · Cited by 7
Li, L., Feng, C. X., & Qiu, S. (2017) · Statistics in Medicine · doi:10.1002/sim.7278 · Cited by 5
“When the number of regions is small, the inverse Gamma for τ2 may be better to replaced by a less restrictive prior such as the half Cauchy [17].”
BackgroundDuncan, E. W., White, N. M., & Mengersen, K. L. (2016) · BMJ Open · doi:10.1136/bmjopen-2015-010253 · Cited by 6
Beyond p-values in the evaluation of brain–computer interfaces: A Bayesian estimation approach
Melinščak, F., & Montesano, L. (2016) · Journal of Neuroscience Methods · doi:10.1016/j.jneumeth.2016.06.008 · Cited by 5
Sullivan, P. J., & Rudstam, L. G. (2016) · ICES Journal of Marine Science · doi:10.1093/icesjms/fsw080 · Cited by 5
“Relatively uninformative priors are used for the slope term a and the regression variance r2Regression following the recommendations of Gelman (2006) under a Gaussian distribution as follows: a Nð0; r2aÞ rRegression HalfCauchyðv; sÞ v Nð0;r2vÞ s Cðc1; c2Þ (see Table 1 for prior parameter values),…”
MethodMultiscale Bayesian state-space model for Granger causality analysis of brain signal
Cekic, S., Grandjean, D., & Renaud, O. (2018) · Journal of Applied Statistics · doi:10.1080/02664763.2018.1455814 · Cited by 6
“The main recommendation in Gelman [15] is to use half-t priors on standard deviation parameters to achieve arbitrarily high noninformativeness.”
MethodCao, C., Chen, M., Zhu, X.-X., & Jin, S. (2017) · Journal of Statistical Computation and Simulation · doi:10.1080/00949655.2017.1349131 · Cited by 5
Comparing multilevel and multiscale convolution models for small area aggregated health data
Aregay, M., Lawson, A. B., Faes, C., et al. (2017) · Spatial and Spatio-temporal Epidemiology · doi:10.1016/j.sste.2017.06.001 · Cited by 5
Wong, C. K., Makalic, E., & Schmidt, D. F. (2018) · Journal of Mathematical Psychology · doi:10.1016/j.jmp.2018.01.001 · Cited by 6
Kozono, J., Nishihara, G. N., Endo, H., et al. (2018) · Phycologia · doi:10.2216/18-26.1 · Cited by 5
A Family-Based Rare Haplotype Association Method for Quantitative Traits
Datta, A. S., Lin, S., & Biswas, S. (2018) · Human Heredity · doi:10.1159/000493543 · Cited by 5
Shertzer, K. W., Bacheler, N. M., Pine, W. E., Runde, B. J., Buckel, J. A., Rudershausen, P. J., & MacMahan, J. H. (2019) · Canadian Journal of Fisheries and Aquatic Sciences · doi:10.1139/cjfas-2018-0356 · Cited by 6
“We used normal and half-Cauchy hyperprior distributions for the means and standard 270 deviations, 271 𝜇𝜇𝑖𝑖~𝑁𝑁(0, 1 3⁄ ) (13) 272 𝛾𝛾𝑖𝑖~Cauchy(0, 1 3⁄ ); 𝜏𝜏𝑖𝑖 = |𝛾𝛾𝑖𝑖| (14) 273 Using the Cauchy distribution for variance parameters (Gelman 2006; Gelman et al. 2008), here 274 with the location parameter equal to 0 and th…”
Manderson, A. A., Rayson, M. D., Cripps, E. J., et al. (2019) · Journal of Atmospheric and Oceanic Technology · doi:10.1175/jtech-d-18-0200.1 · Cited by 5
“Furthermore, if the number of repetitions within the data is small or the standard deviations approach close to zero, uninformative priors, despite their intention, can exert a significant influence on the posterior (Gelman 2006).”
BackgroundR package rjmcmc: reversible jump MCMC using post‐processing
Gelling, N., Schofield, M., & Barker, R. (2019) · Australian & New Zealand Journal of Statistics · doi:10.1111/anzs.12263 · Cited by 5
“We have used weakly informative prior distributions (Gelman 2006) chosen so that the overall variability of the prior predictive distribution is similar between the two models.”
MethodModeling local dependence in latent vector autoregressive models
Tran, T. D., Lesaffre, E. M. E. H., Verbeke, G. N., & Duyck, J. A. J. (2019) · Biostatistics · doi:10.1093/biostatistics/kxz021 · Cited by 5
Tang, Y. (2018) · Statistics in Medicine · doi:10.1002/sim.8062 · Cited by 7
Non-Bayesian Social Learning with Gaussian Uncertain Models
Hare, J. Z., Uribe, C. A., Kaplan, L., & Jadbabaie, A. (2020) · Proceedings of the ... American Control Conference/Proceedings of the American Control Conference · doi:10.23919/acc45564.2020.9147937 · Cited by 5
“Thus, we normalize the uncertain likelihood by another posterior predictive distribution of the measurement sequence, except here we use a noninformative Gaussian-gamma prior [29] having zero prior evidence, i.”
MethodKaluskar, S., Blukacz‐Richards, E. A., Johnson, C. A., He, Y., Langlois, A., Kim, D., & Arhonditsis, G. B. (2019) · Ecosphere · doi:10.1002/ecs2.2976 · Cited by 7
Feeding Ecology of Age-0 Gar at Texoma Reservoir Inferred from Analysis of Stable Isotopes
Snow, R. A., Stewart, D. R., Porta, M. J., et al. (2020) · North American Journal of Fisheries Management · doi:10.1002/nafm.10436 · Cited by 5
Joint analysis of semicontinuous data with latent variables
Wang, X., Feng, X., & Song, X. (2020) · Computational Statistics & Data Analysis · doi:10.1016/j.csda.2020.107005 · Cited by 5
Cho, W., Lambert, D. M., Fornah, A., et al. (2020) · Journal of Agricultural Science · doi:10.5539/jas.v12n10p1 · Cited by 5
“Uniform distribution priors may be preferable when information on individuals, treatments, or group characteristics is limited (Gelman, 2006).”
BackgroundYorozuya, K., Yamane, S., Nobuhisa, M., Owaki, H., Suzuki, T., Okahara, H., Iwamori, W., & Hanaoka, H. (2020) · Medicine · doi:10.1097/md.0000000000022154 · Cited by 5
“The prior distribution was used a half-Cauchy distribution (which is weakly informative prior, and has been recommended).([38,39]) Five Markov Chains simulated 25,000 draws and discarded 5000 warm up draws.”
MethodVector Operations for Accelerating Expensive Bayesian Computations – A Tutorial Guide
Warne, D. J., Sisson, S. A., & Drovandi, C. (2021) · Bayesian Analysis · doi:10.1214/21-ba1265 · Cited by 8
“Here, we consider the selection of weakly informative priors [19] within the framework proposed by Evans and Jang [16] using the implementation by Nott et al.”
BackgroundMethodCavieres, J., Monnahan, C. C., & Vehtari, A. (2021) · Fisheries Research · doi:10.1016/j.fishres.2021.105960 · Cited by 5
“The prior distribution for the other hyperparameters are: θ 2 ∼ N (2 , 2) and for the variance we assume log( σ ) = θ 3 ∼ Cauchy (0 , 2) ([39]).”
BackgroundBerends, K. D., Gensen, M. R. A., Warmink, J. J., et al. (2021) · CivilEng · doi:10.3390/civileng2030032 · Cited by 5
Bayesian Analysis of Nonnegative Data Using Dependency-Extended Two-Part Models
Rodrigues‐Motta, M., & Forkman, J. (2021) · Journal of Agricultural Biological and Environmental Statistics · doi:10.1007/s13253-021-00467-x · Cited by 5
“Among the alternative priors, continuous uniform distributions, U(0, 50) and U(0, 100), were investigated, as recommended by Gelman (2006).”
BackgroundParametric Distributions to Model Numerical Emotion Labels
Bose, D., Sethu, V., & Eliathamby Ambikairajah, A. (2021) · Interspeech (USB) · doi:10.21437/interspeech.2021-1000 · Cited by 7
“…scale parameters as in hierarchical models [20] for every Gaussian-like distribution because Inverse-Gamma prior has a desirable closed form expression due its conditional conjugacy [21] versus Half-Cauchy densities, and the uniform distribution which were implicitly assumed for the MLE estimates.”
BackgroundStarling, J. E., Murray, J. S., Lohr, P. A., Aiken, A. R. A., Carvalho, C. M., & Scott, J. G. (2021) · The Annals of Applied Statistics · doi:10.1214/20-aoas1438 · Cited by 18
“95, and default smoothing parameter κμ = 1, with a half-Cauchy prior on the scale of leaf parameters (Gelman, 2006).”
BackgroundMethodInfluentialA meta‐analytic framework to adjust for bias in external control studies
Incerti, D., Bretscher, M. T., Lin, R., & Harbron, C. G. (2022) · Pharmaceutical Statistics · doi:10.1002/pst.2266 · Cited by 8
“In our aNSCLC example, we used a half-Cauchy prior as recommended by Gelman.(14) To assess the sensitivity of this choice, we also considered a gamma prior (scale = 0.”
MethodInfluentialShinagawa, K., Ito, Y., Tsuji, K., Tanaka, Y., Odaka, M., Shibata, M., Terasawa, Y., & Umeda, S. (2021) · Neuroimage Reports · doi:10.1016/j.ynirp.2021.100060 · Cited by 6
The Inverse G‐Wishart distribution and variational message passing
Maestrini, L., & WAND, M. P. (2021) · Australian & New Zealand Journal of Statistics · doi:10.1111/anzs.12339 · Cited by 7
“This is in keeping with the advice given in Gelman (2006). Moreover, correlation parameters jj′ = ( ) jj′ =( j j′), for each j = j′ pair, have a Uniform distribution over the interval (−1, 1) when HW = 2.”
MethodResultInfluentialThe killing fields. A Bayesian analysis of crop eradication and organized crime violence in Mexico
Muñiz-Sánchez, V., Fuerte-Celis, M. D. P., & Méndez-Ramírez, K. (2021) · Spatial Statistics · doi:10.1016/j.spasta.2021.100553 · Cited by 6
Murayama, K., & Kawano, S. (2021) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2021.3131357 · Cited by 7
“This hyperprior is related to a weakly informative prior [19]–[21].”
BackgroundValidation of numerical predictions for liquefaction phenomenon – Lateral spreading in clean sands
Vargas, R. R., Tang, Z., Ueda, K., et al. (2021) · SOILS AND FOUNDATIONS · doi:10.1016/j.sandf.2021.101101 · Cited by 5
Bayesian inference of spatially correlated random parameters for on-farm experiment
Cao, Z., Stefanova, K. T., Gibberd, M., & Rakshit, S. (2022) · Field Crops Research · doi:10.1016/j.fcr.2022.108477 · Cited by 11
Urdangarin, A., Goicoa, T., & Ugarte, M. D. (2022) · Statistical Methods in Medical Research · doi:10.1177/09622802221079351 · Cited by 6
“Coverage rates improved when using vague uniform priors on the standard deviations as it was recommended by Gelman 31 .”
MethodBedaso, N. G., & Debusho, L. K. (2022) · BMC Infectious Diseases · doi:10.1186/s12879-021-06965-0 · Cited by 5
“In this paper, since empirical information on parameters, β and σb or relevant to the study data is not available, non-informative priors were used [27].”
MethodGreven, S., & Scheipl, F. (2016) · Journal of the American Statistical Association · doi:10.1080/01621459.2016.1250580 · Cited by 5
Nagel, J. B., & Sudret, B. (2014) · AIAA Non-Deterministic Approaches Conference · doi:10.2514/6.2014-1502 · Cited by 9
Li, J., & Han, X. (2022) · International Journal of Environmental Research and Public Health · doi:10.3390/ijerph20010597 · Cited by 5
“The spatial adjacency matrix adopts the first-order ‘Queen’ adjoining form; ε it and ε (p) it represent the Gaussian random terms, and their priors were assigned with Gaussian distribution with zero mean [33].”
BackgroundMielke, M. S., Santos, G. O., Lopes, E. C. S., et al. (2023) · Acta Physiologiae Plantarum · doi:10.1007/s11738-023-03514-x · Cited by 5
A broader flight season for Norway's Odonata across a century and a half
Patten, M. A., & Benson, B. (2023) · Oikos · doi:10.1111/oik.09882 · Cited by 6
Ghaffari, S., Bouchonville, K. J., Saleh, E., Schmidt, R. E., Offer, S. M., & Sinha, S. (2023) · Genome biology · doi:10.1186/s13059-023-03007-7 · Cited by 8
“The Half Cauchy prior was used for the standard deviation as suggested by Gelman [49].”
MethodImproving children's food choices: Experimental evidence from the field
Charness, G., Cobo‐Reyes, R., Eyster, E., Katz, G., Sánchez, Á., & Sutter, M. (2023) · European Economic Review · doi:10.1016/j.euroecorev.2023.104562 · Cited by 5
“yield accurate estimates for such hierarchical models even with as few as 3 clusters (Gelman, 2006).”
BackgroundMethodInfluentialTrends in chronic childhood undernutrition in Bangladesh for small domains
Das, S., Baffour, B., & Richardson, A. (2023) · Population Studies · doi:10.1080/00324728.2023.2239772 · Cited by 7
Bayesian Ideas in Survey Sampling: The Legacy of Basu
Zio, M. D., Liseo, B., & Ranalli, M. G. (2023) · Sankhya A · doi:10.1007/s13171-023-00327-5 · Cited by 5
BEATS : Bayesian hybrid design with flexible sample size adaptation for time‐to‐event endpoints
Bi, D., Liu, M., Lin, J., et al. (2023) · Statistics in Medicine · doi:10.1002/sim.9936 · Cited by 5
Bayesian kernel projections for classifcation of high dimensional data
Domijan, K., & Wilson, S. P. (2011) · University of Limerick Institutional Repository (University of Limerick) · doi:10.34961/852 · Cited by 5
A Bayesian Method for Estimating Parameters in Stochastic Differential
Karimi, H., & McAuley, K. B. (2015) · IFAC-PapersOnLine · doi:10.1016/j.ifacol.2015.08.172 · Cited by 5
Reum, J. C. P., Hovel, R. A., & Greene, C. M. (2014) · Marine Biology · doi:10.1007/s00227-014-2574-8 · Cited by 5
“The scaled inverse-Wishart is noninformative on the correlations but allows a wider range of uncertainty on the variances, improves mixing properties of the chain, and has been suggested as a better alternative (Gelman 2006; Gelman et al. 2014).”
BackgroundFitting Residual Error Structures for Growth Models in SAS PROC MCMC
McNeish, D. (2016) · Educational and Psychological Measurement · doi:10.1177/0013164416652441 · Cited by 5
“…etc.) may be beneficial under such conditions due to a broader selection of prior distributions including the ability to specify the half-Cauchy prior recommended by Gelman (2006) and shown to perform well in a study by McNeish and Stapleton (2016) with sample sizes as low as the mid-single digits.”
MethodHierarchical Bayesian small area estimation for circular data
Hernandez‐Stumpfhauser, D., Breidt, F. J., & Opsomer, J. D. (2016) · Canadian Journal of Statistics · doi:10.1002/cjs.11303 · Cited by 5
“The inverse gamma prior for σ2 sw can potentially lead to issues by being too informative (Gelman, 2006), but that did not appear to be the case for our particular model and data set, as different vague prior specifications resulted in almost identical posteriors.”
MethodResultInfluentialUncertainty Flow Facilitates Zero-Shot Multi-Label Learning in Affective Facial Analysis
Bai, W., Quan, C., & Luo, Z. (2018) · Applied Sciences · doi:10.3390/app8020300 · Cited by 7
“However, these non-informative priors are neither applicable in multiple parameter modelling nor asymptotically inconsistent in deriving the posterior [28].”
Estimating under-observation and the full size of the 2016 Zika epidemic in Rio de Janeiro
Bastos, M. M., & Coelho, F. C. (2018) · PLoS ONE · doi:10.1371/journal.pone.0205001 · Cited by 5
Multi-rubric models for ordinal spatial data with application to online ratings data
Linero, A. R., Bradley, J. R., & Desai, A. (2018) · The Annals of Applied Statistics · doi:10.1214/18-aoas1143 · Cited by 5
Howard, E., Meehan, M., & Parnell, A. C. (2018) · Assessment & Evaluation in Higher Education · doi:10.1080/02602938.2018.1476670 · Cited by 13
Crash Modification Functions for Passing Relief Lanes on Two-Lane Rural Roads
Persaud, B. N., Anarkooli, A. J., Almasi, S., et al. (2020) · Transportation Research Record Journal of the Transportation Research Board · doi:10.1177/0361198120930719 · Cited by 5
“The way that prior distributions were chosen is mainly based on a suggestion in Gelman that “after the model has been fit, one should look at the posterior distribution and see if it makes sense” and then revise the prior distribution if necessary (10).”
BackgroundMethodPosterior impropriety of some sparse Bayesian learning models
Dixit, A., & Roy, V. (2021) · Statistics & Probability Letters · doi:10.1016/j.spl.2021.109039 · Cited by 6
“…1 implies that posterior propriety is assured even if we choose the weakly informative half Cauchy prior on { λ − 1 / 2 i } ni =0 as suggested by Gelman (2006) or the type 2 Gumbel distribution (derived as the penalized complexity prior in Simpson et al. (2017)) on { λ i } ni =0 , along with an…”
MethodPerret, C., & Solier, C. (2022) · Journal of Cognitive Psychology · doi:10.1080/20445911.2022.2029459 · Cited by 6
“A halfCauchy prior distribution was used for each parameter (Gelman, 2006).”
MethodCombining Functional Data Registration and Factor Analysis
Earls, C., & Hooker, G. (2016) · Journal of Computational and Graphical Statistics · doi:10.1080/10618600.2016.1164709 · Cited by 8
“Gelman (2006) also proposed some better alternatives to using Gamma or inverse Gamma priors for variance components that could be used in place of the priors above.”
BackgroundMethodAn Integrated Approach to Empirical Bayesian Whole Genome Prediction Modeling
Chen, C., & Tempelman, R. J. (2015) · Journal of Agricultural Biological and Environmental Statistics · doi:10.1007/s13253-015-0224-3 · Cited by 5
“For all subsequent analyses in this paper, we use the non-informative prior for components of σ as advocated by Gelman (2006),”
MethodIterative posterior inference for Bayesian Kriging
Zhang, Z. (2011) · Stochastic Environmental Research and Risk Assessment · doi:10.1007/s00477-011-0544-y · Cited by 5
Watters, G. M., & Bessey, C. (2008) · Transactions of the American Fisheries Society · doi:10.1577/t06-101.1 · Cited by 5
“We used uniform priors on the hierarchical SDs based on recommendations by Gelman (2006 ), who showed that posterior inference about SDs can be inappropriately constrained by the choice of ϵ when (1) inverse-gamma (ϵ, ϵ) priors are placed on variance parameters (which has frequently been done in hierarchical modeling…”
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.