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 201–300 of 4,089
Couvreux, F., Hourdin, F., Williamson, D., Roehrig, R., Volodina, V., Villefranque, N., Rio, C., Audouin, O., Salter, J., Bazile, E., et al. (2021) · Journal of Advances in Modeling Earth Systems · doi:10.1029/2020ms002217 · Cited by 121
Mixed beta regression: A Bayesian perspective
Figueroa-Zúñiga, J., Arellano-Valle, R. B., & Ferrari, S. L. P. (2012) · Computational Statistics & Data Analysis · doi:10.1016/j.csda.2012.12.002 · Cited by 145
“Gelman (2006) suggests that the prior distribution φ = U 2 with U ∼ U (0 , a ) with large a ( a = 50 for example) is less informative than an inverse gamma prior.”
BackgroundMethodDetermining informative priors for cognitive models
Lee, M. D., & Vanpaemel, W. (2018) · Psychonomic Bulletin & Review · doi:10.3758/s13423-017-1238-3 · Cited by 120
Human‐directed social behaviour in dogs shows significant heritability
Persson, M. E., Roth, L. S. V., Johnsson, M., Wright, D., & Jensen, P. (2015) · Genes, Brain and Behavior · doi:10.1111/gbb.12194 · Cited by 117
“We used a parameter expanded prior for the genetic variance component with V=1, 𝜈 =1, 𝛼𝜇 =0, 𝛼V =1000, which is a half-Cauchy prior for the standard deviation with scale 100 (Gelman 2006), and inverse-Wishart priors (V=1, 𝜈 = 0.002) for the residual variance component.”
MethodInfluentialEvidence for evolution in response to natural selection in a contemporary human population
Milot, E., Mayer, F. M., Nussey, D. H., Boisvert, M., Pelletier, F., & Réale, D. (2011) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1104210108 · Cited by 119
Network meta-analysis of randomized clinical trials: Reporting the proper summaries
Zhang, J., Carlin, B. P., Neaton, J. D., Soon, G. G., Nie, L., Kane, R., Virnig, B. A., & Chu, H. (2014) · Clinical Trials · doi:10.1177/1740774513498322 · Cited by 120
Combining Expert Opinions in Prior Elicitation
Albert, I., Donnet, S., Guihenneuc‐Jouyaux, C., Low‐Choy, S., Mengersen, K. L., & Rousseau, J. (2012) · Bayesian Analysis · doi:10.1214/12-ba717 · Cited by 158
Fogarty, E., Schmitz, S., Tubridy, N. J., Walsh, C., & Barry, M. G. (2016) · Multiple Sclerosis and Related Disorders · doi:10.1016/j.msard.2016.06.001 · Cited by 125
Wang, C., Li, H., Chen, W.-C., Lu, N. T., Tiwari, R. C., Xu, Y., & Yue, L. Q. (2019) · Journal of Biopharmaceutical Statistics · doi:10.1080/10543406.2019.1657133 · Cited by 136
“When there is only one prior study, however, Bayesian hierarchical models are known to have difficulty in estimating study-to-study variation (Gelman 2006), which motivates us to consider the alternative power prior approach.”
BackgroundThe temperature responses of soil respiration in deserts: a seven desert synthesis
Cable, J. M., Ogle, K., Lucas, R. W., Huxman, T. E., Loik, M. E., Smith, S. D., Tissue, D. T., Ewers, B. E., Pendall, E. G., Welker, J. M., et al. (2010) · Biogeochemistry · doi:10.1007/s10533-010-9448-z · Cited by 125
“…specified a hierarchical model for the desertspecific standard deviations for the date random effects (the re = 1/sqrt(se) terms) such that the re‘s arise from a folded Student-t distribution with two degrees of freedom and scale parameter A (Gelman 2006); we assigned a diffuse uniform prior to A.”
MethodLinear mixed models with skew-elliptical distributions: A Bayesian approach
Jara, A., Quintana, F. A., & Martín, E. S. (2008) · Computational Statistics & Data Analysis · doi:10.1016/j.csda.2008.04.027 · Cited by 115
Kefford, B. J., Hickey, G. L., Gasith, A., Ben-David, E., Dunlop, J. E., Palmer, C. G., Allan, K., Choy, S., & Piscart, C. (2012) · PLoS ONE · doi:10.1371/journal.pone.0035224 · Cited by 124
“The hyperpopulation standard deviation was assigned a Uniform (0, 10) distribution based on the reasoning of Gelman [36].”
MethodA temporal shift of the evolutionary principle shaping intratumor heterogeneity in colorectal cancer
Saito, T., Niida, A., Uchi, R., Hirata, H., Komatsu, H., Sakimura, S., Hayashi, S., Nambara, S., Kuroda, Y., Ito, S., et al. (2018) · Nature Communications · doi:10.1038/s41467-018-05226-0 · Cited by 119
Who Wants What? Redistribution Preferences in Comparative Perspective
Rueda, D., & Stegmueller, D. (2019) · Cambridge University Press eBooks · doi:10.1017/9781108681339 · Cited by 201
Kay, M., Nelson, G. L., & Hekler, E. B. (2016) · Proceedings of the 2016 CHI Conference on Human Factors in Computing Systems · doi:10.1145/2858036.2858465 · Cited by 113
“Other accessible articles have also been written about practical concerns in Bayesian analysis, including discussions of how to choose priors [9,11].”
BackgroundHoltmann, J., Koch, T., Lochner, K., & Eid, M. (2016) · Multivariate Behavioral Research · doi:10.1080/00273171.2016.1208074 · Cited by 141
Structural reliability analysis for p-boxes using multi-level meta-models
Schöbi, R., & Sudret, B. (2017) · Probabilistic Engineering Mechanics · doi:10.1016/j.probengmech.2017.04.001 · Cited by 110
“They include Dempster-Shafer’s theory of evidence (Dempster, 1967; Shafer, 1976), Bayesian hierarchical models (Gelman, 2006; Gelman et al., 2009), possibility theory (De Cooman et al.”
Hierarchical calibration of archetypes for urban building energy modeling
Kristensen, M. H., Hedegaard, R. E., & Petersen, S. (2018) · Energy and Buildings · doi:10.1016/j.enbuild.2018.07.030 · Cited by 126
Hone, J., Duncan, R. P., & Forsyth, D. M. (2010) · Journal of Applied Ecology · doi:10.1111/j.1365-2664.2010.01812.x · Cited by 120
Within-subject mediation analysis for experimental data in cognitive psychology and neuroscience
Vuorre, M., & Bolger, N. (2018) · Behavior Research Methods · doi:10.3758/s13428-017-0980-9 · Cited by 107
“For general information on the choice of prior distributions, we refer the readers to excellent textbooks on Bayesian statistics (Gelman, 2006; Gelman et al., 2013; Kruschke, 2014; McElreath, 2016).”
BackgroundMethodInfluentialRandom‐effects meta‐analysis of few studies involving rare events
Günhan, B. K., Röver, C., & Friede, T. (2019) · Research Synthesis Methods · doi:10.1002/jrsm.1370 · Cited by 135
Albrecht, J., Peters, M. K., Becker, J. N., Behler, C., Claßen, A., Enßlin, A., Ferger, S. W., Gebert, F., Gerschlauer, F., Helbig‐Bonitz, M., et al. (2021) · Nature Ecology & Evolution · doi:10.1038/s41559-021-01550-9 · Cited by 142
Does accounting for imperfect detection improve species distribution models?
Rota, C. T., Fletcher, R. J., Evans, J., & Hutto, R. L. (2010) · Ecography · doi:10.1111/j.1600-0587.2010.06433.x · Cited by 132
Bayesian models for comparative analysis integrating phylogenetic uncertainty
de Villemereuil, P., Wells, J. A., Edwards, R. D., & Blomberg, S. P. (2012) · BMC Evolutionary Biology · doi:10.1186/1471-2148-12-102 · Cited by 124
Inequalities in mortality in small areas of eleven Spanish cities (the multicenter MEDEA project)
Borrell, C., Marí-Dell’Olmo, M., Serral, G., Martínez‐Beneito, M. A., & Gotsens, M. (2010) · Health & Place · doi:10.1016/j.healthplace.2010.03.002 · Cited by 142
Meta-analysis of incidence rate data in the presence of zero events
Spittal, M. J., Pirkis, J., & Gurrin, L. C. (2015) · BMC Medical Research Methodology · doi:10.1186/s12874-015-0031-0 · Cited by 110
Whitlock, R. (2014) · Journal of Ecology · doi:10.1111/1365-2745.12240 · Cited by 125
“I used non-informative uniform improper distributions on the standard deviation of random effects as the priors for variance components (Gelman 2006).”
MethodNeural computations of threat in the aftermath of combat trauma
Homan, P., Levy, I., Feltham, E., Gordon, C., Hu, J., Li, J., Pietrzak, R. H., Southwick, S., Krystal, J. H., Harpaz-Rotem, I., et al. (2019) · Nature Neuroscience · doi:10.1038/s41593-018-0315-x · Cited by 101
Bayesian statistics and modelling
van de Schoot, R., Depaoli, S. A., Gelman, A., King, R., Kramer, B. M. R., Märtens, K., Tadesse, M. G., Vannucci, M., Willemsen, J. E., & Yau, C. (2021) · Nature Reviews Methods Primers · doi:10.1038/s43586-020-00003-0 · Cited by 152
Self-citationTravel time estimation for ambulances using Bayesian data augmentation
Westgate, B. S., Woodard, D. B., Matteson, D. S., et al. (2013) · The Annals of Applied Statistics · doi:10.1214/13-aoas626 · Cited by 109
Schuurman, N. K., Grasman, R. P. P. P., & Hamaker, E. L. (2016) · Multivariate Behavioral Research · doi:10.1080/00273171.2015.1065398 · Cited by 123
“These prior distributions, which are usually uninformative with certain hyperparameters, are quite informative when variances are small, resulting in a strong effect of the prior distribution on the parameter estimates (Gelman, 2006; see also Song & Ferrer, 2012, for an example).”
Dynamic Flexibility in Striatal-Cortical Circuits Supports Reinforcement Learning
Gerraty, R. T., Davidow, J. Y., Foerde, K., Galvan, A., Bassett, D. S., & Shohamy, D. (2018) · The Journal of Neuroscience · doi:10.1523/jneurosci.2084-17.2018 · Cited by 105
Bayesian models for subgroup analysis in clinical trials
Jones, H. E., Ohlssen, D. I., Neuenschwander, B. E., Racine, A., & Branson, M. (2011) · Clinical Trials · doi:10.1177/1740774510396933 · Cited by 116
Richardson, D. B., Cardis, E., Daniels, R. D., Gillies, M., Haylock, R., Leuraud, K., Laurier, D., Moissonnier, M., Schubauer-Berigan, M. K., Thierry-Chef, I., et al. (2018) · Epidemiology · doi:10.1097/ede.0000000000000761 · Cited by 105
Single and Multiscale Models of Process Spatial Heterogeneity
Wolf, L. J., Oshan, T. M., & Fotheringham, A. S. (2017) · Geographical Analysis · doi:10.1111/gean.12147 · Cited by 113
Space—time zero‐inflated count models of Harbor seals
Hoef, J. M. V., & Jansen, J. K. (2007) · Environmetrics · doi:10.1002/env.873 · Cited by 106
“Uniform priors on the square root of variance parameters in hierarchical models has recently been suggested by Gelman (2006), and a simulation study (Lambert et al.”
MethodBayesian Population Forecasting: Extending the Lee-Carter Method
Wiśniowski, A., Smith, P. W. F., Bijak, J., Raymer, J., & Forster, J. J. (2015) · Demography · doi:10.1007/s13524-015-0389-y · Cited by 101
Meta-regression analysis of the effects of dietary cholesterol intake on LDL and HDL cholesterol
Vincent, M., Allen, B. C., Palacios, O. M., Haber, L. T., & Maki, K. C. (2018) · American Journal of Clinical Nutrition · doi:10.1093/ajcn/nqy273 · Cited by 121
Zagona, A. L., Kurth, J. A., Turner, E. L., Pace, J. R., Shogren, K. A., Lansey, K. R., Jameson, J. M., Burnette, K. K., Mansouri, M. C., Hicks, T. A., et al. (2022) · Research and Practice for Persons with Severe Disabilities · doi:10.1177/15407969221126496 · Cited by 114
“Minimally informative priors were used for variance components; specifically, we used half-normal priors (M= 0, sd = 10; for example, Gelman, 2006) which provide only positive values, thus avoiding impossible negative variances.”
MethodShaw, B. A., Liang, J., KRAUSE, N. M., Gallant, M. P., & McGeever, K. (2010) · The Journals of Gerontology Series B · doi:10.1093/geronb/gbq073 · Cited by 125
“First, although there is a consensus for weighting data in generating descriptive statistics for a given target population, there is no such agreement in multivariate analyses (Gelman, 2006).”
BackgroundA unified framework for multilevel uncertainty quantification in Bayesian inverse problems
Nagel, J. B., & Sudret, B. (2015) · Probabilistic Engineering Mechanics · doi:10.1016/j.probengmech.2015.09.007 · Cited by 128
“In the statistical literature hyperprior elicitation is exhaustively discussed especially for variance hyperparameters [36–38].”
Clark, N. J., Seddon, J. M., Šlapeta, J. R., & Wells, K. (2018) · Parasites & Vectors · doi:10.1186/s13071-017-2564-z · Cited by 116
“We used parameter expansion (redundant multiplicative reparameterisation of the linear model) for the threat status variance component to reduce dependence among parameters and improve chain mixing [75].”
Mitropoulos, P., Talias, Μ. A., & Mitropoulos, I. (2014) · European Journal of Operational Research · doi:10.1016/j.ejor.2014.11.012 · Cited by 103
How priors of initial hyperparameters affect Gaussian process regression models
Chen, Z., & Wang, B. (2018) · Neurocomputing · doi:10.1016/j.neucom.2017.10.028 · Cited by 101
“The samples are generated using two GP models with two different hyperparameters: θact = [5, 2] and θact = [15, 7], respectively.”
BackgroundMethodRevisiting crash spatial heterogeneity: A Bayesian spatially varying coefficients approach
Xu, P., Huang, H., Dong, N., et al. (2016) · Accident Analysis & Prevention · doi:10.1016/j.aap.2016.10.015 · Cited by 103
“Given that the commonly used inverse‐Gamma ( , ) 39 priors (where is a small number, e.g., 0.01 or 0.001) are sensitive to the value of if the 40 true variance is close to zero, a uniform (0,10) was finally specified for v and k , 41 respectively (Gelman, 2006).”
InfluentialFast Estimation of Ideal Points with Massive Data
Imai, K., Lo, J., & Olmsted, J. (2016) · American Political Science Review · doi:10.1017/s000305541600037x · Cited by 125
“This is a well-known feature of Bayesian hierarchical models (Gelman 2006) and the ideal point models appear to be no exception in this regard.”
A multispecies perspective on ecological impacts of climatic forcing
Mutshinda, C. M., O’Hara, R. B., & Woiwod, I. P. (2010) · Journal of Animal Ecology · doi:10.1111/j.1365-2656.2010.01743.x · Cited by 98
“Finally, we placed Unif(0, 10) priors on the standard deviations di, si and rr. Gelman (2006) gives a justification for this approach to prior specification for variance parameters.”
MethodChen, J. T., Rehkopf, D. H., Waterman, P. D., Subramanian, S., Coull, B., Cohen, B., Ostrem, M., & Krieger, N. (2006) · Journal of urban health · doi:10.1007/s11524-006-9089-7 · Cited by 110
Statistical modelling of railway track geometry degradation using Hierarchical Bayesian models
Andrade, A. R., & Teixeira, P. F. (2015) · Reliability Engineering & System Safety · doi:10.1016/j.ress.2015.05.009 · Cited by 119
“As a side note, it is important to mention that the use of inverse gamma prior distributions for the variance terms is not free of criticism and has been discussed by Gelman [35], who recommended starting with a non-informative uniform prior density, or when more prior information is desired, working within the half-t…”
BackgroundInfluentialApplegate, D., Mantz, A. B., Allen, S. W., von der Linden, A., Morris, R. G., Hilbert, S., Kelly, P. L., Burke, D. L., Ebeling, H., Rapetti, D., et al. (2016) · Monthly Notices of the Royal Astronomical Society · doi:10.1093/mnras/stw005 · Cited by 107
“For this task, we adopt the hierarchical mixture of Gaussians prior employed in Kelly (2007), with uniform priors on Gaussian scale parameters (i.e. σ) following Gelman (2006).”
MethodEarnest, A., Morgan, G. G., Mengersen, K. L., RYAN, L. M., Summerhayes, R. J., & Beard, J. (2007) · International Journal of Health Geographics · doi:10.1186/1476-072x-6-54 · Cited by 129
A Dirichlet Process Mixture of Generalized Dirichlet Distributions for Proportional Data Modeling
Bouguila, N., & Ziou, D. (2009) · IEEE Transactions on Neural Networks · doi:10.1109/tnn.2009.2034851 · Cited by 98
Regression Analysis of Country Effects Using Multilevel Data: A Cautionary TalePreprint
Bryan, M. L., & Jenkins, S. P. (2013) · SSRN Electronic Journal · doi:10.2139/ssrn.2322088 · Cited by 150
Publication Bias in Recent Meta-Analyses
Kicinski, M. (2013) · PLoS ONE · doi:10.1371/journal.pone.0081823 · Cited by 97
Stamp, M. A., & Hadfield, J. D. (2020) · Ecology Letters · doi:10.1111/ele.13565 · Cited by 98
“The ratio-t model was fitted in STAN using diffuse normal priors on the fixed effects, diffuse half-Cauchy priors on the standard deviations (Gelman 2006) and inverse-gamma shape parameter, and an LKJ prior (Lewandowski et al.”
MethodA Bayesian network meta-analysis for binary outcome: how to do it
Greco, T., Landoni, G., Biondi‐Zoccai, G., D’Ascenzo, F., & Zangrillo, A. (2013) · Statistical Methods in Medical Research · doi:10.1177/0962280213500185 · Cited by 106
Meta‐analysis of diagnostic test studies using individual patient data and aggregate data
Riley, R. D., Dodd, S. R., Craig, J. V., Thompson, J. R., & Williamson, P. R. (2008) · Statistics in Medicine · doi:10.1002/sim.3441 · Cited by 98
Spatial location dominates over host plant genotype in structuring an herbivore community
Tack, A. J. M., Ovaskainen, O., Pulkkinen, P., et al. (2010) · Ecology · doi:10.1890/09-1027.1 · Cited by 97
“As it is well known that the estimates for the variance components can be sensitive to the choice of the prior (Gelman 2006), we also computed the posterior densities assuming the values s¼ 0.0001 and s ¼ 1.”
BackgroundInfluentialA general model of detectability using species traits
Garrard, G. E., McCarthy, M. A., Williams, N. S. G., et al. (2012) · Methods in Ecology and Evolution · doi:10.1111/j.2041-210x.2012.00257.x · Cited by 96
“The prior distribution for the standard deviation of the random effects was specified as uniform between 0 and 100 (Gelman 2006).”
BackgroundHierarchical Regression for Analyses of Multiple Outcomes
Richardson, D. B., Hamra, G. B., MacLehose, R. F., Cole, S. R., & Chu, H. (2015) · American Journal of Epidemiology · doi:10.1093/aje/kwv047 · Cited by 102
Iijima, H., Nagaike, T., & Honda, T. (2013) · Journal of Wildlife Management · doi:10.1002/jwmg.556 · Cited by 103
Spatially Adaptive Bayesian Penalized Splines With Heteroscedastic Errors
Crainiceanu, C. M., Ruppert, D., Carroll, R. J., Joshi, A., & Goodner, B. (2007) · Journal of Computational and Graphical Statistics · doi:10.1198/106186007x208768 · Cited by 92
“Alternative to gamma priors are discussed by, for example, Natarajan and Kass (2000) and Gelman (2006).”
BackgroundDevineni, N., Lall, U., Pederson, N., & Cook, E. R. (2013) · Journal of Climate · doi:10.1175/jcli-d-11-00675.1 · Cited by 100
Bazin, É., Dawson, K. J., & Beaumont, M. (2010) · Genetics · doi:10.1534/genetics.109.112391 · Cited by 97
“Note that we do not use an inverse gamma for the su, following the recommendation of Gelman (2006).”
MethodStatistical Methods for Linguistic Research: Foundational Ideas – Part I
Vasishth, S., & Nicenboim, B. (2016) · Language and Linguistics Compass · doi:10.1111/lnc3.12201 · Cited by 119
“For a more advanced treatment of the topic, see Gelman et al. (2014a). Linear mixed models are covered from both the frequentist and Bayesian perspective by Gelman and Hill (2007). For an accessible introduction of Bayesian methods for cognitive modeling see Lee and Wagenmakers (2014).”
BackgroundMethodInfluentialBaldwin, S. A., Larson, M. J., & Clayson, P. E. (2015) · Psychophysiology · doi:10.1111/psyp.12401 · Cited by 91
Power in Bayesian Mediation Analysis for Small Sample Research
Miočević, M., MacKinnon, D. P., & Levy, R. (2017) · Structural Equation Modeling A Multidisciplinary Journal · doi:10.1080/10705511.2017.1312407 · Cited by 108
“The IG is said to be a conditionally conjugate prior for the variance parameter, meaning that if an IG prior is assigned to the variance parameter, then the posterior for the variance, conditional on other model parameters, is also an IG (Gelman, 2006).”
BackgroundExperience with Reviewing Bayesian Medical Device Trials
Pennello, G. A., & Thompson, L. (2007) · Journal of Biopharmaceutical Statistics · doi:10.1080/10543400701668274 · Cited by 102
“For a discussion of alternative, uniform prior distributions for variance component standard deviations, see Gelman (2006).”
BackgroundRöver, C., Bender, R., Dias, S., Schmid, C. H., Schmidli, H., Sturtz, S., Weber, S., & Friede, T. (2021) · White Rose Research Online (University of Leeds, The University of Sheffield, University of York) · Cited by 89
Despotic, high‐impact species and the subcontinental scale control of avian assemblage structure
Nally, R. M., Bowen, M. E., Howes, A., McAlpine, C., & Maron, M. (2011) · Ecology · doi:10.1890/10-2340.1 · Cited by 91
A systematic review of phenotypic responses to between-population outbreeding
Whitlock, R., Stewart, G. B., Goodman, S. J., Piertney, S. B., Butlin, R. K., Pullin, A. S., & Burke, T. A. (2013) · Environmental Evidence · doi:10.1186/2047-2382-2-13 · Cited by 105
Consensus building for interlaboratory studies, key comparisons, and meta-analysis
Koepke, A., Lafarge, T., Possolo, A. M., & Toman, B. (2017) · Metrologia · doi:10.1088/1681-7575/aa6c0e · Cited by 106
“• τ and the σj { } have prior half-Cauchy distributions as suggested by [55] and further supported by [108].”
MethodComparing Bayesian and Frequentist Approaches for Multiple Outcome Mixed Treatment Comparisons
Hong, H., Carlin, B. P., Shamliyan, T. A., Wyman, J. F., Ramakrishnan, R., Sainfort, F., & Kane, R. L. (2013) · Medical Decision Making · doi:10.1177/0272989x13481110 · Cited by 91
“01,2) prior is adopted.(24) For the heterogeneous model, a more complex prior for sBk is introduced; namely, we set log sxy 1⁄4 log s0 1 vxy, where s0;Uniformð0:01; 2Þ; vxy;Nð0;c(2)Þ, and c is a constant chosen to reflect a certain amount of heterogeneity among arms.”
BackgroundMalecki, M. J. (2011) · Journal of European Public Policy · doi:10.1080/13501763.2012.632143 · Cited by 108
“10 Hyperprior parameters are given weakly informative t distributions with low degrees of freedom, and the variance is set at 10 and 100 producing ‘weakly informative priors’ (Gelman 2006; Gelman and Hill 2006; Gelman et al. 2004, 2008).”
BackgroundMulti-level modelling, the ecologic fallacy, and hybrid study designs
Wakefield, J. C. (2009) · International Journal of Epidemiology · doi:10.1093/ije/dyp179 · Cited by 97
Duncan, E. W., White, N. M., & Mengersen, K. L. (2017) · International Journal of Health Geographics · doi:10.1186/s12942-017-0120-x · Cited by 98
Over half of western United States' most abundant tree species in decline
Stanke, H., Finley, A. O., Domke, G. M., Weed, A. S., & MacFarlane, D. W. (2021) · Nature Communications · doi:10.1038/s41467-020-20678-z · Cited by 102
Bayesian parameter estimation for dynamical models in systems biology
Linden-Santangeli, N. J., Krämer, B., & Rangamani, P. (2022) · PLoS Computational Biology · doi:10.1371/journal.pcbi.1010651 · Cited by 113
An introduction to using Bayesian linear regression with clinical data
Baldwin, S. A., & Larson, M. J. (2017) · Behaviour Research and Therapy · doi:10.1016/j.brat.2016.12.016 · Cited by 96
Source Reconstruction Accuracy of MEG and EEG Bayesian Inversion Approaches
Belardinelli, P., Ortiz, E., Barnes, G. R., Noppeney, U., & Preissl, H. T. (2012) · PLoS ONE · doi:10.1371/journal.pone.0051985 · Cited by 92
Probabilistic analysis of gene expression measurements from heterogeneous tissues
Erkkilä, T., Lehmusvaara, S., Ruusuvuori, P., Visakorpi, T., Shmulevich, I., & Lähdesmäki, H. (2010) · Bioinformatics · doi:10.1093/bioinformatics/btq406 · Cited by 88
Choosing priors in Bayesian ecological models by simulating from the prior predictive distribution
Wesner, J. S., & Pomeranz, J. P. F. (2021) · Ecosphere · doi:10.1002/ecs2.3739 · Cited by 89
WinBUGS for Population Ecologists: Bayesian Modeling Using Markov Chain Monte Carlo Methods
Giménez, O., Bonner, S. J., King, R., Parker, R. A., Brooks, S. P., Jamieson, L. E., Grosbois, V., Morgan, B. J. T., & Thomas, L. (2008) · Springer eBooks · doi:10.1007/978-0-387-78151-8_41 · Cited by 102
“Several authors have dealt with important issues regarding the specification of vague priors (Lambert et al. 2005; van Dongen 2006; Gelman 2006), assessment of the sensitivity of the posterior distribution to the specified prior distribution (Millar 2004; Millar and Stewart 2005), parameter…”
BackgroundMavridis, D., Sutton, A., Cipriani, A., & Salanti, G. (2013) · Statistics in Medicine · doi:10.1002/sim.5494 · Cited by 87
Hu, X., Chen, X., Parks, G. T., & Yao, W. (2016) · Progress in Aerospace Sciences · doi:10.1016/j.paerosci.2016.07.004 · Cited by 97
Jordet, G., Aksum, K. M., Pedersen, D. N., Walvekar, A., Trivedi, A., McCall, A., Ivarsson, A., & Priestley, D. (2020) · Frontiers in Psychology · doi:10.3389/fpsyg.2020.553813 · Cited by 120
“As we are unsure of the magnitude of the variance parameter, we set vague uninformative priors on σα, σβ as follows: σα, σβ ∼ Half − Cauchy (β = 25) in accordance with Gelman (2006).”
Bayesian Markov switching models for the early detection of influenza epidemics
Martínez‐Beneito, M. A., Conesa, D., López‐Quílez, A., & López‐Maside, A. (2008) · Statistics in Medicine · doi:10.1002/sim.3320 · Cited by 102
“We choose uniform distributions as priors for the standard deviations of the random effects, as suggested for example in Gelman (2006) [5].”
MethodBayesian Hierarchical Models for aerospace gas turbine engine prognostics
Zaidan, M. A., Harrison, R. F., Mills, A. R., & Fleming, P. J. (2014) · Expert Systems with Applications · doi:10.1016/j.eswa.2014.08.007 · Cited by 89
Multiple defender effects: synergistic coral defense by mutualist crustaceans
McKeon, C. S., Stier, A. C., McIlroy, S. E., & Bolker, B. M. (2012) · Oecologia · doi:10.1007/s00442-012-2275-2 · Cited by 93
Using machine learning to characterize heart failure across the scales
Peirlinck, M., Costabal, F. S., Sack, K. L., Choy, J. S., Kassab, G. S., Guccione, J. M., Beule, M. D., Segers, P., & Kuhl, E. (2019) · Biomechanics and Modeling in Mechanobiology · doi:10.1007/s10237-019-01190-w · Cited by 91
“…a Bayesian setting, we select weakly informative priors and select our hyper-parameters as a k , b k ∼ N(0, 1002) and a k , b k ∼ Half-Cauchy ( = 10) (Gelman 2006), and our remaining parameters as (5) ⎡ ⎢ ⎢ ⎣ log(y i,j ml ) log(y i,j mw) ⎤ ⎥ ⎥ ⎦ ∼ N �� i ml (tj) i mw (tj) � , � , (6)log(yi,jed)…”
Luan, H., Law, J., & Quick, M. (2015) · International Journal of Health Geographics · doi:10.1186/s12942-015-0030-8 · Cited by 87
Multilevel structured additive regression
Lang, S., Umlauf, N., Wechselberger, P., Harttgen, K., & Kneib, T. (2012) · Statistics and Computing · doi:10.1007/s11222-012-9366-0 · Cited by 107
Reich, B. J., Storlie, C. B., & Bondell, H. D. (2009) · Technometrics · doi:10.1198/tech.2009.0013 · Cited by 86
“After a comparative study of several commonly-used priors, Gelman (2006) recommends either a uniform or half-Cauchy prior on the standard deviation.”
MethodDisentangling Bias and Variance in Election Polls
Shirani-Mehr, H., Rothschild, D. M., Goel, S., & Gelman, A. (2018) · Journal of the American Statistical Association · doi:10.1080/01621459.2018.1448823 · Cited by 84
Self-citationSchuemie, M. J., Coloma, P. M., Straatman, H., Herings, R. M. C., Trifirò, G., Matthews, J. N., Prieto‐Merino, D., Molokhia, M., Pedersen, L. H., Gini, R., et al. (2012) · Medical Care · doi:10.1097/mlr.0b013e31825f63bf · Cited by 95
Yuan, L. L., & Pollard, A. I. (2018) · Limnology and Oceanography · doi:10.1002/lno.10955 · Cited by 95
“Weakly informative priors were specified for all model parameters (Gelman 2006).”
MethodApplication of Bayesian Methods in Reliability Data Analyses
Li, M., & Meeker, W. Q. (2014) · Journal of Quality Technology · doi:10.1080/00224065.2014.11917951 · Cited by 93
“Gelman (2006) describes alternative ways to specify di↵use prior distributions for variances and variance components in hierarchical models.”
MethodInfluentialYamaura, Y., Royle, J. A., Kuboi, K., Tada, T., Ikeno, S., & Makino, S. (2010) · Journal of Applied Ecology · doi:10.1111/j.1365-2664.2010.01922.x · Cited by 90
Vasilev, M. R., & Angele, B. (2016) · Psychonomic Bulletin & Review · doi:10.3758/s13423-016-1147-x · Cited by 100
“The only notable finding was that the Gamma prior tended to underestimate τ (cf. Gelman, 2006), but even that did not result in a dramatic difference in the results.”
ResultFactors influencing citizens’ co-production of environmental outcomes: a multi-level analysis
Alonso, J. M., Andrews, R., Clifton, J., & Díaz‐Fuentes, D. (2019) · Public Management Review · doi:10.1080/14719037.2019.1619806 · Cited by 92
“It should be highlighted that Bayesian methods involve choosing a prior probability distribution for the parameters before analysing the data, such choice being the object of substantial debate in the related literature (see, e.g., Berger 2006; Browne and Draper 2006; Efron and Morris 1972; 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.