BayesTools_refit_required errorAIC() and BIC() stop with an explanation instead of returning numeric(0)seed argument of update()radial() stops for likelihood-weighted fits, and cluster-unit LOO and WAIC of binomial and Poisson GLMMs stop (use unit = "estimate")mods = ~ g + g:x); use mods = ~ g * x, an independent prior for the term, or treatment contrastsprior_mods, prior_scale, and prior_heterogeneity_allocation stop when the model has no corresponding component instead of being ignoredweights are rejectedprior_lograte is Normal(log(pooled crude rate), 1), independent of the exposure-time unit (set its SD with RoBMA.options(default_lograte.sd = ))BMA.glmm() uses mean-difference contrasts by default, as the other model-averaging constructors doplot() of zplot() results no longer draws the extrapolated curve by default (plot_extrapolation = FALSE)predict() gains conditioning_depth ("marginal" by default, "cluster", or "estimate"), which changes the default same-data type = "estimate" predictions and the clustered and GLMM response predictionszplot() targets the marginal posterior predictive distribution by default (conditioning_depth)funnel() draws plug-in contours; the posterior-averaged contours of 4.0.0 are drawn by the new bfunnel()NA with the reasonas_draws() returns the parameters shown by print() and summary() by default; auxiliary parameters require include_auxiliary = TRUEg{j}BMA() and RoBMA() summaries is labelled exp(intercept), and a location intercept fixed at zero is omitted from meta-regression outputbrma.mv() for meta-analysis with a known sampling covariance matrix V and formula random effects (random, e.g., ~ 1 | study / esid) with diagonal, unstructured, compound-symmetry, AR(1), heterogeneous, and continuous-time AR(1) structures, and known random-effect group covariance (R)BMA.mv(), RoBMA.mv(), bPET.mv(), bPEESE.mv(), and bselmodel.mv() for model-averaged and publication-bias-adjusted multivariate and multilevel modelsprior_random() interface for structured random-effect priors (variance allocations, inclusion gates, and LKJ priors), completed with RoBMA's default priorstau, tau2, rho, rho_total, tau_total, tau2_prop, ...) are named consistently across summaries, plots, hypotheses, and as_draws(); a mean-difference or orthonormal factor in a random-effect formula has one SD and, under us(), is correlated with the other coefficients by level (rho(intercept, f[a])) and as a whole (rho_total(intercept, f)), and random-effect inclusion is reported in the component inclusion tablecomponent to summary_heterogeneity(), pooled_heterogeneity(), ranef() (also expand and simplify), plot(), plot_prior(), print_prior(), and plot_diagnostic()prior_ordered() for ordered-factor moderatorsmods, scale, and yi ~ formulas accept ^ crossing (e.g., (x + z)^2)selection = selection_model(...), with product or best p-value weight rules, publication groups, and deterministic quadrature with a controlled QMC fallback (set_selection_likelihood_control())selection_cache_info(), remove_selection_cache()) and a coarse-corrected selection sampler (RoBMA.options(selection.sampler = "coarse_corrected"))selection_sensitivity_diagnostics() and add_selection_sensitivity_diagnostics()hypothesis() (aliases bf_hypothesis() and BF_hypothesis()) for point, interval, and directional hypotheses on parameters, factor levels, and their linear combinations in fitted models and marginal means, and hypothesis_quantities(), which lists the testable quantitiesdensity_method, density_control) for plots, hypotheses, and marginal means, with density_diagnostics()check_indicators, monitor, and allow_not_assessable to convergence checks and autofitlog_lik(), a loo_model_weights() method, and parallel, cores, repetitions, method, maxiter, and silent to add_marglik()pooled_effect()max_samples to hatvalues(), residuals(), rstandard(), qqnorm(), and vif(), and component and parameter to dfbetas() and covratio()vif() computes design-based VIF/GVIF in each posterior draw, with optional credible intervals (probs) and drawsbfunnel(), a metafor::forest() method with as_metafor_forest(), and lines() methods for fitted models and marginal meanstransform = "EXP" and "LOG" apply to individual coefficients and heterogeneity parameters in plotsregplot() accepts moderators that appear only in the scale formula, and radial() supports brma.mv() modelsregplot(), zplot(), and plot_pet_peese() use Harrell-Davis quantilesas.data.frame() converts printable results to long data framesupdate() accepts named NULL values to clear convergence and autofit settingsset_contrast_factor_predictors = "independent" defines one coefficient per factor levelHoogeveen2023 data setRoBMA.options(native_threads = ) and RoBMA.options(jags.worker_output = )cooks.distance() uses the unscaled squared Mahalanobis distance, matching metafor (values are multiplied by the fixed-effect rank relative to 4.0.0)... instead of ignoring themsubset, missing values) drops unused factor levels, except for the index factor of structured random-effect termsprior_informed_field = "medicine", default interaction priors that would depend on a mean-difference or orthonormal basis use the multivariate t version of the informed priorrstandard() reports NA instead of 0 for the standardized residuals of exactly fitted estimates, and qqnorm() omits thempredict() stops when newdata lacks a formula variable instead of filling outcome-named columns with zerosyi ~ 0) fixes the location intercept at zerosummary() no longer failsplot_weightfunction() marks two-sided p-values for two-sided weight functionsxlim on the displayed scaleranef(component = "total") keeps the estimate namesvif(), hatvalues(), and related diagnostics of RoBMA() ensembles use the bias design of each memberdfbetas(), influence(), and vif() omit a fixed-zero intercept, and covratio() is finite for mean-difference and orthonormal contrastsas_zplot() and zplot() validate significance_levelxlim, count length.out as bins, and keep the to limit when by does not divide the rangesummary_heterogeneity() omits I2 and H2 for saturated meta-regressions instead of failingradial() passes targs to transf as its second argument, as in metaforprior_effect, prior_heterogeneity, and predictor priors reject PET, PEESE, and non-point discrete (e.g., Bernoulli) prior distributionsinterpret() reports errors of the pooled summaries instead of omitting the sentencesylimforest() accepts predstyle = "polygon"covratio(), dfbetas(), and posterior-sample summaries return the printed objectprint() and default summary() of fitted modelsRoBMA.options(native_threads = )) and use AVX2/AVX-512 for selection normalizersbrma class hierarchy. Single-model fits now use brma(), brma.glmm(), bselmodel(), bPET(), and bPEESE(); model-averaged fits use BMA(), BMA.glmm(), and RoBMA().RoBMA.reg(), NoBMA(), NoBMA.reg(), BiBMA(), and BiBMA.reg() constructors. Use mods, scale, and cluster in the new constructors, BMA() for no-bias normal-likelihood model averaging, and BMA.glmm() for GLMM model averaging.d, r, logOR, OR, z, y, se, v, n, study_names, study_ids, weight, and transformation with yi, vi/sei, ni, slab, cluster, weights, measure, output_measure, and transform.combine_data(), check_setup(), extract_posterior(), marginal_summary(), marginal_plot(), plot_models(), adjusted_effect(), as_zcurve(), and the old z-curve plotting methods.measure for fitted models. Use measure = "GEN" for generic effect sizes without a known unit-information scale.update() for brma objects now focuses on extending MCMC samples, updating labels, and refreshing cached quantities, not changing model structure.set_convergence_checks() no longer accepts the old remove_failed and balance_probability arguments.brma() / brma.norm() for single normal-likelihood Bayesian meta-analysis, including random-effects, meta-regression, multilevel, and location-scale models.brma.glmm() for binomial-normal and Poisson-normal GLMM meta-analysis from raw two-arm counts (measure = "OR" and "IRR").bselmodel(), bPET(), and bPEESE().BMA() / BMA.norm() for Bayesian model averaging without publication-bias adjustment.BMA.glmm() for Bayesian model averaging of GLMM meta-analyses without publication-bias adjustment.RoBMA() as a product-space model-averaged ensemble over effect, heterogeneity, moderator, scale, and publication-bias components.prior_weightfunction(), wf_cumulative(), wf_fixed(), and wf_independent() for BayesTools-backed selection-weightfunction priors.prior_PET(), prior_PEESE(), prior_none(), prior_factor(), prior_informed(), and BayesTools contrast helpers as package-level prior utilities.posterior package interfaces via as_draws(), as_draws_array(), as_draws_df(), as_draws_list(), as_draws_matrix(), and as_draws_rvars() for fitted models and brma_samples.brma_samples posterior-sample class with print, summary, matrix, and posterior conversion methods.predict.brma() for posterior predictions of fixed terms, cluster effects, latent true effects, observed responses, and scale terms, with newdata, conditional, bias_adjusted, output_measure, and transform support.fitted(), pooled_effect(), pooled_heterogeneity(), blup(), true_effects(), and ranef() for brma objects.add_loo(), loo(), loo_compare(), loo_weights(), check_loo(), add_waic(), and waic() using the loo package.brma fits via add_marglik(), bridge_sampler(), logml(), bf(), bayes_factor(), and post_prob().residuals(), rstandard(), rstudent() / LOO-PIT, hatvalues(), influence(), dfbetas(), dffits(), cooks.distance(), covratio(), and vif().brma objects: posterior/prior plots, funnel(), regplot(), qqnorm(), radial() / galbraith(), MCMC diagnostic plots, weightfunction plots, and PET-PEESE plots.marginal_means() with summary and plotting methods for moderator models.summary_models() for marginal and individual model-weight summaries of product-space RoBMA, BMA, and BMA.glmm objects.interpret() for concise textual interpretation of fitted brma and model-averaged objects.as_zplot() and adds the direct plotting wrapper zplot(), with plot(), hist(), lines(), summary(), and print methods for zplot objects.RoBMA.options() and RoBMA.get_option() package options for defaults such as core count, automatic LOO/WAIC/marginal-likelihood computation, prior scaling defaults, and selection-bias defaults.cluster.slab, matching metafor naming.weights for supported constructors and applies them consistently to posterior fitting, log-likelihoods, LOO, WAIC, and diagnostics; brma.mv() currently rejects likelihood weights.measure, output_measure, and transform for effect-size scale handling. Supported conversions include SMD, COR, ZCOR, and OR; transform = "EXP" exponentiates log ratio measures for display.predict.brma() default to type = "terms". GLMM type = "response" predictions return continuity-corrected effect-size estimators by default via as_measure = TRUE.unit from conditioning_depth for residuals, fitted values, LOO, WAIC, and related diagnostics.brma.mv() and estimate-/cluster-level LOO/WAIC targets for multilevel brma() models, with target metadata to prevent invalid comparisons.brma objects; product-space RoBMA, BMA, and BMA.glmm objects rely on product-space only.bias_indicator and branch-aware selected-normal contexts for RoBMA publication-bias mixtures instead of inferring selection branches from omega.10000 samples and accepts Inf where full posterior evaluation is requested.max_samples controls to expensive funnel, regplot, and zplot summaries.vignette("v00-introduction", package = "RoBMA").bridgesampling, loo, MASS, and parallel as imports and posterior as a suggested package.src/selnorm/ and updates Makevars*, native registration, cleanup rules, and JAGS distribution registration.Explanation vignette that helps navigate users through the vignettessummary() function now provides publication bias model type summary (type = "models") for models fitted using algorithm = "ss"funnel() plot to visualize residuals vs the expected sampling distribution for RoBMA() and RoBMA.reg() models when using the algorithm = "ss"residuals() method for RoBMA() and RoBMA.reg() models when using the algorithm = "ss"as_zplot() function to transform meta-analytic models into a zplot object, only available for RoBMA() and RoBMA.reg() fitted using the algorithm = "ss"plot(), summary(), and print() functions for the as_zplot objectssummary() function now supports a standardized_coefficients argument to report either standardized (default) or raw meta-regression coefficientsextract() function to extract the posterior samples of the model parameterstrue_effects() function to summarize the true effect size estimates of RoBMA() and RoBMA.reg() models when using the algorithm = "ss"predict() method for RoBMA() and RoBMA.reg() models when using the algorithm = "ss"RoBMA() and RoBMA.reg() functions with the cluster argument when using algorithm = "ss"BiBMA and BiBMA.reg functions with the cluster argument when using algorithm = "ss"pooled_effect() function to compute the pooled effect size from the RoBMA.reg, NoBMA.reg, and BiBMA.reg modelsadjusted_effect() function to compute the adjusted effect size from the RoBMA.reg, NoBMA.reg, and BiBMA.reg modelssummary_heterogeneity() for BiBMA modelscluster and study_labels argumentsrescale_priors argumentsummary_heterogeneity() is now conditional on the presence of the effectalgorithm = "ss"summary_heterogeneity() with only a single study does not produce relative heterogeneity instead of crashingBiBMA.reg functionalgorithm = "ss" argument for BiBMA modelsset_default_binomial_priors() functionalgorithm = "ss" argument (see a new vignette for more details)set_default_priors() functionrescale_priors argument allows to conveniently re-scale the prior distributions for the effect, heterogeneity, and bias simultaneouslysummary_heterogeneity() function to summarize the heterogeneity of the RoBMA models (prediction interval, tau, tau^2, I^2, and H^2)check_RoBMA_convergence() function to check the convergence of the RoBMA modelsupdate() function re-evaluates convergence checks of individual models (https://github.com/FBartos/RoBMA/issues/34)BiBMA functionNoBMA and NoBMA.reg() functions as wrappers around RoBMA RoBMA.reg() functions for simpler specification of publication bias unadjusted Bayesian model-averaged meta-analysisupdate.RoBMA() function (only non-converged models by default or all by setting extend_all = TRUE)RoBMA.reg() functionRoBMA.reg models with summary_marginal() and plot_marginal() functionsweight argument in the RoBMA and combine_data functions in order to pass custom likelihood weightsweighted_type = "inverse_sqrt" argumentpriors_hierarchical and priors_hierarchical_null arguments instead of priors_rho and priors_rho_null. The model summary now shows Hierarchical component summary.cluster argument in RoBMA() and setting weighted = TRUE. The likelihood contribution of estimates from each study is down-weighted proportionally to the number of estimates in that study. Note that this experimental feature is supposed to provide a conservative alternative for estimating RoBMA in cases with multiple estimates from a study where the multivariate option is not computationally feasible.cluster argument in RoBMA. However, note that this is (1) an experimental feature and (2) the computational expense of fitting selection models with clustering is extreme. As of now, it is almost impossible to have more than 2-3 estimates clustered within a single study).informed_prior() function (from the BayesTools package) that allows specification of various informed prior distributions from the field of medicine and psychologysave = "min"interpret() function (issue #11)output_scale argument to plot() and plot_models() functionsPlease notice that this is a major release that breaks backwards compatibility.
priors_mu -> priors_effect, priors_tau -> priors_heterogeneity, and priors_omega -> priors_bias),prior(distribution = "two.sided", parameters = ...) -> prior_weightfunction(distribution = "two.sided", parameters = ...)),prior_none()),prior_PET(distribution = "Cauchy", parameters = ...) and prior_PEESE(distribution = "Cauchy", parameters = ...)),model_type argument allowing to specify different "pre-canned" models ("PSMA" = RoBMA-PSMA, "PP" = RoBMA-PP, "2w" = corresponding to Maier et al., in press , manuscript),combine_data function allows combination of different effect sizes / variability measures into a common effect size measure (also used from within the RoBMA function),autofit = FALSE)prior_scale and transformation arguments),plot_models() function and the forest plot can be obtained with the forest() function,plot.RoBMA() function and parameter = "weightfunction" and parameter = "PET-PEESE".