FBartos/BayesTools: BayesTools 0.3.0

Features

• major refactoring and speed-up of unit tests
• adds support for __default_factor and __default_continuous priors in JAGS_formula() - when specified in the prior_list, these are used as default priors for factor and continuous predictors that are not explicitly specified
• adds automatic standardization of continuous predictors via formula_scale parameter in JAGS_formula() and JAGS_fit() - improves MCMC sampling efficiency and numerical stability
• adds transform_scale_samples() function to transform posterior samples back to original scale after standardization
• adds transform_prior_samples() function to generate and transform prior samples using the same matrix transformation as posterior samples - enables correct visualization of priors on the original (unscaled) predictor scale, including proper handling of the intercept which depends on multiple coefficient priors
• adds transform_scaled argument to plot_posterior() for visualizing prior and posterior distributions on the original (unscaled) scale when using formula-based models with auto-scaling
• adds exp_lin transformation type for log-intercept unscaling in density/plotting functions: exp(a + b * log(x))
• adds log(intercept) formula attribute for specifying models of the form log(intercept) + sum(beta_i * x_i) - useful for parameters that must be positive (e.g., standard deviation) while keeping the intercept on the original scale. Set via attr(formula, "log(intercept)") <- TRUE. Supported in JAGS_formula(), JAGS_evaluate_formula(), and marginal likelihood computation
• adds advanced parameter filtering options to runjags_estimates_table():
  • remove_parameters = TRUE to remove all non-formula parameters
  • remove_formulas to remove all parameters from specific formulas
  • keep_parameters to keep only specified parameters
  • keep_formulas to keep only parameters from specified formulas
  • when bias is specified in remove_parameters or keep_parameters, the corresponding bias-related parameters (PET, PEESE, omega, alpha, pi_null, and phack_kind) are automatically included based on the bias prior type
• adds probs argument to runjags_estimates_table() and runjags_estimates_empty_table() for custom quantiles (default: c(0.025, 0.5, 0.975))
• adds effect_direction argument to plot_posterior(), plot_prior_list(), lines_prior_list(), and geom_prior_list() for PET-PEESE regression plots - use "positive" (default) for mu + PET*se + PEESE*se^2 or "negative" for mu - PET*se - PEESE*se^2
• redesigns prior_weightfunction() around a unified side, steps, and weights specification, with wf_cumulative(), wf_fixed(), and wf_independent() constructors for cumulative Dirichlet, fixed, independent, and log-independent weightfunction priors
• adds p-hacking and composed selection-bias priors via prior_phacking(), prior_bias(), calibration helpers, and selection_backend_spec() for compiling active step/p-hacking backend parameters
• adds error % for inclusion BF calculation

Changes

• changes quantile column names in runjags_estimates_table() and stan_estimates_table() from lCI/Median/uCI to numeric values (e.g., 0.025/0.5/0.975) for consistency with ensemble summary tables
• implied prior distributions for estimated marginal means, unstandardized coefficients, and PET-PEESE no longer require prior samples
• implied prior distributions for weightfunction weights now use analytical forms for cumulative Dirichlet, fixed, independent, and log-independent priors, including mixture and model-averaged weightfunctions where possible
• independent weightfunction priors now allow non-reference weights above one via non-negative omega-scale priors or unrestricted log-omega priors
• replaces the legacy dot-named weightfunction prior specifications with the unified weightfunction prior API and updates JAGS generation, marginal likelihood computation, posterior extraction, diagnostics, and summary tables to use the new component-local omega representation
• composed selection-bias priors and publication-bias mixtures now support prior sampling and explicit unsupported-operation errors for ambiguous scalar prior generics

Fixes

• reports inclusion Bayes factors as NA when the prior assigns probability 0 or 1 to inclusion, while keeping finite-sample bounds for posterior inclusion probabilities of 0 or 1
• fixes incorrect ordering the printed mixture priors
• fixes formula with no intercepts coded as 0 (instead of only -1)
• fixes bug in .is.wholenumber with NAs and na.rm = TRUE
• fixes ggplot prior spike layers for marginal factor plots with density and point components