Skip to contents

Robust Bayesian Model Averaging for Meta-Analysis

Source: R/robma.R
robma.Rd

Fits a Robust Bayesian Meta-Analysis (RoBMA) model using Bayesian model averaging across models with and without an effect, heterogeneity, and publication bias. Use stan_code(model) to inspect the generated Stan programs after fitting.

Usage

robma(
  data,
  studyvar,
  event_ctrl = NULL,
  event_int = NULL,
  mean_ctrl = NULL,
  mean_int = NULL,
  sd_ctrl = NULL,
  sd_int = NULL,
  n_ctrl = NULL,
  n_int = NULL,
  likelihood = c("binomial", "gaussian", "poisson"),
  priors_effect = NULL,
  priors_effect_null = NULL,
  priors_heterogeneity = NULL,
  priors_heterogeneity_null = NULL,
  priors_bias = NULL,
  priors_bias_null = NULL,
  rescale_priors = 1,
  method = c("bridge", "ss"),
  bias_indicator = c("bias_corrected", "pet_peese", "selection_weight"),
  null_range = NULL,
  b_prior = NULL,
  p_bias_prior = NULL,
  p_cutoffs = c(0.025, 0.05),
  parallel = FALSE,
  chains = 4,
  iter_warmup = 1000,
  iter_sampling = 1000,
  adapt_delta = 0.95,
  seed = 1234,
  quiet = FALSE,
  custom_model = NULL,
  custom_data = NULL,
  format = TRUE,
  ...
)

Arguments

data

A data frame with one row per study.

studyvar

Character. Column name of the study identifier.

event_ctrl, event_int

Character. Column names of event counts (binomial / Poisson likelihoods).

mean_ctrl, mean_int, sd_ctrl, sd_int

Character. Column names of arm means and SDs (Gaussian likelihood).

n_ctrl, n_int

Character. Column names of arm sample sizes.

likelihood

Character. One of "binomial", "gaussian", "poisson".

priors_effect, priors_effect_null, priors_heterogeneity, priors_heterogeneity_null, priors_bias, priors_bias_null

Lists of prior objects for the effect, heterogeneity, and publication-bias components (alternative and null). If NULL, RoBMA defaults are used.

rescale_priors

Numeric. Scale factor applied to default priors. Default 1.

method

Character. "bridge" (default) uses bridge sampling across the full model grid; "ss" uses a single spike-and-slab Stan model.

bias_indicator

Character. Spike-and-slab bias mechanism: "bias_corrected", "pet_peese", or "selection_weight".

null_range

Numeric vector of length 2 giving the null range on the log scale (e.g., c(-0.1, 0.1) for log OR). Effects within this range are considered practically equivalent to zero. Defaults to NULL (point null at exactly zero). For OR/RR, c(-0.1, 0.1) corresponds to OR/RR in [0.905, 1.105].

b_prior

Prior on the b slope for spike-and-slab bias correction.

p_bias_prior

Prior on the bias inclusion probability.

p_cutoffs

Numeric vector of one-sided p-value cutoffs for selection-weight models. Default c(0.025, 0.05).

parallel

Logical. Fit the bridgesampling grid in parallel.

chains, iter_warmup, iter_sampling, adapt_delta, seed

MCMC settings.

quiet

Logical. Suppress per-step progress messages.

custom_model

Optional Stan program(s) that override code generation. For method = "bridge", a named list of character scalars keyed by model label. For method = "ss", a single character scalar.

custom_data

Optional Stan data overrides merged onto the auto-built data list(s). Same shape conventions as custom_model.

format

Logical. If TRUE (default), auto-format generated Stan programs with stanc --auto-format.

...

Additional arguments passed to cmdstanr::CmdStanModel$sample().

Value

An object of class c("bayesma_robma", "bayesma").