Publication-bias prior specification

Publication-bias prior distributions are specified separately from the meta-analytic prior distributions documented in prior_specification. They define selection-model weights, PET/PEESE regression coefficients, or the publication-bias model space in RoBMA.

Arguments

prior_bias

prior distribution for publication-bias adjustment. For bselmodel, this is usually a weightfunction prior created by prior_weightfunction. For bPET, use prior_PET. For bPEESE, use prior_PEESE. For RoBMA, this can be a single publication-bias prior distribution or a list of publication-bias prior distributions. In the single-model bias-adjustment constructors, omitted or NULL uses the corresponding default prior distribution.

prior_bias_null

prior distribution(s) for null publication-bias component(s) in RoBMA, usually prior_none(). A single prior distribution object creates one null component, a list creates multiple null components, and NULL or FALSE omits the null publication-bias component.

model_type

character string specifying predefined publication-bias model ensembles for RoBMA. One of "PSMA", "6w", "2w", or "PP".

steps

numeric vector of one-sided p-value cut points for the default bselmodel weightfunction prior. If prior_bias is supplied, the prior distribution carries its own p-value cut points.

Details

Single-model bias-adjustment priors

bselmodel, bPET, and bPEESE fit one publication-bias adjustment at a time. The prior_bias argument must match the fitted bias-adjustment type.

Constructor	Prior constructor	Default prior distribution
bselmodel()	prior_weightfunction	one-sided cumulative weightfunction with steps = 0.025
bPET()	prior_PET	positive Cauchy centered at 0
bPEESE()	prior_PEESE	positive Cauchy centered at 0, with UISD-adjusted scale

The default PET prior distribution uses RoBMA.get_option("default_bias_PET.scale"). The default PEESE prior distribution uses RoBMA.get_option("default_bias_PEESE.scale") after rescaling to the analyzed effect-size measure. The default weightfunction prior distribution uses RoBMA.get_option("default_bias_weightfunction.alpha") for each cumulative-Dirichlet alpha parameter.

Model-averaged publication-bias priors

RoBMA averages over publication-bias components. By default, prior_bias_null is prior_none() and model_type determines the alternative components:

"PSMA"	six weight functions, PET, and PEESE
"6w"	six weight functions
"2w"	two two-sided weight functions
"PP"	PET and PEESE

Custom prior_bias replaces the preset alternative components. If prior_bias is omitted, model_type still supplies the default alternative components even when prior_bias_null is customized or omitted. Setting prior_bias = NULL or prior_bias = FALSE omits all alternative bias-adjustment models. Setting prior_bias_null = NULL or prior_bias_null = FALSE omits the no-bias component.

Each publication-bias component has a prior model weight. User-specified components receive equal weights unless their prior objects set prior_weights. The "2w" and "6w" presets split the alternative bias mass equally across their weight functions, "PP" splits it equally across PET and PEESE, and "PSMA" assigns half of the alternative bias mass to the six weight functions combined and one quarter each to PET and PEESE.

Publication-bias prior distributions are not rescaled by rescale_priors. The exception is the default PEESE prior distribution, whose scale is adjusted to the effect-size measure via the unit-information standard deviation.

See also

prior_weightfunction, prior_PET, prior_PEESE, prior_none, RoBMA_prior_specification, prior_specification