Rubin-style shrinkage trace plots for a CAMS fit

Draws bayesmeta::traceplot() diagnostics from the native bayesmeta / bmr objects stored on a Bayesian CAMS fit (trial-specific shrinkage means vs heterogeneity τ, with the τ posterior underneath). Available type values are interaction (the interaction-block bayesmeta fit) and combined (the combined-population effects represented by the overall-block bmr fit).

Usage

# S3 method for class 'interactmeta'
traceplot(
  x,
  type = c("subgroups", "interaction", "combined"),
  ylab = NULL,
  xlab = NULL,
  tau.prior = "sqrt",
  X = NULL,
  ...
)

Arguments

x

A Bayesian interactmeta from cams() (engine = "bayesmeta").

type

"interaction" or "combined". "modifiers" / "subgroups" are rejected: Bayesian CAMS does not store separate per-subgroup meta-analyses.

ylab

y-axis label passed to bayesmeta::traceplot(). Default is "Interaction" for type = "interaction" and "Effect" for "combined".

xlab

x-axis label used in place of bayesmeta's hardcoded "heterogeneity τ". Default is "heterogeneity τ_γ" for interaction and "heterogeneity τ" for combined.

tau.prior

Ignored. Retained for compatibility; the stored native objects already use the CAMS priors from the original cams() fit.

X

Optional design row for bayesmeta::traceplot.bmr. Default is intercept plus mean trial information fraction from the stored CAMS trial blocks.

...

Further arguments passed to bayesmeta::traceplot() (mulim, taulim, ci, prior, infinity, colours, …).

Value

Invisibly, a named list of the native bayesmeta::bayesmeta() or bayesmeta::bmr() objects that were plotted.

Details

This method does not fit auxiliary bayesmeta::bayesmeta() models. Frequentist CAMS fits have no native bayesmeta objects; traceplot() returns an informative error rather than substituting another implementation. Per-subgroup traceplots are not stored by the orthogonal CAMS factorization (type = "modifiers" or "subgroups" errors).

See also

forestplot(), bayesmeta::traceplot(), plot_heterogeneity(), native_fit(), vignette("cams")

Examples

if (FALSE) { # \dontrun{
library(interactmeta)
es <- escalc(measure = "IRR", yi = edss$yi, vi = edss$vi, data = edss)
fit <- cams(es, reference = "Lower")
traceplot(fit, type = "interaction")
traceplot(fit, type = "combined", prior = TRUE)
} # }