Side-by-side treatment-by-subgroup interactions from DA, AD, SWADA, and CAMS

Under the default Interaction RE-weights, the SWADA-consistent interaction equals ad(); it is listed once as SWADA rather than as two duplicate SWADA methods. da() and ad() are different constructions (Panaro, Röver & Friede 2026) and generally differ because they use different study weights.

Usage

compare_interactions(
  da,
  ad = NULL,
  swada = NULL,
  cam = NULL,
  exponentiate = FALSE,
  type = "Overall"
)

Arguments

da

A da() or conventional() fit, or a cams() fit when comparing two CAMS approaches.

ad

An ad() fit. Optional when da is a conventional() bundle, or a second cams() fit.

swada

A swada() fit.

cam

A cams() CAMS fit.

exponentiate

If TRUE, report on the exponentiated (ratio) scale. Default FALSE (fitted log scale).

type

CAMS layer ("Overall" default).

Value

A tibble with class interactmeta_compare_interactions.

Details

The first argument may be a conventional() bundle, in which case DA and AD are read from that object and ad may be omitted when the remaining arguments are swada and cam (historical three-argument call).

Warns if mixed constructions do not share the same engine and method. A like-for-like frequentist comparison uses metafor / REML throughout (da() / ad() / swada() defaults, plus cams(..., method = "REML")). Two CAMS fits of the same model with different methods are labelled (CAMS — Bayesian vs CAMS — REML) and are not treated as a mismatch.

Two or more cams() fits may be passed instead of DA / AD / SWADA / CAMS. plot() draws the compared interaction estimates.