pmed() now accepts a medfit::ParallelMediationData object and computes the
joint P_med for k parallel mediators — the probability the outcome is
higher with all mediators at their treated levels than at their control
levels: Phi(delta * sum(a*b) / sqrt(2*sum(b^2*Vm) + 2*Vy)), recovering the
single-mediator formula at k = 1. All four methods are supported (plugin,
parametric_bootstrap, nonparametric_bootstrap, mbco), with the total
indirect effect sum(a_j b_j). Gaussian outcome and mediators only. See
vignette("parallel-mediation").
pmed(..., method = "mbco") adds a deterministic Model-Based Constrained
Optimization interval (Tofighi & Kelley, 2020): a likelihood-ratio interval
for both P_med and the indirect effect a * b, obtained by inverting the
constrained-likelihood test rather than by resampling. Gaussian outcome and
mediator (with covariates) and any contrast x_ref != x_value; seed-free and
grid-resolution-independent. Non-Gaussian models still use the bootstrap
methods.
print(PmedResult) interpretation line now shows the mediation estimand
P(Y(1, M(1)) > Y(1, M(0))) (manuscript Definition 1) instead of the stale
direct-effect notation P(Y_{X*, M_X} > Y_{X, M_X}). The computed value was
already correct; only the displayed notation was wrong. A plain-language gloss
was added beneath it. README cached output updated to match.First GitHub release (non-CRAN).
pmed() computes P_med — a scale-free probabilistic effect size for causal
mediation — from a formula or a medfit::MediationData object, with plugin,
parametric-bootstrap, and nonparametric-bootstrap methods, alongside the
indirect effect (a * b).pmed() now computes the mediation estimand
P(Y(x, M(x)) > Y(x, M(x*))) + 0.5 P(=) (manuscript Definition 1): treatment
held fixed, mediator varied between its treated and control levels with the
tie term. Previously it computed a direct-effect contrast that depended on the
direct effect c' and could land on the wrong side of 0.5. Verified against
the closed form Phi(ab / sqrt(2 (b^2 sigma_M^2 + sigma_Y^2))) and the
manuscript memory_exp example (P_med = 0.68).a/b); nonparametric bootstrap refits on the correct
family (was always Gaussian).litera theme with a clean, academic design matching the rmediation package style..qmd) for the README and vignettes, providing modern publishing capabilities.pmed() function to compute the probabilistic effect size $P_{med}$ for mediation analysis.extract_mediation() support for SEM models fitted with the lavaan package, including FIML and robust estimators.mediation::mediate() objects.pmed() now reports the Indirect Effect (product of coefficients) alongside $P_{med}$, including bootstrap confidence intervals.lavaan and mediation package integrations.