NEWS
rmedsem 1.1.1
- New backend for PLS-SEM and PLSc-SEM models estimated with plssem
rmedsem 1.1.0 (2026-09-16)
Breaking changes
summary() returns a summary.rmedsem object
as.data.frame() returns a plain data frame
rmedsem() stops with an error for models that previously ran but gave
wrong or empty results
ci.two.tailed now determines all intervals
- blavaan: the direct effect's
pval is now the posterior probability of the
opposite sign
mcreps defaults to 5000 and is no longer raised to the sample size
- RID is not reported if the direct effect is not
significant;
RIT() warns if the total effect is too small
New features
summary() method
- New methods
coef(), confint() and nobs() for rmedsem objects.
rmedsem.blavaan() gains ci.two.tailed and hdi (highest-density intervals)
- Printing is now an extensible S3 method system
- Result objects gain the elements
zlc.method
- All exported functions check their arguments and stop with informative
messages
- lavaan: a warning is given if the model did not converge.
- cSEM: new
seed argument; results are reproducible with set.seed()
Bug fixes
- lavaan and modsem models with labelled regression paths (e.g.,
read ~ a*math) now work
- blavaan: the rows of the printed table were shifted relative to their
labels
- modsem: the moderation effects are printed in causal direction
- cSEM: the bootstrap interval of the indirect effect was taken from the incorrect row
digits and indent are passed correctly to the effect-size output.
plot_coef() no longer emits a position_dodge() warning
- Very small p-values are no longer printed as
0
plot_effect() title and caption are wrapped to fit small figures
summary() and plot_effect() flag RIT and RID values that should not be
interpreted
- Clearer printed output
- Dependencies on dplyr and purrr removed
Documentation
- Help pages were merged into a compact set:
?rmedsem, ?rmedsem-methods, ?effect-sizes and ?plot.rmedsem
- Examples for all backends
- Corrected and completed the documentation of the datasets (incl. sources
and response scales)
- New article "Working with the results"
- Added CRAN badge to README
rmedsem 1.0.0 (2026-03-16)
- Initial CRAN release.
- Support for mediation analysis with lavaan (CB-SEM), cSEM (PLS-SEM),
blavaan (Bayesian SEM), and modsem (moderated mediation).
- Implements Baron and Kenny (1986) and Zhao, Lynch & Chen (2010) approaches.
- Reports indirect effects via Sobel, Delta, and Monte-Carlo methods.
- Effect size measures: RIT and RID.
- Summary and plot methods for results.
- Fix standard error calculation for standardized coefficients in
rmedsem.lavaan and rmedsem.modsem (contributed by Kjell Slupphaug).
- New
ci.two.tailed parameter for configurable confidence interval width.