NEWS
plavaan 0.0.2.9000
penalized_est() gains eps control for built-in penalties, including eps = "telescoping" over a decreasing epsilon sequence. Telescoping reuses the original starting values by default; set telescoping_control = list(warm_start = TRUE) to warm-start later stages from prior estimates.- New
effective_df() for penalized fits: reports the effective number of parameters per penalty component and the effective model degrees of freedom, using the soft-count property of the penalties.
effective_df() (and the fit-evaluation degrees of freedom) now count sample statistics correctly for models with ordinal/categorical indicators (e.g. WLSMV fits with thresholds). The count now uses lavaan's own lav_pt_ndat() where available (with a structural fallback), so it accounts for ordinal thresholds, composites, and the correlation metric, and always matches lavaan's own model df. The previous count used a continuous-variables formula that miscounted such models. Results for purely continuous models are unchanged.- Fits returned by
penalized_est() are now plavaan objects. summary() reports the effective number of parameters and degrees of freedom, and — when fit evaluation is enabled with test = "Chisq" (or "SatorraBentler") — fitmeasures() and the chi-square test in summary() additionally report fit indices at the effective degrees of freedom via a frozen refit at the penalized estimates. This fit evaluation is experimental and disabled by default (test = "none", the new default for the test argument); an experimental notice is shown whenever it is used.
summary() on a penalized fit prints standard errors (and the resulting z-values and p-values) when penalized_est() was called with se = "robust.huber.white". A development-version change that computed the summary() from a frozen refit had suppressed the standard-error column even when standard errors were requested; this is now corrected. parameterEstimates() was unaffected.fitmeasures() on a penalized fit now honours its arguments for selecting and formatting fit indices (e.g. fitmeasures(fit, "rmsea"), output = "matrix", fm_args). A development-version change had declared the method as (object, ...), so lavaan's fitmeasures() generic's named arguments were silently dropped and every call returned the full set of fit measures; the method now declares and forwards them.
plavaan 0.0.2 (2026-07-28)
- Added
penalized_est_multistart() and support for custom optimizer starting values via start in penalized_est().
- Minor changes to speed up the optimization process.
Breaking Changes
- Penalty metric for factor loadings: The penalty for cross-group differences in factor loadings is now computed on the original scale ($\lambda_{g1} - \lambda_{g2}$) rather than the log scale ($\log(\lambda_{g1}) - \log(\lambda_{g2})$).
plavaan 0.0.1 (2025-12-30)