RESI 1.3.3
Bug Fixes
- Fixed error when passing a
dataargument containingNAs toresi()for GEE models (geeglm). Rows withNAin any model variable are now silently stripped before bootstrapping, matching the complete cases used to fit the model and preventing cluster-size mismatches during re-fitting (#51). - Fixed misleading error in
summary.resi()when a differentalphalevel is requested but confidence intervals cannot be recomputed. The message now clearly explains that either (a) bootstrapping is not supported for the model type, or (b)resi()was not run withstore.boot = TRUE, and instructs the user to re-runresi()with the desiredalphadirectly (#53). - Fixed
resi()failing to compute the overall Wald test when the model response is a computed expression (e.g.log10(charges)orI(charges > 10000)). Previously the intercept-only reduced model could not be fitted inside forked parallel workers becauseupdate()tried to re-evaluate the expression in an environment where the underlying variable was not in scope. The fix constructs the reduced-model formula using the already-evaluated column name frommodel.frame(), avoiding any re-evaluation. - Added an informative error when
vcov.args = list(type = "const")is passed together withvcovfunc = sandwich::vcovHC. The message explains thattype = "const"is the OLS sandwich (not robust) and directs users to usevcovfunc = stats::vcovfor parametric variance estimation instead (#50). -
resi_penow reports CS-RESI and L-RESI forlmer.
New Features
- Added support for
robustbasemodels (lmrobandglmrob) via newresi_pe.lmrob,resi_pe.glmrob,resi.lmrob, andresi.glmrobmethods. Both default tovcovfunc = stats::vcov, which uses the model’s built-in robust sandwich variance.glmrobredirectssandwich::vcovHCtostats::vcovwith a warning sincevcovHCdoes not supportglmrob(#12).
RESI 1.3.2
CRAN release: 2025-07-29
- Added RESI estimation for
emmeansobjects - Added support for
glmgeemodels fromglmtoolbox - Added support for Gaussian models from
glmmTMB - Added pdf vignette (
vignette("RESI_paper")) - Fixed bug with
geeglmobjects not assigning weights properly with missing data - Documentation fix for Linux systems
RESI 1.3.0
CRAN release: 2025-03-24
- Added citation for Journal of Statistical Software paper doi:10.18637/jss.v112.i03
- Bug fix in cluster bootstrapping: Now correctly assigns the clustered IDs in the bootstrap
- Minor documentation updates
RESI 1.2.0
CRAN release: 2023-06-09
- Implemented
bootpackage for bootstrapping-
boot.resultselement ofresiobject contains fullbootobject - Allows parallelization
-
- Updated
geeglm/geemethods - Revamped
plot.resifunction - Added
ggplotmethods - Added
omnibusfunction to extract overall Wald test fromresiobject - Combined
t2Sandt2S_altintot2Sandz2Sandz2S_altintoz2S - Substantially reduced code duplication between methods
- Expanded error checking and messaging
- Fixed error in
resiwhen using a model with only one predictor - Various typo fixes and cleaned up documentation
RESI 1.1.0
CRAN release: 2023-02-10
- Expanded longitudinal methods
-
geeandgeeglmreporting valid confidence intervals -
geeandgeeglmreport both a longitudinal RESI estimate and a cross-sectional RESI estimate -
lmeandlmerModreports point estimates for longitudinal RESI only (confidence intervals in development) -
anovaoption added forgeeglm,lme, andlmerMod - Updated bootstrap sampling to ensure the correct number of clusters
-
- Error in
f2Sandt2S_altformula corrected (denominator is now n instead of the residual degrees of freedom) - Added bootstrap fail counter on
resifornlsandgeeglmmodel types - Changed class of
summaryon aresiobject tosummary_resifor consistency - Small typos and inconsistencies fixed
