RESI 1.4.2

RESI 1.4.2

New Features

• Added analytic confidence intervals based on quadratic form, Cornish-Fisher, and Normal approximations. qf is now the default confidence interval method for lm and glm.
• Added support for robustbase models (lmrob and glmrob) via new resi_pe.lmrob, resi_pe.glmrob, resi.lmrob, and resi.glmrob methods. Both default to vcovfunc = stats::vcov, which uses the model’s built-in robust sandwich variance. glmrob redirects sandwich::vcovHC to stats::vcov with a warning since vcovHC does not support glmrob (#12).

Bug Fixes

• Fixed error when passing a data argument containing NAs to resi() for GEE models (geeglm). Rows with NA in 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 different alpha level 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 with store.boot = TRUE, and instructs the user to re-run resi() with the desired alpha directly (#53).
• Fixed resi() failing to compute the overall Wald test when the model response is a computed expression (e.g. log10(charges) or I(charges > 10000)). Previously the intercept-only reduced model could not be fitted inside forked parallel workers because update() 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 from model.frame(), avoiding any re-evaluation.
• Added an informative error when vcov.args = list(type = "const") is passed together with vcovfunc = sandwich::vcovHC. The message explains that type = "const" is the OLS sandwich (not robust) and directs users to use vcovfunc = stats::vcov for parametric variance estimation instead (#50).
• resi_pe now reports CS-RESI and L-RESI for lmer.

RESI 1.3.2

• Added RESI estimation for emmeans objects
• Added support for glmgee models from glmtoolbox
• Added support for Gaussian models from glmmTMB
• Added pdf vignette (vignette("RESI_paper"))
• Fixed bug with geeglm objects not assigning weights properly with missing data
• Documentation fix for Linux systems

RESI 1.3.0

• 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 1.2.4

• Minor bug and documentation fixes
• Website deployed

RESI 1.2.0

• Implemented boot package for bootstrapping
  • boot.results element of resi object contains full boot object
  • Allows parallelization
• Updated geeglm/gee methods
• Revamped plot.resi function
• Added ggplot methods
• Added omnibus function to extract overall Wald test from resi object
• Combined t2S and t2S_alt into t2S and z2S and z2S_alt into z2S
• Substantially reduced code duplication between methods
• Expanded error checking and messaging
• Fixed error in resi when using a model with only one predictor
• Various typo fixes and cleaned up documentation

RESI 1.1.1

• Minor argument consistency fix

RESI 1.1.0

• Expanded longitudinal methods
  • gee and geeglm reporting valid confidence intervals
  • gee and geeglm report both a longitudinal RESI estimate and a cross-sectional RESI estimate
  • lme and lmerMod reports point estimates for longitudinal RESI only (confidence intervals in development)
  • anova option added for geeglm, lme, and lmerMod
  • Updated bootstrap sampling to ensure the correct number of clusters
• Error in f2S and t2S_alt formula corrected (denominator is now n instead of the residual degrees of freedom)
• Added bootstrap fail counter on resi for nls and geeglm model types
• Changed class of summary on a resi object to summary_resi for consistency
• Small typos and inconsistencies fixed

RESI 1.0.5

• First version