Binder IPCW method was not being calculated correctly. It could have
impacted the intercept (and hence predictions), but coefficient
estimates have always been consistent. Now predictions and the intercept
should also work correctly. (Thanks @alexpate30)
eventglm 1.4.4
Bug fixes:
Handle cases where last observation is an event leading to NaN
(thanks @trinhdhk)
eventglm 1.4.2
Bug fixes:
Fixed case-cohort sampling example thanks @tagteam
eventglm 1.4.1
Bug fixes:
Handle missing data correctly with gee.fit/multiple time points
eventglm 1.4.0
New features:
New pseudo module for infinitesimal jackknife that allows for
delayed entry/left truncation.
eventglm 1.3.0
New features:
Allow id argument in cumincglm and rmeanglm. When id is present,
geese.fit will be used instead of glm.fit, with clusters indicated by
the id. This is to account for clustered data, for example.
Informative error message if there is missing data in the censoring
model
eventglm 1.2.2
Add references to JSS publication
eventglm 1.2.1
Remove tidyverse lifecycle badge
eventglm 1.2.0
New features:
Time vector allowed in cumincglm. Use this to model multiple
timepoints simultaneously, and allow for time varying covariate effects
using tve() (for time varying effect) in the right side of the formula.
Inverse probability of censoring weights are returned by the
fitting functions in the list element called “ipcw.weights”. * Small
edits and improvements to the documentation.
eventglm 1.1.1
Bug fixes:
Check and fix name clashes for reserved variables pseudo.vals, .Tci,
and .Ci
Minor efficiency updates in corrected covariance estimation
Improve documentation
eventglm 1.1.0
New features and vignette:
Methods for computing pseudo observations are now in modules.
Users can define their own modules for computing pseudo
observations
Option to use survival instead of cumulative incidence in standard
models
eventglm 1.0.2
Update tests/jackknife-agree.R to do tolerance based comparison
rather than floating point comparison