Changelog • RobinCar2

RobinCar2 0.2.4.9000

Under development

CRAN release: 2026-08-22

The new surv_control argument in robin_surv() allows to fine-control the root finding algorithm used for the hazard ratio estimation.

Previously robin_surv() gave small numerical differences to survival::coxph() for the hazard ratio estimate. This is now fixed.
Fixed robin_surv() silently combining distinct joint strata when their pasted factor labels were identical.

Adapted a unit test to a change in stats::model.frame.glm() in R-devel, which now enforces the factor levels recorded in the model fit, consistent with stats::model.frame.lm(). As a consequence, predict_counterfactual() requires the factor levels of the treatment variable in data to match those of fit. Previously, mismatched levels were silently accepted for glm fits; this restriction has always applied to lm fits.

CRAN release: 2026-07-02

Fixed a bug in predict_counterfactual that the ordering of the treatment levels affect the order of the marginal mean.

CRAN release: 2026-01-09

The user interface for robin_surv has been improved for specifying analysis stratification variables directly as part of the model formula, similar to how it is done in survival::coxph. In addition, the randomization scheme is now specified as for the other RobinCar2 functions for consistency. A warning will be issued when the randomization strata are not adequately included in the analysis model. See the updated vignette for details.
It is now possible to only perform the log rank test, without estimating the log hazard ratio, in robin_surv by setting the argument contrast = "none". This can be useful e.g. when performing simulation studies focusing only on the log-rank test operating characteristics, because the log-rank test is computationally less expensive than estimating the hazard ratio.

Fixed a bug in covariate-adjusted stratified survival function estimation in robin_surv which could occur when there are character covariates with values only appearing in one stratum, which could have failed or lead to incorrect results.
Fixed a issue in robin_lm that variance method does not apply correctly.
Fixed a issue in robin_glm that vcovHC could previously be used for non-Gaussian family.
Fixed another bug in covariate-adjusted stratified survival function estimation in robin_surv, which resulted from design matrices separately derived per stratum. Now the design matrix is created once including the stratum indicator, and then the stratum-specific parts are extracted as needed.

Changed from sp to sr which is easier to read as “simple randomization”.
The print output for robin_surv objects has been improved for better readability.

CRAN release: 2025-09-11

Now multiple stratification variables are supported in robin_surv by adding them on the right-hand side of the treatment formula.

There will be no more spurious warnings in robin_surv during the hazard ratio estimation coming from the variance calculation: Now the variance is only calculated after the hazard ratio estimate is obtained.
When adjusting for a single factor covariate, or when the covariate is strongly correlated with a strata variable in robin_surv, the function now works correctly and does not fail with a “singular design matrix” error any longer.
Fixed a bug in covariate-adjusted survival function estimation in robin_surv which could occur when there are character covariates with values only appearing in one treatment group, which could have led to incorrect results.

CRAN release: 2025-09-09

Add survival analysis with log-rank test and hazard ratio estimation.
Add log_odds_ratio and log_risk_ratio function for inference.
Add contrast_mat to treatment_effect object.
Add confint to obtain confidence interval for prediction_cf and treatment_effect object.

Previously in robin_glm, vcovG is always used regardless the variance type argument. This is fixed now.

CRAN release: 2025-04-03