User Guide to RobinCar and RobinCar2
Source:vignettes/articles/robincar-comparison.Rmd
robincar-comparison.Rmd
if (!requireNamespace("RobinCar", quietly = TRUE)) {
install.packages("RobinCar")
}
#> Installing package into '/usr/local/lib/R/site-library'
#> (as 'lib' is unspecified)
#> also installing the dependencies 'gtools', 'caTools', 'gplots', 'listenv', 'gargle', 'ids', 'rematch2', 'ROCR', 'future', 'truncnorm', 'conflicted', 'dtplyr', 'googledrive', 'googlesheets4', 'reprex', 'nnls', 'gam', 'cvAUC', 'future.apply', 'progressr', 'Rsolnp', 'emulator', 'tidyverse', 'fastDummies', 'SuperLearner', 'AIPW'
library(RobinCar2)
library(RobinCar)RobinCar was developed as a comprehensive library of existing and future methods on covariate adjustment. RobinCar2 aims to become a standard R package for industry and regulatory agencies to implement covariate adjustment using selected methods with clear documentation and validation for GxP compliance.
See the American Statistical Association (ASA) BIOP Covariate Adjustment Scientific Working Group (SWG) Software Subteam page here for more information.
As such, RobinCar and RobinCar2 share many methods. Please use this vignette as a guide to the similarities and differences between RobinCar and RobinCar2.
RobinCar2 and RobinCar produce the same results when run with the same settings on the same data
RobinCar2 has a testing suite that ensures that the outputs are
identical between RobinCar and RobinCar2. The comparisons use the
dummy_data dataset included in RobinCar2. The outputs for
the point estimates and variance estimates are compared under a variety
of settings, including:
- Marginal means and treatment contrasts, for the difference, relative risk, odds ratio, and a custom contrast.
- Under Gaussian, binomial, Poisson, and negative binomial (known and unknown dispersion parameters) working models.
- Mean model specifications of ANOVA, ANCOVA, and ANHECOVA with up to three covariates.
- Under simple, Pocock-Simon, and permuted block randomization schemes.
Please see the following testing file for details: https://github.com/openpharma/RobinCar2/blob/main/tests/testthat/test-validate-robin_glm.R or view the report at RobinCar2 Validation Report.
Differences between RobinCar and RobinCar2 argument usage
The interface for users is very similar between RobinCar and RobinCar2, though RobinCar2 has some more streamlined use. Users of RobinCar should find RobinCar2 at least as user-friendly and not have trouble switching between packages.
Here we provide an example of using a logistic regression working
model with stratified permuted block randomization, and
the dummy_data provided in the package. We output the
treatment contrast.
head(dummy_data)RobinCar
Here is the call in RobinCar, and the output:
robincar_glm(
df = dummy_data,
treat_col = "treatment",
response_col = "y_b",
formula = y_b ~ treatment * s1 + covar,
car_strata_cols = "s1",
car_scheme = "permuted-block",
g_family = binomial(link = "logit"),
contrast_h = "diff"
)
#> $main
#> Treatment group mean estimates from a GLM working model of family binomial and link logit using formula:
#> response ~ treat * s1 + covar
#>
#>
#> Used HC0-type of heteroskedasticity-consistent variance estimates
#> and adjusted variance-covariance matrix for randomization car_strata s1
#> consistent with the permuted-block design.
#>
#> Estimates:
#> # A tibble: 3 × 4
#> treat estimate se `pval (2-sided)`
#> <chr> <dbl> <dbl> <dbl>
#> 1 pbo 0.356 0.0336 3.03e-26
#> 2 trt1 0.581 0.0344 7.25e-64
#> 3 trt2 0.621 0.0340 1.54e-74
#>
#> Variance-Covariance Matrix:
#> [,1] [,2] [,3]
#> pbo 1.128902e-03 1.856234e-05 1.333885e-05
#> trt1 1.856234e-05 1.184599e-03 2.178112e-05
#> trt2 1.333885e-05 2.178112e-05 1.157268e-03
#>
#> $contrast
#> Treatment group contrasts using linear contrast
#>
#> Contrasts:
#> # A tibble: 2 × 4
#> contrast estimate se `pval (2-sided)`
#> <chr> <dbl> <dbl> <dbl>
#> 1 treat trt1 - pbo 0.225 0.0477 0.00000251
#> 2 treat trt2 - pbo 0.265 0.0475 0.0000000239
#>
#> Variance-Covariance Matrix for Contrasts:
#> [,1] [,2]
#> [1,] 0.002276376 0.001118782
#> [2,] 0.001118782 0.002259492RobinCar2
In RobinCar2, all of the randomization information, including the
name of the treatment variable, the stratification variable, and the
randomization scheme info is included in the treatment
argument. This contrasts with RobinCar, where the information is
required in treat_col, car_strata_cols, and
car_scheme.
robin_glm(
data = dummy_data,
formula = y_b ~ treatment * s1 + covar,
treatment = treatment ~ pb(s1),
family = binomial(link = "logit")
)
#> Model : y_b ~ treatment * s1 + covar
#> Randomization: treatment ~ pb(s1) ( Permuted-Block )
#> Variance Type: vcovG
#> Marginal Mean:
#> Estimate Std.Err 2.5 % 97.5 %
#> pbo 0.356097 0.033599 0.290243 0.4219
#> trt1 0.580696 0.034418 0.513238 0.6482
#> trt2 0.621386 0.034019 0.554711 0.6881
#>
#> Contrast : h_diff
#> Estimate Std.Err Z Value Pr(>|z|)
#> trt1 v.s. pbo 0.224599 0.047711 4.7075 2.508e-06 ***
#> trt2 v.s. pbo 0.265290 0.047534 5.5810 2.391e-08 ***
#> trt2 v.s. trt1 0.040691 0.047941 0.8488 0.396
#> ---
#> Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1The structure of the output of RobinCar2 more closely mirrors the
structure of the output that one would obtain using
glm().
Feature availability in RobinCar v. RobinCar2
Here we list any differences in the features that are available in RobinCar and RobinCar2. This will be continually updated as the packages are updated.
Last Updated: February 21, 2025
Features available in RobinCar2 that are not available in RobinCar
Variance function
RobinCar only has one option for the variance, which RobinCar2 calls
vcovG. vcovG is the influence function-based
variance estimator. RobinCar2 has the option of specifying a
vcov other than vcovG.
Currently, the only other vcov that is allowed in
RobinCar2 is the sandwich variance, called vcovHC and it is
only valid for ANCOVA with a linear contrast. Please see the table in
the following blogpost from the ASA-BIOP Covariate Adjustment SWG here
for more information about the methodology behind the sandwich variance
versus influence-based variance.
Features available in RobinCar that are not available in RobinCar2
Time-to-event analysis
Currently, RobinCar has functions for time-to-event data analysis and RobinCar2 does not. This is planned for future releases of RobinCar2. Please see here for usage of RobinCar time-to-event methods.