Computes the Variance-Covariance Matrix of Multidimensional Parameters Using M-Estimation

Provides a flexible framework for estimating the variance-covariance matrix of estimated parameters. Estimation relies on unbiased estimating functions to compute the empirical sandwich variance. (i.e., M-estimation in the vein of Tsiatis et al. (2019) .

Mestim

This package provides a flexible framework for estimating the variance-covariance matrix of estimated parameters. Estimation relies on providing unbiased estimating functions to compute the empirical sandwich variance. _{^{what is this?}} :monkey:

Installation

For latest release

devtools::install_github("fcgrolleau/Mestim")

For stable release

install.packages("Mestim")

Implementation

library(Mestim)

# Put estimated parameters in a list
thetas_hat <- list(theta_1=coef(mod)[1], theta_2=coef(mod)[2])

# Build a list of unbiased estimating functions
# NB: parameters' names must be consistent with the previous list
M_1 <- expression( ((1/(1+exp( -( theta_1 * x_1 + theta_2 * x_2 ) ))) - y ) * x_1 )
M_2 <- expression( ((1/(1+exp( -( theta_1 * x_1 + theta_2 * x_2 ) ))) - y ) * x_2 )
est_functions <- list(M_1, M_2)

## Pass arguments to get_vcov and obtain what you are looking for
res <- get_vcov(data=dat, thetas=thetas_hat, M=est_functions)

Find more information in the introduction vignette.

Authors

This package is written and maintained by François Grolleau ([email protected]).

References

• Boos DD. and Stefanski, LA. Essential Statistical Inference. 2013. [Springer]
• Tsiatis, AA., Davidian, M., Holloway, ST. and Laber, EB. Dynamic Treatment Regimes: Statistical Methods for Precision Medicine. 2019. [CRC Press]