Relaxation of the parameter independence assumption in the `bootComb` R package

TL;DR

The paper introduces an extension to the bootComb R package, enabling it to handle dependent parameters in confidence interval calculations by incorporating Gaussian copulas and user-specified correlation matrices.

Contribution

The authors extend bootComb to account for dependence among parameters, moving beyond the previous independence assumption, and provide a way to perform sensitivity analyses.

Findings

Enables handling of dependent parameters in bootComb

Uses Gaussian copulas to model dependence

Allows sensitivity analysis for dependence impact

Abstract

Background. The bootComb R package allows researchers to derive confidence intervals with correct target coverage for arbitrary combinations of arbitrary numbers of independently estimated parameters. Previous versions (< 1.1.0) of bootComb used independent bootstrap sampling and required that the parameters themselves are independent - an unrealistic assumption in some real-world applications. Findings. Using Gaussian copulas to define the dependence between parameters, the bootComb package has been extended to allow for dependent parameters. Implications. The updated bootComb package can now handle cases of dependent parameters, with users specifying a correlation matrix defining the dependence structure. While in practice it may be difficult to know the exact dependence structure between parameters, `bootComb` allows running sensitivity analyses to assess the impact of parameter dependence on the resulting confidence interval for the combined parameter. Availability. bootComb is available from the Comprehensive R Archive Network (https://CRAN.R-project.org/package=bootComb).