A new distribution-free approach to constructing the confidence region for multiple parameters

Area of Science:

• Statistical inference
• Nonparametric statistics
• Computational biology

Background:

• Confidence intervals/regions are crucial for statistical inference.
• Existing methods often assume known distributions or have limitations like biased coverage rates.
• Accurate characterization of sampling distributions for biological parameters is frequently challenging.

Purpose of the Study:

• To develop a novel distribution-free method for constructing confidence regions.
• The method should be based on geometrical principles and account for data distribution shape.
• To implement the method in an R package for practical application.

Main Methods:

• A new distribution-free confidence region construction method was developed.
• The method was implemented in the R package 'distfree.cr/R'.
• Statistical properties were evaluated and compared using Monte Carlo simulations.

Main Results:

• The new method demonstrated superior performance compared to existing parametric and nonparametric approaches.
• Outperformance was consistent across normal and non-normal bivariate distributions.
• Superiority was maintained across varying sample sizes and correlation levels.

Conclusions:

• The developed distribution-free method offers a robust and accurate approach for confidence region construction.
• It effectively addresses limitations of traditional methods, particularly for biological data.
• The R package provides a valuable tool for genomics and other biological research requiring reliable statistical inference.