A comparative assessment of methods for constructing confidence intervals for the population coefficient of variation

Abstract

The coefficient of variation (CV) is an important measure of relative variability used across many fields. Constructing accurate confidence intervals (CIs) for the population CV remains challenging, particularly for small samples and skewed distributions. This study compares classical, modified, median-based, and bootstrap CI methods for the CV using simulation under symmetric and skewed distributions. Performance is assessed through coverage probability and average interval width, and real-life datasets from health sciences, business, and manufacturing are used for illustration. The results show that classical methods perform well for normal data, median-based methods are more robust under skewness, and bootstrap methods offer a good balance between accuracy and precision across diverse settings.