Abstract

This study investigates the construction of confidence intervals for the common coefficient of variation (CV) of multiple delta-inverse Gaussian (delta-IG) distributions, which are suitable for modeling skewed and zero-inflated data. Existing methods for CV inference are generally limited to single populations or standard distributions and do not adequately address the combined challenges of skewness, excess zeros, and multiple populations. To overcome these limitations, five approaches are considered: generalized confidence intervals (GCI), adjusted GCI (AGCI), parametric bootstrap percentile confidence intervals (PBPCI), Bayesian credible intervals (BCI), and highest posterior density intervals (HPDCI). A Monte Carlo simulation study is conducted to evaluate performance in terms of coverage probability (CP) and average width (AW) across various parameter settings and population sizes. The results reveal a clear trade-off between coverage accuracy and interval efficiency. Among the methods, HPDCI provides coverage probabilities closest to the nominal level, though it still exhibits some undercoverage, whereas BCI offers reasonably stable performance. In contrast, PBPCI produces the narrowest intervals but suffers from severe undercoverage, and GCI and AGCI fail to achieve a satisfactory balance. An application to traffic accident data further supports the simulation findings. Overall, the HPDCI methods are recommended for reliable inference in delta-IG settings.

IPC Classification

Keywords