Abstract The overlap coefficient (OVL) is widely used to quantify the similarity between two probability distributions, with applications in agriculture, quality control, and related fields. This study evaluates four OVL estimators: two nonparametric methods, the histogram-based estimator and the kernel density estimator (KDE), and two parametric estimators under the binormal model, with and without Box–Cox transformation, in terms of bias and root mean squared error (RMSE). A simulation study examined estimator behavior under normal, approximately normalizable, and strongly non-normal distributions. Results indicate that the KDE provides more stable and accurate nonparametric estimates than the histogram method, while parametric estimators perform best when normality assumptions hold. The Box–Cox transformation provides a robust approach to transform to normality allowing for the use of the binormal model even when normality assumptions fail. The methods were also applied to a dataset on apple quality attributes to illustrate their performance in a real agricultural quality-control context. The empirical findings support the simulation results: KDE yields stable nonparametric estimates, and parametric estimators produce larger OVL values when normality is satisfied. These results offer practical guidance for selecting appropriate OVL estimators in agricultural and quality-control applications.
Alexandri et al. (Mon,) studied this question.