A simulation-based study of Zero-inflated Bernoulli model with various models for the susceptible probability

TL;DR

This study evaluates the robustness of parameter estimation in Zero-Inflated Bernoulli models with various susceptible probability models using simulation and real data, highlighting the probit model's suitability for fishing data.

Contribution

It introduces a maximum likelihood estimation approach for different susceptible probability models within the ZIBer framework and assesses their performance through simulations and real data analysis.

Findings

MLE provides accurate and reliable inferences.

Probit-ZIBer model is most suitable for fishing data.

Results are consistent and practically meaningful.

Abstract

In this work, we are interested in the stability and robustness of the parameter estimation in the Zero-Inflated Bernoulli (ZIBer) model, when the susceptible probability (SP) model is modeled by numerous different binary models: logit, probit, cloglog and generalized extreme value (GEV). To address this problem, we propose the maximum likelihood estimation (MLE) method to check its performance when different SP models are considered. Based on numerical evidences through simulation studies and the analysis of a real data set, it can be seen that the MLE approach has provided accurate and reliable inferences. In addition, it can also be seen that for the empirical analysis, the probit-ZIBer model is probably more suitable for the fishing data set than the other models considered in this study. Besides, the results obtained in the experimental analysis are also very consistent, compatible and very meaningful in practice. It will help us to understand the importance of increasing production while fishing.