ALTERNATIVES TO THE CLASSICAL FREQUENTIST CONFIDENCE INTERVAL FOR DESCRIBING ZERO-INFLATED LEAF DISEASE SEVERITY

Main Article Content

Jhessica Letícia Kirch
http://orcid.org/0000-0002-9143-8037
Brena Geliane Ferneda
http://orcid.org/0000-0001-8928-211X
Fernando Henrique Silva Garcia
http://orcid.org/0000-0002-2947-9604
Sonia Maria de Stefano Piedade
http://orcid.org/0000-0003-1336-1246
Idemauro Antonio Rodrigues de Lara
http://orcid.org/0000-0002-1172-9855

Abstract

This paper presents the bootstrap percentile interval and the Bayesian credible interval as alternatives to the classical frequentist confidence interval for analysis of zero-inflated data. The indicated methods were applied to soybean downy mildew severity data obtained by stratified sampling in two municipalities in the state of São Paulo: Estiva Gerbi and Piracicaba. The amplitudes of the frequentist and bootstrap percentile confidence intervals were similar. For the Bayesian approach, the credible intervals of the posterior predictive distribution were considered using the zero-inflated beta distribution as likelihood. The credible intervals showed a wider range and included values in the upper bounds of the intervals greater than those observed in the data. We conclude that Bayesian inference is more complex, but allows incorporation of prior information regarding regional and seasonal aspects, contributing to better disease management in the field. When this information is not known, nonparametric bootstrap resampling is a simple alternative to construct intervals for zero-inflated data without assuming the distribution function.

Article Details

How to Cite
Kirch, J. L., Ferneda, B. G., Garcia, F. H. S., Piedade, S. M. de S., & Lara, I. A. R. de. (2022). ALTERNATIVES TO THE CLASSICAL FREQUENTIST CONFIDENCE INTERVAL FOR DESCRIBING ZERO-INFLATED LEAF DISEASE SEVERITY. Brazilian Journal of Biometrics, 40(2). https://doi.org/10.28951/bjb.v40i2.563
Section
Articles

Most read articles by the same author(s)