Proposição de estimadores para funções lineares binomiais com amostras inflacionadas de zero

Abstract

With the purpose of obtaining statistical inferences in different research areas, this paper aims to propose an improvement by incorporating robust estimators, which substitute the maximum likelihood estimator of binomial samples with excess zeros on the Wald and Wilson methods, usually used to estimate binomial linear functions. In order to validate the methods, we used the Monte Carlo simulation combining different scenarios involving parametric values, the number of Bernoulli trials, sample size and different percentages of null values in the sample. We recommend the use of the Wald method in situations in which the estimates of the robust binomial proportions present excess of zeros, which maximize the variance of the binomial linear functions using the ρ1sistematic component and orthogonal coefficients. Regarding the Wilson method, we do not recommend its use due to presenting the possibility of incoherent coverage at the nominal level of confidence.