Multiple comparison for binomial parameters using bootstrap / ComparaÃÃes mÃltiplas para parÃmetros binomiais utilizando bootstrap

AUTOR(ES)
DATA DE PUBLICAÇÃO

2006

RESUMO

The multiple comparisons methods and the analysis of variance are not reliable alternatives for comparing two or more binomial proportions, when the experiments have only Bernoulli trails. Although, this comparison can be made using the intensive computational techniques named infinite bootstrap. This work aimed to evaluate the performance of two binomial proportions bootstrap tests computing the experimentwise type I error rates and the power. These two infinite bootstrap tests distinguished on the estimators of pi. One of these tests considered the maximum likelihood estimator (ML) and the other the Pan s estimator (Pan, 2002) and they were evaluated in different configurations considering the number of populations and the parameters values, resultant of 2000 Monte Carlo simulations. In the first stage the experimentwise type I error rates were evaluated under complete null and partial H0 hypotheses. The simulations under complete H0 were done in all combinations between parameters values p= 0.1; 0.5 and 0.9, number of populations k=2, 5 and 10 and sample sizes n=10, 30 and 100. The experimentwise type I error rate was also evaluated under partial H0 considering a difference of between values of p of distinct groups. In a second stage the powers of the tests were evaluated under partial H0 and alternative hypotheses. Both simulations were done using 1% and 5% significance level. PanÂs and ML bootstrap tests showed excellent performance, because experimentwise error rate were always under their nominal levels. Powers of both procedures were high and they know best performance with extreme proportions and small sample sizes n<10, when PanÂs bootstrap test is preferable.

ASSUNTO(S)

bootstrap binomial proportion estatistica mÃtodo monte carlo bootstrap monte carlo method proporÃÃo binomial

ACESSO AO ARTIGO

Documentos Relacionados