Resumo

MANGEAUD, Arnaldo  e  VIDELA, Martín. Looking for the lost independence: using Mixed Generalized Linear Models in choice tests. Ecol. austral [online]. 2005, vol.15, n.2, pp.199-206. ISSN 1667-782X.

The preference of organisms for resources is usually evaluated through multiple-choice tests, without accounting the lack of independence present in the data. This study presents several statistical models which explicitly consider such dependence structure comparing their performance by using real non-simulated data. Four types of models were used: Analysis of Deviance (Generalized Mixed Linear Model), Analysis of Variance with a factor, One-way ANOVA with random block effect (both General Mixed Linear Models) and Non-Parametric ANOVA with block effect (Friedman Test). A covariable and an offset variable were also added to the Mixed GLM model. Results suggest that the most powerful model for the counting-type variable (with Poisson distribution), is the Mixed GLM; whereas for the relativized variable (count/surface), is the Mixed GLM with an offset variable.

Palavras-chave : Multiple choice test; Feeding preference; Statistical models; Deviance.

        · resumo em Espanhol     · texto em Espanhol     · Espanhol ( pdf )