Estadísticos para la detección de observaciones anómalas en modelos de elección binaria:una aplicación con datos reales

Resumo

This paper considers the problem of outliers in binary response models. Based on the statistic proposed by Gracia-Díez y Serrano (1994) which measures the influence of each observation on the estimated parameler vector, we derive other statistics in order to measure the influence of each observation as well as the influence of a group of observations on i) the vector of estimated probabilities and ii) subsets and linear combinations of tbe parameters in the model. Also, the method proposed by Peña y Yohai (1991) to deal with the masking problem in linear models has been generalised to the case of binary choice models. Lastly, we propose a diagnostic strategy to detect outliers in this type of models. The application of this strategy is ilustrated by estimating the probit model used by Dhillon et. al (1987).