Observaciones anómalas en modelos de elección binaria
Loading...
Download
Official URL
Full text at PDC
Publication date
1992
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Facultad de Ciencias Económicas y Empresariales. Decanato
Citation
Abstract
En este trabajo se trata el problema de la existencia de observaciones anómalas en modelos de elección binaria. Se demuestra que la presencia de estas observaciones afecta a la consistencia de los estimadores de máxima verosimilitud. En cuanto a su detección: se muestra que el análisis de residuos no es un instrumento adecuado, debido a la censura de la variable dependiente y se deriva un estadístico, similar al propuesto por Cook (1977) para modelos lineales, que resulta apropiado para la detección de anomalías en este tipo de modelos. Los resultados teóricos se contrastan con datos simulados.
This paper deals with the problem of outliers in binary response models. It is proved that the existence of these observations in the sample affects the consistency of maximum likelihood estimators. Regarding to detection of outliers: it is shown that residual analysis is not a useful diagnostic tool, due to the censoring of the dependent variable and it is derived a statistic, analogous to the one proposed by Cook (1977) for linear models, which seems to be appropiated for outlier detection in this type of models. The theoretical results are tested using simulated data.
This paper deals with the problem of outliers in binary response models. It is proved that the existence of these observations in the sample affects the consistency of maximum likelihood estimators. Regarding to detection of outliers: it is shown that residual analysis is not a useful diagnostic tool, due to the censoring of the dependent variable and it is derived a statistic, analogous to the one proposed by Cook (1977) for linear models, which seems to be appropiated for outlier detection in this type of models. The theoretical results are tested using simulated data.