Publication: Problemas fundamentales del análisis logarítmico
lineal (I): El colapsamiento de variables y su
influencia en el ajuste de modelos
Authors
Correa Piñero, Ana Delia
item.page.secondaryauthor
item.page.director
Publisher
Universidad de Murcia
publication.page.editor
publication.page.department
DOI
item.page.type
info:eu-repo/semantics/article
Description
Abstract
En este trabajo analizamos algunos de los problemas que pueden surgir en el ajuste de
modelos logarítmicos lineales cuando se toman decisiones de colapsamiento (recombinar categorías,
eliminar categorías o variables, categorizar datos continuos, ... ). Ésta es una práctica
frecuente entre los investigadores cuando la tabla de contingencia presenta un exceso de casillas
con valores esperados muy bajos. La influencia del colapsamiento puede manifestarse de
diversas maneras. En unos casos, la modificación de las categorías es tan drástica que la
variable subyacente, en cierta medida, cambia. En otros, pueden verse afectados los valores de
los parámetros y la calidad de ajuste de los modelos. Finalmente, se presentan una serie de
recursos mediante los cuales minimizar el efecto distorsionador del colapsamiento.
This paper analyze sorne problems that can raise in the application of log-linear models when the researcher make decisions of collapsing (recombining categories, deleting categories or variables, treating continuous data as if they were discrete, ... ). This is a common practice among researchers when they obtain too many cells with small expected values. The effects of collapsing can be observed in different ways. In sorne cases, a drastic modification of categories can lead to changing the underlying variable. In other cases, the value ofparameters and models goodness of fit can be affected. Finally, it is presented a serie of recourses to reduce the distorting effect of collapsing.
This paper analyze sorne problems that can raise in the application of log-linear models when the researcher make decisions of collapsing (recombining categories, deleting categories or variables, treating continuous data as if they were discrete, ... ). This is a common practice among researchers when they obtain too many cells with small expected values. The effects of collapsing can be observed in different ways. In sorne cases, a drastic modification of categories can lead to changing the underlying variable. In other cases, the value ofparameters and models goodness of fit can be affected. Finally, it is presented a serie of recourses to reduce the distorting effect of collapsing.
publication.page.subject
Citation
item.page.embargo
Ir a Estadísticas
Este ítem está sujeto a una licencia Creative Commons. http://creativecommons.org/licenses/by-nc-nd/4.0/