Beta regression is the standard method to explore how a response assuming values in (0;1) depends on a set of covariates. With respect to standard regression, in this case, the parametric model requires two systems of equations: one for the mean and the other for the precision parameter that can be based on the same set of covariates. Therefore for two different sets of covariates and the same value of the linear predictor for the mean we could have different precisions. Nevertheless a linear model for the precision parameter could not be good enough to capture all the heterogeneity in the data. We will extend the characteristic approach of cluster weighted linear models to the beta regression problem in order to obtain a flexible model both in analyzing relations between means and covariates and in evaluating prediction precision.
CLUSTER WEIGHTED BETA REGRESSION
Nieddu L
2014-01-01
Abstract
Beta regression is the standard method to explore how a response assuming values in (0;1) depends on a set of covariates. With respect to standard regression, in this case, the parametric model requires two systems of equations: one for the mean and the other for the precision parameter that can be based on the same set of covariates. Therefore for two different sets of covariates and the same value of the linear predictor for the mean we could have different precisions. Nevertheless a linear model for the precision parameter could not be good enough to capture all the heterogeneity in the data. We will extend the characteristic approach of cluster weighted linear models to the beta regression problem in order to obtain a flexible model both in analyzing relations between means and covariates and in evaluating prediction precision.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
