QIC: Quasilikelihood Information Criterion

In repolr: Repeated Measures Proportional Odds Logistic Regression

Description

The quasilikelihood information criterion (QIC) developed by Pan (2001) is a modification of the Akaike information criterion (AIC) for models fitted by GEE. QIC is used for choosing the best correaltion structure and QICu is used for choosing the best subset of covariates. The quasilikelihood (QLike) is also reported for completeness. When choosing between two or more models, with different subset of covariates, the one with the smallest QICu measure is preferred and similarly, when choosing between competing correlation structures, with the same subset of covariates in both, the model with the smallest QIC measure is preferred.

Usage

QIC(object, digits = 3)

Arguments

object	is a fitted model using repolr.
digits	the number of decimal places to display in reported summaries.

Value

QLike	model quasilikelihood.
QIC	model QIC.
QICu	model QICu.

References

Pan W. Akaikes information criterion in generalized estimating equations. Biometrics 2001; 57:120-125.

Examples

data(HHSpain)
mod.0 <- repolr(HHSpain~Time, data=HHSpain, categories=4, subjects="Patient",
            times=c(1,2,5), corr.mod="independence", alpha=0.5)
QIC(mod.0)
QIC(update(mod.0, formula = HHSpain~Time + Sex))$QICu
QIC(update(mod.0, formula = HHSpain~Time * Sex))$QICu

repolr documentation built on May 1, 2019, 9:20 p.m.