CUB (version 0.1)

CUBE: Main function for CUBE models

Description

Main function to estimate and validate a CUBE model for given ratings, explaining uncertainty, feeling and overdispersion.

Usage

CUBE(ordinal, m=get('m',envir=.GlobalEnv), Y = 0, W = 0, Z = 0, starting, maxiter,
toler, makeplot = TRUE, expinform=FALSE, summary=TRUE)

Arguments

ordinal

Vector of ordinal responses

m

Number of ordinal categories (if omitted, it will be assigned to the number of categories specified in the global environment)

Y

Matrix of covariates for explaining the uncertainty component. If omitted (default), no covariate is included in the model

W

Matrix of covariates for explaining the feeling component. If omitted (default), no covariate is included in the model

Z

Matrix of covariates for explaining the overdispersion component. If omitted (default), no covariate is included in the model

starting

Initial parameters estimates to start the optimization algorithm. If missing, the function calls specific routines computing the best initial estimates available

maxiter

Maximum number of iterations allowed for running the optimization algorithm (default: maxiter=1000)

toler

Fixed error tolerance for final estimates (default: toler = 1e-6, except for the model including covariates for all the three parameters, in which case toler=1e-2)

makeplot

Logical: if TRUE (default) and no covariate is included in the model, the function returns a graphical plot comparing fitted probabilities and observed relative frequencies

expinform

Logical: if TRUE and no covariate is included in the model, the function returns the expected variance-covariance matrix (default is FALSE)

summary

Logical: if TRUE (default), summary results of the fitting procedure are displayed on screen

Value

An object of the class "CUBE" is a list containing the following results:

estimates

Maximum likelihood estimates: $(\pi, \xi, \phi)$

loglik

Log-likelihood function at the final estimates

varmat

Variance-covariance matrix of final estimates

niter

Number of executed iterations

BIC

BIC index for the estimated model

Details

It is the main function for CUBE models, calling for the corresponding functions whenever covariates are specified: it is possible to select covariates for explaining all the three parameters or only the feeling component. The program also checks if the estimated variance-covariance matrix is positive definite: if not, it prints a warning message and returns a matrix and related results with NA entries. The optimization procedure is run via "optim". If covariates are included only for feeling, the variance-covariance matrix is computed as the inverse of the returned numerically differentiated Hessian matrix (option: hessian=TRUE as argument for "optim"), and the estimation procedure is not iterative, so a NULL result for $niter is produced. If the estimated variance-covariance matrix is not positive definite, the function returns a warning message and produces a matrix with NA entries.

References

Iannario M. (2014). Modelling Uncertainty and Overdispersion in Ordinal Data, Communications in Statistics - Theory and Methods, 43, 771--786 Piccolo D. (2015). Inferential issues for CUBE models with covariates, Communications in Statistics. Theory and Methods, 44(23), 771--786. Iannario M. (2015). Detecting latent components in ordinal data with overdispersion by means of a mixture distribution, Quality & Quantity, 49, 977--987 Iannario M. (2016). Testing the overdispersion parameter in CUBE models. Communications in Statistics: Simulation and Computation, 45(5), 1621--1635.

See Also

probcube, loglikCUBE, loglikcuben, inibestcube, inibestcubecsi, inibestcubecov, varmatCUBE

Examples

data(relgoods)
m<-10
ordinal<-na.omit(relgoods[,37])  
model<-CUBE(ordinal,starting=c(0.1,0.1,0.1),summary=TRUE)  
model$estimates        # Final ML estimates
model$loglik           # Maximum value of the log-likelihood function
model$varmat         
model$niter
model$BIC
######################## 
ordinal<-relgoods[,40]
cov<-relgoods[,2]
nona<-na.omit(cbind(ordinal,cov))
modelcovcsi<-CUBE(nona[,1],W=nona[,2],summary=TRUE)
modelcov<-CUBE(nona[,1],Y=nona[,2],W=nona[,2], Z=nona[,2])
modelcov$BIC
modelcovcsi$BIC
#######################################
data(univer)
m<-7
ordinal<-univer[,8]
starting<-inibestcube(m,ordinal)
model<-CUBE(ordinal,starting=starting,summary=TRUE)