cubm package in R to fit CUB models

Abstract

The class of CUB models is commonly used by practitioners to model ordinal data, in this paper we propose the cubm package which provides the class of CUB models in the R system for statistical computing. The cubm package allows to specify a formula for each parameter of the model, the Maximum Likelihood (ML) estimation is performed by optimization via the functions nlminb, optim and DEoptim and the variance-covariance matrix can be obtained by numerical approximation of the Hessian matrix or by bootstrap method. The utility of the package is illustrated by an application and a simulation study.

Fitting mixture models for feeling and uncertainty for rating data analysis

“…The choice of the logit link is mainly motivated by the ease of interpretation of the covariates' effect. In fact, any admissible link may be used; for instance, the probit link is considered as an option in Hernández Barajas, Usuga Manco, and García Muñoz (2018), where the estimation steps are carried out without resorting to expectation-maximization procedures.…”