Multivariate Ordinal Regression Models

A flexible framework for fitting multivariate ordinal regression models with composite likelihood methods. Methodological details are given in Hirk, Hornik, Vana (2020) .


This file documents updates and changes in the package mvord

changes in version 0.3.5 (2019-03-06)

  • added suppressWarnings(RNGversion("3.5.0")) in tests

changes in version 0.3.4 (2019-02-20)

  • two additional arguments have been added to the error structures: value = numeric(0) (for initial values) and fixed = FALSE (for allowing the user to fix the parameters of the error structure to the values specified in argument value)
  • efficiency improvements by using, paste0, seq_len, seq_along where possible

changes in version 0.3.3 (2018-10-30)

  • fixed bug standard errors with threshold.constraints

changes in version 0.3.2 (2018-10-03)

  • function polycor() for polychoric correlations is now available
  • added packages minqa, BB, ucminf and dfoptim as dependencies

changes in version 0.3.1 (2018-06-12)

  • out-of-sample predictions are now available
  • additional examples are now available in an additional vignette

changes in version 0.3.0 (2018-05-03)

  • new model design with multiple measurement objects MMO and MMO2 (former mvord2())
  • removed function mvord2()
  • removed argument scale (covariates are scaled internally for optimizer)
  • fixed bug in standard errors with binary outcomes
  • removed argument response.names (if specific ordering is desired this can be performed by an ordered factor for the multiple measurement index)
  • additional input types are now applicable for ordinal response variables
  • fixed bug trace
  • control = mvord.control()
  • additional checks
  • new function name marginal_predict() (instead of marginal.predict())
  • new function name joint_probabilities() (instead of get.prob()); type "class" instead of "class.max"
  • new function name error_structure() (instead of get_error_struct())

changes in version 0.2.1 (2017-11-29)

  • adapted predict(), get.prob(), marginal.predict()
  • added contrasts as argument to mvord() and mvord2()
  • fixed bug in VGAM design of coef.constraints
  • changed internal design matrices

changes in version 0.2.0 (2017-11-10)

  • implemented nobs.mvord()
  • implemented vcov.mvord()
  • implemented terms.mvord()
  • implemented model.matrix.mvord()
  • implemented fitted.mvord()
  • implemented logLik.mvord()
  • AIC() and BIC() are now available
  • additional argument scale
  • new (additional) design for coef.constraints in analogy to constraints in VGAM
  • category-specific regression parameters
  • additional argument offset
  • new class "mvlink"
  • implemented coef.constraints()
  • logPL(), claic() and clbic() are not exported anymore. use logLik(), AIC() and BIC() instead.
  • renamed cor_general, cor_ar1, cor_equi, cov_general
  • implemented names_constraints()

changes in version 0.1.0 (2017-10-17)

  • changed function name from multord() to mvord()
  • changed function name from multord2() to mvord2()
  • implemented predict function()
  • implemented predict.marginal function()
  • implemented get.prob function()
  • changed link probit to mvprobit()
  • changed link logit to mvlogit(df = 8L); default value of degrees of freedom is 8
  • implemented a more flexible multivariate logistic distribution with logistic marginals and t copula (df are settable)
  • additional argument PL.lag

Reference manual

It appears you don't have a PDF plugin for this browser. You can click here to download the reference manual.


1.1.1 by Rainer Hirk, 7 months ago

Browse source code at

Authors: Rainer Hirk [aut, cre] , Kurt Hornik [aut] , Laura Vana [aut] , Alan Genz [ctb] (Fortran Code)

Documentation:   PDF Manual  

GPL-3 license

Imports MASS, pbivnorm, stats, optimx, mnormt, numDeriv, Matrix

Depends on minqa, BB, ucminf, dfoptim

Suggests knitr, rmarkdown, xtable, colorspace

See at CRAN