fpc (version 2.2-13)

lcmixed: flexmix method for mixed Gaussian/multinomial mixtures

Description

lcmixed is a method for the flexmix-function in package flexmix. It provides the necessary information to run an EM-algorithm for maximum likelihood estimation for a latent class mixture (clustering) model where some variables are continuous and modelled within the mixture components by Gaussian distributions and some variables are categorical and modelled within components by independent multinomial distributions. lcmixed can be called within flexmix. The function flexmixedruns is a wrapper function that can be run to apply lcmixed.

Note that at least one categorical variable is needed, but it is possible to use data without continuous variable.

There are further format restrictions to the data (see below in the documentation of continuous and discrete), which can be ignored when running lcmixed through flexmixedruns.

Usage

lcmixed( formula = .~. , continuous, discrete, ppdim,
                     diagonal = TRUE, pred.ordinal=FALSE, printlik=FALSE )

Value

An object of class FLXMC (not documented; only used internally by flexmix).

Arguments

formula

a formula to specify response and explanatory variables. For lcmixed this always has the form x~1, where x is a matrix or data frome of all variables to be involved, because regression and explanatory variables are not implemented.

continuous

number of continuous variables. Note that the continuous variables always need to be the first variables in the matrix or data frame.

discrete

number of categorical variables. Always the last variables in the matrix or data frame. Note that categorical variables always must be coded as integers 1,2,3, etc. without interruption.

ppdim

vector of integers specifying the number of (in the data) existing categories for each categorical variable.

diagonal

logical. If TRUE, Gaussian models are fitted restricted to diagonal covariance matrices. Otherwise, covariance matrices are unrestricted. TRUE is consistent with the "within class independence" assumption for the multinomial variables.

pred.ordinal

logical. If FALSE, the within-component predicted value for categorical variables is the probability mode, otherwise it is the mean of the standard (1,2,3,...) scores, which may be better for ordinal variables.

printlik

logical. If TRUE, the loglikelihood is printed out whenever computed.

Author

Christian Hennig https://www.unibo.it/sitoweb/christian.hennig/en

Details

The data need to be organised case-wise, i.e., if there are categorical variables only, and 15 cases with values c(1,1,2) on the 3 variables, the data matrix needs 15 rows with values 1 1 2.

General documentation on flexmix methods can be found in Chapter 4 of Friedrich Leisch's "FlexMix: A General Framework for Finite Mixture Models and Latent Class Regression in R", https://CRAN.R-project.org/package=flexmix

References

Hennig, C. and Liao, T. (2013) How to find an appropriate clustering for mixed-type variables with application to socio-economic stratification, Journal of the Royal Statistical Society, Series C Applied Statistics, 62, 309-369.

See Also

flexmixedruns, flexmix, flexmix-class, discrete.recode, which recodes a dataset into the format required by lcmixed

Examples

  set.seed(112233)
  options(digits=3)
  require(MASS)
  require(flexmix)
  data(Cars93)
  Cars934 <- Cars93[,c(3,5,8,10)]
  cc <-
  discrete.recode(Cars934,xvarsorted=FALSE,continuous=c(2,3),discrete=c(1,4))
  fcc <- flexmix(cc$data~1,k=2,
  model=lcmixed(continuous=2,discrete=2,ppdim=c(6,3),diagonal=TRUE))
  summary(fcc)