simulate.mclogit {mclogit}

Simulating responses from baseline-category and conditional logit models

Description

The simulate() methods allow to simulate responses from models fitted with mclogit() and mblogit(). Currently only models without random effects are supported for this.

Usage

## S3 method for class 'mblogit'
simulate(object, nsim = 1, seed = NULL, ...)
## S3 method for class 'mclogit'
simulate(object, nsim = 1, seed = NULL, ...)

# These methods are currently just 'stubs', causing an error
# message stating that simulation from models with random
# effects are not supported yet
## S3 method for class 'mmblogit'
simulate(object, nsim = 1, seed = NULL, ...)
## S3 method for class 'mmclogit'
simulate(object, nsim = 1, seed = NULL, ...)

Arguments

object

an object from the relevant class

nsim

a number, specifying the number of simulated responses for each observation.

seed

an object specifying if and how the random number generator should be initialized ('seeded'). The interpetation of this argument follows the default method, see link[stats]{simulate}

...

other arguments, ignored.

Value

The result of the simulate method for objects created by mclogit is a data frame with one variable for each requested simulation run (their number is given by the nsim= argument). The contents of the columns are counts (or zero-one values), with group-wise multinomial distribution (within choice sets) just like it is assumed for the original response.

The shape of the result of the simulate method for objects created by mblogit is also a data frame. The variables within the data frame have a mode or shape that corresponds to the response to which the model was fitted. If the response is a matrix of counts, then the variables in the data frame are also matrices of counts. If the response is a factor and mblogit was called with an argument from.table=FALSE, the variables in the data frame are factors with the same factor levels as the response to which the model was fitted. If instead the function was called with from.table=TRUE, the variables in the data frame are counts, which represent frequency weights that would result from applying as.data.frame to a contingency table of simulated frequency counts.

Examples

library(MASS)
(house.mblogit <- mblogit(Sat ~ Infl + Type + Cont, 
                          data = housing,
                          weights=Freq,
                          from.table=TRUE))
sm <- simulate(house.mblogit,nsim=7)

housing.long <- housing[rep(seq.int(nrow(housing)),housing$Freq),]
(housel.mblogit <- mblogit(Sat ~ Infl + Type + Cont,
                           data=housing.long))
sml <- simulate(housel.mblogit,nsim=7)

housing.table <- xtabs(Freq~.,data=housing)
housing.mat <- memisc::to.data.frame(housing.table)
head(housing.mat)

(housem.mblogit <- mblogit(cbind(Low,Medium,High) ~
                               Infl + Type + Cont,
                           data=housing.mat))
smm <- simulate(housem.mblogit,nsim=7)

str(sm)
str(sml)
str(smm)

head(smm[[1]])