| mmlcrcomponentinit {mmlcr} | R Documentation |
Used internally by mmlcr to initialize an mmlcr component. The function is generic.
mmlcrcomponentinit(object, n.groups, prob, data, grouping)
object |
see details below |
n.groups |
the number of classes. Provided by mmlcr.
|
prob |
a data.frame of (initial) posterior probabilities of class
membership. Provided by mmlcr.
|
data |
a data.frame. Provided by mmlcr.
|
grouping |
the variable of data indicating subject id. Provided by mmlcr.
|
The object argument is a list, originally provided to the mmlcr function as a
component of a list in the components argument. At a minimum, that list consists
of a formula of the form resp ~ covars, where resp is the response variable for
that component and covars are any additional covariates (resp ~ 1 is commonly
used), and a class. For longitudinal responses, the covariates should include
the chronological variable, e.g., resp ~ poly(age, 2). Implemented classes include
cnormlong, cnormonce, multinomlong, multinomonce, nblong, nbonce,
normlong, normonce, poislong, and poisonce. The *long classes are for longitudinal
responses, while the *once classes are for one-time responses. So that a rectangular
data structure can be used for the data argument in mmlcr, the *once classes use
just the first occurence of a response for each subject.
The cnorm* classes are for censored normal distributions. They take a min and max
argument for the minimum and maximum possible observed value. The multinom* classes
are for multinomial, or categorical responses. The response variable should be
a factor. The nb* classes are for
negative binomial models (the usual overdispersed Poisson model). The norm* classes
are for normal models, and the pois* classes are for Poisson models. For the cnorm*
and norm* classes the minimum and maximum variances are constrained to be within a
factor of 100 of each other. For the pois* classes an overdispersion estimate is
calculated but not used.
an mmlcr component object.