logpost.Loglin(object, theta = <<see below>>, prior = <<see below>>)
"missmodel"
whose
paramIter
component is of
class
"Loglin"
. I.e., an object resulting from the use of the
mdLoglin
,
daLoglin
, or
emLoglin
functions. Alternatively, a class
"preLoglin"
object.
paramIter
component
of a
missmodel
object. If a
class
"preLoglin"
object is input, the cell probabilities must be
specified.
"Loglin"
object with multiple rows; then calculations
are performed for each row, and a vector is returned.
"priorLoglin"
, or an array of
hyperparameters.
"ml"
(maximum likelihood) or
"noninformative"
.
String matching is used,
so the characters
"m"
or
"n"
are sufficient. The values
of the hyperparameters changes with the algorithm (see
for details). E.g.
"noninformative"
means a common value of 1 for
EM, and a common value of 0.5 for DA.
"priorLoglin"
object is created by routine
priorLoglin
.
start
for the order to use in specifying a vector of
hyperparameters. If a single numeric value is input, its value is
replicated for all cells in the table.
The hyperparameters for a data dependent prior (following an
independence model) can be generated using routine
dataDepPrior
.
See
for details.
"noninformative"
. When a class
"missmodel"
object is input, any value specified in a previous call has priority
over the default value (but not over any currently specified value).
NA
) when a vector of
hyperparameters is input as argument
prior
.
theta
.
This is the log-likelihood or log-posterior density that ignores the missing-data mechanism.
fit <- emLoglin(object = crime, margins = ~Visit.1+Visit.2) logpost.Loglin(fit)