priorLoglin
object, which represents the
hyperparameters of the Dirichlet distribution used as a prior
distribution for the multinomial model, or the constrained Dirichlet
used for the loglinear model. In
priorLoglin
the prior is specified
by a character string. There is one hyperparameter for each cell of
the contingency table formed by the levels of the factors. Often the
hyperparameters are set to a common value. The value of the
hyperparameters eventually used depends on the algorithm used.
priorLoglin(type = "data.dependent", nPriorObs)
a data dependent prior. Estimate prior hyperparameters from
the data assuming that the factors are independent. Under
independence, the probability of a combination of factor levels is the
product of the marginal probabilities. Estimate the marginal
probabilities by the observed proportions. Then, for data
augmentation, each hyperparameter equals the estimated probabilities
multiplied by
nPriorObs
(see below). For
the EM algorithm, add one to each hyperparameter.
"data.dependent"
prior is used. Intuitively,
nPriorObs
is the
number of prior observations. Not used for other prior types.
"priorLoglin"
giving a specification of the
prior. For
"data.dependent"
priors, the routine
dataDepPrior
is used to compute the actual prior in the context of the data and
algorithm. See
.
Schafer, J. L. (1997), Analysis of Incomplete Multivariate Data, Chapman & Hall, London.
priorLoglin(type = "data.dependent", nPriorObs = 5)