dmod: Discrete interaction model (log-linear model)

Description

Specification of log--linear (graphical) model. The 'd' in the name dmod refers to that it is a (graphical) model for 'd'iscrete variables

Usage

dmod(
  formula,
  data,
  marginal = NULL,
  interactions = NULL,
  fit = TRUE,
  details = 0,
  ...
)

Arguments

formula

Model specification in one of the following forms: 1) a right-hand sided formula, 2) as a list of generators, 3) an undirected graph (represented either as a graphNEL object or as an adjacency matrix). Notice that there are certain model specification shortcuts, see Section 'details' below.

data

Either a table or a dataframe. In the latter case, the dataframe will be coerced to a table. See 'details' below.

marginal

Should only a subset of the variables be used in connection with the model specification shortcuts

interactions

A number given the highest order interactions in the model, see Section 'details' below.

fit

Should the model be fitted.

details

Control the amount of output; for debugging purposes.

...

Additional arguments; currently no used.

Value

An object of class dModel.

Details

The independence model can be specified as ~.^1 and the saturated model as ~.^.. Setting e.g. interactions=3 implies that there will be at most three factor interactions in the model.

Data can be specified as a table of counts or as a dataframe. If data is a dataframe then it will be converted to a table (using xtabs()). This means that if the dataframe contains numeric values then the you can get a very sparse and high dimensional table. When a dataframe contains numeric values it may be worthwhile to discretize data using the cut() function.

The marginal argument can be used for specifying the independence or saturated models for only a subset of the variables. When marginal is given the corresponding marginal table of data is formed and used in the analysis (notice that this is different from the behaviour of loglin() which uses the full table.

The triangulate() method for discrete models (dModel objects) will for a model look at the dependence graph for the model.

Examples

Run this code

# NOT RUN {
 
## Graphical log-linear model
data(reinis)
dm1 <- dmod(~ .^., reinis)
dm2 <- backward(dm1, k=2)
dm3 <- backward(dm1, k=2, fixin=list(c("family", "phys", "systol")))
## At most 3-factor interactions
dm1<-dmod(~ .^., data=reinis, interactions=3)
# }