gRim (version 0.1-17)

stepwise.iModel; backward; forward: Stepwise model selection in (graphical) interaction models

Description

Stepwise model selection in (graphical) interaction models

Usage

"stepwise"(object,  criterion = "aic", alpha = NULL, type ="decomposable", search="all", steps = 1000, k = 2, direction = "backward", fixinMAT=NULL, fixoutMAT=NULL, details = 0, trace = 2, ...)
backward(object, criterion = "aic", alpha = NULL, type = "decomposable", search="all", steps = 1000, k = 2, fixinMAT=NULL, details = 1, trace = 2,...)
forward(object, criterion = "aic", alpha = NULL, type = "decomposable", search="all", steps = 1000, k = 2, fixoutMAT=NULL,  details = 1, trace = 2,...)

Arguments

object

An iModel model object

criterion

Either "aic" or "test" (for significance test)

alpha

Critical value for deeming an edge to be significant/ insignificant. When criterion="aic", alpha defaults to 0; when criterion="test", alpha defaults to 0.05.

type

Type of models to search. Either "decomposable" or "unrestricted". If type="decomposable" and the initial model is decompsable, then the search is among decomposable models only.

search

Either 'all' (greedy) or 'headlong' (search edges randomly; stop when an improvement has been found).

steps

Maximum number of steps.

k

Penalty term when criterion="aic". Only k=2 gives genuine AIC.

fixinMAT

Matrix (p x 2) of edges. If those edges are in the model, they are not considered for removal.

fixoutMAT

Matrix (p x 2) of edges. If those edges are not in the model, they are not considered for addition.

direction

Direction for model search. Either "backward" or "forward".

details

Controls the level of printing on the screen.

trace

For debugging only.

...

Further arguments to be passed on to testdelete (for testInEdges) and testadd (for testOutEdges).

Value

An iModel model object.

See Also

cmod dmod mmod testInEdges testOutEdges

Examples

data(reinis)
## The saturated model
m1 <- dmod(~.^., data=reinis)
m2 <- stepwise(m1)
m2