XploRe Help : glmbackward

Keywords - Function groups - @ A B C D E F G H I J K L M N O P Q R S T U V W X Y Z

Library:	glm
See also:	glmselect glmforward glmest glmcore glmstat

Quantlet:		glmbackward
Description:		glmbackward performs a backward model selection by searching the best of all subset models w.r.t. the AIC or BIC criterion. Optionally, a number of columns can be given, which are always included in the submodels.

Link:

Tutorial: GLM in XploRe

Usage:	`sel = glmbackward(code,x,y,opt)`
Input:
	`code`	text string, the short code for the model (e.g. "bilo" for logit or "noid" for ordinary PLM).
	`x`	n x p matrix, the predictor variables.
	`y`	n x 1 vector, the response variables.
	`opt`	optional, a list with optional input. 'glmbackward' allows the same optional parameters as 'glmest'. Additionally, the following are supported.
	`opt.fix`	r x 1, r < p, numbers of columns which should be always included in the model.
	`opt.shm`	integer, if exists and =1, some output is produced which indicates how the selection is going on.
	`opt.crit`	string, either "aic" or "bic", the selection criterion to use. If not given "aic" is used.
Output:
	`best`	p x 5, five best models.
	`bestcrit`	list, containing criteria for five best models:
	`bestcrit.aic`	1 x 5, AIC's for five best models.
	`bestcrit.bic`	1 x 5, BIC's for five best models.
	`bestordcrit`	list, containing criteria for five best models:
	`bestordcrit.aic`	1 x p or 1 x (p-r), AIC's for five best models for every order.
	`bestordcrit.bic`	1 x p or 1 x (p-r), BIC's for five best models for every order.
	`bestfit`	GLM fit for best model, contains b, bv and stat as computed by "glmest".

Note:

The number of estimated models is np*(np+1)/2, when np denotes the number of variables to select.

Example:

library("glm") ;========================== ; simulate data ;========================== n=200 b=1|2|0|0 p=rows(b) x=normal(n,p) y=x*b+normal(n) ;========================== ; GLM model selection ;========================== shm=1 x=matrix(n)~x opt=glmopt("shm",1) ; show selection going opt=glmopt("fix",1:2,opt) ; always variables 1,2 modsel=glmbackward("noid",x,y,opt) modsel.best modsel.bestcrit

Result:

The best out of 7 possible submodels is determined. The 5 best
models and their AIC's and BIC's are shown.

Author: M. Mueller, 20010228