R help - Mar 2010 - Forced variable retention in regsubsets in 'leaps' package

Hi,

In working with model selection algorithms in other stats packages, there are
often the following type of selection procedures:

1) Forward selection:  Variables are added one at a time, starting with the most
significant [according to your selection criterion.]  Variables are not removed.
2) Stepwise selection:  Like forward selection, but at each stage, variables can
be dropped or added.
.
.
etc.

The regsubsets function in leaps offers a) exhaustive search, b) forward
selection, c) backward selection or d) sequential replacement. I thought forward
would work according to definition 1) above, but I am pretty sure I have
encountered a situation where, for example:

The one-factor model includes variable A
The two-factor model includes variables D, E
The three-factor model included variables C,D,E
.
.
etc.

Is there a way to coerce the behavior in definition 1? That is, if the best one
factor model contains A, all multi-factor variants will also? In other words,
variables will be added but never removed?

Thank you very much.

James C. McGrath


	[[alternative HTML version deleted]]