Forward vs. backward selection. Advantages and Disadvantages

#1
I guess there are automated methods but I am referring to manual selection. Like: first finding out which variables have significant individual effects at a certain cut-off (say 0.25) and then putting those signficant variables in the model. Backward selection would mean (from what I understand), putting all these signficant ones in the model and eliminating one by one, the variables that are now insignificant.

Forward selection would mean I suppose, putting in one variable with significant individual effect at a time and removing a variable as and when it falls out of signficance.

To my mind, one big advantage of forward selection seems to be that it will enable you to know which variable is responsible for another variable falling out of signficance. For example, if A had signficant individual effects and was okay (signficant) after B was added but fell out after C was added- it would suggest that A was signficant because of C and C being same, A is not signficant. (So you remove A and keep C). Backward selection it seems to me would not give this kind of information. Is this right?

Are there any (other- if the above was right) advanatages/disadv. to using one method over another?
Thanks.
 

trinker

ggplot2orBust
#3
The answer to this question is the same as another you posted. It depends on what you want to use the model for. You haven't given us this vital piece of information. It's like asking a carpenter (my father's trade) "Should I use the saw or the hammer?" The first question he'd ask you is the same I'll ask of you.
 

Dason

Ambassador to the humans
#4
The answer to this question is the same as another you posted. It depends on what you want to use the model for. You haven't given us this vital piece of information. It's like asking a carpenter (my father's trade) "Should I use the saw or the hammer?" The first question he'd ask you is the same I'll ask of you.
Psh. Always use the saw.