4 continuous IVs predicting a single binary DV

[fizix]

Like the title says...

I have 4 continuous IVs, whose combined effects (i.e. interactions possible) should predict a binary DV (possible values 1 and 0).

What kind of model/tests should I use? Where can I get more info?

Thanks

 

[victorxstc]

you should use a binary logistic regression. A small googling or search within this site will find you tons of useful information.

 

[fizix]

thanks for your response

I notice that this is a stepwise sort of procedure.

Since I'm only interested in main effects and interaction effects when everything else is controlled for, I can enter all four IVs and their possible interactions all at once right?

Or should I enter one interaction effect at a time?

Sorry, that was probably a very unclear question...

edit:
I guess I just don't understand all the stepwise stuff. Why can't you just put in all the variables and interactions you're interested in at once? Could someone explain that part more succintly to me?

 

[bugman]

You can use backward or forward proceedures, which start with either all terms and sequentially eliminate terms or start with the bare minimum and sequentially add terms until the optinimal model is found.

 

[Dason]

You can use backward or forward proceedures, which start with either all terms and sequentially eliminate terms or start with the bare minimum and sequentially add terms until the optinimal model is found.

You could but you shouldn't.

 

[bugman]

Want to elaborate?

 

[Dason]

Elaboration: Stepwise procedures are not something I am fond of and I find the results to be misleading.

 

[bugman]

Fair enough.

 

[Karabiner]

You can use backward or forward proceedures, which start with either all terms and sequentially eliminate terms or start with the bare minimum and sequentially add terms until the optinimal model is found.

If not a model is sought which is optimal
for the present sample , but a model which
is generalizable beyond sample data, then
stepwise procedures are ususally considered
of limited value (e.g.
http://www.philender.com/courses/linearmodels/notes4/swprobs.html
http://os1.amc.nl/mediawiki/images/Babyak_-_overfitting.pdf )

With kind regards

K.

 

[fizix]

So if you'd rather not use a stepwise procedure, is there a better sort of test/model I should conduct?

edit:
also, what kind(s) of post hoc test(s) should I use?