Multiple regression

S

Hello

I'd thought I'd done all my analysis etc only to realise I completely forgot to include one of my variables, silly I know, but deadline is soon and I'm very stressed. Anyway, so I've started redoing the whole thing from scratch only to find out that I might not be able to use multiple regression anyway... so I'd thought I'd ask these:

1) Do the predictor variables need to be normally distributed? I'm getting mixed information on these, some say yes, others no

2) Can you do a multiple regression when the criterion variable is nominal? I'm trying to come up with a model that will predict a phenomenon, but the phenomenon either happens or it doesn't (although I can probably change it to a ration variable if it needs be).

3) Any other tests I should be thinking about in additional to (hopefully!) multiple regression?

Any help would be greatly appreciated.

Skig

Avatar for sneaks

you're looking at logistic regression.

Your residuals have to be normally distributed - and if asked for SPSS will provide details on this.

S

Thank you!

Avatar for sneaks

just to add - if you do have non-normal data you can look into bootstrapping. SPSS 18 can do this, but it has to be a separate add-on which can be bought. Pm me if you need to do that.

19172