Date:         Sat, 22 Jun 2002 09:33:22 +1000
Reply-To:     Bob Green <bgreen@dyson.brisnet.org.au>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         Bob Green <bgreen@dyson.brisnet.org.au>
Subject:      comparing multiple regression & logistic regression
Content-Type: text/plain; charset="us-ascii"; format=flowed

I am interested to know if anyone is aware of literature or studies
analysing the same data set using multiple regression and logistic
regression. I understand that the former requires a continuous dependent
variable and the latter a dichotomous dependent variable - however it would
be possible to dichotomise a continuous variable to achieve a form of
comparison.

I ask this because I have seen a variety of approaches to reporting
logistic regression results and am curious regarding to what extent
variables with various odds ratios might be significant predictors (or not)
in a multiple regression. One issue is that often variables have an OR
greater than 1 and these are reported as significant (though there may not
be included any indication of statistical significance).
In contrast, multiple regression results generally appear more clearcut in
terms of whether variables contribute significantly to the regression, and
R squared provides an indication of how useful the model is overall.

Any thoughts on this are appreciated,

Bob Green