|Date: ||Wed, 22 Feb 2006 12:08:18 -0700|
|Reply-To: ||Matthew Pirritano <email@example.com>|
|Sender: ||"SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>|
|From: ||Matthew Pirritano <firstname.lastname@example.org>|
|Subject: ||Re: Logistic Regression Interaction|
|Content-Type: ||text/plain; charset=ISO-8859-1; format=flowed|
Thanks for the reply.
The way that I understand it you need to hand calculate the odds ratios
for the different levels of each dichotomous variable while holding the
other constant. I've actually already calculated these odds ratios
using formulas in Chen (2003). I'm banking on this method being
correct. From what I've read I understand that you cannot simply do
separate logistic regressions at each level of the dichotomous variables
in order to interpret the interaction effect. Correct me if I'm wrong
about this. Now I need to determine the significance levels of the odds
ratios I have calculated. So I was trying to use formulas in Hosmer &
Lemeshow (1989) to create a 95% confidence interval, but the formulas
require the covariance, and logistic regression in SPSS seems to only
provide a correlation matrix. Hence, my dilemma.
Matthew Pirritano, Ph.D.
National Science Foundation Post-Doctoral Fellow
College of Education
Department of Individual, Family & Community Education
1 University of New Mexico
Albuquerque, NM 87131-0001
Hector Maletta wrote:
> Interaction between dichotomous variables is just another dichotomous
> variable which equals 1 when both the other two equal 1. If you have
> predictors X and Z, and interaction term W=X*Z, you get the log odds
> increasing by a coefficient B(x) when X=1, by a coefficient B(z) when Z=1,
> and a further increase of B(w) when W=1, i.e. when it happens that both the
> other variables are 1. Jut treat the interaction term as an additional
> predictor and I guess your interpretation would be more straightforward. In
> terms of slopes, the increase of the log odds when X=1 is B(x) when Z=0 and
> is B(x)+B(z)+B(w) when Z=1. The same mutatis mutandis for Z=1.
> -----Mensaje original-----
> De: SPSSX(r) Discussion [mailto:SPSSX-L@LISTSERV.UGA.EDU] En nombre de
> Matthew Pirritano
> Enviado el: Wednesday, February 22, 2006 1:46 PM
> Para: SPSSX-L@LISTSERV.UGA.EDU
> Asunto: Logistic Regression Interaction
> I'm conducting a logistic regression in which I have 2 dichotomous
> variables with one continuous covariate.
> I'm under the impression that I need to recalculate by hand the odds
> ratios for the different levels of my dichotomous variables in the
> presence of a significant interaction. You apparently cannot just rerun
> separate logistic regressions at the different levels of the dichotomous
> variable as you would do with a significant interaction in an ANOVA. Am
> I correct about this?
> I have ordered a copy of Jaccard (2001), which I'm hoping will give me
> some guidance. But I can't really wait for it to arrive. I need to get
> this done post haste! Does anyone know of a website or article that
> would provide the equations that I need to follow up on this significant
> Also, SPSS gives the correlation matrix for my design. How is this
> helpful? As far as I can tell, in order to calculate the significance
> of the follow up odds ratios, that I'm calculating by hand, I need the
> covariance matrix. Easy enough to convert I suppose. Just an aside.
> Matthew Pirritano, Ph.D.
> National Science Foundation Post-Doctoral Fellow
> College of Education
> Department of Individual, Family & Community Education
> MSC05 3040
> 1 University of New Mexico
> Albuquerque, NM 87131-0001
> Telephone (505)277-7115
> FAX (505)277-8361