Quick question about logistic regression

[Charlotte-9]

Dear all,

As some of you will know from my previous posts, I am currently trying to
model uptake of a screening test using logistic regression
(outcome ‘screened’ or ‘not screened’).  My independent variables are
ethnicity (5 categories), age, deprivation (continuous) and gender.  As
well as fitting a model with the four independent variables, my employer
insists that I look at the variables individually (e.g. what effect does
age alone have on screening?).  In each of these cases, I am sure that the
effects are potentially modified by other variables.

As to my question, I would like to know if it makes sense to use logistic
regression look at the effects of the individual variables, so that I can
then introduce one other variable in each case to check for interactions.
For instance, when looking at deprivation, say, I would like to check for
an interaction with ethnicity.  I’m just not sure whether this makes sense
given that I am ultimately building a logistic model containing all
variables.  At the moment, I have just generated basic cross-tabs for each
variable.  Any suggestions as to how I should proceed would be welcomed.

Thanks,

Charlotte

[Hector Maletta]

         Charlotte,
         You are right that a logistic regression with only one of the variables is probably confounded by the effect of some omitted variable.
         You can add other variables gradually, to test for either additive or interaction effects (i.e. adding the new variable on its own or in an interaction/multiplicative term). This can be done either by you in the order and manner you prefer, or through the stepwise method provided by SPSS, which adds new variables according to their contribution to the explanatory power of the equation.

         Hector

         -----Mensaje original-----
De: SPSSX(r) Discussion [mailto:[hidden email]] En nombre de Charlotte
Enviado el: 04 June 2007 07:47
Para: [hidden email]
Asunto: Quick question about logistic regression

         Dear all,

         As some of you will know from my previous posts, I am currently trying to
         model uptake of a screening test using logistic regression
         (outcome ‘screened’ or ‘not screened’).  My independent variables are
         ethnicity (5 categories), age, deprivation (continuous) and gender.  As
         well as fitting a model with the four independent variables, my employer
         insists that I look at the variables individually (e.g. what effect does
         age alone have on screening?).  In each of these cases, I am sure that the
         effects are potentially modified by other variables.

         As to my question, I would like to know if it makes sense to use logistic
         regression look at the effects of the individual variables, so that I can
         then introduce one other variable in each case to check for interactions.
         For instance, when looking at deprivation, say, I would like to check for
         an interaction with ethnicity.  I’m just not sure whether this makes sense
         given that I am ultimately building a logistic model containing all
         variables.  At the moment, I have just generated basic cross-tabs for each
         variable.  Any suggestions as to how I should proceed would be welcomed.

         Thanks,

         Charlotte