|
|
Dear all,
I am running a logistic regression using GENLIN command. I have 2 independent variables: 2 factors (dummy variables) and 1 continuous: Sex (dummy), Nationality (dummy) and Age (continuous)
When I use EMMEANS, SPSS gives me the predicted mean response for categories in one factor assumming the distribution of the other factor is 50% of the sample belongs to each category. I mean EMMEANS calculates the mean response for Sex from the predicted equation weighting the other factor (nationality) coefficient by 0.5 (e.g. assumming I have same number of nationals than foreigners).
I would rather use the centroid. I mean, as I have 90% of nationals and 10% of foreigners I would like to calculate the mean response for men like this:
mean response (for men) = alpha + 0,1*beta2(foreigner) + mean_age*beta3
So, I will be assuming a "centroid" man who is 90% national and 10% foreigner (of course, this is a fiction as no man is like this, but I find it very informative to present the results).
Does anyone know how to specify a subcommand for EMMEANS so it creates an "artificial" centroid for other independent categorical variables in my model?
PS: I know I can do it if I say to SPSS that nationality is continuous and then I grand-center the variable (so I get a coding -0.1 for nationals and 0.9 for foreigners). The problem in doing so is that I get a smaller confidence interval that the one I calculate by hand. This happens as the "continuous variable" specification changes the covariance of the coefficients reducing the variance of the mean response.
|
Locked
