In a cross-sectional study, we use 'binary-logistic regression' function to calculate odds ratio, or do univariate analysis, and later construct multivariate model for variables with P-value
First it appears that your strategy is confused. Please read the attached papers and then return with a modified question If you should still have one Best wishes David Booth
Using a univariable p-value threshold to screen variables for inclusion in a multivariable model is not a good method of variable selection.
To answer your question, you can produce risk ratios rather than odds ratios in binomial regression by replacing the canonical logit link function with a log link. In r using the base glm function, this would simply be:
I concur with Jack Henry's statement, "Using a univariable p-value threshold to screen variables for inclusion in a multivariable model is not a good method of variable selection." If you need another reference to give to your boss or supervisor, I like this article by Mike Babyak:
Also, this UCLA page has examples of the kind of modeling Jack described (for obtaining RR estimates rather than OR estimates). The examples use Stata, not SPSS. But I believe the GENLIN command in SPSS can estimate all of the -glm- models shown. The key thing is to pay attention to the link functions (and error distributions).