Relationship between categorical variables

[jhs2171]

Hello, 

I am not sure if I should post this here as this technically isn't a SAS question..(but somewhat related). I have about 180 subjects in my dataset, and primarily interested in examining the relationship between three modifiable behavioral factors (two binary and one three-level nominal) that could be potentially associated with the outcome (death; all 180 subjects). Because the data wasn't collected for research per se, I do not have a reference or a control group to compare. My question is: is there any statistical analyses I can do to examine the relationship between these three modifiable risk factor variables? I tried chi-sq and it was helpful, but I am curious to find out if there are other tests I could perform. Can tetrchoric or polychoric (PROC FREQ) corr or performing factor analyses (using the matrix from the correlation procedure) be meaninful in any way?  

Thank you!

[Rick_SAS]

In addition to tests for association in PROC FREQ, you might look at correspondence analysis, which is the discrete/categorical analogue of principal component analysis.  In SAS, you can carry out correspondence analysis by using the CORREP procedure.  

[jhs2171]

I see. I didn't know about correspondence analysis, so I appreciate the suggestion! Does this mean the statistics I got from PLCORR such as Tetrachoric values (in PROC FREQ) are not valid?

[Rick_SAS]

?? I don't see why they wouldn't be valid.  Correspondence analysis is a multivariate technique that attempts to reveal relationships in categorical variables. But it doesn't replace or invalidate other statistical methods.

[jhs2171]

Oh ok. I said that not because you suggested a different method, but I read somewhere that standard methods of performing FA (using a tetra/polychoric correlations matrix) assume that factors are continuous. I also noticed that having a three-level nominal var in the matrix generate some funky numbers (compared to only having binary vars).

Thanks again for the suggestion!

[ballardw]

Logistic regression is a common way of examining data with two outcomes (it looks like death/alive in your case?) with one or more factors that are typically not continuous such as  smoker/nonsmoker, low/middle/high value indicator, gender or such.

[jhs2171]

Hello,
Unfortunately, I only have one outcome (death) otherwise I would have totally tried logistic reg!