What is best method to calculate correlation for 2 nominal variables?

#1
Hello,

I am analyzing data that looks at several nominal variables (e.g. employment status, race, gender, etc.) to determine if it plays a factor in the person's adherence to medication (also a nominal variable).

What is the best way to do that? From what I've read on the forums and other places, Cramer's is the way to go?

Thank you for any advice or suggestions!
 

CB

Super Moderator
#2
Please don't post duplicates of the same question. I've deleted your other two posts.

Multiple IVs, nominal DV - logistic regression?
 
#3
I apologize for the duplicate. I posted it by accident under psych and then I realized the broader category and posted there and came back here to delete here but couldn't.

Also, my 3 question was a non-duplicate, which is basically how do you measure if there is any correlation between a continuous independent variable and a nominal dependent variable?
 

CB

Super Moderator
#4
I apologize for the duplicate. I posted it by accident under psych and then I realized the broader category and posted there and came back here to delete here but couldn't.

Also, my 3 question was a non-duplicate, which is basically how do you measure if there is any correlation between a continuous independent variable and a nominal dependent variable?
Ah ok then I apologise in turn - I didn't notice that the third question was different. There are a few ways you can look at the relationship between a continuous IV and a nominal DV. One way would simply be to use an ANOVA (or t-test if the DV is binary). Sure, we usually think of ANOVA as having a nominal IV and continuous DV, but hypotheses about causality aren't built into the test - so with a bit of thought about interpretation you can do things this way. Another more general method that would allow you to add multiple IVs would again be logistic regression.

Re headings - don't worry too much about which subforum you post in. Most of the regulars keep track of posts via the recent posts feature, so posts tend to get picked up even if they're not quite in the right subforum.
 
#5
gotcha, thank you!

I've been looking into relationships between 2 nominal variables using Cramer's V, but I guess I gotta read up on logistic regression.
 

CB

Super Moderator
#6
Cramer's V is fine, but the fact that it allows only two variables could be a limitation depending on your project. It's also not that commonly used (at least in my experience). So having a look into logistic regression may indeed be worthwhile if only to at least check out your options :)