Multiple regression of categorical variables

#1
Hi everybody. I am almost new to statistics. I have some categorical features of a group of patients. some of these features are nominal( with 2 Levels) and some are ordinal (mostly with more than 4 levels). I analyzed the association between every two features by "Pearson chi-square test of independence" and Cramer's V and Phi value. Now I have two questions and I appreciate it if anyone can help me:
1- to Expand my analysis to multivariate analysis what should I do and which test or models should I use?

2- Since I'm checking the association of all these features should I apply multiple comparison correction? If yes should I apply for example FDR on ALL P-values or it is required to apply FDR separately on P-values corresponding to comparing every feature with others?

Thanks in advnace
 

hlsmith

Less is more. Stay pure. Stay poor.
#2
What is your sample size and how many candidate and kept variables do you have?

What is the study question(s)?

Like you should move onto multiple regression. How is the dependent variable formatted? You traditionally correct alpha level for all candidate variable that could have been in the model, not the final subset.
 
#3
Thanks for your reply.
I have 150 patients. There are 8 Tumor features (4 binary, 1 nominal with 3 levels, and 3 ordinal with 5 levels), 5 Epileptic features (3 binary and 2 ordinal with 4 levels), 3 Seizure frequency features (ordinal with 4 levels), and 1 Overall survival feature (Continous) recorded from every patient. The study questions are: Is there any association between Tumor features and 1- Epileptic features, 2-Seizure frequency features 3-OS features, and the same questions for Epileptic features.
I checked the association of every two features in these groups by "Pearson chi-square test of independence".
Now I want to expand the analysis to multivariate. The response variable in multiple regression can be nominal with more than 2 levels or an ordinal variable. what model should I use for these types of response variables? the predictors are 4 categorical variabls.
 

gianmarco

TS Contributor
#4
I think you should look into Logistic Regression, if your dependent variable is categorical. The independent variables (predictors) can be either continous or categorical, or both. But consider that there are some rules of thumb when it comes to the number of predictors vis-a-vis the number of cases.
 
#5
I think you should look into Logistic Regression, if your dependent variable is categorical. The independent variables (predictors) can be either continous or categorical, or both. But consider that there are some rules of thumb when it comes to the number of predictors vis-a-vis the number of cases.
Thanks
 

gianmarco

TS Contributor
#6
Last edited: