Regression Analysis Help

#1
Hi All, first post here, glad to be part of this forum.

After surveying my population, I am currently experimenting what factors (about 10 explanatory variables) are influencing a dependent variable, the data is as follows:

Dependent variable is Discrete, it is ordinal, and has 5 values.

Explanatory variables (10) are Discrete, ordinal, and have 5 values.

Questions: is Logistic Regression Analysis suitable for the above data? I want to reveal the relationship between the dependent variable and several explanatory variables.

I will be using Minitab.

Thanks
Mohammed
 

WeeG

TS Contributor
#2
Ordinal Logistic Regression is the model you are looking for.

But, I would take it carefully if I were you. Your independent variables are ordinal, and have 5 values each, which is 4 dummy variables for each one, 10 times, it's like having 40 variables. If your ordinal variables come from a survey, check if they are correlated, if so, try to use principal components, and use the new components as your new independent variables.

Minitab support all that.
 
#3
Ordinal Logistic Regression is the model you are looking for.

But, I would take it carefully if I were you. Your independent variables are ordinal, and have 5 values each, which is 4 dummy variables for each one, 10 times, it's like having 40 variables. If your ordinal variables come from a survey, check if they are correlated, if so, try to use principal components, and use the new components as your new independent variables.

Minitab support all that.
Hi WeeG

Thanks for your reply.

How do I check if they are correlated, and what is "principal components".

a bit new to all this, thanks for your help on this matter.

Regards,
Mohammed
 

WeeG

TS Contributor
#4
examine the correlation between all pairs of independent variables. If you see some suspicious correlations, run a linear regression model (not logistic) with ANY dependent variable (if needed, create one) and calculate the VIF for each independent variable. If you find VIF values of 5 or more (some say 3 or more), your variables are correlated.

Principle components analysis is a method of taking p correlated independent variables, and creating k (<<p) independent variables that are not more than linear combinations of the original variables. These components can then be entered as "new" independent variables to a regression model.

if your IV's are representing the same thing, maybe you can build some summarizing measure from them (like a mean of them all)