topic Checking Multicollinearity in Logistic Regression model in New SAS User

Checking Multicollinearity in Logistic Regression model

Malathi13 — Tue, 16 Jul 2019 21:31:18 GMT

Hi SAS gurus, I'm trying to check multicollinearity between independent variables (all categorical including dependent variable which is obesity with yes/no categories) using proc logistic regression command. How do I check for multicollinearity using this command, what options work in proc logistic? proc logistic data=test; model Obesity= age, sex, BMI, height, weight; run; I know how to use VIF and TOL or CoLLIN options in Proc Reg but I don't know what option can be used in proc logistic. Can you please help! Thank you, M

Re: Checking Multicollinearity in Logistic Regression model

PaigeMiller — Tue, 16 Jul 2019 21:52:35 GMT

Multicollinearity can be checked in PROC REG with made-up Y variable, as these calculations do not depend on Y.

Re: Checking Multicollinearity in Logistic Regression model

Ksharp — Wed, 17 Jul 2019 11:20:16 GMT

There are no such command in PROC LOGISTIC to check multicollinearity .

1) you can use CORRB option to check the correlation between two variables.

model good_bad=x y z / corrb ;

You will get a correlation matrix for parameter estimator, drop the correlation coefficient which is large like > 0.8

2) Change your binary variable Y into 0 1 (yes->1 , no->0) and use PROC REG + VIF/COLLIN .

Re: Checking Multicollinearity in Logistic Regression model

Malathi13 — Wed, 17 Jul 2019 15:05:22 GMT

Hi Ksharp,

Thank you for the solution, both of your suggestions worked except that for Proc Reg, I had to convert the character values to numeric types to run Proc Reg.

Thank you,