I will do negative binomial regression analysis using R by glm.nb function. But, I want to check collinearity first. In linear regression, we can use SPSS to check collinearity with steps: Analyze > Regression > Linear > Statistics > Collinearity diagnostic, and then we will obtain VIF and TOL from that output.
Can we use same steps above to check collinearity on negative binomial regression? All independent variables are ratio.
Multicollinearity statistics like VIF or Tolerance essentially give the variance explained in each predictor as a function of the other predictors. The fact that the outcome is a count does not affect how VIF or Tolerance is calculated because the outcome variable is not involved in the calculation of these statistics.
So, In SPSS, you can run your model as a linear model (even though the outcome is a count) and use the VIF/Tolerance values that SPSS outputs (do not look at the regression coefficients or standard errors, however. They are almost certainly incorrect). Having a discrete outcome affects how the predictors relate to the outcome but does not affect how the predictors relate to other predictors (in which multicollinearity is interested), especially if they are all measured on a ratio scale.
Hi,
your question: How to detect multicollinearity on negative binomial regression using SPSS?
Kindly see the attachment.
Multicollinearity statistics like VIF or Tolerance essentially give the variance explained in each predictor as a function of the other predictors. The fact that the outcome is a count does not affect how VIF or Tolerance is calculated because the outcome variable is not involved in the calculation of these statistics.
So, In SPSS, you can run your model as a linear model (even though the outcome is a count) and use the VIF/Tolerance values that SPSS outputs (do not look at the regression coefficients or standard errors, however. They are almost certainly incorrect). Having a discrete outcome affects how the predictors relate to the outcome but does not affect how the predictors relate to other predictors (in which multicollinearity is interested), especially if they are all measured on a ratio scale.
Hi Awwalina
Daniel is quite correct. The easiest way in SPSS is to first run your model as a linear regression. In the Statistics option, tick Collinearity Diagnostics. In the output, the last table should show Collinearity Diagnostics. Look at the last value of the Condition Index (CI). For log-linear models (your case), you want CI
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression