While running regression models, I have to include variables (binary, categorical as well as continuous) whose pairs produce correlation coefficients equal to 0.50 similar values. These values are obtained between a one level of a categorical variable (binary) and a continuous variable, between two binary variables, between two levels of same categorical variable. However, after building models, the multicollinearity is within the limits (Variance Inflation Factor < 5). Is it fine to interpret the coefficients of the variables from the models. Some people told me that correlation coefficient must be below 0.4 for interpretation.
Thanks in advance.
Just a side note, feature selection is NP complete and, therefore, such arbitrary rules come up to simplify the feature selection problem (IMHO).
I have read that highly correlated predictors would result in wrong standard errors. I want to know if it is fine to interpret the coefficients if some categorical variables have correlation coefficient of around 0.5
Hi Srijith
I think most of statistician and researcher they consider sever multicollinearity if VIF>15. So your model does not suffer from multicollinearity and you may go a head to interprit you coefficients results.
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts