Dear freinds,
I am doing MMM on the time series data, my dependent variable is sales and independent variables are investemt on different format of Ads(Like Functional, thematic, promotional etc.). Here is finding high correlation(.58) between these format of Ads.
All my independent variable are important, please suggest how I take these variables into the model or how to tackle multicollinearity.
Thanks in Advance.
A very comprehensive reply by Mr. Paul.
Mare the fact that two IVs have correlation does not mean they have multicollinearity. If IVs have a significant functional form then multicollinearity exists. A simple way to check it to compute Variance Inflationary Factor (VIF):
VIF = 1/ Tolerance (T)
T = 1 - R^2
The option is available in SPSS, you have to put a check on collinearity diagnostics while running MLR. If coefficient of VIF < 5, it means no multicollinearity.
VIF < 1 strict criteria
VIF < 5 moderate criteria
VIF < 10 relax criteria
Even if VIF is greater than 10 yet your IVs are important then go ahead as Mr. Juehui suggested.
Finally, you can not overcome multicollinearity especially when you are dealing with time series data BUT you can reduce it.
Hope it helps.
Dear Juehui, Paul & Nawaz,
Thank you so much for detail on avaluating the Multicollinearity.
Now if the multicolliearity exist the data then how can I reduce it without droping any IVs.
What are the remedial measures to reduce multicollinearity.
Is Ridge Regression can useful on this ?
Thank you so much once again :)
Danish
- Badges
- Science topic