userTester
asked on
Model towards Normal
With Simple Linear Regression, is the goal to transform towards a near normal bell-shaped model?
In other words, if I find some skewness, residual autocorrelation, outliers, and heteroskedasticity in residuals, should I apply techniques to transform the data towards the ideal?
I applied the following:
Hope you can help!
Thank you!
In other words, if I find some skewness, residual autocorrelation, outliers, and heteroskedasticity in residuals, should I apply techniques to transform the data towards the ideal?
- Should the transformation be applied to the independent (X) variable only or both X and Y?
- when should this be applied to residuals only?
- should these techniques be applied early on to X and Y?
I applied the following:
- log(10) to X and the histogram looks more normal than before.
- "residual shift" technique which improved the autocorrelation substantially (autocorrelation now -0.045, Durbin-Watson now 2.09)
- excluded data outside the "InterQuartile Range (IQR)" - not sure if this did much
Hope you can help!
Thank you!
ASKER CERTIFIED SOLUTION
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
ASKER
Thanks, so much!
ASKER
Thanks, so much!
ASKER
So, if I find that residual autocorrelation is at 0.55, I should find out why the residuals are so correlated? How do I do that?
Isn't transformation for the purpose of finding the best fit?