Non normallity and multicollinearity, is this true?
Posted: Wed Feb 01, 2017 8:46 am
Non normallity and muticollinearity are not problems!
is this true?
is this true?
Multiple regression is built precisely to handle multicollinearity. So long as the classic assumptions hold, estimates are unbiased, standard errors are correct, and the estimates are efficient.
Normality of the errors has nothing to do with bias or efficiency of regression estimates. Normality does matter for the small sample properties of test statistics, but with larger samples the central limit theorem replaces any need for assuming normal errors.