1

In linear regression (linearity assumption had been checked), what is the effect if distribution of predictor is skewed but errors are normally distributed? Is there a risk for estimation of coefficients and confidence intervals ?

I founded this paper that seems to answer that it's not a problem:

Assumptions of Multiple Regression: Correcting Two Misconceptions Williams, Matt N.; Gómez Grajales, Carlos Alberto; Kurkiewicz, Dason Practical Assessment, Research & Evaluation;Sep2013, Vol. 18 Issue 11/12, p1

whuber
  • 281,159
  • 54
  • 637
  • 1,101
Emmanuel W
  • 37
  • 1
  • 4
  • Even if errors are not approximately normally distributted there does not have to be "risk" for coefficients estimates, e.g. https://stats.stackexchange.com/questions/29731/regression-when-the-ols-residuals-are-not-normally-distributed/29748#29748 – Tim May 12 '16 at 13:40
  • @Tim Thank you! You are right. I edit my question. – Emmanuel W May 12 '16 at 13:44
  • See [Logistic regression diagnostics when predictors all have skewed distributions](http://stats.stackexchange.com/q/67078/17230) & search on [leverage](https://stats.stackexchange.com/search?q=[regression]+leverage). – Scortchi - Reinstate Monica May 13 '16 at 15:06

0 Answers0