0

First of all, I have to admit I am here because I am struggling with my dissertation. With this disclaimer out of the way, here is my problem:

I am trying to carry out multiple regression analysis but I cannot meet the assumptions for normal distribution of the residuals and for homoscedasticity. I transformed the dependant variable by inverting it which resulted in a normal distribution of the residuals. However, the variance of the residuals is heteroscedastic. I tried to dig around and came across weighted least square regression (WLS). My question is:

Do I need to do both WLS and data transformation? Can I do just WLS/does WLS require normal distribution of residual?

I apologise if these are silly questions but I am having to teach all of this myself and I am struggling.

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
mv94
  • 1
  • 5
    What is the purpose of your model ? Normally distributed residuals are *nice* but not necessarily important. – Robert Long Jul 31 '21 at 15:03
  • 3
    Part of your post is ambiguous: what exactly has a Normal distribution? The raw residuals or the weighted residuals? It sounds like your question might not be addressing the underlying problem, which is to diagnose and cope with possible problems with the multiple regression. Consider changing your question to describe those problems rather than focusing on difficulties arising from your attempts to deal with them. – whuber Jul 31 '21 at 15:15
  • Just a comment that normality and homoskedasticity are not requirements for OLS, though you lose some efficiency and will have to correct your standard errors. If you don't have constant variance then you can correct with White standard errors in many circumstances. See chapter 3 of Angrist and Pischke Mostly Harmless Econometrics, or some relevant answers to questions like https://stats.stackexchange.com/questions/29731/regression-when-the-ols-residuals-are-not-normally-distributed?rq=1. – Matterhorn Aug 01 '21 at 23:36

0 Answers0