1

I'm trying to build a multivariable least-square linear regression model, and there is heteroscedasticity in my model. I saw many articles suggesting transformations such as log transformation or Box-Cox transformation. However, because my response variable contains many negative values, I cannot use these transformations.

In this case, How can I deal with this heteroscedasticity? Is there any transformation I can use for response variables with negative values?

Residual plot and linear regression scatter plot graphs between the dependent and one of the independent variables are given below. On the linear regression plot, the x-axis is one of my independent variables, and the y-axis is the dependent variable.

enter image description here

enter image description here

Ahmet Atilla Colak
  • 21
  • 1
  • It lloks like your independent variables are positive. Why are you not taking log only of them? You do not have to transform both sides. – Misius Jun 13 '21 at 17:47
  • 1
    That is not useful because transformation in independent variables doesn't affect the impact of heteroscedasticity. For this, I have to transform the dependent variable. – Ahmet Atilla Colak Jun 13 '21 at 18:26
  • Oh yeah, that is true, I forgot about your overall goal. I would suggest using weighted linear regression for dealing with heteroskedasticity but since it does not involve transformation of the data, it does not really answer your question... – Misius Jun 13 '21 at 18:38
  • There are several methods for dealing with heteroscedasticity listed here: [Alternatives to one-way ANOVA for heteroskedastic data](https://stats.stackexchange.com/a/91881/7290) (although not all will be applicable in your situation). To use Box-Cox transformations of Y data w/ negative values, first add a constant to make them all positive. However, a transformation is probably not your best option. – gung - Reinstate Monica Jun 17 '21 at 15:22
  • In addition, you seem to have a rather substantial misfit due to using an improper functional form (what I'm guessing are all straight lines). – gung - Reinstate Monica Jun 17 '21 at 15:23

0 Answers0