@Maya Yes it can be used. For detailed information refer to the following link https://towardsdatascience.com/when-and-how-to-use-weighted-least-squares-wls-models-a68808b1a89d

Your best bet is probably Box-Cox transformation.full information as well as package implementation in stat packages can be had by repeating the appropriate searches from the screenshot.. Best wishes, David Booth

Dear Maya,

I agree with Imran Anwar. Yes, you can use it.

With best regards,

Firdos

Maya -

You could first model the heteroscedasticity using weighted least squares (WLS) regression, where regression weights are part of the regression coefficients. The estimated residuals do not have to be too nearly normally distributed, so you may he fine with that. You could use a "graphical residual analysis" to check on fit to your sample, and a "cross-validation" to try to avoid fitting too closely to that particular sample at the expense of fitting less well to the rest of the population or subpopulation to which you wish to apply your model.

You might want to look into the following:

https://www.researchgate.net/project/OLS-Regression-Should-Not-Be-a-Default-for-WLS-Regression, and the various updates there, in reverse chronological order.

Also:

https://www.researchgate.net/publication/352134279_When_Would_Heteroscedasticity_in_Regression_Occur

Cheers - Jim

PS Maya -

I suggest avoiding transformations unless really needed. Since you said an inverse transformation helped, scatterplot graphics might be particularly interesting, though I suggest they always be used. Note that the WLS regression I referenced above is good for any regression of form

Yi = predicted-y + epsilon

By the way, please note that OLS regression is just a special case of WLS regression where regression weights are all equal, which is often not a very good fit. It is often a convenient, but not very good option.

Data transformation is a bad idea. See why: Method Non-linear regression -Why you shouldn't take the logarithms...