I am working a cross section data. I am planning to follow 3-FGLS technique to handle the problem of heteroscedasticity. So, in the first step I am running a simple OLS and achieving the error terms. In the 2nd step, I am regressing the squared error terms from the 1st step on the same independent variables that I used in the 1st step. Then in the 3rd step I am weighting the equation in the 1st step by the square root of the predicted value in the 2nd step.

Now the problem is that I have dummy variables in my original equation. So, in the 3rd step when I am using the weights from the 2nd step on the explanatory variables, the dummy variables no more remain in the form of 0 and 1, as for each observation I have a different weight in form of predicted values from the 2nd step (So I have more than two values for the dummies after the adjustment). In such case, how will I be able to interpret the dummy variables? OR What should I do with the dummies? Please help me with this problem.