BOX-COX transformation - Correct method

Dear All,

I fitted a quadratic model to a set of data.
I am trying to apply the box-cox transformation to check if my data should be transformed. I use a software (Design Expert) that do the transformation for me but I am not able to repeat the results by doing the calculations by hand (using mathematica to be precise).

Anyway my method is the following:
1. Apply the boxcox transformation to my data (I use this formula:
2. Fit a new quadratic model to my transformed data. The new quadratic model has the same form than the original one.
3. Generate an ANOVA Table and calculate the Sum of square of the residuals.
4. I do that for several lambda values and generate the plot Log(SSresiduals)vslambda.

I guess the method is incorrect. It would be greatly appreciated if someone could provide me with the correct method, a "protocol" to use the box cox transformation.

Thanks in advance and sorry for my approximate English.

I have a question for you. I'm looking at a probit model and would like to know

1) whether it would be better if my IV's were approx. normally distributed?

2) if so, why shouldn't I simply use transformations (e.g. Box-Cox) to come closer to a normal distribution?

I'm not sure about this one and I can see that, normally, one may be interested in getting interpretable coeff's. yet, with probit, I'm calculating predicted values anyways - so why not transform all of my continuous IV's?

Thank you for your help!