From OLS to Box Cox to ?


I have just found this website - thank you so much for all the contributions made here. I hope you can provide me with some quick help.

Using SPSS I have run an OLS regression finding out later that the
assumptions of multivariate normality and normal distribution of the
dependent variable are violated. (Using Shapiro Wilk and multivariate
tests such as Srivastava's und Mardia's).

I decided to try out Box-Cox Transformation using SPSS syntax provided
by Michael Speed of the Texas A & M University, getting a Lambda-value
that I can now use to transform my dependent variable.

I just don't know what to do now. I have seen that after
transformation, people report Loglikelihood values, Chi2-values and so
on - something I am not familiar with at all. What kind of regression
should I use given the transformed values? What steps need to be taken
after transforming the DV?

It would be so great if you could help me - I'm about to finish my
final thesis but got stuck with this analysis. I really don't know
what to do about this...

Thanks a lot!




TS Contributor
Since the overall purpose of a Box-Cox transformation is to restore normality, then I would use OLS on the transformed variable(s).

Thanks for replying. That's what I thought, too. -

But why does my prof (and Stata) switch to the so called "Box-Cox-Model", reporting log-Likelihood-values, p>Chi2's etc?

And, what kind of regression is used to compute these values?

I'd be more than happy if somebody knew more about this...