Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: RE: boxcox and panel data?


From   "Nick Cox" <n.j.cox@durham.ac.uk>
To   <statalist@hsphsun2.harvard.edu>
Subject   st: RE: boxcox and panel data?
Date   Mon, 19 Oct 2009 13:51:32 +0100

As to "can", why not try it and find out? (Yes, you can.) 

As to "should", which may be your underlying question, my impression is
that Box-Cox neither knows nor cares about any panel aspect to data. The
effect is to lump together different panels regardless and to ignore any
associated error structure. You may get a reasonable indication of what
transform your data could bear, but confidence intervals and P-values
are dubious. 

The question is puzzling insofar as the usual point of Box-Cox
transformation is to find a transformation, not to compare
transformations. 

In my view Box-Cox, despite its splendid name, is distinctly overrated.
The idea of letting the data tell you what transformation you should use
is brilliantly simple, but being told that powers of 0.345 are optimal
is often difficult to interpret scientifically and/or practically. In
most practical data analysis, if a simple transform like logarithm,
square root, square or reciprocal is appropriate then that is usually
apparent from simpler methods. 

-boxcox2- is a generalisation written by David Drukker and available
from the StataCorp website. Please remember to explain where
user-written software such as -boxcox2- which you refer to comes from. 

Nick 
n.j.cox@durham.ac.uk 

Philomela@gmx.net

i want to do a box-cox transformation to compare different functional
forms. Can I use -boxcox or -boxcox2 for that when I have (unbalanced)
panel data?

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index