Re: st: ln transform and box cox

[David Hoaglin <dchoaglin@gmail.com>]

Tom,

You should carefully consider whether a quadratic is an appropriate
and adequate summary of the contribution of age.  Many nonlinear
relations are not well approximated by a quadratic.  The choice of
"overall quadratic growth" on Slide 5 in the presentation by Gutierrez
is not very convincing.  Also, on Slide 32, part of "the problem with
the fixed-effects approach" is the choice of 60 linear segments.

Also, you should examine the choice of functional form for age in the
context of the model that contains all the explanatory variables that
you plan to use.  The adjustments for those explanatory variables may
affect the apparent pattern of the relation of the dependent variable
to age.

Applied Longitudinal Analysis by Fitzmaurice, Laird, and Ware is a
very useful book.  The second edition came out in 2011.  Chapter 13 in
the first edition became Chapter 16 in the second edition.  The
companion website has, among other resources, Stata code for many of
the examples in the book.

David Hoaglin

On Wed, Mar 6, 2013 at 12:10 PM, Thomas Norris <T.Norris2@lboro.ac.uk> wrote:
> Rebecca,
> Thanks for your input and for the slides.
>
> It is natural to observe increased variation between (human) fetuses as gestation progresses and this is why I logged the weight variable (and to help the models converge). The fracpoly command returned powers of 1 and 2 when weight was on the log scale. My analysis thus far has been to try and obtain the best fitting age terms for the data and then I was going to add explanatory variables to the model (ethnicity, sex, maternal height & weight).
>
> Tom

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