R help - Aug 2008 - Adjusting for initial status (intercept) in lme growth models

Hi everyone, I have a quick and probably easy question about lme for this
list.

Say, for instance you want to model growth in pituitary distance  as a
function of age in the Orthodont dataset.

fm1 = lme(distance ~ I(age-8), random = ~ 1 + I(age-8) | Subject, data
Orthodont)

You notice that there is substantial variability in the intercepts (initial
distance) for people at 8 years, and that
this variability in initial distance is related to growth over time:

R# summary(fm1)
...
Random effects:
 Formula: ~1 + I(age - 8) | Subject
 Structure: General positive-definite, Log-Cholesky parametrization
            StdDev Corr
(Intercept) 1.8866 (Intr)
I(age - 8)  0.2264 0.209
Residual    1.3100

Now 2 questions:

1.  With lme, how can you get a fit of the growth model accounting for the
relationship between initial status (intercept) and growth?
Some texts call this latent variable regression or something or other, which
seems to basically boil down to adding the random effects
intercept as a predictor in the growth model.  Is this done in lme by simply
adding the intercept results from ranef(fm1) to the model?
This two-step process seems wrong to me for some reason, perhaps because it
seems too simple.  Anyone know the proper way to do
in lme?

2.  In addition, suppose you see that there are significant differences in
initial status by Sex:

fm2 = lme(distance ~ I(age-8) + Sex, random = ~ 1 + I(age-8) | Subject, data
= Orthodont)

R# summary(fm2)
Fixed effects: distance ~ I(age - 8) + Sex
             Value Std.Error DF t-value p-value
(Intercept) 22.917    0.5134 80   44.64   0.000
I(age - 8)   0.660    0.0713 80    9.27   0.000
SexFemale   -2.145    0.7575 25   -2.83   0.009

Along the lines of question #1, how would you get a growth model adjusting
for these Sex differences in initial status?  I am looking for something
similar to adjusting for baseline differences between Sexes in ANCOVA.  I
know Lord would not approve, but this is just by way of example...  Thanks
so much for your help, and this wonderful program Dr. Bates.

- DC

	[[alternative HTML version deleted]]

D Chaws <cat.dev.urandom <at> gmail.com> writes:
> Say, for instance you want to model growth in pituitary distance  as a
> function of age in the Orthodont dataset.
> 
> fm1 = lme(distance ~ I(age-8), random = ~ 1 + I(age-8) | Subject, data >
Orthodont)
> 
> You notice that there is substantial variability in the intercepts (initial
> distance) for people at 8 years, and that
> this variability in initial distance is related to growth over time:
Looks like a perfect example to use parameter weight=varPower(something)  in
lme; you could use some power function of the initial distance. See Chapter 5.2
in Pinheiro-Bates.

Dieter