R help - Oct 2009 - Question on mixed effect models with LME

Good afternoon

Using R 2.9.2 on a machine running Windows XP

I have a longitudinal data set, with data on schools and their test scores over
a four year period.  I have centered year, and run the following


m1.mod1 <- lme(fixed = math_1 ~ I(year-2007.5)*TFC_,
               data = long,
               random = ~I(year-2007.5)|schoolnum,
               na.action = "na.omit")

where math_1 is a percentage of students in a given school that are at the
lowest math achievement level, year is year, TFC_ is a categorical variable for
a treatment I wish to evaluate, and schoolnum is an identifier.

When I run summary on this model, I get a strong negative correlation (-0.91)
between the intercept and I(year-2007.5), despite the fact that the mean of year
is 2007.5.

I am puzzled, as I thought centering the time variable should eliminate, or at
least strongly reduce, this correlation.

Any insights appreciated

thanks

Peter


Peter L. Flom, PhD
Statistical Consultant
Website: www DOT peterflomconsulting DOT com
Writing; http://www.associatedcontent.com/user/582880/peter_flom.html
Twitter:   @peterflom

On Wed, Oct 21, 2009 at 11:06 AM, Peter Flom
<peterflomconsulting at mindspring.com> wrote:
...
> I have a longitudinal data set, with data on schools and their test scores
over a four year period. ?I have centered year, and run the following
>
>
> m1.mod1 <- lme(fixed = math_1 ~ I(year-2007.5)*TFC_,
> ? ? ? ? ? ? ? data = long,
> ? ? ? ? ? ? ? random = ~I(year-2007.5)|schoolnum,
> ? ? ? ? ? ? ? na.action = "na.omit")
>
> where math_1 is a percentage of students in a given school that are at the
lowest math achievement level, year is year, TFC_ is a categorical variable for
a treatment I wish to evaluate, and schoolnum is an identifier.
>
> When I run summary on this model, I get a strong negative correlation
(-0.91) between the intercept and I(year-2007.5), despite the fact that the mean
of year is 2007.5.
>
Hi Peter,

For the "what's going on here?" questions it's very helpful to
have a
reproducible example.  I tried to create data fitting your
description, but the correlation disappeared as expected:

set.seed(777)
library(nlme)
school <- factor(rep(1:20, each=4))
year <- rep(2006:2009, 20)
year.c <- year - mean(year)
tmt <- sample(0:1, 20, replace = TRUE)[school]
math <- rnorm(80, 2 + tmt + .001*year + .0001*tmt*year, 1.5) +
rnorm(20)[school]
tmt <- factor(tmt)
dfr <- data.frame(math, school, tmt, year, year.c)
rm(math, school, year, tmt)
f1 <- lme(math ~ year*tmt, data = dfr, random=~1|school)
f2 <- update(f1, . ~ year.c*tmt)
summary(f1)$corFixed['year', '(Intercept)']
# [1] -0.9999997
summary(f2)$corFixed['year.c', '(Intercept)']
# [1] 0


A possibility is that the data are not of the expected classes.  What
does str(long) report?

hth,
Kingsford Jones







> I am puzzled, as I thought centering the time variable should eliminate, or
at least strongly reduce, this correlation.
>
> Any insights appreciated
>
> thanks
>
> Peter
>
>
> Peter L. Flom, PhD
> Statistical Consultant
> Website: www DOT peterflomconsulting DOT com
> Writing; http://www.associatedcontent.com/user/582880/peter_flom.html
> Twitter: ? @peterflom
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>