[R] interpreting an LME regression result...

[Andrew Robinson]

Hi Johan,

it's not clear to me that we have enough information to answer your
question reliably, but the following thoughts might be useful to you.

1) in your lme() model, all the coefficients are closer to zero than
   in your lm() model.

2) the change in Z is not very large relative to its standard error
   under either model.

3) one possible explanation for the change is that the contributions
   of individual observations to the parameter estimates are weighted
   differently under lme() than under lm(), and this can be noticeable
   when the data are unbalanced.  Specifically, there might be a large
   group of individuals who are influencing the parameter estimates
   strongly in lm(), but who are all from the same family.  As
   individuals, their contribution will lessen in lme().  If your data
   are unbalanced, then I would look here for the effect.

4) there might be other explanations for the change.

I hope that this helps,

Andrew

On Sat, Nov 10, 2007 at 12:58:21AM -0700, Johan Jackson wrote:
> Any help would be most appreciated.  (Don't make me get down on my
> hands and knees and beg for help, cause I'll do it!!) My boss has me
> learning R and doing nested regression with the report due Mon (Friday
> night statistics...fun. ). Anyway, here's my problem:
> 
> In a regression equation not accounting for the fact that people are
> nested in families, the result for Z variable is VERY strong (beta =
> -4511), but this result  is much weaker when I use lme and account for
> people nested in families (beta = -2613). I'm struggling with a verbal
> interpretation of this result. Is it because the effect of Z within
> families is not very strong, but between family variation is high?
> 
> 
> 
> 
> -----------------------------------------------------------------
> 
> If my R syntax would be helpful, it's below:
> 
> BIG EFFECT OF Z (using lm)
> summary(lm(Y ~ X + Z + age ,data=pharma.data ,na.action='na.omit'))
> 
>             Estimate Std. Error t value Pr(>|t|)
> (Intercept)   102.27      24.98    4.09    7e-05 ***
> X     -629.76     267.87   -2.35    0.020 *
> Z    -4511.96    2032.39   -2.22    0.028 *
> age             1.88       1.42    1.32    0.188
> 
> 
> SMALLER EFFECT OF Z (using lme)
> summary(lme(Y ~ X + Z +  age,  random = ~1|Family.ID,
> data=pharma.data, method="ML", na.action='na.omit'))
> 
>               Value Std.Error DF t-value p-value
> (Intercept)   103.9      20.0 85   5.200  0.0000
> X.   -417.3     179.3 85  -2.327  0.0223
> Z   -2613.0    1845.1 85  -1.416  0.1604
> age             1.3       1.2 85   1.126  0.2632
> 
> ______________________________________________
> 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.

-- 
Andrew Robinson  
Department of Mathematics and Statistics            Tel: +61-3-8344-9763
University of Melbourne, VIC 3010 Australia         Fax: +61-3-8344-4599
http://www.ms.unimelb.edu.au/~andrewpr
http://blogs.mbs.edu/fishing-in-the-bay/