[R] standardized residuals (random effects) using nlme and ranef

Doran, Harold HDoran at air.org
Mon Jul 31 17:20:32 CEST 2006

 > To sum up, I can't figure out how MLWin calculates the 
> standardized residuals. But I understand that this is not a 
> question for the R list. 
> Nevertheless, it would help if someone could point me to some 
> arguments why not to use them and stick to the results 
> obtainable by ranef().

Hi Dirk:

Well, it is interesting that mlWin and lmer generate the same exact
random effects but different results for the standardized random
effects. Now, my prior post showed exactly how lme calculates the
standardized random effects, so this is now totally transparent.

What I would recommend you do is this

1) Calculate the unstandardized random effects in mlWin
2) Calculate the standardized random effects in mlWin
3) Divide the mlWin unstandarized random effects by the standarized
random effects. This will show what denominator is used to standardize
the random effects.

Basically, just replicate what I did in my prior email using the mlWin

Clearly, since R and mlWin generate the exact same unstandardized random
effects, but different standardized results, the difference lies in what
denominator is used. In R, we know what denominator is used and it is
the standard deviation of the variance component for the random effect.

If you do the steps above, you will solve for the denominator used in
mlWins calculation. This will solve your problem.


More information about the R-help mailing list