[R-sig-ME] sigmoid residual distribution for a random effect

[Romain.Piault]

Hi everybody!

Using a linear mixed-effects model (library nlme), I have found a
sigmoïd pattern in the distribution of residuals for my random factor,
and I wonder whether some of you know potential causes of this
phenomenon.  

The model is as follow: 
model<-lme(MeanBand~Body_Condition+Brood_Size,random=~1|Nid)

where * MeanBand is a measure of the width of the sub-terminal black
band on the tail feathers of young kestrels
      * Body_Condition is a continuous measure of the amount of
resources nestlings received during their development
      * Brood_Size is a factor (with 2 levels) indicating whether the
brood where a nestling was raised was reduced or increased by one
nestling at hatching
      * Nid is the nest where a nestling was raised. Because there are 2
or more nestlings in one nest, Nest enters in the model as a random
factor.

Using qqnorm(model,~ranef(.,level=1)), I find that residuals do not
follow a line but describe a sigmoïd, revealing that the distribution of
the residuals for my random factor is not normal.

My question is therefore: do you know potential causes for such
distribution of residuals for a random effect?

Here is the output of the model:

 
Linear mixed-effects model fit by REML
 Data: NULL 
       AIC      BIC    logLik
  322.0317 332.9037 -156.0159

Random effects:
 Formula: ~1 | Nid
        (Intercept) Residual
StdDev:    1.523250 2.059382

Fixed effects: MeanBand ~ condition + Brood 
                Value Std.Error DF   t-value p-value
(Intercept) 19.050443 0.6243577 48 30.512066  0.0000
condition    0.093862 0.0263644 48  3.560170  0.0008
BroodRed     1.730359 0.8764315 17  1.974323  0.0648
 Correlation: 
          (Intr) condtn
condition  0.080       
BroodRed  -0.717 -0.109

Standardized Within-Group Residuals:
        Min          Q1         Med          Q3         Max 
-1.81195733 -0.68685432 -0.02572958  0.65085007  1.77797726 

Number of Observations: 68
Number of Groups: 19 

Furthermore, I wonder whether the high correlation shown in the
"Correlation" argument is indicative of any particular problem.

Thanking you in advance for your answers!

Best regards,

Romain