[R] How to set up starting values in lme

[Dieter Menne]

zhongmiao wang <zhongmiao <at> gmail.com> writes:

> 
> I keep getting the error message "Error in lme.formula(fixed = Score ~
> factor(time) - 1, data = ldata, random = list(dumid = mat),  :
>         iteration limit reached without convergence (9)"
> Is there a way to set up starting values, especially the starting
> values for the variance covariance of random effect? Thanks in
> advance!

For models that simple in the fixed part, lme is rather robust when the random 
part is correctly specified. Using starting value IS important in nlme, but in 
this case I assume that your externally computed "mat" is ill-conditioned. Did 
you try one of the supplied versions (e.g. pdDiag(~....)) first? These cover 
quite a large ground, and if you are not, better try these first.

Dieter