Problem assigning nlsList call to an object

[Dave Auty]

Hello Forest-R users,

Following the modelling steps in Pinheiro and Bates (chapters 6 & 8),
I’m trying to assign an nlsList call to an object using the following
groupedData object (Silviscan data):

>MFA.new <- groupedData(Mean_MFA ~ Disc_Ring_Nr | Tree/Position, data=MFA.data) #12 trees with 5 positions (sample heights) in each tree

Call:

>MFA.lis<-nlsList(Mean_MFA~ b0/(1+exp(b1*Disc_Ring_Nr))+ b2, start=c(b0=30, b1=0.07, b2=9), data = MFA.new)

#But this gives the following (long) error message:

Error in nls(formula = formula, data = data, start = start, control =
control) :

step factor 0.000488281 reduced below 'minFactor' of 0.000976562

Error in nls(formula = formula, data = data, start = start, control =
control) :

number of iterations exceeded maximum of 50
...
Error in nls(formula = formula, data = data, start = start, control =
control) :

singular gradient


#However, if I don’t assign it to an object and just run:

>nlsList(Mean_MFA~ b0/(1+exp(b1*Disc_Ring_Nr))+ b2, start=c(b0=30, b1=0.07, b2=9), data = MFA.new)

The output is normal i.e. parameter estimates are given for all
positions in all 12 trees.

The reason I want to assign it to an object is so I can look at
boxplots of residuals and plot 95% confidence intervals for the
coefficients for the nls object versus the nlsList object, in order to
show the effect of taking the grouping structure into account:

#plot(MFA.lis, Tree ~ resid(.), abline=0, ylab="Tree")

#plot( intervals (MFA.lis) )

There is definitely a “Tree” effect in MFA data and most likely an
effect of “Position” (sample height), hence the grouping structure:
Mean_MFA ~ Disc_Ring_Nr | Tree/Position.

If anyone can spot what’s going wrong your help/comments would be much
appreciated.

Thanks,

Dave