[R] understanding the verbose output in nlme

[Greg Distiller]

Thanks for the reply...will give some thought to your suggestion about 
stepping through the function.
I have read the Pinheiro and Bates book, in fact its my primary reference 
for getting into the nonlinear mixed models with R.
Lastly wrt the bit under subjects 1-6, I had thought about it being an 
estimated random effect but in this model there are 2 random effects so not 
sure if that holds...
thanks again...

----- Original Message ----- 
From: "Spencer Graves" <spencer.graves at pdf.com>
To: "Greg Distiller" <gregd at stats.uct.ac.za>
Cc: <r-help at stat.math.ethz.ch>
Sent: Sunday, June 04, 2006 8:49 PM
Subject: Re: [R] understanding the verbose output in nlme


>   I don't know, but if it were my question, I think I could find
> out by making local copies of the functions involved and stepping
> through the algorithm line by line using "debug" (see, e.g.,
> "http://finzi.psych.upenn.edu/R/Rhelp02a/archive/68215.html").
>
>   Have you read Pinheiro and Bates (2000) Mixed-Effects Models
> in S and S-Plus?  If no, I encourage you to do so.  Over the past 4
> years or so, I've probably spent more time with this book and referred 
> more people to it than any other.  Doug Bates is a leading original 
> contributor in this area, and I believe you will find this book well worth 
> your money and  your time.
>
>   Regarding "the numbers under subjectno1-6", I'm guessing that
> these may be the current estimates of the random effects for the first 6 
> of the 103 subjects.  The purpose of "verbose" is NOT to dump everything
> but only enough to help you evaluate whether the algorithm seems to be
> converging.
>
>   hope this helps.
>   Spencer Graves
>
> Greg Distiller wrote:
>> Hi
>> I have found some postings referring to the fact that one can try and 
>> understand why a particular model is failing to solve/converge from the 
>> verbose output one can generate when fitting a nonlinear mixed model. I 
>> am trying to understand this output and have not been able to find out 
>> much:
>>
>> **Iteration 1
>> LME step: Loglik: -237.4517 , nlm iterations: 22
>> reStruct  parameters:
>>   subjectno1   subjectno2   subjectno3   subjectno4   subjectno5 
>> subjectno6
>>  -0.87239181   2.75772772  -0.72892919 -10.36636391   0.55290322 
>> 0.09878685
>>
>> PNLS step: RSS =  60.50164
>>  fixed effects:2.59129  0.00741764  0.57155
>>  iterations: 7
>>
>> Convergence:
>>    fixed reStruct
>> 5.740688 2.159285
>>
>> I know that the Loglik must refer to the value of the log likelihood 
>> function, that the values after "fixed effects" are the parameter 
>> estimates, and that the bit after Convergence obviously has something to 
>> so with the convergence criteria for the fixed effects and the random 
>> effects structure. I did manage to find a posting where somebody said 
>> that the restruct parameter is the log of the relative precision of the 
>> random effects? The one thing that is a bit confusing to me is that it 
>> appears as if the fixed effects convergence must be zero (or close to it) 
>> as one would expect but in one of my converged models the output showed a 
>> restruct value of 0.72 ?
>>
>>
>>
>> Then I have no idea what the numbers under subjectno1-6 are, especially 
>> as I have 103 subjects in the data!
>>
>>
>>
>> Can anyone help shed some light on this output and how it can be used to 
>> diagnose issues with a model?
>>
>>
>>
>> Many thanks
>>
>>
>>
>> Greg
>>
>> ______________________________________________
>> R-help at stat.math.ethz.ch mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-help
>> PLEASE do read the posting guide! 
>> http://www.R-project.org/posting-guide.html
>
>