[R-sig-ME] nlme question...

[Doran, Harold]

Jeff

I still think something is wrong here, but maybe it is just me and someone else might see it differently. First off, your model notation is y = XB + e which is not a mixed model. Second, if k is a parameter, then it cannot be in your *known* model matrix. Your model matrix has two columns and your parameter vector has two unknowns (b0 and b1).

If your model matrix were (or something like it), I can see how you would estimate b0 and b1 as you would estimate those from the observed data. But, upon what data is k estimated?

 X= |  1   0  |
      |  1   1  |
      |  1  2  |
      |  1  3  |

If I am understanding your problem correctly (and it seems I may not be), your design matrix has two columns of known values, but your parameter vector has 3 unknowns. So, how can this be?

Also, are you hoping to estimate a mixed model or a least squares model?

-----Original Message-----
From: Jeffrey Harring [mailto:harring at umd.edu] 
Sent: Monday, August 30, 2010 3:23 PM
To: Doran, Harold
Subject: Re: [R-sig-ME] nlme question...

  Harold,

Thanks for responding. Yes, k in the design matrix is a parameter to be 
estimated. If you think lmer is a viable option for this type of problem 
I will certainly look into it.

Jeff

On 8/30/2010 2:31 PM, Doran, Harold wrote:
> This is confusing. First, how is your model non-linear? It looks like it is linear in the parameters. A design matrix is known and parameters are estimated. So, is k known or is it part of the parameters to be estimated.
>
> Moreover, can you use lmer and not lme since nlme is not supported much anymore
>
> -----Original Message-----
> From: r-sig-mixed-models-bounces at r-project.org [mailto:r-sig-mixed-models-bounces at r-project.org] On Behalf Of Jeffrey Harring
> Sent: Monday, August 30, 2010 1:48 PM
> To: r-sig-mixed-models at r-project.org
> Subject: [R-sig-ME] nlme question...
>
>    Hi all,
>
> Can the algorithm in nlme handle a nonlinear function that is written as
> a design matrix and linear coefficients like the following
>
> The model is a nonlinear growth model with five time points: y = X*b +
> e, where design matrix X is defined as
>
> X= |  1   0  |
>      |  1   1  |
>      |  1   k  |
>      |  1  2k  |
>      |  1  3k  |
>
> and parameter vector b = (b0, b1). And where "k" is a parameter to be
> estimated. Of course I also want to estimate the intercept (b0), slope
> (b1). Error variances (e) with 5 free parameters and random effects
> covariance matrix  (2x2: for b0 and b1).
>
> If anyone has concrete suggestions I would love to hear from you.
>
> Thanks for your consideration,
> Jeff
>
>
>
>

-- 
**********************************************************
Jeffrey R. Harring, Assistant Professor
Department of Measurement, Statistics&  Evaluation (EDMS)
1230 Benjamin Building
University of Maryland
College Park, MD 20742-1115

Phone: 	301.405.3630
Fax: 	301.314.9245
Email: 	harring at umd.edu
Web:  	http://www.education.umd.edu/EDMS/fac/Harring/webpage.html