[R] SAMM package for mixed models

[kwright@eskimo.com]

First, a disclaimer.  I am not affiliatied with the SAMM package.  I am
only a user of the package, but I have been contacted (off list) by people
requesting information about SAMM and so I am posting this information
here.

SAMM is software for fitting mixed models.  Versions are available for
both S-Plus and R.  More information and downloads of the software (and
manual) are available here:
   http://www.dpi.qld.gov.au/fieldcrops/14715.html
This URL appears not to be indexed by Google, which is part of the reason
I am posting this.


Here are some personal, random notes about the package:

SAMM is commercial software and requires a (non-free) license.  You can
test the software for free for 30 days.

SAMM is short for Spatial Analysis Mixed Models.

SAMM estimates variance components under a general linear mixed model by
REML.  In particular, the Average Information REML algorithm is used along
with a sparse-matrix representation of matrices.

For some types of problems, I have seen SAMM converge 100 to 1000 times
faster than PROC MIXED or lme, which makes analysis of large datasets /
complex models possible (sometimes in nearly real-time).  'Amazing' is a
word that comes to mind.  (Side note: I have heard rumors that SAS has
hired a developer to look at the Average Information REML technique...)

SAMM can fit two-dimensional spatial structures (such as AR1xAR1) and can
plot two-dimensional variograms.

The 'engine' for the mixed-models in SAMM is the same one used by Genstat
and ASREML.

Like all mixed-models software, SAMM has quirks such as convergence
issues, degrees of freedom, model-specification, etc.

The user community is small, so resources like email lists are limited.

Some types of linear models can be fit using either lme/lme4 or SAMM. 
There are some big differences between SAMM and lme, however (cost,
graphics, support, community, types of tests of fixed effects, etc.).

Using both SAMM and lme to fit a model can be an experience that is
tedious/painful but ultimately rewarding in a deeper understanding of the
modelling process.

SAMM has its origins in ASREML, which comes from a plant-breeding
background.  Although SAMM can be a general-purpose package, the focus is
on evaluation of field experiments.  For that purpose, it is an excellent
tool.


Kevin Wright