[R--gR] R--gR: Model specification in gR

[DED (David George Edwards)]

Some ideas about specifying models in gR.

Quick Win
----------
1. For graphical models (ie where the graph is sufficient to define the
model, the distributional part being implicit), just use the graph as it is
(will be) defined in R (a clique list, for example). For high dimensional
models, model formulae get so long as to be completely unmanageable. 
2. For DAGs and other types of model constructed from univariate response
models, use  Wilkinson and Roger's notation for each vertex, as Giovanni
suggests. 

Slow Haul
---------
3. For the other (non-graphical, non-univariate response) models we need to
be careful. It would be helpful if model formula handling in R could be
'objectified' and so made accessible to the programmer.   

Regards
David