An XML format for block designs.

[John P. Morgan]

Nam-Ky Nguyen wrote ...

>
>My concern about an on-line database of (block) designs is that we will
>eventually encourage the experimenters to adopt the old approach of
>designing experiments, i.e. experiment for the design (in the
>database/design catalogues). In the computer age, we should encourage
>experimenters to adopt the new approach of designing experiments, i.e.
>design for the experiment (and do not experiment for the design). This
>approach is made feasible by the use of DOE software such as Cycdesign
>(http://www.ffp.csiro.au/tigr/software/cycdesign/cycdes.htm) or Gendex
>(http://www.designcomputing.net/gendex/). Hopefully, one day the
>mentioned software will follow the open source model so that the second
>approach of designing experiments will be promoted.
> 

The corresponding concern with the "new" approach is that programs generating 
designs by search and/or cyclic methods, while usually finding highly efficient 
designs, do not assure an optimal design. The even newer approach is to harness 
computing power in conjunction with optimal design theory to determine optimal 
designs for as great a number of settings as is possible (this number will 
always be growing), then catalog them on-line. They become permanently available 
and of course supercede the need for search software. The search software will 
still have its role for settings where the newer approach has still not found a 
definitive solution. But even then, any design produced by a search software can 
be cataloged in the database and labelled as such, so that the search need not 
be repeated.

Each design in this database will report whether or not it is a known optimum
for each of the standard criteria. What's more, since values will be reported on 
a great number of optimality criteria, an experimenter can search the database 
for a design which optimizes her/his own weighting of said criteria. The 
efficacy of such a search will of course be directly related to the completeness 
of the database, which will also be growing through time.

John P Morgan
Department of Statistics
Virginia Tech
Blacksburg, VA 24061-0439
540-231-9701 (office)
540-231-3863 (fax)
jpmorgan at stat.vt.edu