Thomas Lumley
DSC 2001: Cooperation between programs (and programmers)
********************************************************


The second international conference on Distributed Statistical
Computing was held in March 2001 in Wien, Austria.  Hosted by the
Center for Computational Intelligence of the Vienna University of
Technology (Technische Universit\"at Wien), it was substantially
larger than the first Vienna DSC conference held two years earlier.
There were about 30 presentations over the three days of the
conference, with speakers from universities, industrial research
groups, and software companies in twelve countries. Despite the
programmed sessions' running about 30\% over time there was lively
discussion over coffee and lunch and in during the conference's social
program.

DSC 2001 emphasized the value of high quality open source software as
a statistical research platform but also for the ultimate goal of
delivering modern statistical methods to those with little knowledge
of (or interest in) statistical computing.

One clear theme of the conference was the leverage to be gained by
joining together specialized tools. Statistical computing environments
(most speakers preferring dialects of the S language) need to handle
large data sets: databases can do this. They should provide
interactive dynamic graphics for displaying complex, high-dimensional
data: programs such as GGobi and Orca do this. The must supply
familiar interfaces, from menus and dialog boxes to web pages and
spreadsheets, and again there are both free and proprietary programs
that do this.

Now that modern programming languages, and applications from Excel to
Oracle, make much of their functionality available to external callers
it is unnecessary for statistical environments to duplicate these
features. The traditional retort to unnecessarily complicated feature
requests, based on Dennis Ritchie's ``If you want PL/I you know where
to find it.'', becomes much more helpful when you can link ``it'' in to
what you currently use.  As always, this works better in theory than
in practice, but there is now real hope that the problems can be
solved.

Another theme of the conference was the value of object-based
programming, partly in inter-process communication but also in static
and dynamic graphics, documentation, and statistical modelling. This
is hardly new in the computer world, but statisticians have not
traditionally thought in OOP. Some of these uses, such as a more
flexible notation for statistical models, will require more
understanding of the benefits of an object-based approach by the wider
statistical community. Other uses will be invisible to users, apart
from the increased convenience they will produce.

Literate Programming, and its new cousin, Literate Data Analysis, made
an appearance, and it may be that tools will soon (finally) be
available to make these more than intellectually interesting. They are
among the many areas that may benefit from the use of XML and XSL to
store and manipulate text, code, and data.

These proceedings cannot capture the full content of DSC 2001, but we
hope they will be a useful resource for statistical computing users
and researchers.