Future Software for Statistical Models

This is a proposal for discussion and eventually for joint work on software for statistical models. The goal is to bring together people interested in as many as possible of the major current research directions in this area, with the hope that we can design software that provides broader, better designed, and more useful support for those using statistical models.

The proposal is being made initially as part of the Omega project for statistical computing, a joint research effort aimed generally at producing open software for statistical computing. Anyone with an interest in the topic is invited to contribute suggestions and comments to the discussion.

A Mailing List

The mailing list omega-models@omegahat.net has been established to be forum for discussions of future directions in software for statistical models. Anyone interested in the topic is encouraged to join and contribute to the list. To join, send e-mail to omega-models-request@omegahat.net, with "subscribe" as the body of the message.

A Meeting in Vienna

On March 19-23, an informal meeting will be held in Vienna on The Future of Statistical Computing. Future directions for statistical models will be one of the sessions at a that meeting.

Some Topics

The following topics are among those for discussion. They are not intended to be at all restrictive, just some areas that seem particularly interesting.

New classes of models: One way software for statistical models has grown is by introducing new types of models, either extending our abilities into new areas or integrating various models under a new, general approach. New or newly extended types of models from the last five years or so include mixed-effects models, spatial models, and survival models.

New classes of models challenge us to provide good new software. In addition, bringing the new classes of models together with each other and with older classes of models can give users easier access.

Bayesian techniques: Great advances have been made in recent years in the application of Bayesian inference about statistical models. The advances have introduced both large areas of new computational techniques and new conceptual frameworks for specifying models, for example using graphical model descriptions.

Bringing these techniques together with other aspects of software for statistical models would be of great value to users.

Distributed computing for models: software on the internet and the use of languages that can take advantage of distributed software, such as Java, open many new opportunities. These range from enhanced user interfaces to algorithmic techniques such as high-level parallelism in fitting models.

New software organization for distributed computing will be needed. Along the way, we will want to incorporate exisiting software as much as we can.

Omega project for statistical computing
Last modified: Wed Feb 17 10:33:22 EST 1999