Bayesian Modeling and Data Analysis in S-plus

Period of Performance: 04/17/2002 - 06/30/2004

$379K

Phase 2 SBIR

Recipient Firm

Insightful Corporation
INSIGHTFUL CORPORATION, 1700 WESTLAKE AVE N, STE 500
Seattle, WA 98109
Principal Investigator

Abstract

The ultimate goal of this project is to provide an extensive suite of Bayesian statistical software tools, utilizing a fast and effective Markov chain Monte Carlo (MCMC) computation engine. The overall implementation will be in the 5-PLUS object-oriented language and system for statistical modeling and data analysis, and the MCMC engine will be implemented in C++, with an efficient interface to 5-PLUS. The implementation will emphasize an ease-of-use paradigm that strongly encourages routine use of Bayesian methods as well as research-oriented exploration for new Statistical techniques by 5-PLUS. Bayesian methods will be developed not only for the most widely used statistical models such as hierarchical models based on exponential family, and linear regression models, but also for more complicated models such as generalized linear mixed models, missing data models and models for robust inference. A large percentage of statisticians in the United States are employed in biostatistics and allied "bio" industries, and a considerable amount of statistical education and research occurs in medical and health related fields. The availability of a broad range of Bayesian statistical methods in a commercially viable data analysis product such as 5-PLUS, will provide an important service to these industries, and to the research and educational needs by supporting and advancing the emerging paradigm of Bayesian modeling and data analysis. PROPOSED COMMERCIAL APPLICATIONS: There is no commercial software offering contemporary Bayesian statistical techniques by MCMC computation. By offering an extensive suite of Bayesian modeling and data analysis toolkit ranging from simple hierarchical models based on exponential families, linear regression models as well as more sophisticated models such as generalized linear mixed models, missing data models and models for robust estimation and interference, we will garner a significant market edge relative to competitors.