Dan Rice is President of Rice Analytics. Dan founded this business in early 1996 as a sole proprietorship, but it was incorporated into its current structure in 2006. Prior to 1996, he was an assistant professor at the University of California-Irvine and the University of Southern California. Dan has almost 25 years of research project and advanced statistical modeling experience for major organizations that include the National Institute on Aging, Eli Lilly, Anheuser-Busch, Sears Portrait Studios, Hewlett-Packard, UBS, and Bank of America. He has a Ph.D. from the University of New Hampshire in Cognitive Neuroscience and Postdoctoral training in Applied Statistics from the University of California-Irvine. Dan is a previous recipient of an Individual National Research Service Award from the National Institutes of Health and is author of more than 20 publications, many of which are in conference proceedings and peer-reviewed journals in cognitive neuroscience and statistics. Dan was the lead author of two peer reviewed papers in the 1990's that together presented the first explanation based upon quasi-experimental evidence involving temporal lobe brain imaging and recent memory deficit correlations that the causal process in Alzheimer's disease must have a preclinical stage of at least 10 years. This idea was controversial at the time as many physicians simply did not accept such a long preclinical period for Alzheimer's, but has now become widely accepted in the biomedical research community. In 2010-2011, the National Institute on Aging issued new diagnostic guidelines for Alzheimer's disease to incorporate this long preclinical phase. This is because a large number of brain imaging studies have now replicated and extended the original finding of "Alzheimer-like" temporal lobe brain abnormalities and associated recent memory deficits in non-demented elderly (Rice et al., Journals of Gerontology, 1991) and the predictive explanatory models from such studies have also now consistently projected that Alzheimer's disease must have at least a 10 year average preclinical phase .
Since the mid 1990's and as a part of his consultancy activities, Dan has been interested in the development of a reduced error form of regression - Reduced Error Logistic Regression (RELR) - to automate the discovery of similar putative causal explanatory models for business and science applications which do not have easy access to experimental data. As an automated means to explanatory predictive modeling, RELR reflects his long standing interest in overcoming problems in standard statistical modeling and machine learning related to error, instability, multicollinearity, dimensionality, and lack of parsimony. In the past few years, he has given invited addresses on RELR at the SAS M2007, Shop America, and Classification Society conferences, and contributed papers on this same topic at Psychometric Society, Joint Statistical Meetings, and SAS MWSUG. RELR has been used to build explanatory models in applications ranging from clinical and pharmaceutical research to marketing and marketing research to finance and credit scoring to text mining and natural language processing and Dan's career spans all of these applications.