Management of regression-model data

摘要

Statisticians are becoming increasingly ambitious in the size and complexity of the data populations they are analyzing. Database support is becoming increasingly important to them, and in particular, they are demanding archiving of analysis results for possible later use, a facility not generally provided in statistical packages. We discuss the key database issues of managing regression-data models, one such analysis result, and we propose data structures including multiple partial indexes to support model-inference methods. We suggest a role for inference methods analogous to those used in artificial intelligence (especially inheritance inferences), to provide a generalization of the techniques of analysis of variance and covariance. A key feature of our approach is the quantification of inferences to resolve ‘multiple inheritance’ conflicts between models applicable to a new situation.