Models for Data

Abstract

We point out that models based on probability theory, and the statistical techniques derived from them, have limited applicability, at least in exploratory multivariate situations. Prior knowledge, if avail- able, must be incorporated into the analysis to yield greater stability. If prior knowledge is not available, however, it must not be invented. This applies to both the structural and the replication framework aspects of a model. The methods of gauging and stability analysis are introduced as alternatives. They use the notion of a technique as the pivot of data anal- ysis, not that of a model. Homogeneity analysis is used as an example.