"Predictive Accuracy in Model Assessment"


Predictive Validation, Stochastic vs Algorithmic Models;
Approaches of Data Miners, Machine Learners and Statisticians


The predictive accuracy of a statistical model is a pertinent
consideration, even for applications where there are other
considerations.  Its use does not offer a way to avoid model
assumptions!  Simple forms of empirical assessment, such as are
widely advocated, can be misleading.  The issues are subtler than
indicated in most of the texts.  Assumptions are required that
relate a specific set of model predictions to the population that is
the target for predictions.  There are implications for the teaching
of data analysis.  Specific data analysis problems will illustrate
these points.