Stephen Downes

This is a great article, 15 years old, but still popular, as reported in today's Data Science Newsletter. It argues at length and in considerable depth that explanation is not the same as prediction. To explain is to have a model that describes the causes of an event; to predict is to use that model to, well, predict future events. Galit Shmueli shows how they are not the same across four dimensions: causation-association, theory-data, retrospective-prospective, and bias-variance. People who read education research articles will find themselves nodding at a lot of what's discussed here. Though still a heady read, you can get the essence of the paper by reading sections 1 and 4. Section 1 describes the four dimensions. Section 2 outlines how the distinction plays out in each stage of the statistical modelling process, while Section 3 offers two examples of the distinction. Section 4 contains conclusions and recommendations.