I thought this would be about Bayesian vs. Frequentist, but it really isn't.<p>The sort of statistical problems he is talking about are basically predictive: what is the distribution of variable y if we know the values of variables x_i? How high can we expect ozone levels to be, given the previous weeks meteorological data?<p>His claim is that statistics is dominated by people who begin with a data model already decided upon. These models tend to be simplistic, but can be tuned to fit any data set by including enough parameters. His suggested replacement seems to involve decision trees and neural nets, what he calls "algorithmic modeling" (on the other hand, he doesn't seem to like MCMC, which is usually based on a data model, albeit a more sophisticated one.) This is not exactly a frequentist position, since it assumes that data can be modeled by an NN.<p>Anyways, the article is from 2001, and I don't think statistics people have abandoned data models because such models are quite in vogue among AI people.