I should clarify the point I made in my piece on approaches for assessing change.
Most notably, it is not about the substance of Prof. Hake’s work, which I agree is important-—important in taking prior knowledge into account, important in the comparisons he makes, important in implications for instruction, and important in using *a* method (namely <g>) that mitigates certain problems with ceiling effects with test scores.
My point is, perhaps, more in the nature of a "sociology of science" comment concerning a methodological choice. The substantive aspect of the work would be equally valuable, and provide the same essential results, with an alternative (and in the relevant aspects analogous) method for dealing with those effects. The reason that psychometricians today would be prone to use an alternative such as IRT rather than <g> is that the alternative accomplishes the same purpose but is grounded in the methodological web of a generative program of research. Specifically, it is founded on probability-based inference, is invariant (with the appropriate qualifications) across testing situations, and is extensible to broader models for the nature of knowledge, learning, and change.