As a more general observation, it's hard to comment on this row without having some idea about the local office politics. These, of course, tend to be dominated by jockeying for power/status/prestige and not by discussions of effective teaching methods.

The political fault lines he's describing exist at every flagship state public university, and so I'm not at all surprised to hear that a quake has happened along those lines at Berkeley.

But also most performers have a flair for the dramatic, and Coward's excellent student reviews seem to come in part from his talent at performance. So his interpretations are likely massaged in some form, and the object-level claims could be easily exaggerated.

But he claims that he and the department differ on a fairly simple statistical claim--how to estimate the effect of his courses on students' future performance. The related email correspondence is here, and well worth reading, both to judge that specific matter yourself, and get a sense of how defensive Coward can seem. (He's definitely escalating emotionally, but justifiably is harder to know.)

My summary: In a report, Stark, a statistician, makes a three-way comparison between the three 1A classes (two of which were taught by Coward), and finds that they are not statistically significantly different. Coward asks why a three-way comparison is done, instead of comparing the Coward group to the non-Coward group. Stark replies that since the students were assigned non-randomly, we can't separate the direct effect of instruction from any confounding variables.

Which is, of course, correct--it's very likely that the students who got into the class with the instructor widely believed to be superior by students are more competent than the students who didn't, and so should be expected to do better in future classes--but an equally valid point against the three-way comparison.

This seems strange to me. Engineering departments should have faculty that are perfectly capable of teaching the math and physics that their students will need. And this happens to a limited extent. For example, at UC Berkeley, the computer science department offers its own discrete math course instead of telling students to take the roughly equivalent discrete math course offered by the math department. Is there something preventing this from becoming more widespread?

This is good for the engineering department but bad for the math department--instead of eight professors all teaching one non-major course each, you could have two lecturers teaching four non-major courses each, with the attendant loss of prestige, funding, and political clout for the department.

Would the university really stop subsidizing Math and Physics dept's to the same degree if it weren't for their "service" obligations? I don't think this is right - I think the administration is broadly happy with the status quo, in terms of prestige, etc. If the department has two full-time lecturers, the only consequence is that they will also hire a bunch of full-time researchers to balance things out. By contrast, the "service" role is probably a lot more important politically for departments which teach lots of fluffy GenEd courses.

favoring his boss's boss (i.e. the university as a whole) instead of his boss (the math department)

Favoring the "goals" of the organization as an abstraction over the actual punishment/reward structure of the living, breathing, and interacting cogs of the organization.

I've come to look at bureaucracies as parasites on the host organization.

Aligning the goals of the bureaucracy with the goals of the org is actually a very hard, very interesting, and very important problem.