Its quite straightforward.
Most performance variation among students is indeed entirely due to the nature of the student. If a school has students that are programmed by family and tradition to be excellent (100% Chinese say) it takes extremely bad teaching indeed for them to produce bad test scores.
If one wants to rate schools, in terms of what value they add to educational results, you have to equalize for the students. You will otherwise end up rating a stupidly run school employing an entirely drunken crew of ignorant teachers, but 100% Chinese students, above a tight ship operated by a totally disciplined crew, with 100% black children of single parents whose siblings are all gangsters.
The California system was to assign schools to a batch of “Similar Schools”, based on a demographic regression model that tried (quite well I thought) to extract all the student-based performance factors.
It doesn’t make sense to me yet.
What is being compared (exactly), then how is it “normalized”.
What is the reference? I get suspicious when a reference isn’t needed to grade.