[Dobbs]

> Another big problem is that private schools seem to have made more realistic predictions of their students' results than state schools

This might be partially true, but there are also issues in the algorithm that helped cause this skew. For example depending upon the size of the glass the teacher's estimates were adjusted more or less. Very small classes (~8 students IIRC) had no adjustment, small classes (<20 students IIRC) had moderate adjustment, large classes (>= 20 students IIRC) had heavy adjustment. State schools almost never have small classes, while private and public schools (public != state school) were much more likely to fall into the small or very small class size buckets.