Could you explain why all statisticians involved in education are unable to include this specific "huge statistical[ly] significant difference" in a predictive model?
They can and do. However, when it comes to teacher pay and student performance such things are politically untenable. No Child Left Behind does not mean except for when your peer group is full of truants.
The best evidence for this is actually from tracking randomly assigned edge cases. Often good schools will accept X numbers of students from another area and when they pull randomly from the pool of available students it's not hard to track what's going on and compare crossovers performance with students from each area that stayed in that area.
They can and do. However, when it comes to teacher pay and student performance such things are politically untenable. No Child Left Behind does not mean except for when your peer group is full of truants.
The best evidence for this is actually from tracking randomly assigned edge cases. Often good schools will accept X numbers of students from another area and when they pull randomly from the pool of available students it's not hard to track what's going on and compare crossovers performance with students from each area that stayed in that area.