Huh? Value Added Modelling explicitly takes home environment into account, at least as much as is possible. (I.e., it usually controls for income, race, whether parents are married, and similar statistically available facts.)
The teacher discussed in this article was not a teacher of students with poor home environment. Home environment and past test results suggested her students should perform at the 97'th percentile. They actually scored considerably lower, at the 89'th.
What the the statistical chances that a particular group of students with those backgrounds would turn out to yield a performance at the 89th percentile? I've no idea, but say it's 1% for the sake of argument. That would mean that 1% of similar classes with averagely competent teachers would achieve that score, all other things being equal. That's many thousands of classes at the national level.
It's like the argument made in a UK court not long ago that someone must be guilty because the chances of the evidence being a coincidence were millions to one. Of course in a population of millions such a coincidence becomes almost inevitable.
I've no idea if this teacher was good or bad, but that's something that should be determined by expert, informed evaluation not mindless statistical witch hunt.
All metrics have errors. If we fire 99 bad teachers and 1 good one, we've improved education. The goal here is to improve education, not provide a permanent job for education grads.
I've no idea if this teacher was good or bad, but that's something that should be determined by expert, informed evaluation...
Why do you believe this method has an error rate lower than VAM? I agree the error rate is harder to characterize, but that isn't the same thing.
The teacher discussed in this article was not a teacher of students with poor home environment. Home environment and past test results suggested her students should perform at the 97'th percentile. They actually scored considerably lower, at the 89'th.