"Differences in teaching effectiveness was the single largest factor affecting student academic progress," Sanders said.
In other words, teachers had more of an impact on how much a student learned than any of the out-of-classroom factors like race or poverty. That's because, Sanders says, when two students enter the same teacher's class at the same level, they tend to exit at the same level, too.
"And I can't tell you how much that was considered to be a heresy!" Sanders said. The year was 1982.
Analyzing that Knox county data, Sanders created the Tennessee Value-added Assessment System (TVASS). "Value-added" data aims to measure how much students learn over time, and then use that measure to evaluate the teachers who taught them. As the United States slips in the educational rankings of developed nations, value-added data has come to shape the reform debate, and Tennessee has the deepest, richest value-added data of any state, because Tennessee was the first to collect the data statewide.
Sanders' method of data analysis created controversy, because, at the time, the thinking about education was largely shaped by cross-sectional data. Cross-sectional data looks at broad swaths of students--for example, poor kids as a group, black kids as a group, white kids as a group--and how they scored on tests they took in a single year. And this data showed that nationwide poor kids and black kids do worse on standardized tests than their middle-class and white counterparts. That was true in 1982 and it's still true today.
"If you do the analysis that way, that leaves very little left for schooling," Sanders said. "It basically says, 'You have got to fix the homes, you have got to fix the poverty if you are wanting to improve student achievement.'"
Sanders has traveled around the country explaining the way he prefers to collect and analyze student test-score data.
"Imagine a kid's physical growth curve they're so tall when they're two, and they're three, and they're four, and they're five," Sanders says. He always uses height as his analogy for learning, and this is the way he explains TVASS to lay audiences.
"If you've been testing each kid each year, you could plot a math growth curve, just like you could a height growth curve," Sanders said.
And Sanders says his data can be used to identify the best teachers, and the worst ones.
"If you were to find that most kids are having downturns in their math growth curve when they are in the classroom of a specific teacher, this is very strong, powerful evidence that something inappropriate in instruction is going on in that classroom," Sanders said.
Before Sanders came along there wasn't a lot of data, like Sanders', which tracked individual students' growth over time. And Sanders system has come to shape the education landscape, to a point.
"All kinds of folks still believe that to be fair to teachers and schools you've got to adjust for race and poverty and things like that," Sanders said. "I am adamant that that is exactly the wrong thing to do."
Sanders-esque value-added data is beginning to gain more and more traction nationally. Federal Race to the Top grants require that states keep track of it, and require that states use the data in their evaluation of teachers. But not everyone thinks that's a good idea.
Diane Ravitch is a professor at New York University and a former Assistant U.S. Secretary of Education.
"The root cause of low performance is not bad teachers. The root cause of low academic performance is poverty. It is just a fact that being hungry, homeless, sick, moving frequently and having all kinds of issues in your home-life affects attitude, motivation and attendance and all of those things affect performance," Ravitch said.
"These corporate reformers have unleashed this very mean-spirited approach into education," Ravitch continued. "I don't know that education is going to be improved simply by numbers."
At the moment the Tennessee legislature seems to be siding with Sanders--starting July 1st, 50 percent of teachers' evaluations in Tennessee will be based on student test scores. Thirty-five percent of that test data will be value-added.
And Sanders test-based data will be in for a test of its own.