Value-Added Models, Student Sorting, and the Measurement of Teacher Quality

• Activity Number: 218
• Type: Invited
• Date/Time: Monday, August 1, 2011 : 2:00 PM to 3:50 PM
• Sponsor: General Methodology
• Author(s): Jesse Rothstein*+
• Companies: University of California at Berkeley
• Keywords: value added ; teacher quality

Abstract

A great deal of effort has gone into correctly specifying the educational production function for use in value added models (VAMs). But the intended purpose of VAMs is to estimate the causal effect of having a particular teacher. The assumptions under which a VAM identifies the teacher's causal effect are very strong, and depend as much on the process by which students are assigned to classrooms as on the specifics of the educational production function. Each of the common models relies on an implicit assumption about the assignment process that is falsified by panel data on actual student achievement histories - each model indicates that current teachers have "effects" on students' past test scores. Thus, VAM-based systems confound teacher effectiveness with differences in student assignments, with potentially important consequences for the identification of effective and ineffective teachers. Moreover, whatever effects VAMs identify appear to fade away extremely quickly, raising important questions about the value of measured achievement gains.