Sending Teacher Value-Added into Tailspin: A Simulation Study of Measurement Error and Nonrandom Sorting

As a result of federal efforts to hold individual teachers accountable for student learning based on the standardized test results of their students (Race to the Top, U.S. Department of Education, 2009), the use of value-added models of teacher performance as a component of teacher evaluation has been spreading rapidly. Often researchers and policymakers alike feel comfortable with imprecision in the models because they are usually employed primarily to identify teachers in the high and low ends of the distribution in effectiveness. Using these models to reward teachers at the top and sanction those at the bottom seems a relatively attainable goal. However, the question of whether we can “get the tails of the distribution right” has not been adequately resolved in the research literature. Related to this is the issue of measurement error in student test scores. This paper addresses the issue of measurement error and its potential to introduce bias in the tails of the effectiveness distribution and sound a warning note that identifying teachers at the top and bottom of the scale may be