Co-Principal Investigator: Zeyu Xu

Purpose: With the added emphasis on incorporating evidence of student learning into teacher evaluations, many states are using value-added modeling as part of their approach to evaluating teachers. The evidence typically comes in the form of year-to-year changes in student scores on annual standardized tests administered in spring, with the spring teacher assumed to be the only teacher in the intervening year. This approach misattributes a teacher's value-added for at least three reasons. First, there is still some instruction by the previous year's teacher after the test. Second, students may or may not have had a teacher during the summer. Third, students could have transferred between classrooms during the year. This study seeks to quantify the magnitude of the misattribution, to examine policy effects of it, and to develop a modeling approach that will properly account for it.

Project Activities: The research team has 7 years of test data, teacher location information, and student enrollment data for all of Florida's public schools. They will use these data to conduct a battery of value-added analyses using models that account differently for teacher exposure and that vary in whether they include other student- and school-level data. The researchers will use the results of those analyses to illustrate differences in the personnel decisions which would result from using the different modeling approaches.

Products: Products from this study include evidence on the differences in statistical results and personnel outcomes across various value-added modeling approaches. Peer-reviewed publications will also be produced.

Structured Abstract

Setting: The data come from students enrolled in public schools in Florida.

Sample: The dataset contains records for all K–12 students in Florida public schools from the 2002–03 through 2008–09 academic years.

Intervention: The results of the study will inform the conversation about the application of value-added modeling to teacher evaluation.

Research Design and Methods: The researchers are using secondary data from Florida that they already have in-house to examine the differences among multiple approaches to attributing student-teacher contact time in value-added models. They will carry out a series of value-added analyses that vary in terms of how they account for teacher exposure, and in the data they include in the model. The team will then compare the results from these different models.

Comparison Condition: The typical approach to teacher attribution in value-added modeling (i.e., assuming the current teacher has been the only teacher since the previous year's test) can be thought of as the comparison condition, as it is the "business-as-usual" value-added model.

Key Measures: The data from Florida contain the aforementioned information on student enrollment and teacher assignments, along with student demographics. The data contain test scores on the Florida Curriculum Assessment Test for all grade levels during all school years included in the dataset and Stanford-9 or Stanford-10 scores for grades 3–10 through 2008. Data on some teacher-level variables, such as certification status and experience, are also available to the researchers.

Data Analytic Strategy: Value-added models assuming imperfect persistence of learning will be used to model the various approaches to attributing teacher value-added. Each attribution approach will be run once without student/family characteristics or school-level variables and once with those variables in the model. The results of the models will be compared in terms of differences in teachers who would be laid off, with analyses of whether certain approaches favor or punish certain subgroups of teachers (e.g., those in low-performing schools, those in schools in low-income areas).

Products and Publications

Journal article, monograph, or newsletter

Özek, U., and Xu, Z. (2017). Misattribution of Teacher Value-Added. Education Finance and Policy , 1–57.