In order to estimate specific effects of schooling and instruction, state-wide proficiency tests like VERA (VERgleichsArbeiten) often report not only the state’s mean score, but also an adjusted school score for a so-called fair comparison. The average performance of similar schools is estimated for the fair comparison with the aim of controlling for influences that lie outside the latitude of teachers and school management. Adjustment strategies include comparisons with subgroup mean scores, mean scores of similar existing schools, and model-based expected values. Using the large-scale assessments VERA 3 (over 2300 schools) and VERA 8 (over 1200 schools) in Baden-Württemberg from 2019 and 2021, adjustment strategies were compared with regard to their fairness (coefficient of determination R²) and standard errors. Variables for contextualization were gender, everyday language, migration background, socio-cultural capital, and, additionally in VERA 8, type of school and previous knowledge. Model-based expected values yielded the largest R² values and the smallest standard errors across all subjects. This advantage in terms of fairness is not to the disadvantage of the test efficiency, because the context variables are needed for every adjustment strategy. The results also show that the heterogeneity of test performances is considerably influenced by the social context. Against this backdrop, VERA featuring a fair comparison feedback can be a helpful element for school development.

Keywords
fair comparisons, Vergleichsarbeiten, value-added models, heterogeneity