Teacher bias by student SES or measurement error in ability? A cautionary tale for observational studies

Teachers act as judges of academic merit, but unfair evaluations beyond students' true abilities may perpetuate inequality based on socioeconomic status (SES). This article investigates if student SES biases teacher grades and track recommendations, after accounting for measurement error (test scores) and omitted variables (noncognitive skills) in ability—a widespread issue that can lead to overestimating teacher bias in observational studies. Using the German NEPS panel across elementary education, we seek to identify student ability through various cognitive (standardised test scores) and noncognitive (effort) measures, along with an instrumental variable (IV) design. We also examine heterogeneity across the ability distribution to test whether teacher bias is most pronounced among low performers. First, after approximating latent student ability with the IV approach and adjusting for noncognitive skills, the residual effect of student SES drops by over 40 %. This reduction indicates that estimates of teacher bias are substantially inflated when relying solely on snapshot test scores. Second, despite this reduction, the effect size of student SES remains considerable at 6 % of a standard deviation for GPA and five percentage points for track recommendations. Third, residual SES effects are disproportionately seen among low-to-average-performing students, suggesting that high-SES parents use compensatory strategies to influence teacher assessments. We discuss the theoretical and methodological implications of our findings for estimating and interpreting teacher bias as a mechanism of educational reproduction in observational research.