Mechanical TA 2: Peer Grading with TA and Algorithmic Support

Hedayat Zarkoob, Kevin Leyton-Brown

Abstract: Mechanical TA 2 (MTA2) is an open-source, distributed peer grading system that boosts performance by leveraging both trusted TAs and computationally intensive algorithms. The system provides a unified platform for submission of assignments, grading by both peers and TAs, and reporting of feedback. It also supports dividing students into different pools based on their peer-grading prowess; mechanisms for automated calibration and spot checking; and the ability for students to appeal grades and to give feedback about individual reviews. Bayesian inference and mixed-integer programming algorithms perform interpretable aggregation of peer grades and estimate students' grading performance, providing feedback, incentivizing high-quality grading, and directing TA spot checks appropriately. Analysis of data from four offerings of a large undergraduate class provides empirical evidence of MTA2's effectiveness.