Unfolding the Network of Peer Grades: A Latent Variable Approach

• DOI: 10.48550/arxiv.2410.14296
• Publication Date: 2024-10-18
• Authors: Giuseppe Mignemi; Yunxiao Chen; Irini Moustaki
• Keywords: FOS: Computer and information sciences; Applications (stat.AP); Statistics - Applications

Peer grading is an educational system in which students assess each other's work. It commonly applied under Massive Open Online Course (MOOC) and offline classroom settings. With this system, instructors receive a reduced workload, enhance their understanding of course materials by others' data have complex dependence structure, for all the peer grades may be dependent. This structure due to network grading, where student can viewed as vertex network, grade serves edge connecting one grader another examinee. paper introduces latent variable model framework analyzing develops fully Bayesian procedure its statistical inference. has several advantages. First, when aggregating multiple grades, average score other simple summary statistics fail account effects and, thus, biased. The proposed approach produces more accurate parameter estimates therefore, aggregated modeling heterogeneous behavior with variables. Second, method provides way student's performance grader, used identify pool reliable graders or generate feedback help improve grading. Third, our further provide insights into answering questions such whether who performs better coursework also tends grader. Finally, thanks approach, uncertainty quantification straightforward inferring student-specific variables well structural parameters model. two real-world datasets.