Coursera HCI peer grading dataset: 63,199 peer grades from 7,240 students across two course offerings

Product or Programme · All products and programmes
Evidence · 6 claims (6 for) · 1 study (1 associational), q2 · 0 of 1 report an effect size · 6 claims rest on one study

Description

A research dataset released by Chris Piech and colleagues containing 63,199 peer grades from two Coursera Human–Computer Interaction course offerings using calibrated peer grading.

Components

• Coursera HCI peer grading dataset: 63,199 peer grades from 7,240 students across two course offerings: Two consecutive Coursera offerings of the Human Computer Interaction course taught by Scott Klemmer, using a calibrated peer grading system. Per assignment, each student evaluated five randomly selected submissions (one a ground-truth submission) on a rubric and was evaluated by four classmates; the final score was the median of peer grades. The dataset comprises 3,607 students (HCI1) and 3,633 students (HCI2), 13,972 submissions, and 63,199 peer grades — described as "the largest peer grading networks analysed to date". (Chris Piech et al. (2013))

Claims

• Probabilistic peer grading models that estimate grader bias and reliability reduce RMSE against ground-truth grades by 33% and 31% on two HCI course offerings [+M]
• Higher-scoring students tend to be more reliable graders (except the very best), and students are least biased when grading peers with similar scores [+M]
• Including grader bias and reliability as features improves prediction of future course participation from AUC 0.93 to 0.98, capturing an engagement dimension missed by assignment grade [+M]
• Estimating each grader's bias alone accounts for 95% of the RMSE reduction, while modeling grader reliability contributes little with only four grades per student [+M]
• Grading time shows a sweet spot: students grading around 20 minutes (time z-score near -0.25) have significantly lower residual standard deviations than slow graders, while snap graders are unreliable and inflate grades [+M]
• The model's posterior confidence estimates are conservative: over 95% of predictions the model claims 90–95% confidence in are correct [+M]

Related Products and Programmes

-

Key Sources

• Chris Piech, Jonathan Huang, Zhenghao Chen, Chuong Do, Andrew Ng, Daphne Koller. (2013). Tuned Models of Peer Assessment in MOOCs. arXiv. https://arxiv.org/abs/1307.2579

Page metadata

Field	Value
type	product
id	coursera-hci-peer-grading-dataset-63199-peer-grades-from-7240-students-across-two-course-o
title	Coursera HCI peer grading dataset: 63,199 peer grades from 7,240 students across two course offerings
description	A research dataset released by Chris Piech and colleagues containing 63,199 peer grades from two Coursera Human–Computer Interaction course offerings using calibrated peer grading.
status	review
product_kind	dataset
generated	by process:settle-candidates on 2026-10-16
trust	unverified — no one has confirmed this page's content

Frontmatter is the page's machine-readable record — see the Schema & Guide.