Understanding the what and when of peer feedback benefits for performance and transfer

• Published in: Computers in human behavior 2023-10, Vol.147, p.107857, Article 107857
• Main Authors: Yu, Qiuchen, Schunn, Christian D.
• Format: Article
• Language: English
• Subjects: Heterogeneity; Knowledge transfer; Learning; Peer feedback
• ISSN: 0747-5632; 1873-7692
• DOI: 10.1016/j.chb.2023.107857

The impact of online peer feedback on learning outcomes has been well-established in previous meta-analyses. However, considerable variations remain unexplained, emphasizing the need to better understand the underlying reasons for heterogeneity. In this study, a large-scale and cross-context dataset, comprising 20,879 students enrolled in 505 assignments within 243 courses at 76 different institutions, was employed. Meta-regression, multi-level modeling, and ANCOVA were used to examine the effect size of each peer feedback experience and explore the associated heterogeneity of effects across courses and assignments within courses. Results revealed that learning benefits are more closely associated with providing feedback rather than receiving it, and are influenced more by the length of peer feedback rather than its helpfulness. Furthermore, heterogeneity exists at the assignment level rather than the course level, and only two peer feedback experiences (i.e., provided length and received length) exhibit particularly large variations in effect size. Particularly novel to our study, we found that learning benefits of both peer feedback experiences diminished as knowledge transfers further away. Similarly, providing feedback demonstrates robust learning compared to receiving feedback, primarily in the form of smaller downward trends and higher learning gains at more further transfer level. Instructors are recommended to design logically-structured consecutive assignments within a course and to provide more guidance to students on giving detailed feedback. •Tested the relative benefits of providing and receiving peer feedback in a large naturalistic dataset.•Learning was more connected to providing rather than receiving, and with comment length rather than perceived helpfulness.•Large effect-size heterogeneity was predominantly at the assignment rather than course level.•Developed a novel method for examining knowledge transfer levels.•Effect sizes varied as function of relative level of knowledge transfer distance.