Measurement Models for Group-Peer Assessment of Project-Based Learning in Software Engineering

Over two decades we have been developing and testing alter-native frameworks for Group-Peer-Assessment (GPA) in software engineering education. The focus lies on skills as-sessed through observation of students’ behavior and/or ex-amination of the results of project assignments. Assessment for such instructional types is not covered satisfactorily by traditional approaches. Assessment at the group as well as individual level must be able to cope with multiple quality criteria on scales of various types, and multiple assessors responsible for assessing distinct quality aspects. We offer two-parameter scoring models for GPA, by which individual student scores are derived from a group score and mutual peer ratings. The two parameters are: (1) a constraint on the spread of student scores and (2) the relative impact of peer ratings. GPA imposes no artificial restrictions on group size, type or number of quality criteria, or other context-specific aspects of GPA. We briefly describe our experience with GPA for software engineering courses at a university in the UK