Geintra

This paper introduces a new methodology to analyze grading discrepancies in PBL software-engineering courses with a high student-to-faculty ratio. The methodology is based on a quantitative analysis of the more relevant software features according to the grades assigned by each instructor. In order to reduce the discrepancies detected, a new grading consensus has to be built, and automatic analysis tools must assist the instructors when grading. The evaluation of the methodology in two academic years revealed a 62% reduction of the grading discrepancies, achieving an average inter-instructor discrepancy of 0.10 in a scale from 0 to 1.