Abstract
While autograding systems are prevalent in universities, their application in K-12 education remains underexplored. Building on automated feedback research, this study investigates how K-12 students’ prior programming knowledge and working behaviors within the Artemis autograder influence learning success.
Data from 76 10th-grade computer science students were analyzed, comparing their performance on repetition exercises (testing 9th-grade prior knowledge) with that on advanced exercises introducing reference attributes, inheritance, and arrays. Using multiple linear regression and Spearman’s rank correlation, student success metrics (score and error quotient) and working behaviors (submissions and build failures) were evaluated against a paper-and-pencil posttest.
Results indicate that prior knowledge strongly correlates with success in advanced object-oriented topics, though this effect diminishes for algorithmic concepts like arrays. Furthermore, behavior metrics revealed complex interactions: identical measures yielded both positive and negative learning outcomes, depending on the specific topic. These findings suggest that future investigations require a complete reworking of exercises tailored to research requirements, rather than adapting existing materials.
Grade
Graded 1.3 (second-best grade) by Prof. Dr. Tilman Michaeli