Evaluating the Effectiveness of Feedback in SQL-Tutor Antonija Mitrovic, Brent Martin Intelligent Computer Tutoring Group Computer Science Department University of Canterbury Christchurch, New Zealand
Overview SQL-Tutor Feedback types Evaluation study Results
Architecture of the standalone version of SQL-Tutor
SQLT-Web: a Web-enabled SQL tutor Interface developed in CL HTTP-server Used in COSC courses since 1999 Open for outside users on March 29, 2000
Architecture of SQLT-Web
Feedback types Positive/negative Error flag Hint All errors Partial solution Complete solution
Hypothesis: Positive/negative and Error flag too general Low-level feedback contraproductive (partial and complete solution) Constraint-based feedback most effective (hint and all errors)
Evaluation study COSC313, May 1999 (Web) Single 2-hour session 6 lectures + 8 hours of labs 33 students
Three groups: Detailed (complete and partial solution) General (hint and all errors) Limited (pos/neg and error flag)
Results 1 GroupSolved (%)NoAttemptsTime/attempt Detailed s General s Limited s Difference in times is significant
Results 2 FeedbackConstSuccFailLearnCorr (%) Pos/neg %78.0 Error flag %81.8 Hint %74.4 All errors %80.0 Partial %91.6 Full sol %44.2
Initial learning rate All errors0.44 Error Flag0.40 Pos/Neg0.29 Hint 0.26 Partial0.15 Full solution0.13
Conclusions General feedback is most effective Detailed feedback is detrimental to learning Feedback level needs to be adapted