I have often thought that the academic grading system should provide two characteristic numbers corresponding to two fundamental measures of information:
1) Comprehension (C) — What breadth of explanation, creativity, or ambition has the student expressed?
2) Percent Error (PE) — How much of the student’s expressed knowledge contradicts accepted knowledge?
This is how the proposed system would work. On a given test or assignment, a student who answers every question gets all the Comprehension points. Answer half the questions and get half the Comprehension points. Then, Percent Error is calculated based on number of questions answered and the number of mistakes.
A simple example would be a student who answers 4 out of 10 questions and 3 answers are correct (1 answer is incorrect). Using the two-axis grading system, the resulting scores would be C=40 and PE=25. This provides a point in a fundamentally meaningful two-dimensional space. For those who are familiar with information theory, this grading system measures the student’s Receiver Operating Characteristic (ROC).
This new system actually simplifies grading in many cases. Late work or missing answers get penalized in their C score, without affecting PE. Creative assignments become easier to grade, because the overall creativity or lack thereof goes into the C score without concern for technical flaws, which show up in PE. Classes with mixed levels of students can require students to show different levels of comprehension in order to achieve the same numerical score. Most importantly, it lets the student express directly how much he claims to know without encouraging Bogus Solutions (BS).
Teachers often ask “What if I have to give a single score? Won’t that score be the same as I would have given using the old system?” The answer is that you can reduce C and PE to a Single Score (SS), but the result will be different only if you grade on a curve. The formula is SS=(C/100)*(100-PE). If you curve, then the curve is only applied to C based on its mean value across all students. This method provides a better measure of information than curving SS based on the mean value of SS across all students.