Listening to the radio this morning, my attention was drawn to a new medical test with an accuracy of ‘more than 99%’. I was left thinking ‘well, is that good enough?’ and then ‘so is a marking accuracy of 99% good enough for e-assessment questions?’ (Actually the situation with the medical test is not as simple as I’d thought – the idea is that this test would just be used to give a first indication. That’s better.)
Returning to e-assessment. Well no, on one level 99% accuracy is not good enough. What about the 1%? But the e-assessment is likely to be considerably more accurate than human markers. And students are more likely to be confused by something in the question (which may be as simple as reading an addition sign as a division) than they are to be marked inaccurately by the system. Clearly this is something that we must continue to monitor, and to do our best to improve accuracy of marking (whether by humans or computers). And it matters even in purely-formative use, because if our answer matching is not correct we will not be giving the correct feedback message to the students.