Khan Academy developer David Hu writes a very interesting piece about how we’re running a number of experiments to improve our understanding of student mastery and encourage more learning from our students:
In high level terms, we increased overall interest — more new exercises attempted, fewer problems done per proficiency — without lowering the bar for proficiency — P(next problem correct | just gained proficiency) was roughly the same for both groups. Further, it seemed that overall learning, as measured by the distribution of accuracies obtained, went up slightly under the new model.
Optimistically, we hypothesise that our gains are from moving students quicker off exercises they’re good at, while making them spend more time on concepts in which they need more practice.