Do students really learn an equal amount independent of whether they get an item correct or wrong?

The field of educational data mining has been paying attention to Knowledge Tracing (KT) for a long time. Corbett and Anderson assumed the amount of learning that students do does not depend on whether students get items right or wrong. Ohlsson and others argued that the student should learn more from a previous incorrect performance. We decided to investigate a Bayes Network similar to KT but that allows us to have learning rates that are different according to whether students get items correct or not. While the idea of allowing learning rates from previous incorrect performances to be higher seems intuitive, our experiments showed that this way does not always lead to better predictions. Of course reasoning from a null result is dangerous, our contribution is that this intuitive idea is not one that other researchers should waste time in working on, unless they come up with a different model from the model we used (which is the naïve way of modifying KT).