In a typical supervised learning experiment, one might present visual stimuli, e.g. faces, one after another and ask participants to classify each one into one of two categories, e.g. A and B. Usually participants are not required to DO anything after getting feedback - they just wait, and eventually the next trial appears. Does learning improve if, when the response is incorrect, participants are required to change their response to the correct one before proceeding?
To provide some context - unlike the "typical" experiment described above, I am using educationally relevant materials - mathematics story problems. They are still binary forced choice, but responses are given using radio buttons instead of the usual "press P or Q", so to correct an incorrect response, participants would have to go change the radio button response.
More specifically, since details were requested, all of the problems are story problems involving the concept of "sampling with replacement". Before doing the problems, participants read an instructional passage explaining the general formula for sampling with replacement problems. This passage explains that each problem involves two numbers, a and b, where a indicates the number of sampling events and b indicates the size of the sampled set. The problems ask for the number of different possible outcomes, which is given by the formula a^b. Thus, to solve the problems, it is only necessary to figure out which number is the number of sampling events and which the size of the sampled set. Here is a sample problem:
A group of friends is eating at a restaurant. Each person chooses a meal from the menu. (It is possible for multiple people to choose the same meal.) In how many different ways can the friends choose their meals, if there are 4 friends and 5 meals?
I can think of at least two reasons why they MIGHT learn better from doing this. First, it might improve motivation - they know it's a pain in the neck when they answer wrong, so they're more motivated to answer right. Second, it might have a cognitive benefit, in that they would encode the correct and incorrect answers more strongly because they have to actually respond in some way.
On the other hand, there could be a negative effect in that it interrupts the flow of the trials. Doing a large number of trials quickly, pop pop pop, could increase learners' fluency, and such benefits could be decreased by the proposed requirement.
On the third hand, it might be even more beneficial to use an even more extreme version of the proposed requirement, e.g. present several questions at once and only tell them that SOME are wrong, not WHICH ones, so they have to think about it even more. Of course this could encourage them to figure out ways to play the system and could be even worse with respect to the possible negative effect I mentioned.
Anyone have any evidence or speculations about this?