A review from the mid-1990s pulled together the existing experiments on this issue and reported that, in 22 experiments using test questions that demanded students recall information (for instance, “What years in U.S. history are often called the Gilded Age?”), learning loss was about 28 percent. Retention was even better when questions required recognizing the correct answer, as on a multiple-choice test. For such tests, the average learning loss across 52 experiments was just 16 percent.
Tests taken a year later, with multiple choice answers showed only 16% learning losd