ABSTRACT
The results of training studies about statistical reasoning reported in the last chapter were meager to modest, but showed that training can be effective. Overall, the results speak against the view held by supporters of the heuristicsandbiases approach, which states that it is impossible to teach people statistical reasoning. The modest results found for abstract-rules training regimens might not be the last word about what can be achieved by statistical training, however. Surprisingly, previous training regimens did not yet explicitly apply important results from instructional theory. This chapter specifies two conditions which are informed by instructional theory, with which better training results should be found. The first is learning by doing, which has been proposed to be an essential ingredient of successful teaching, but which was largely neglected in prior training studies. The second is the use of flexible tutoring systems, which are implied and specified by results from research about computer-based instruction. These two conditions should boost the effectiveness of any kind of statistical training. However, tutoring programs derived from the abstract-rules approach differ from those derived from the adaptive-algorithms approach. Whereas for the abstract-rules approach, the representational format does not matter, its form is crucial for the adaptivealgorithms approach. This difference between the two approaches is decisive in the studies reported in following chapters. Although the pragmatic-
implications approach does not specifically take representational for-mats into account, it is arguably closer to the adaptive-algorithms approach in this respect.
