Attempts to teach statistical thinking using corrective feedback or a ``rule-training'' approach have been only moderately successful. A new training approach is proposed which relies on the assumption that the human mind is naturally equipped to solve many statistical tasks in which the relevant information is presented in terms of absolute f requencies instead of probabilities. In an investigation of this approach, people were rained to solve tasks involving conjunctive and conditional probabilities using a requency grid to represent probability information. It is suggested that learning by doing, whose mportance was largely neglected in prior training studies, has played a major role in the current training. Study 1 showed that training that combines external pictorial epresentations and learning by doing has a large and lasting effect on how well people an solve conjunctive probability tasks. A ceiling effect prevented comparison of the requency grid and a conventional pictorial representation (Venn diagrams) with respect o effectiveness. However, the grid representation was found to be more effective in Study 2, which dealt with the more difficult topic of conditional probabilities. These results uggest methods to optimize the teaching of statistical thinking and the presentation of statistical information in the media.
The CAUSE Research Group is supported in part by a member initiative grant from the American Statistical Association’s Section on Statistics and Data Science Education