How to improve statistical thinking: Choose the task representation wisely and learn by doing

Attempts to teach statistical thinking usingcorrective feedback or a “rule-training” approach havebeen only moderately successful. A new trainingapproach is proposed which relies on the assumptionthat the human mind is naturally equipped to solvemany statistical tasks in which the relevantinformation is presented in terms of absolutefrequencies instead of probabilities. In aninvestigation of this approach, people were trained tosolve tasks involving conjunctive and conditionalprobabilities using a frequency grid to representprobability information. It is suggested that learningby doing, whose importance was largely neglected inprior training studies, has played a major role in thecurrent training. Study 1 showed that training thatcombines external pictorial representations andlearning by doing has a large and lasting effect onhow well people can solve conjunctive probabilitytasks. A ceiling effect prevented comparison of thefrequency grid and a conventional pictorialrepresentation (Venn diagrams) with respect toeffectiveness. However, the grid representation wasfound to be more effective in Study 2, which dealtwith the more difficult topic of conditionalprobabilities. These results suggest methods tooptimize the teaching of statistical thinking and thepresentation of statistical information in themedia.