Making heads or tails of probability: An experiment with random generators


  • Kinga Morsanyi is now working at the University of Cambridge.

Correspondence should be addressed to Kinga Morsanyi, Centre for Neuroscience in Education, Department of Experimental Psychology, University of Cambridge, Downing Street, Cambridge, CB2 3EB, UK (e-mail:


Background. The equiprobability bias is a tendency for individuals to think of probabilistic events as ‘equiprobable’ by nature, and to judge outcomes that occur with different probabilities as equally likely. The equiprobability bias has been repeatedly found to be related to formal education in statistics, and it is claimed to be based on a misunderstanding of the concept of randomness.

Aims. The aim of the present study was to examine whether experimenting with random generators would decrease the equiprobability bias.

Sample. The participants were 108 psychology students whose performance was measured either immediately after taking part in a training session (n= 55), or without doing any training exercises (n= 53).

Method. The training session consisted of four activities. These included generating random sequences of events, and learning about the law of large numbers. Subsequently, the participants were tested on a series of equiprobability problems, and a number of other problems with similar structure and content.

Results. The results indicated that the training successfully decreased the equiprobability bias. However, this effect was moderated by participants’ cognitive ability (i.e., higher ability participants benefitted from the training more than participants with lower cognitive ability). Finally, the training session had the unexpected side effect of increasing students’ susceptibility to the representativeness heuristic.

Conclusions. Experimenting with random generators has a positive effect on students’ general understanding of probability, but the same time it might increase their susceptibility to certain biases (especially, to the representativeness heuristic). These findings have important implications for using training methods to improve probabilistic reasoning performance.