An experiment on choices between single and compound lotteries is presented, and results are calibrated with neural network models. Many subjects tend to average out probabilities, though behaviour becomes more rational with more exposure to compound lotteries in the practice stage. The Prior...