The paper Imperfect Learning and Error Cascades in Sequential Guessing Games: An Experiment describes an experiment in which students played Chinos, a Spanish parlor game. In this game each player hides a number of coins or other small items in his or her hand. Then each player must guess the number of items in each person’s hand. This is done sequentially such that if you are player t, you have also heard the guesses of (t-1) other players. This resembles an information cascade because you have your own private information (your instinctual guess) and you know the guesses of those who have guessed before you (who based their information instinct and on those who have guessed before them).

Four sessions of a Chinos-like-game was played with 12 students per session. Each session contained 20 rounds. Each session had four groups of three such that through the course of the experiment 16 groups of 20 rounds were observed. Players were anonymous to each other but consistent throughout the game (guessing order was kept constant). Students were paid to participate and there was a small monetary prize for each round.

The results of the experiment were averaged together and show that as player position (t) increases the probability of a correct guess increases. It was also noticed that these probabilities were not as high as the theoretical probabilities. This abnormality suggests players deviated from equilibrium strategies and made errors in judgment. These error cascades were noticed in several rounds. The authors of this paper describe an error cascade as “the situation where the deviation from learning by a player increases with respect to the deviations from learning by preceding players.” Section 6 of this paper describes the modeling of error cascades.

It was concluded that if preceding players started to deviate from the correct guess, there would be imperfect learning and this deviation would continue to cascade through the game. The authors argue that imperfect learning results in error cascades. As more rounds of the game are played, the first guesser in the sequence realizes that he or she has a low probability of winning; therefore it is less costly for that player to deviate from optimal behavior. I imagine myself playing this game as the first guesser. I would probably get upset or frustrated and want to either make different guesses to try to win (compared to following my instinct which has caused me to be wrong so far) or make absurd guesses in order to make other players loose.