Experimental Design Details
Preliminary results of the theoretical model analysis show that the presence of non-risk-neutral individuals in the population slows down convergence speed and may change the probability of convergence to different options. In the laboratory experiment, I empirically test the theoretical predictions, comparing the cascading and herding decisions of participants in settings where they have to choose between two symmetrically risky options or between two asymmetrically risky options.
In the laboratory experiment, I plan to split the sample randomly into three groups: control group, treatment group 1, and treatment group 2. Before the main experimental part, all the participants will take part in a Bomb Game (Crosetto, Filippin, 2013), aimed at their risk preferences elicitation. After the Bomb Game, the participants will receive a description of different types of risk preferences, based on the possible behavior types in the Bomb Game. Then they will proceed to the main experimental part.
This experiment part will last for fourteen rounds. In all rounds, the individuals will operate in subgroups of five individuals. The individuals will have to make choices in fourteen different situations, one choice per round.
In the first ten rounds, the individuals will have position three in the subgroups, and they will learn the choices of the first two players. The decisions of the first two players will be taken as histories of choices from the pilot experiment (where the individuals were in the same experimental group: control or one of the treatment groups). In the pilot experiment, there were four experimental sessions, with 27 participants in each of them, and there was variation in the distribution of risk preferences between the sessions. Thus, to exploit this variation in the main experiment, I will use histories of choices from different pilot experiment sessions in different rounds of the main experiment.
In the first four choices, the first two individuals will be randomly chosen from all the histories of choices ``AA" from different pilot sessions, a first pilot experimental session in choice one, a second -- in choice two, and so on. In choices five, six, and seven, the first two individuals will be randomly chosen from all the histories of choices ``BB" from pilot experimental sessions one, three, and four, correspondingly (the second pilot experimental session did not have a history of choices ``BB" one of the experimental groups, so it is not used in these choices).
In choices eight, nine, and ten, the first two individuals will be randomly chosen from all the histories of choices ``AB" and ``BA" from different pilot sessions. In the choice eight, they will be chosen from all the pilot experimental sessions. In choice nine, they will be chosen from all the pilot experimental sessions except the one chosen in choice eight. In choice ten, they will be chosen from all the pilot experimental sessions except the ones chosen in choices eight and nine.
When all the first ten choices for the individuals will be randomly picked from the corresponding histories of choices in the pilot experiment, their order will be randomized.
In rounds eleven and twelve the individuals will have position four in the subgroups. In each of the rounds, their subgroup will be randomly chosen from the subgroups that were formed in the first seven rounds (inside the same experimental group), one of them -- from the first four rounds, and another -- from the next three rounds, the order of these choices will be random. In rounds thirteen and fourteen all the individuals will have position five in the subgroups, following the subgroups, randomly chosen from rounds eleven and twelve, correspondingly (inside the same experimental group), the order of the choices will be random.
In all the rounds the participants will learn the choices of the individuals preceding them in their subgroups. Additionally, the participants will know that they will receive a personal signal that will tell them additional information about their potential earnings from different options. Nevertheless, they will learn their signals only after making a choice, and before making a choice they will be presented with the menu of possible signals, and they will be asked to make choices for all possible signal realizations.
The agents will be informed about the presence of fourteen rounds in the game, their structure, and the random distribution into the subgroups. Additionally, they will be truthfully informed that they will receive the payments associated with their choices only from one of the fourteen rounds that will be chosen randomly after their choices are made. The results of their choices will be known only for this round, so they will not learn their signals and earnings from the other rounds during the experiment, and they will be informed about it.
The settings of choice in the control group mimic the settings of the classical information cascades model (Banerjee, 1992; Bikhchandani et al., 1992), where the agents have to choose between two symmetrically risky assets. The agents have to choose between Option A and Option B. Before the beginning of each round, a coin is flipped, and it defines the payments from each of the options. If it falls heads, option A gives the players 300 CZK and option B gives them 100 CZK. If it falls tail, option A gives the players 100 CZK, and option B gives them 300 CZK. The coin was flipped before the beginning of each round in the pilot experiment, and the possible payments from the options are the same for all players in the group for the round. The possible payments from the options will be known to the players only after the end of all experimental rounds, as well as their payments from the game.
Before making a decision each player will learn that they may receive one of the two possible personal signals, and they will be asked to make a choice for each possible realization of the signal. They will learn a realization of the signal only for the round randomly chosen for payment and only after making all the decisions in the main experiment part. This personal signal will be shown as a result of drawing a ball from a box with three balls with replacement. If the coin falls heads and option A gives higher payments, two balls in the box will have the letter ``a" on them, and one ball will have the letter ``b" on it. If the coin falls tails and option B gives higher payments, two balls in the box will have the letter ``b" on them, and one ball will have the letter ``a" on it.
In each round, the players will not know the result of the coin flip before the end of all rounds, and they will not know the composition of the balls in the box from which the ball was drawn but they will see the letter on the ball. Nevertheless, the realization of the signal will be known to the participants only for the round randomly chosen for payment, and only after they finish making all the choices in the main experimental part.
Moreover, all players are informed that each agent learns the choices of all agents preceding them in their subgroup. They will learn that the first two players in their subgroups played the game in the past in different experimental sessions, and they will have to continue some of the games from the sessions as players with positions three, four, or five in the subgroups. Finally, the agents are informed about the distribution of risk preferences in the corresponding sample (constituted of the sample of the corresponding pilot experimental session and three players randomly chosen for the subgroup in the main experimental session) based on their choices in the Bomb Game.
The participants of treatment group 1 will have the same settings of choice and the same information as the participants of the control group, except for the possible outcomes from the options. In contrast to the control group, treatment group 1 participants will have to choose between two asymmetrically risky assets, where option A always gives participants 200 CZK, but option B brings them 100 CZK if the coin fell heads, and 300 CZK otherwise.
Treatment group 2 participants will make the same choice as participants of treatment group 1 but, in contrast to that group, they will have perfect information on the risk preferences of agents preceding them in the sequence. This approach would allow me to distinguish between the effect of the risk preferences of the agents on their choices and the effect of uncertainty regarding the risk preferences of the other agents in the group.
Based on the theoretical predictions, I test the following hypotheses:
1. The percentage of actual herding (cascading) in potential herding (cascading) cases would be lower in treatment 1 group than in control and treatment 2 group;
2. The percentage of incorrect decisions in actual herding (cascading) decisions would be higher in the treatment 1 group than in control and treatment 2 group.
Exclusion criteria: if a person did not understand instructions for at least one of the experimental parts.
Additional variables (balance check across groups will be conducted for all of them, and the ones with significant differences across groups will be used as control variables): gender, age, occupation, education level, presence of mathematical background (if specialization of work or studies includes mathematical skills), household monthly income, risk preference type, previous participation in experiments, prior participation in experiments similar to the information cascades game.