**4. Description of the Data**

All of the games investigated are of the 2 × 2 form, as shown in Figure 1. The experiments from these games were collected by Selten and Chmura [8] and discussed in Chmura et al. [9]. Chmura et al. [9] investigate a series of learning models and determine which rules characterize individual and aggregate performance better. They find that self-tuning EWA fit the data best yet impulse-matching learning also fit the data well. The twelve 2 × 2 games include both constant sum and non-constant sum games.

The experiments were performed at the Bonn Lab with 54 sessions and 16 subjects in each session. 864 subjects participated in the experiment. The subjects in a given session were only exposed to one game. Games were 200 rounds long and subjects were randomly matched by round in groups of eight during the sessions. Knowledge of the game structure, payoffs, and matching protocols were public at the outset. The subject's role of row or column player are fixed during the experiment, so four subjects in the group of eight were assigned to be a row player and the others were assigned to be a column player. At the end of each round, the subjects were told their current round payout, the other player's choice, the round number, and their cumulative payout. The experiments lasted between 1.5 and 2 h and subjects received at 5 Euro show-up fee plus an average of 19 Euros in additional payouts.


Note: Payoffs for row (r) and column (c) players are given (r,c) in the matrix. Abbreviations of for Up, Down, Left, and Right are given as U,D,L, and R.

**Figure 1.** 2 × 2 Games.
