Additional Material
Supplement toMagic Mirror on the Wall, Who Is the Smartest One of All?
Here you can find examples of the interfaces subjects saw during the experiment. Participants faced four games in random order.
To improve participants' experience and to assist in selecting an action, we implemented a highlighting tool that used two colors: yellow and light green. When a participant moved their mouse over a row in their matrix (``Your Earnings''), the action was highlighted in yellow color in both matrices: a row in their matrix, and a column in Player Z's matrix (``Player Z's Earnings''). By left clicking the mouse over a row it remained highlighted, and participants could unhighlight it by clicking their mouse again or clicking another row. Similarly, when participants moved their mouse over a row that corresponds to an action of Player Z in ``Player Z's Earnings,'' the row was highlighted in light green and the corresponding column was highlighted in light green in ``Your Earnings.'' Clicking the mouse over the row kept it highlighted, and clicking it again (or clicking another action) unhighlighted it.
The ``DS'' Game
Experimental Implementation:
The ``IR'' Game
Experimental Implementation:
The ``MS'' Game
Experimental Implementation:
The ``NE'' Game
Experimental Implementation:
Additional Material
Supplement toThe Streetlight Effect in Data-Driven Exploration
Here you can find examples of the interfaces participants saw during the experiment. The experiment consisted of independent "rounds." Mimicking our theoretical framework, each round was composed of two "stages" – the two time periods over which player payoffs were calculated. Participants take the role of an individual engaged in a hunt for precious gems. There are five mountains and each of them hides one type of gem, which can only be uncovered by exploring the mountain. There are three types of gems of varying rarity and value hidden in the mountains: three topazes, one ruby, and one diamond. The diamonds are always worth more than the rubies and the rubies are always worth more than the topazes.
All five players are anonymous to each other and cannot directly interact or communicate. Players select which mountain to explore sequentially, based on a random order that changes every round. A dynamic instruction element on their screen turns green and indicates that it is their turn to make a choice (otherwise they must wait). None of them has any initial private information about the location of the gems, which changes every round (but not between the first and second stage of the same round). While waiting for their turn, players can see which mountains are being selected by their co-players. When it is their turn, players choose one mountain to explore. They can pick the same or different mountain as other players and their payoff is independent of whether or not their choice has already been selected by someone else. In other words, if participants overlap in their choice of mountain, each of them still receives the entire value of the gem uncovered since payoffs are non-rival.
Participants received detailed written instructions about the experiment and watched a compulsory six-minutes video that reiterated the main instructions while also familiarizing them with the experimental platform.
You can watch the no-data treatment video here
and the data treatment video here.
No-Data Treatment
Data Treatment with Topaz Signal
Data Treatment with Ruby Signal
Data Treatment with Diamond Signal
Additional Material
Supplement toInformation and the Bandit: The Good, the Bad and the Ugly
Here you can find examples of the interfaces participants saw during the game, depending on the environment and the observability of actions. Participants could see their fellow group members’ action choices and payoffs on their computer screens.
In both bad news and good news environments with observable actions, participants playing "safe" was displayed as 
In the bad news environment, when playing "risky", they saw either 
or 
depending on whether a breakdown had occured or not.
With hidden ations and conditional on no breakdown having occured, the other player's action was displayed as 
, making the inference more difficult and corresponding to the “the ugly” in the reference to Sergio Leone's spagetti western film from the year 1966.
In the good news environment, the experimental implementation was similar with the difference that we implemented 
for breakthroughs instead of in the bad news environment.
Similarly, we displayed 
in the good news environment instead of in the bad news environment.
Bad News with Observable Actions
Bad News with Hidden Actions
Good News with Observable Actions
Good News with Hidden Actions
Additional Material
Supplement toCoordination in the Network Minimum Game
Here you can find examples of the interfaces participants saw during the experiments. At the start of the experiment, groups were randomly formed and participants were randomly allocated a position within a network. Groups and positions were fixed throughout the experiment. Our experimental implementation not only indicated a participant's position within the network but also highlighted their personal "watch-list" (neighbourhood). The position was highlighted in red color and the corresponding watch-list was circled in red color as well.
You can watch 5-6 minute videos participants saw before starting the experiment. The videos explain each step (screen) of the experiment with an emphasis on the concept of neighbourhood (called "watch-lists") as well as the coordination game.
3-player network: sparse & acyclic
Experimental implementation:
4-Player network: sparse & acyclic
Experimental implementation:
6-Player network: sparse & acyclic
Experimental implementation:
Experimental implementation:
4-Player network: sparse & cyclic
Experimental implementation:
6-Player network: sparse & cyclic
Experimental implementation:
6-player network:dense & acyclic
Experimental implementation:
6-Player network:dense & cyclic
Experimental implementation:
12-Player network:dense & acyclic
Experimental implementation:
Additional Material
Supplement toBandits in the Lab
Here you can find examples of the interfaces participants saw during the game, showing the evolution of the screen over time. In the top half (third) of her screen, a participant could see their own past actions and payoffs, while the bottom half (two thirds) of the screen showed their fellow group members’ actions and payoffs. A blue (red) part of the payoff curve indicated that the player used the safe (risky) arm over the corresponding period. The x-axis represented calendar time, while the y-axis gave the player’s cumulated total earnings up to each point in time. There was no prior indication of the point in time the game would end.
All four heatmaps show the total number of fixations. The accumulated number of fixations is calculated for an entire game. Each fixation made has the same value and is indepentent of its duration. A color gradient is used to indicate the areas with more fixations (low=green to high=red).
Strategic Treatment with 2 Players
Watch the eye-tracking video here.
Control Treatment with 2 Players
Watch the eye-tracking video here.
Strategic Treatment with 3 Players
Watch the eye-tracking video here.
Control Treatment with 3 Players
Watch the eye-tracking video here.
