# Task Setup The general idea for the task behind the data is that there are 4 different monster families with a set of monsters contained in each. These monsters vary along different dimensions. For example, there is a family of monsters where each has a different height, or different widths – or both (like shown in **Figure 1**). |**Figure 1.** Examples of monster families with (**a**) only 1 variable dimension, and (**b**) with two variable dimensions| |:-:| |<img src="https://github.com/flowersteam/Humans-monitor-LP/blob/master/images/monster_dimensions.png?raw=true">| Thus, the monsters can vary along 1 or 2 dimensions. For the 4 monster families, these dimensions determine what food preference a monster has. For example, the short monsters in a family could prefer pancakes while the tall monsters prefer waffles. Participants would then need to determine, based on a monster’s features, what food they preferred. # Monster families and food Each monster family was randomly assigned how many dimensions determined their food preferences. In the data file, these are referred to as categories. These categories are: * A1: the monster family only varies along one dimension, and this dimension determines a food preference * A2: the monster family varies along 2 dimensions, but only one dimension determines a food preference * A3: the monster family varies along 2 dimensions, and both dimensions determine a food preference * A4: the monster family varies along 2 dimensions, but food preferences are random As mentioned, there are also always 4 monster families. In the raw data files, they are named as: * Bear * Bunny * GreenMonster * Squid There are also 4 pairs of food options: * pancakes/waffles * bananas/oranges * broccoli/carrot * grilled_cheese/tacos The food pairings were randomly assigned to each monster family. During the task, one monster would be presented on the screen, and two food choices would appear as buttons. The participant would then need to select one food option as a guess if that was the food the monster preferred. They would then receive feedback if they were correct or not. Over time, they could then learn a monster’s preference (except for the random condition monster family, as described below). **Figure 2** depicts how the trials were played by the participants. First (1), a participant was required to choose one family from the set of 4. This would results in a random sampling of a family member which the participant was prompted to label with one of the presented food choices (2). Finally, (3) the participant received positive of negative feedback, after which the cycle was repeated. |**Figure 2.** Schematic description of the task.| |:-:| |<img src="https://github.com/flowersteam/Humans-monitor-LP/blob/master/images/fig1a.png?raw=true">| # Raw data files Data from the study can be found in the [Files][1] of the project. The raw data (in 'raw\_data.zip') is organized into 2 pairs of .csv files, one for each instruction group. The 'eg\_main' and 'eg\_extra' files contain data from the external goal condition, and the 'ig\_main' and 'ig\_extra' files contain data from the internal goal condition. The 'main' files contain data from the guessing game trials, and the 'extra' data files contain participants' responses to the questionnaire (see Self-reports below). ## Column meanings for the “main” files: - sid: unique participant identifier - group: instruction group (0 = internal goal; 1 = external goal) - stage: one of three stages of the experiment (train = familiarization stage; free = free choice stage; test = testing stage) - trial: the trial number starting from the first trial of the familiarization stage - blockTrial: the trial number starting from the choice (or presentation) of a new monster family - trialStartTime: milliseconds since the onset of the task - monster: a string identifying the individual monster (stimulus) presented to the participant (family_dimension1_dimension2) - family: monster family (i.e. stimulus category) - activity: difficulty level of the sampled stimulus (see Monster Families and Food above) - preferredFood: ground truth of the sampled monster's food preference - response: participant's guess for the sampled monster's food preference - correct: whether or not participant's response was correct (True = correct, False = incorrect) - rt: time in milliseconds between the presentation of the stimulus and participant's response (guessing of food preference) ## Column meanings for the “extra” files: - sid: unique participant identifier - group: instruction group (0 = internal goal; 1 = external goal) - age: age of participant in years - gender: gender of participant (male; female; <NA> = undisclosed) - ethnicity (hisp = hispanic; nonhisp = nonhispanic; <NA> = undisclosed; not reported = omitted) - thoughts: (optional) open-ended answer to the question regarding participant's thoughts on what they experiment was about - comments: (optional) additional comments by participant about the experiment The remaining columns in the 'extra' files correspond to the questions described below (see Self Reports). Each question has a shortened question ID that is listed in the square brackets next to the question text (future-learn-0, interested, complex, time, progress, rule, future-learn-1). All question responses are on a 10-point likert scale. In the data columns, each question ID is also followed by which family the question was asking about. For example, interested-Bear would contain the responses given to the [interested] question for the Bear family. # Processed data files We also provide the processed data files (in 'processed_data.zip') and modeling data (in 'model_data.zip') that were prepared with Python (mostly Pandas and Numpy). The corresponding code can be found on [Github][2]. These files were generated using the raw data described above and contain the data used to perform analyses reported in the article. # Self-reports There 7 self-reported ratings of the experimental stimuli: **Post-familiarization question:** 1. Before continuing, please rate each monster family based on how much you think you can learn about its food preferences during the rest of the task: [future-learn-0] Definitely Cannot Learn More [1] - Definitely Can Learn More [10] **Post-task questions:** 1. Rate each monster family based on how much you were interested in discovering what they preferred eating: [interested] Less Interested [1] - More Interested [10] 2. Rate each monster family based on how complex you thought they were: [complex] Less Complex [1] - More Complex [10] 3. Rate each monster family based on how much time you spent on them: [time] Less Time [1] - More Time [10] 4. Rate each monster family based on how much progress you felt you made for learning their food preferences: Less Progress [1] - More Progress [10] 5. Rate each monster family based on how likely you think it had a rule for food preferences: [rule] Definitely No Rule [1] - Definitely a Rule [10] 6. Rate each monster family based on how much more you think you could learn if you had more time to play with it: [future-learn-1] Definitely Could Not Learn More [1] - Definitely Could Learn More [10] [1]: https://osf.io/yg5k3/files/ [2]: https://github.com/alex-ten/Humans-monitor-LP