Foraging models explain human exploration in uncertain tasks.

Experiment 1: 2-armed restless bandit task (258 participants ; online experiment on amazon mturk)In Experiment 1, participants were required to repeatedly choose one of two options. Each option was associated with a probability of reward, which changed across trials and independently across options (step size = 0.1, hazard rate = 0.1). Each participant's reward schedule was randomly generated to avoid environmental biasesall_sub_2ab: behavior formatted for fitting + analyses in pythonsingleiti_202203011023_lightweight: behavior formatted for fitting + analyses in matlabExperiment1_fits : models fits found in main figures Experiment1_fits_supplementary : models fits found in supplementarysimulationOutputs_Experiment1 (2/2) : simulations with environment matched to participantsExperiment 2: Linked 2-option bandit task (95 participants ; online experiment on amazon mturk)To understand how fixing richness and decreasing uncertainty in unchosen option value influences behavior, participants performed a modified version of the two-armed restless bandit task in which option values were linked. One option walked according to the same statistics used in Experiment 1, while the value of the other option was set to 1 minus this value. Because the values always summed to 1, there was no variation in the richness of the environment across trials.Experiment2_data: behaviorExperiment2_fits : traditional RL, foraging-RL fitssimulationOutputs_Experiment2 : simulations with environment matched to participantsExperiment 3: Volatile 3-option bandit task. (270 participants ; online experiment on amazon mturk)To explore how increased uncertainty influences behavior, participants performed a 3-option restless bandit task under multiple different levels of volatility in the environment. All volatility levels were higher than that used in Experiment 1. Specifically, both hazard rates and step sizes were variable (hazard rates were between 0.1 and 1; step sizes ranged from 0.05 to 0.667).Experiment3_data: behaviorExperiment3_fits : traditional RL, foraging-RL fitssimulationOutputs_Experiment3 : simulations with environment matched to participantsExperiment 4: Multi-option bandit task. (150 participants ; online experiment on amazon mturk)To understand how the number of alternatives options influenced behavior, participants performed either 2-, 3- or 4- option restless bandit task (50 participants in each group). Experiment 4 also used a slightly higher hazard rate than Experiment 1 (0.2) to encourage exploration.Experiment4_data: behaviorExperiment4_fits : traditional RL, foraging-RL fitssimulationOutputs_Experiment4 : simulations with environment matched to participantsExperiment 5: In-person 2-option bandit task. (29 participants, in person)The experiment consisted of 20-25 trials practice block to familiarize with the task followed by 2 testing blocks (1 for each modality: eye movements and touch screen presses). Experiment5_data: behaviorExperiment5_fits : traditional RL, foraging-RL fitssimulationOutputs_Experiment5 : simulations with environment matched to participants