Purchase this article with an account.
James Tee, Laurence Maloney; Separating Noise from Suboptimal Inference in Choice Behavior Variability. Journal of Vision 2015;15(12):975. doi: https://doi.org/10.1167/15.12.975.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
INTRODUCTION: Choice behavior can vary from trial to trial, even with near-identical stimuli. This is commonly attributed to errors caused by internal neural noise (Faisal et al, 2010, Nat Rev Neurosci). An alternative theory (Beck et al, 2012, Neuron) is that variability could also arise from suboptimal inference. We designed two complementary tasks to separate noise from suboptimal inference. METHODS: During each trial in Task 1 (No Spin), a subject was presented with two roulette wheels where a fraction of each wheel is colored orange. The subject was asked to choose the wheel that has the larger proportion of orange. Making the correct choice won a monetary prize. In Task 2 (Spin), a subject was presented with the same pair of wheels and was asked to choose the wheel that has the larger chance of winning. After choosing, the wheels were spun and the subject won a monetary prize if the chosen wheel stopped in the orange. A staircase procedure was used to vary one wheel while the other acted as the test condition. There were 20 test conditions, uniformly distributed between [0.025,0.975] with 30 trials per condition. 20 naïve subjects performed both tasks. RESULTS: While the total number of wise (i.e. correct) choices differed from subject to subject, 20/20 subjects made more wise choices in the No Spin task (median=489/600) than the Spin task (median=444/600). For 19/20 subjects, the median angular difference between both wheels in unwise choice trials is significantly larger for the Spin task. CONCLUSIONS: Both tasks have the same optimal solution: choose the wheel with more orange. The No Spin task established a baseline for errors caused by internal noise while the Spin task measured the incremental error arising from suboptimal inference, allowing for separation of both types of errors.
Meeting abstract presented at VSS 2015
This PDF is available to Subscribers Only