Abstract
According to the diffusion-to-bound decision model (DDM), decisions are made by a gradual process of noisy evidence accumulation until a bound is reached, resulting in variability of decision times. The “starting point” of evidence accumulation is the prior, so processing is faster when prior is consistent with evidence. A less intuitive model prediction is that prior-derived decision bias gradually decrease with processing time, which, for difficult decisions and moderate prior, leads to close to zero bias in slow (>median) trials. Here, we tested the interaction of bias and processing time by manipulations of prior probability (exp. i–ii) and of visual context (exp. iii–iv). Specifically, we tested the following tasks. (i) Detection biased by target probability (25%, 50%, or 75%; using low-contrast Gabor patches, d′=~1.5, duration=50ms, σ=0.42°, λ=0.3°). (ii) Orientation discrimination biased by orientation probability (25%, 50%, or 75%; using near-vertical Gabor patches, d′=~1, duration=50ms, σ=0.42°, λ=0.3°). (iii) Orientation discrimination biased by a previously seen oriented adaptor (tilt aftereffect; using Gabor patches for both adaptation and test, ISI=600ms, duration=50ms, σ=0.42°, λ=0.3°). (iv) Orientation discrimination biased by the orientation of a surrounding ring (tilt illusion; using circular sine-wave gratings). Reaction time (RT) was used as the standard proxy for decision time. Results showed interaction of bias and time: trials with faster (median>) RT measured response bias in accordance with the experimental manipulation (prior-derived beta of ~2, induced tilt of ~2.5°), while trials with slower RT (>median) measured either zero bias (prior manipulations, beta=~1, p< 0.001) or a reduced bias (~35% reduction of induced tilt, p< 0.001). We observed a minor reduction in task accuracy for slower RT, which could not trivially explain the effect of bias. In conclusion, reduced bias with longer processing time is predicted by a standard decision model assuming temporal integration of noisy evidence, suggesting the interaction is ubiquitous in brain systems.
Acknowledgement: Supported by the Israel Science Foundation (ISF)