Fig 1.
Experimental design and computational framework.
(A) Shown here is the experimental design used by [23] in their EEG study, which differed in several minor ways from the design used by [22] in their fMRI study (see Materials and methods). Subjects were instructed to respond to a target stimulus (white circle) by either pressing a button (Go) or witholding a button press (NoGo). Subjects had to learn the optimal action based on stimulus cues (shapes) and reward or punishment feedback. For all conditions, the optimal action yielded reward delivery or punishment avoidance with 70% probability; this probability was 30% for the suboptimal action. (B) Pavlovian and instrumental prediction and valuation combine into a single integrated decision value based on a weighting parameter (w) that represents the evidence for the uncontrollable environment (i.e., in favor of the Pavlovian predictor). Figure adapted from [18], with permission. See Materials and methods for technical details.
Fig 2.
Top: Protected exceedance probabilities (PXPs) for 3 computational models fit to the EEG data set (A) and the fMRI data set (B). Bottom: Go bias (difference in accuracy between Go and NoGo trials) computed as a function of the Pavlovian weight for the EEG data set (C) and the fMRI data set (D). Lines show model fits, circles show means with standard errors.
Fig 3.
(A) Montage showing region of interest, derived from [23]. (B) Stimulus-locked midfrontal theta power as a function of the Pavlovian weight. Error bars show standard error of the mean.
Fig 4.
(A,B) Regions of interest. IFG: inferior frontal gyrus; vmPFC: ventromedial prefrontal cortex. (C) IFG response to the stimulus cue as a function of Pavlovian weight, separated by Go and NoGo conditions. (D) vmPFC response to the stimulus cue as a function of Pavlovian weight, separated by Win and Avoid conditions.