Experimental design and behavior. (A) Subjects were presented with two stimuli (in every trial pseudorandomly selected among three possible stimuli) in horizontal arrangement (screen 1, for a variable time between 3 and 5 s). Stimuli then flipped to a vertical arrangement indicating the actions required to obtain each stimulus (making a saccade or pressing a button, screen 2). When a response had been registered, the screen was immediately cleared for a short delay and subsequently the outcome was revealed (screen 4 at 6 s after screen 2), indicating receipt of reward or no reward. There were two conditions: an SC as just described and an AC in which the first screen was not shown and subjects immediately saw the stimulus–action pairing. Red, green, and cyan arrows indicate onset times for regressors used to estimate the effects reported in and (which are illustrated using the same color code). (B) Example reward probability paths for the three stimuli as a function of the trial number. The probabilities of being rewarded following choice fluctuated slowly and independently for each stimulus across the experiment. (C) Actual choice probability plotted against fitted model choice probability (binned 0.1 wide), averaged across subjects (lines represent SEM). (D) Reaction time (after the action pairing is revealed in screen 2) is significantly lower in SC trials than in AC trials (paired t test, P < 10⁻¹¹, vertical lines represent SEM).

Neural correlates of chosen value. (A) Activity in vmPFC showed significant correlation with the value of the stimulus that was subsequently chosen before the stimulus–action pairing was revealed (SC trials, red). The value chosen signal in AC trials was represented slightly more ventrally (green). Voxels overlapping in both contrasts are shown in yellow. Activations survive correction for multiple comparisons as described in . (B) A comparison of effect size at the overlapping region confirms that, in SC trials, the value chosen is represented only at the time of the stimulus screen (red) but not at the time of the following stimulus–action pairing (cyan). In AC trials, the value chosen is represented at the coinciding stimulus/action screen (green). Bars indicate SEM. (C) Event-related blood oxygen level-dependent responses in SC trials time locked to the stimulus presentation (Left), the stimulus–action pairing (Center), and in AC trials time locked to the coinciding stimulus–action pairing (Right). Time courses are plotted separately for trials in which the chosen values were small (dashed, V < 0.5) and large (solid, V > 0.5). Note that, consistent with the effect sizes shown in B, time courses split in SC trials after stimulus presentation (Ts, Left) but already approximately 4 s before the action pairing (Ta, Center). Dashed lines represent SEM.

Neural correlates of stimulus value. (A) Activity in vmPFC showed significant correlation with the average value of the two stimuli that were presented in SC trials at the time of screen 1 (red). Activations survive correction for multiple comparisons as described in . (B) Comparison of effect sizes at the peak region between conditions. Neural activity in this region correlated with stimulus values in SC trials but not in AC trials. Bars indicate SEM.