Neural Response to Reward Anticipation under Risk Is Nonlinear in Probabilities

The experimental sequence of events. (1) A single gamble, consisting of the probability p₁ of receiving some dollar amount $x₁ (or 0 otherwise). (2) In 12 of the 120 trials, subjects are then asked to indicate whether the probability in the previous screen was greater or less than 40/100 (to engage attention to screen 1). (3) Subjects see a choice screen showing the gamble shown in 1 and a new gamble.

A, The analysis decomposes expected reward responses into two terms, the linear component in p (left, dashed line) and the nonlinear π(p) − p component (right). B, Glass brain and coronal section of activations to both linear p and nonlinear π(p) − p = 0.77. Red, Regions where both linear and nonlinear terms are activated at p < 0.001, excluding regions where linear and nonlinear terms are significantly different at p < 0.9; Yellow, Regions where both linear term is activated at p < 0.001 and nonlinear term at p < 0.005, excluding regions where linear and nonlinear terms are significantly different at p < 0.5. For additional coronal sections, see supplemental Figure S2 (available at www.jneurosci.org as supplemental material). C, Normalized GLM β coefficients of BOLD signal activation for extracted voxels (blue dots) in the left and right striatum coronal section shown in B (yellow) and Prelec function with group behavioral parameter (α = 0.77) inferred from choices (solid black line).