Divisive normalization predicts value modulation. Shown is predicted versus observed firing rate data under different proposed value coding algorithms: A, fractional; B, difference; C, simple divisive normalization; D, full divisive normalization. Left column, Predictions when model parameters were fit separately for each individual neuron, for all neurons across the LOW-RF, MED-RF, and HIGH-RF conditions. Middle column, Predictions when model parameters were fit to the population data in the MED-RF condition. Right column, Predictions when model parameters were fit to the population data across all conditions. Each point represents predicted and observed data under a specific choice set condition.