Figure 3. Characteristics of value coding in DLPFC neurons.

(a) Mean normalized activity for value responses in different trial periods (identified by multiple linear regression, P<0.05, calculated for each response), sorted into value terciles. Yellow shaded periods were used for analysis. N refers to number of responses. (b) Percentages of value responses (among task-related responses) in different trial periods. (c) Recording locations in upper and lower principal sulcus banks. Numbers indicate anterior-posterior distance from inter-aural line. Sections taken from a stereotaxic atlas⁶⁴. Inset: photomicrograph of a cresyl violet-stained coronal section of the frontal lobe in monkey A. The lesion in the principal sulcus marks a typical electrode track. (d) Object-specificity of value coding. Linear regression of population activity (273 value-coding responses taken across task periods) on object value for preferred and alternative object. Data points indicate means of 11 equally populated value bins±s.e.m. (e) Classification of value responses (across task periods) based on position of regression coefficients in value space. The figure shows the classification into object value coding or relative (sum/difference) value coding depending on the polar angle (θ) of coefficients in value space (coloured areas), calculated as four-quadrant arc-tangent of the coefficients. Yellow data point: example neuron from Fig. 2. (f) Proportion of value coding responses across task periods reflecting object value (only value A or value B) and responses reflecting relative value (value sum/difference). Black/orange bars: results obtained from conventional multiple regression and axis-invariant method. *P<0.05 (χ²-test). (g) Distribution of regression coefficients (273 value-coding responses; fixed-window analysis across all task periods) for object value and choice. Most value responses had non-significant choice coefficients (Kolmogorov-Smirnov test).