Fig. 1.

Schematic illustration of principled differences between objective and subjective prediction errors, both of which are subject to measurement and predictive of behavior. Objective value refers to observable states, be it monetary reward, food, or a ballet. We can entirely generate expectations (X₁) about, but usually at best influence (e.g., poker) or trigger (e.g., slot machines), objective outcome states (X₂). Subjective value is the result of the appraisal of objective value via idiosyncratic compositions of external and internal states. On the subjective side, we can generate both expectations about how a given objective outcome will be valued subjectively (Y₁) and our actual subjective valuation (Y₂) following the actual outcome. Across expectations and outcomes, subjective value is influenced—but not strictly determined—by objective value. For both objective and subjective value, prediction errors (ΔX and ΔY) are defined as the difference between outcomes and expectations. The constituent parts of prediction errors, except for objective outcome value, are not directly observable and necessitate measurement. In this context, “measurement” primarily pertains to self-report. Whereas the measurement of objective-value expectations is relatively constrained by context (e.g., reported dollar amounts), the measurement of subjective value—approximated via affect—may rely on a wider range of measures, such as feeling ratings. Based on such measurement, both objective and subjective prediction errors about current outcomes have been shown to predict future behavior (e.g., repetition of behaviors that previously led to positive prediction errors).