Neural Correlates of Value Are Intrinsically History Dependent

Lee, S., Lerman, C., & Kable, J. W.

Series of studies in consumer neuroscience has shown that taking simple fMRI measurements of subjective value from areas such as ventral striatum can predict, over and above traditional measures, population-level behaviors such as market-level advertising success, album sales, future purchases, crowdfunding, and information sharing. However, neural measurements in such studies may be miscalibrated for behavior prediction. Studies at the single neuron level has found that some neurons’ encoding of value are heavily influenced by the valuation of past trials: if the subjective value of the last trial was high, the value signal of the current trial was low, and vice versa. In this paper, we show that even at the fMRI level, neural signal of subjective value is intrinsically history dependent while choice behavior does not depend on valuation of previous trials. This miscalibration between neural signal and behavior can lead to systematic errors that lower the predictive powers of neural measurements. Surprisingly, we also found that a whole-brain multivariate predictor of choice can cancel out this neural history dependency by subtracting signals from multiple brain regions. Consequently, our whole-brain predictor shows far superior predictive performance compared to region of interest (ROI) based approaches, which has been the norm of prediction in consumer neuroscience. Therefore, when using neural correlates to predict behavior, researchers should consider using a whole-brain approach, such as the one we provide here, which will yield higher predictive powers.

Sangil LeeUnder Review