Assessing consumer demand with noisy neural measurements

Recent studies have used the random utility framework to examine whether neural data can assess and predict demand for consumer products, both within and across individuals. However the effectiveness of this methodology has been limited by the large degree of measurement error in neural data. The resulting “error-in-variables” problem severely biases the estimates of the relationship between neural measurements and choice behaviour, thus limiting the role such data can play in assessing marginal contributions to utility. In this article, we propose a method for controlling for this large degree of measurement error in value regions of the brain. We propose that additional neural variables from areas of the brain that are unrelated to valuation can serve as “proxies” for the measurement error in value regions, substantially alleviating the bias in model estimates. We demonstrate the feasibility of our proposed method on an existing dataset of fMRI measurements and consumer choices. We find a substantial reduction in the bias of model estimates compared to existing baseline methods (the estimated coefficients roughly double), leading to improved inference and out-of-sample demand prediction. After controlling for measurement error, we also find a considerable reduction in the variation of model estimates across consumers.