When Neural and Behavioral Forecasting Works: the Critical Role of Matching

Over the last decade, decision neuroscientists have documented several stylized examples of neuroforecasting -- the idea that neuroimaging data collected from a small sample of subjects can be used to estimate how populations will behave. Frequently, these studies demonstrate significant neural forecasts even when forecasting is not possible with choice. Additionally, only three brain regions have been demonstrated to forecast: the Nucleus Accumbens (NAcc), Medial PreFrontal Cortex (MPFC), and the Anterior Insula (AIns). However, these studies vary on which brain regions forecast the market outcome of interest. To date, two concepts have been proposed that may help explain discrepant neuroforecasting results. According to a "partial scaling" account, the generalizability of a choice component depends on how basic or integrative the process in question is. Acording to a "market matching" account, choice components that are closely matched to the market outcome of interest will provide better forecasts. This dissertation comprises three studies that aim to test generalizability and matching in forecasting. Chapters 2 and 3 demonstrate that components that generalize well do not always outperform components that are well matched to the market outcome of interest. Chapter 4 demonstrates that when market matching is achieved by matching demographic features between the sample and the population, less generalizable components forecast better. Chapter 5 discusses implications of this research for theory and practice.