Statistical modeling: Harnessing uncertainty and variation in neuroimaging data

Statistics has become an integral part of neuroimaging research. It is used to explain variation in the data and make decisions in the face of uncertainty about whether or not effects are significant. This chapter focuses on statistical modeling and its role in neuroimaging research and decision-making. We highlight examples of successful statistical models and how they have been applied to neuroimaging data, and the types of conclusions they have allowed. Further, we describe how to build, test, and interpret statistical models in the context of neuroimaging data and the necessary assumptions needed to make valid inferences and conclusions. We particularly highlight pitfalls when developing statistical models and controversies in their application to neuroimaging data. These include issues related to multiple comparisons correction and sample size, as well as controversies related to circularity and pipeline variability. We conclude with a discussion of future directions and open research problems, such as large n studies, translational neuroimaging, and causal inference and brain stimulation.