Using Hierarchical Bayesian Models to Test Complex Theories About the Nature of Latent Cognitive Processes
From a computational perspective, the primary goal of cognitive science is to infer the influence of unobservable psychological constructs on observed behavioral data. Cognitive models can facilitate this inference by more directly expressing a theoretical cognitive processes through relationships among psychologically interpretable parameters. If cognitive models are developed in a hierarchical Bayesian framework, significantly more nuanced and complex theories may be expressed, thereby allowing for deeper insight into the nature of latent cognitive processes. Here, I highlight three specific benefits of a hierarchical Bayesian approach to cognitive modeling, with a special emphasis on model-based theory testing. First, after a brief introduction to Bayesian methods, I discuss how hierarchical modeling permits one to coherently analyze data from highly complex experimental designs. To demonstrate this first benefit, I describe how the development of a hierarchical Bayesian metaregression model inspired a new technique for quantitative assessment of the robustness of a psychological theory. Next, I describe how hierarchical modeling enables simultaneous implementation of multiple different styles of cognitive models, which permits theory testing at a higher level of abstraction. I demonstrate this through a cognitive latent variable model-based comparison of theories of attention ability. Finally, I discuss how hierarchical modeling can be used to express highly complex neurocognitive processes, as demonstrated through a new approach to joint modeling of neural and behavioral data that is better suited to hypothesis testing than previous approaches. Ultimately, I contend that hierarchical Bayesian cognitive modeling is an ideal way to perform more powerful and informative analyses of behavioral data by radically expanding the scope of questions we may ask about the nature of latent cognitive processes.