Decision science evolve towards the use of formalized mathematical and computational models of choice (such as sequential accumulator or drift
diffusion models, DDMs). These may represent a key step forward, but
only if they properly incorporate the conceptual and theoretical richness of the affective sciences.
Computational models increasingly inform our understanding of decision processes, but the influence of affect on these computations remains in its infancy.
This review lays the groundwork for integration based on the strengths of both the affective sciences and computational modeling.
Theories often posit the existence of hypothetical concepts. However, these concepts are frequently abstractly defined, which means that they must be operationalized before predictions can be tested. In the context of computational models, operationalization can involve mapping a concept onto the computational parameters of the relevant model. This mapping can take at least one of two forms: direct concept-to-parameter mapping, or indirect mapping through a mediating mechanism that is not directly represented in the model. Once a concept has been mapped to a parameter, the predictions can be tested, which then informs the theory.
However, a conceptual framework is only one piece of the puzzle. Researchers will still need to grapple with some of the concrete challenges that arise. Figuring out how best to address these issues will depend on careful theorizing that draws on all available empirical methods. Building clear linkages to neural and physiological mechanisms represents an important avenue for future research. There is, of course, a long history of research on the neuroscience of both decision-making and affect, with many different perspectives.
Although introducing a further field of work to the mix brings its challenges, it also provides additional methods that can inform and constrain the development of formalized models. For example, neural and physiological measures could reveal the focus of internal attention in the same way that eye gaze reveals external attention. Thus, physiological measures, such as interoceptive sensitivity, could help to determine when people attend to their affective states or interoceptive information, providing a way to measure the influence of these states on evidence accumulation.
Despite these strengths, it is important to acknowledge a few limitations of computational modeling.
First, caution is warranted when drawing connections between concepts, operational variables, and computational parameters. For example, measures that appear at face value to provide support for dual systems may be inconclusive owing to issues with construct validity (i.e., correspondence between a conceptual variable and its operationalization) and internal validity (i.e., confounding explanations).
A manipulation such as time pressure may not solely influence the extent of processing (i.e., decision threshold), and could also influence how people allocate attention, which could, in turn, influence other decision processes (e.g., drift rate). Unless such processes are accounted for, the model may identify spurious effects. Therefore, computational modeling provides a powerful tool, but it cannot substitute for experimental ingenuity and careful observation.
Second, … many important concepts from affective science are unlikely to be captured by a computational parameter. As Marr’s levels of analysis suggest, computations can be understood in terms of their purpose, algorithmic solution, and biological implementation. Computational models are best understood at the algorithmic level, but other levels of analysis are equally important. For example, the purpose of evidence accumulation (e.g., maximizing the rate of reward outcomes) can in some cases suggest important revisions to the standard DDM.
Computational models should be informed by work at other levels of analysis (e.g., the functions of emotions), but may not directly address the important debates and issues.
Surtil : 