This week’s speaker in the SWE colloquium is Peter Juslin, who visits on 6 December from Uppsala University, Sweden.

Precise or non-precise: Modeling error distributions to distinguish between intuitive and analytic cognitive processes

The Precise/Not Precise (PNP) model: A Brunswikian conception of intuitive and analytic cognitive processes. Authors: Joakim Sundh, Philip Millroth, August Collsiöö, & Peter Juslin, Uppsala University, Uppsala, Sweden

One of the most influential distinctions in theorizing on human thought is between intuitive and analytic processes (James, 1890), a distinction very much alive in present-day so called dual-systems theories (Evans, & Stanovich, 2013; Kahneman & Frederick, 2002; Sloman, 1996). While this research draws on a popular notion, it has been repeatedly criticized (Gigerenzer, 2011, Keren & Schul, 2009; Kruglanski & Gigerenzer, 2011; Melnikoff & Bargh, 2018) for confusing different distinctions, such as implicit vs. explicit, automatic vs. controlled processes, impulsiveness vs. reflection, and intuition vs. analysis. In this talk, I will describe our ongoing work exploring a more technical (operational) definition of intuitive and analytic cognitive processes inspired by the work by Egon Brunswik (1956). This definition has its roots in the distinction between perceptual and conceptual processes, emphasizing the stochastic nature of these processes and the associated error distributions. Consistently with Brunswik’s claims, the PNP-model assumes that intuitive processes involve a homogenous Gaussian noise around the output of a cognitive algorithm, whereas analytic processes are characterized by a heterogeneous error distribution, effectively sampling from two different distributions: error-free application of the algorithm, and responses affected by errors in execution of the algorithm. We demonstrate that the PNP-model can a) identify the intuitive and analytical nature of cognitive processes, b) can often recover the true parameters of the process more effectively than a model with the standard assumption of a Gaussian error, and c) often provides a better fit to data, suggesting that the standard Gaussian assumption in cognitive modeling is sometimes a misspecification. We propose that this sort of more limited operational definition of intuition and analysis can serve to enlighten the debate on the properties of different cognitive processes, but also that it can allow for more accurate description of cognitive processes characterized by heterogeneous error.