How cognitive modeling can benefit from hierarchical Bayesian models

https://doi.org/10.1016/j.jmp.2010.08.013Get rights and content

Abstract

Hierarchical Bayesian modeling provides a flexible and interpretable way of extending simple models of cognitive processes. To introduce this special issue, we discuss four of the most important potential hierarchical Bayesian contributions. The first involves the development of more complete theories, including accounting for variation coming from sources like individual differences in cognition. The second involves the capability to account for observed behavior in terms of the combination of multiple different cognitive processes. The third involves using a few key psychological variables to explain behavior on a wide range of cognitive tasks. The fourth involves the conceptual unification and integration of disparate cognitive models. For all of these potential contributions, we outline an appropriate general hierarchical Bayesian modeling structure. We also highlight current models that already use the hierarchical Bayesian approach, as well as identifying research areas that could benefit from its adoption.

Introduction

Bayesian statistics provides a compelling and influential framework for representing and processing information. Over the last few decades, it has become the major approach in the field of statistics, and has come to be accepted in many or most of the physical, biological and human sciences. This paper, and this special issue, are about what one particular niche within Bayesian statistics, in the form of hierarchical models, can contribute to cognitive modeling.

It would be wrong to claim that there is complete agreement on exactly how Bayesian analyses should be conducted and interpreted. Like any powerful and fundamental idea, it can be conceived and formulated in a variety of ways. At the basic theoretical level, the ‘objective Bayesian’ approach expounded by Jaynes (2003) encourages a different style of thinking about Bayesian analysis than the ‘subjective Bayesian’ approach of de Finetti (1974). At the practical level of conducting Bayesian analyses, there is also a spectrum, ranging from work that closely follows the objective viewpoint (e.g., Gregory, 2005, Sivia, 1996), to work that is more agnostic or adopts a naturally subjective position (e.g., Congdon, 2006, Gelman et al., 2004, Gelman and Hill, 2007). There are many additional subtleties and perspectives in the excellent accounts provided by Bernado and Smith (2000), Lindley (1972) and MacKay (2003) and others.

But Bayesian statistics is in agreement on the very basic issues. Knowledge and uncertainty about variables is represented by probability distributions, and this knowledge can be processed, updated, summarized, and otherwise manipulated using the laws of probability theory. These commitments distinguish Bayesian statistics from other competing frameworks, especially those based on frequentist views of probability, and sampling distribution approaches to handling uncertainty. What Bayesian statistics offers is a remarkably complete, coherent and intuitive method for understanding what is known, based on the assumptions being made, and the information that is available.

Because Bayesian statistics provides a formal framework for making inferences, there are different ways it can be applied in cognitive modeling. One way is to use Bayesian methods as a statistician would, as a method for conducting standard analyses of data. Traditionally, the framework for statistical inference based on sampling distributions and null hypothesis significance testing has been used. Calls for change, noting the clear superiority of Bayesian methods, date back at least to the seminal paper of Edwards, Lindman, and Savage (1963), and have grown more frequent and assertive in the past few years (e.g., Gallistel, 2009, Kruschke, 2010a, Lee and Wagenmakers, 2005, Wagenmakers, 2007). It seems certain Bayesian statistics will play a progressively more central role in the way cognitive science analyzes its data.

A second possibility is to apply Bayesian methods to cognitive modeling as a theoretician would, as a working assumption about how the mind makes inferences. This has been an influential theoretical position for the last decade or so in the cognitive sciences (e.g., Chater et al., 2006, Griffiths et al., 2008). Most existing work has focused on providing ‘rational’ accounts of psychological phenomena, pitched at the computational level within the three-level hierarchy described by Marr (1982). These models generally use Bayesian inference as an account of why people behave as they do, without trying to account for the mechanisms, processes or algorithms that produce the behavior, nor how those processes are implemented in neural hardware. More recently, however, there have also been attempts to apply computational sampling approaches from Bayesian statistics as a theoretical metaphors at the algorithmic and implementation levels. In this work, models are developed in which people mentally sample information (e.g., Sanborn, Griffiths, & Shiffrin, 2010). These uses of Bayesian statistics as theoretical analogies have led to impressive new models, and raised and addressed a range of important theoretical questions. As with all theoretical metaphors–including previous ones like information processing and connectionist metaphors–“Bayes in the head” constitutes a powerful theoretical perspective, but leaves room for other complementary approaches.

A third way to use Bayesian statistics in cognitive science is to use them to relate models of psychological processes to data (e.g., Lee, 2008, Rouder et al., 2005, Wetzels et al., 2010). This is different from the data analysis approach, because the focus is not generic statistical models like the generalized linear model. Instead the goal is relate a detailed model of some aspect of cognition to behavioral or other observed data. One way to think of the distinction is that data analysis typically does inference on the measured dependent variables from an experimental design–measures of recall, learning, response times, and so on–whereas modeling applications typically do inference on latent psychological parameters–memory capacities, learning rates, decision criteria, and so on–that control the behavioral predictions of the model. It is also different from the use of Bayesian inference as a metaphor for the mind (Kruschke, 2010b). There is no requirement that the cognitive models being related to data make Bayesian assumptions. Instead, they are free to make any sort of processing claims about how cognition works. The goal is simply to use Bayesian statistical methods to evaluate the proposed model against available data.

This third approach is the focus of the current special issue. We think it is an especially interesting, important, and promising approach, precisely because it deals with fully developed models of cognition, without constraints on the theoretical assumptions used to develop the models. The idea is to begin with existing theoretically grounded and empirically successful models of cognition, and embed them within a hierarchical Bayesian framework. This embedding opens a vista of potential extensions and improvements to current modeling, because it provides a capability to model the rich structure of cognition in complicated settings.

In the remainder of this paper, we identify four major new capabilities offered by the hierarchical Bayesian extensions of cognitive models. We discuss each capability, focusing on how it can help theory and model development, and identifying places where they have already been applied, or could and should be applied soon.

Section snippets

Benefits of using hierarchical Bayes in cognitive modeling

Before discussing the potential contribution hierarchical Bayesian methods can make to cognitive modeling, we need to say what we mean by ‘hierarchical Bayes’. We do that in the next section–by characterizing its complement, in the form of the currently dominant non-hierarchical modeling approach–and then discuss the advantages of the hierarchical approach.

Conclusion

Non-hierarchical approaches to understanding cognitive processes dominate the current landscape. The basic approach can probably fairly be caricatured as one of identifying a psychological phenomenon (e.g., generalization, memory, decision-making), finding an interesting task relating to some aspect of the phenomenon (e.g., similarity judgments, recall, two-alternative forced-choice decisions) and building a model that can fit empirical data from the task using a few psychologically meaningful

References (68)

  • D. van Ravenzwaaij et al.

    Cognitive model decomposition of the BART: assessment and application

    Journal of Mathematical Psychology

    (2011)
  • A.N. Sanborn et al.

    Uncovering mental representations with Markov chain Monte Carlo

    Cognitive Psychology

    (2010)
  • J. Vandekerckhove et al.

    A cross random effects diffusion model for speeded semantic categorization decisions

    Acta Psychologica

    (2010)
  • W. Vanpaemel

    Constructing informative model priors using hierarchical methods

    Journal of Mathematical Psychology

    (2011)
  • R. Wetzels et al.

    Bayesian parameter estimation in the expectancy valence model of the Iowa gambling task

    Journal of Mathematical Psychology

    (2010)
  • M.D. Zeigenfuse et al.

    Finding the features that represent stimuli

    Acta Psychologica

    (2010)
  • M.D. Zeigenfuse et al.

    A general latent assignment approach for modeling psychological contaminants

    Journal of Mathematical Psychology

    (2010)
  • J.M. Bernado et al.

    Bayesian theory

    (2000)
  • I. Borg et al.

    Multidimensional similarity structure analysis

    (1987)
  • N.L. Cassimatis et al.

    Ability, parsimony and breadth in models of higher-order cognition

    Cognitive Science

    (2008)
  • P. Congdon

    Bayesian statistical modeling

    (2006)
  • B. de Finetti
    (1974)
  • M.R. Dougherty et al.

    The psychological plausibility of fast and frugal heuristics

    Psychological Review

    (2008)
  • W. Edwards et al.

    Bayesian statistical inference for psychological research

    Psychological Review

    (1963)
  • C.R. Gallistel

    The importance of proving the null

    Psychological Review

    (2009)
  • A. Gelman et al.

    Bayesian data analysis

    (2004)
  • A. Gelman et al.

    Data analysis using regression and multilevel/hierarchical models

    (2007)
  • G. Gigerenzer et al.

    Simple heuristics that make us smart

    (1999)
  • G. Gillund et al.

    A retrieval model for both recognition and recall

    Psychological Review

    (1984)
  • D.M. Green et al.

    Signal detection theory and psychophysics

    (1966)
  • P.C. Gregory

    Bayesian logical data analysis for the physical sciences

    (2005)
  • T.L. Griffiths et al.

    Bayesian models of cognition

  • T.L. Griffiths et al.

    Topics in semantic representation

    Psychological Review

    (2007)
  • E. Heit et al.

    Are there two kinds of reasoning?

  • Cited by (239)

    View all citing articles on Scopus
    View full text