Project

Back to overview

Parameter Intercorrelations in Cognitive Models - Prevalence, Causes, and Solutions

English title Parameter Intercorrelations in Cognitive Models - Prevalence, Causes, and Solutions
Applicant Scheibehenne Benjamin
Number 172806
Funding scheme Project funding
Research institution Faculté d'économie et de management Université de Genève
Institution of higher education University of Geneva - GE
Main discipline Psychology
Start/End 01.09.2017 - 29.02.2020
Approved amount 245'989.00
Show all

All Disciplines (3)

Discipline
Psychology
Economics
Science of management

Keywords (4)

Judgment and decision making; Experimental psychology; Hierarchical Bayesian statistics; Cognitive modelling

Lay Summary (German)

Lead
Mathematische Modelle nehmen ein immer grössere Rolle in den empirischen Sozialwissenschaften ein. Jedoch haben viele dieser Modelle gemein, dass ihre zentralen Parameter nicht eindeutig identifiziert sind, das verringert ihre Interpretierbarkeit und Vorhersagegenauigkeit. Das Projekt untersucht und testet moegliche Lösungsansätze.
Lay summary

Hintergrund

Viele in den empirischen Sozialwissenschaften angewendete Modelle beschreiben wie Personen eine Auswahl zwischen unterschiedlichen Alternativen treffen. Die Modelle messen verschiedene latente Variablen, um so individuelles Verhalten in unterschiedlichen Situationen vorhersagen zu können. Obwohl es sich dabei um teilweise sehr unterschiedliche Kontexte, zum Beispiel aus der Wahrnehmungs-, Erinnerungs- oder Entscheidungsforschung handelt, gibt es Funktionen und Parameter welche viele dieser Modelle gemeinsam haben. Für die Interpretierbarkeit der Modellparameter ist es wichtig, dass diese möglichst voneinander unabhängig sind. Wenn einige Parameter stark miteinander korrelieren, dann ist es schwierig, diese sinnvoll zu interpretieren. Hohe Parameterkorrelationen weisen ausserdem darauf hin, dass die Modelle unnötig kompliziert sind. Erste Untersuchungen zeigen, dass selbst bei Prominenten Modellen in der Literatur die Parameter häufig stark miteinander Korreliert sind.

Ziel

Das Projekt hat zwei zentrale Ziele. In einem ersten Schritt soll die Prävalenz und die Auswirkung des Problems von Parameterkorrelationen anhand von drei zentralen Modellen der Kognitionswissenschaften erfasst werden. In einem nächsten Schritt sollen mögliche Lösungsansätze identifizert und getestet werden. Dieser Ansatz zielt darauf ab, die Bedeutsamkeit der Problematik aufzuzeigen und gleichzeitig leicht anwendbare Werkzeuge zur Problemlösung anzubieten.

Bedeutung

Der Einfluss kognitiver mathematischer Modelle in den empirischen Sozialwissenschaften hat in den letzten Jahren immer mehr zugenommen. Wenn jedoch selbst populäre Modelle, die nicht nur in der Forschung sondern auch in der Praxis zu Anwendung kommen, Probleme in der Interpretierbarkeit und der Vorhersagegenauigkeit aufweisen, kann die Anwendung dieser Modelle zu Fehlschlüssen führen. Unser Projekt soll dazu beitragen, kognitive Modellierungsansätze zu verbessern.

Direct link to Lay Summary Last update: 05.04.2017

Responsible applicant and co-applicants

Employees

Project partner

Publications

Publication
A new way to guide consumer's choice: Retro-cueing alters the availability of product information in memory
Krefeld-Schwalb Antonia, Rosner Agnes (2020), A new way to guide consumer's choice: Retro-cueing alters the availability of product information in memory, in Journal of Business Research, 111, 135-147.
Empirical comparison of the adjustable spanner and the adaptive toolbox models of choice.
Krefeld-Schwalb Antonia, Donkin Chris, Newell Ben R., Scheibehenne Benjamin (2019), Empirical comparison of the adjustable spanner and the adaptive toolbox models of choice., in Journal of Experimental Psychology: Learning, Memory, and Cognition, 45(7), 1151-1165.
Hypothesis-Testing Demands Trustworthy Data—A Simulation Approach to Inferential Statistics Advocating the Research Program Strategy
Krefeld-Schwalb Antonia, Witte Erich H., Zenker Frank (2018), Hypothesis-Testing Demands Trustworthy Data—A Simulation Approach to Inferential Statistics Advocating the Research Program Strategy, in Frontiers in Psychology, 9, 1-24.

Scientific events

Active participation

Title Type of contribution Title of article or contribution Date Place Persons involved
Lab meeting Center for Decision Sciences Individual talk Structural Parameter Interdependencies in Computational Models of Cognition 20.09.2019 New York CIty, United States of America Krefeld-Schwalb Antonia;
Subjective Probability Utility and Decision Making Conference (SPUDM) Talk given at a conference Structural Parameter Interdependencies in Computational Models of Cognition 18.08.2019 Amsterdam, Netherlands Scheibehenne Benjamin; Krefeld-Schwalb Antonia;
Gästekolloquium Individual talk Hypothesis-testing demands trustworthy data-a simulation approach to inferential statistics advocating the research program strategy 05.11.2018 Zürich, Switzerland Krefeld-Schwalb Antonia;
Tagung Experimentell Arbeitender Psychologen (TEAP) Talk given at a conference Parameter collision in probabilistic models of cognition: How to separate parameterization of evidence strength and choice noise 08.03.2018 Marburg, Germany Scheibehenne Benjamin; Krefeld-Schwalb Antonia;


Associated projects

Number Title Start Funding scheme
160962 Modelling cognitive processes of judgment and decision making within a Bayesian hierarchical framework 01.06.2015 International short research visits
149846 Modeling Cognitive Choice Processes in the Health Domain 01.03.2014 Project funding
130149 Testing Cognitive Process Models of Choice in a Consumer Context 01.09.2010 Project funding

Abstract

Using mathematical models of cognitive processes to explain and predict behavior is an increasingly popular approach that has great potential to boost new scientific insights across various disciplines in the behavioral and social sciences, including psychology, economics, and marketing (Bourgine & Nadal, 2004; Lee & Wagenmakers, 2013). Cognitive models aim at identifying underlying psychological processes of overt behavior. They have been developed for a wide range of cognitive functions, especially (but not exclusively) in research on memory, judgment, and decision making (Busemeyer & Diederich, 2010; Lewandowsky & Farrell, 2011). These models commonly feature several free parameters that can be estimated based on empirical data and that have a psychological interpretation (e.g., they relate to specific aspects such as memory capacity, learning rate, risk preferences, or response errors, to name but a few). The estimated parameters allow to disentangle processes contributing to the observable behavior and thus can reveal important insights regarding the underlying cognitive processes. In addition, they allow to measure individual differences or the effect of experimental manipulations. For understanding variability on the parameters, they are often linked to other variables such as personality traits, physiological measures, genetic, or neural measures (fMRI or EEG) (D’Esposito, 2007; Gevins, Smith, McEvoy, & Yu, 1997; Riefer & Batchelder, 1988; Yechiam, Busemeyer, & Stout, 2004). For these purposes it is crucial that the estimates of the model parameters are accurate, reliable, and valid.The interpretability and reliability of the parameter estimates can be jeopardized, however, to the extent that parameters within a cognitive model are correlated. While sometimes such correlations are to be expected on theoretical grounds (e.g., due to their shared relationship with general cognitive functioning, such as intelligence, parameters of outcome sensitivity and noise are often positively correlated), parameter correlations may also arise from the formal architecture of the models. In the latter case, such unwarranted correlations make it harder to interpret the parameters in isolation and they blur the model’s predictions; eventually, they can even lead to wrong inferences about the underlying cognitive processes (see van Ravenzwaaij, Dutilh & Wagenmakers, 2011, and the comment on this in Heathcote, Brown & Wagenmakers, 2015, for an example) .Recent research indicates that unwarranted correlations among parameters may be quite prevalent in existing cognitive models (Scheibehenne & Pachur, 2015). Furthermore, parameter correlations are often difficult to identify and thus may go unnoticed due to limitations in modeling practices (Heathcote et al., 2015). However, up to now there has been no comprehensive investigation of the problem, nor any suggestion for generally applicable solutions to parameter correlations.The research proposal at hand aims to address this problem in two steps: First, we will identify and quantify the extent of parameter correlations across current and widely used cognitive models. Second, we will develop and propose applicable solutions for reducing these correlations. Given the significance of the investigated models across different fields of research in the behavioral sciences (including applied fields such as health sciences, marketing, or economics), it is important to understand these correlations more closely and to develop solutions to reduce them. Focusing on different models, from different fields allows to generalize the solutions to other models as well, which has a potential impact for the cognitive science as a whole. To achieve these goals, the present research plan incorporates a two-year research project that consists of three work packages. In the first work package, we will provide an extensive overview on the problem with a quantitative review on parameter correlations in different cognitive models. In the second work package, we will develop formal and experimental solutions to reduce these correlations based on reanalyzing existing data and by means of parameter recovery studies. In the third work package, we will test different solutions empirically, using newly collected data.
-