Dynamical Models of Decision Confidence in Visual Perception: Implementation and Comparison
Sebastian Hellmann, Michael Zehetleitner, Manuel Rausch, Catholic University of Eichstätt-Ingolstadt, Germany
Posters 2 Poster
Pacific Ballroom H-O
Fri, 26 Aug, 19:30 - 21:30 Pacific Time (UTC -7)
Confidence refers to the degree of belief that a decision is correct. Formal models based on the idea of sequential sampling offer the possibility of relating confidence judgments to the dynamics of the decision making process. The models are able to account for the joint distribution of choice, response time and confidence and the effect of task difficulty. We compared sequential sampling confidence models, which are either based on the drift diffusion or the race model account for decision making, including the new dynamic evidence and visibility model (dynWEV). DynWEV assumes that confidence is not solely based on task relevant evidence but also on information about stimulus reliability conveyed by task irrelevant stimulus features. We fitted the models to data from three different experiments including two orientation discrimination tasks and a motion direction identification task. A quantitative model comparison reveals that diffusion based dynWEV outperforms race models in all experiments and 2DSD in two of the three experiments. Moreover, dynWEV was the only model that could reproduce all qualitative relationships between discriminability, response time and confidence judgments. All models were implemented in the R package dynConfiR which is available on CRAN (https://CRAN.R-project.org/package=dynConfiR).