This repository accompanies the following paper:
Li, H.H. and Ma, W.J. (2020). Confidence reports in decision-making with multiple alternatives violate the Bayesian confidence hypothesis. Nature Communications. DOI: 10.1038/s41467-020-15581-6
The repository contains data and MATLAB code for fitting three main models to the data. Please refer to this paper when using the data or the code in this repository.
The data of each participant are saved in .mat files in the 'data' folder.
Data are saved in a structure variable named 'trl' that contains trial-by-trial information.
trl.config: the configuration (stimulus distribution)
trl.estC: confidence reports
trl.estD : category decisions
trl.target_x: horizontal locations of the target dot
trl.target_y: vertical locations of the target dot
trl.theta: the color (angle in unit of radian on a color wheel) of the leftmost category.
Execute wrapper_fitmodel.m to run the optimization. In this script, one can determine which model, experiment and subject to fit. wrapper_fitmodel.m calls fitmodel_on_individual.m to set up and start the optimization.
One can choose to use either BADS or patternsearch for optimization in fitmodel_on_individual.m.
nll_diff_is.m, nll_map_is.m and, nll_en_is.m compute the values of the objective functions for the Difference, Max and Entropy models respectively. In these three functions, response distributions of category decisions and confidence reports are estimated by Monte-Carlo simulations, and the the negative log-likelihood of the model is computed.