This repository provides experimental data, analysis scripts as well as the computational models for the following publication:
The value of confidence: Confidence prediction errors drive value-based learning in the absence of external feedback (https://doi.org/10.31234/osf.io/wmv89).
It is subdivided into several folders:
-
data contains the entire behavioral dataset [data.pkl] as well as an [extraction.py]-file through which variable-specific .npy-arrays can be extracted. Moreover, simulated data and experimental protocols are saved in the sim and para_experiment folder, respectively.
-
model contains different versions of the computational models either including [rl_simple_simchoice.py] or excluding [rl_simple.py] simulated choices as well as optimization scripts [maximum_likelihood.py].
-
plot contains visualization scripts for our publication. The resulting figures are saved under figures. Please refer to the publication for further detail.
-
revision2 contains scripts for calculating and aggregating results of parameter recovery, model recovery and the models' generative performance.
-
run_model contains model fitting scripts with parameter bounds. Model-specific parameter estimates and model evidences are saved in .pkl-format under results/fittingData/.
-
stats contains data analysis scripts. Please refer to the publication for further detail.