PredictAutoIm_HGF

This repo includes the analysis code for an experiment on automatic imitation behaviour under different conditions of predictability. Analysis focuses on computational modelling of precision weighted learning to provide insight into the behavioural effects (reaction time differences between congruent and incongruent trials), and comparing Rescorla-Wagner (RW) to hierachical perceptual-response models that incorporate beliefs about levels of uncertainty (Hierachical Gaussain Filter). HGF implemented with TAPAS toolbox: see https://github.com/translationalneuromodeling/tapas

For the experimental paradigm task scripts see: "PredictAutoIm_Exptask" :-)

The dataset of reaction times from a study using this task with 28 healthy adult participants is available as a matlab file in this repository in the extractAll_RTdata.zip

Analysis scripts: perceptual/learning models on log-reaction time

Requires:

1. MATLAB2018+
2. TAPAS toolbox version, HGF version 5-3 used for RW and HGF
3. SPM12 for spm_bms "bayesian model selection" used in An1_HGF2_model_comparision.m

Three sets of scripts here:

1. SRC1_data*.m = data prep. Starting from raw log files, to raw RT checked for missing trials, and trial orders, to logRT as inputs for HGF/RW.

• Output provided in 'extractAll_RTdata.zip'; the directory where these scripts were run with the outputs for summary data scripts (but not raw log files) The file 'P_Alln28.mat' contains a structure 'P' with fields 'trial' and 'ID' for each participant. Within 'trial' there is the details of each trial: 'probabilty', 'cond', 'btrial', 'block', 'keyRT' (see the task scripts for definition of these trial variables). The reaction times from this trial log structure have been organised as matrices of 'responses' and 'inputs' for entering into the HGF within the file 'data_n28.mat'.

2. An1_HGF*.m = HGF modelling An1_RW1_setup_modelfit.m = Rescorla-Wagner An2_*.m = model comparision and evidence Note: that the script 'tapas_hgf_binary_config_MC_SRCstudy.m' is a modified copy of the config script from TAPAS/HGFv5-3 with the prior mean and variance for the perceptional model resulting from variational Bayes optimisation based on response-free model of experimental inputs. This script was saved into the HGF directory so as not to modify the original version of the script.

• Output for the analysis run on the data referenced above (see 'data_n28.mat') is in the 'run_HGFv5-3_n28_logRT.zip' subject-wise outputs includes .fig of individual's parameter trajectories across trials.

3. Further response modelling with a simple case of a drift diffussiond decision model is also included here. This is a supplement to the TAPAS toolbox not scripts that call those toolbox functions. This secondary modelling using MATlAB built in functions from statistical toolbox. The wrapper: run_DD_ResponseModel_MinGenLL.m calls 2 functions 1) "rm_chooseparams2minise.m" while select the parameters that are fixed and those that are optimised; and 2) "rm_loglikelihoodRT.m" will run minimisation of LL, and then generate estimated reaction times.