# Time Bandits
## Table of Contents
* [General info](#general-info)
* [Data](#experimental-data)
* [Behavioral Analyses](#behavioral-analyses)
* [RL model](#rl-model)
* [Model-based Analyses](#model-based-analyses)
* [Other](#other)
## General info
This repository contains all the code and (anonymized) data necessary for reproducing [Wu, Schulz, Pleskac, & Speekenbrink (*bioRxiv* 2021)](https://charleywu.github.io/downloads/wu2021time.pdf). The following text provides a summary of each file and it's function, with further comments provided throughout the code.
## Experiment data
* `Data/fullData.csv` contains the anonymized participant data
## Behavioral Analyses
* `behavioralPlots.R` contains the behavioral plots presented in the paper
* `behavioralRegressions.R` contains the mixed effects regressions performed on the behavioral data
## RL model
* `models.R` defines the RL models, specifically the Bayesian mean tracker (BMT)
* `generatebmtpredictions.R` reads in participant data and generates predictions at each timepoint about the expected reward and uncertainty of each arm, based on the RL model (BMT). The predictions are saved in `modelPreds/`. In addition, `BMTpredictionsPlot.R` examines the pattern of predictions, separated by time pressure and payoff condition
## Model-based Analyses
* `hierarchicalSoftmax.R` runs a softmax regression, using the model(s) defined in `STANmodels/*.stan` to predict choices as a function of the posterior means and variances estimated from `generatebmtpredictions.R`
* `hierarchicalRT.R` runs a hierarchical regression predicting reaction time based on the posterior means and variances estimated from `generatebmtpredictions.R`. The model posteriors are stored in `brmsModels/`.
* `lba.R` estimates a linear ballistic accumulator model using code from [Annis, Miller, \& Palmeri (2017)](https://link.springer.com/article/10.3758/s13428-016-0746-9), which is located in `LBA/lba_single.stan`. The parameter estimates of the LBA model are then analyzed and plottedin `lbaParams.R`.
* `analyseLBA.R` fits a Bayesian mixed effects regression using the posterior means and variances estimated from `generatebmtpredictions.R` to predict the estimated drift rates estimated from `lba.R`, The model posteriors are stored in `brmsModels/`
* `posteriorPlots.R` puts together the results of all model-based analyses into a single plot (in addition to more detailed SI figures)
* `modelComparison.R` is compares the different softmax models
* `modelInteractions.R` visualized the different interactions of the RT and LBA regression models
* `modelsVsBehavior.R` links the outcomes of the model-based analysis with behavioral patterns
## Other
* `statFormat.R` is used to provide LaTeX formatted outputs for the statistical tests
