Analysis repository for Silent Failures in Automation. Experiment Repository: https://github.com/callummole/SP_18-19

Preprint: Mole et al., 2020, Predicting Takeover Response to silent automated failures (https://psyarxiv.com/bv2pt/).

Data information about raw and processed data can be found in the Data folder README.

The key analysis files are in the folder manuscript_analysis. Also in this folder are the saved model fits.

The folder Processing contains code for eye-tracking, most of which has not been applied to the current dataset so is not worth looking at. The folder Post-processing contains a range of part-baked scripts of different analysis adventures - also not worth looking at.

The analysis pipeline from the Raw_Data folder hosted on the OSF (https://osf.io/aw8kp/) to the manuscript figures is as follows:

• Extract the Raw_Data folder into the local repo Data folder.

To generate the condition onset times and steering angle biases, and generate the simulated time-to-line-crossings (all relevant csvs are also found in the Raw Data folder so you can skip these steps), do the following:

1. run TrackSimulation.py to output 'simulated_roadcrossing.csv'.
2. run TrackSimulation_sobol.py to obtain the relevant steering angle biases for the random conditions. This file outputs 'SimResults_samplesobol_onsettimes.csv'.
3. run plot_failures_perspective.py to plot all the failures and output 'simulated_ttlcs.csv', used in the analysis.

To run the analysis and output figures found in the manuscript, do the following:

1. run processing_steering_only.py to output a collated steering csv in the Data folder.
2. optional, run save_as_rds.R to save time for loading csvs into R.
3. run manuscript_figures_ttlc.rmd for TTLC results (put REFIT = TRUE to refit model).
4. run manuscript_figures_swa.rmd for steering results (put REFIT = TRUE to refit model).
5. run cogtask_performance.rmd for the cognitive task analysis.