Omission

This project contains the materials for the paper "A counterfactual simulation model of causation by omission" by Tobias Gerstenberg and Simon Stephan.

Project structure

.
├── code
│   ├── R
│   │   └── cache
│   ├── flash
│   │   ├── experiment1
│   │   ├── experiment2
│   │   └── experiment3
│   └── python
│       └── figures
├── data
│   ├── empirical
│   └── simulations
├── docs
├── figures
│   ├── diagrams
│   │   ├── experiment1
│   │   ├── experiment2
│   │   ├── experiment3
│   │   └── model
│   ├── experiment_slides
│   │   ├── experiment1
│   │   ├── experiment2
│   │   └── experiment3
│   └── plots
└── videos
    ├── experiment1
    │   ├── mov
    │   └── swf
    ├── experiment2
    │   ├── mov
    │   └── swf
    └── experiment3
        ├── mov
        └── swf

code

R

Analysis and plotting script. You can view a rendered html file of the analysis here.

flash

Adobe Animate files that were used to create the video clips in the different experiments.

python

dependencies

tested with pygame v1.9.6, pymunk v5.6.0
pip install pygame pymunk (reminder that this is Python 3)
quick check: in the code/python/ directory, python simulations.py trial_config.json 2 should show an animation

main files

model.py contains the physics simulation and animation code.
simulations.py contains functions that collect data from running model.py's simulations.
trial_config.json contains a list of objects that specify the initial configuration of a trial.
figures/ contains the image files that model.py loads for animations.
precomputed simulations:
- all_trial0.json and all_trial1.json contain results for the first and second entries in trial_config.json, the hinderer and helper case, respectively. Delay ranges over range(0, 100), angle ranges over range(100, 260, 2), and magnitude ranges over range(10, 30).
- all_obstacle.json and all_non_obstacle contain results for the obstacle (marble A) and non-obstacle (marble B) case, respectively, from the fourth entry in trial_config.json. Delay and magnitude don't matter and are set to 0 and 20 respectively; angle ranges over range(1000, 2600) / 10.

The python files have inline comments explaining implementation details!

running simulations

(Working directory assumed to be code/python/.) To quickly visualize a single trial from a configuration file such as trial_config.json, run python simulations.py trial_config.json <config index> and you should see the animation.

Alternatively, in the Python interpreter, run

>>> import simulations
>>> trials = simulations.load_trials('trial_config.json')

The module provides a number of functions to collect data from simulations.

run_trial runs a single trial and returns the outcome for specified marbles, i.e. whether it went through the exit and what the minimum distance to the exit was, and also the recorded paths for specified marbles.
run_all simulates all possible cases for a given configuration, iterating over specified delay, angle, and magnitude ranges for the var marble. No noise is added regardless of the configuration. It saves the results in a file and returns a count of trials in which the target marble did or did not go through the exit.
get_ideals extracts ideal cases from the output produced by run_all and saves them to a different file. What to consider as "ideal" is specified by passing in a Boolean function, e.g. ideal_helper or ideal_obstacle.
run_ideals_file uniformly samples from the ideal cases output by get_ideals, runs the simulations with noise, and returns a count of trials in which the target marble did or did not go through the exit.
noise_vs_failure repeats run_ideals_file for various settings of the noise parameter in model.py, and saves a tab-separated table of results to a file. This is used to produce the noise vs. failure rate plots at the end of code/R/omission.Rmd.

data

empirical

Raw data files of the three experiments reported in the paper.

simulations

Model simulations for Experiment 1 and Experiment 3.

figures

diagrams

Diagrams illustrating the experiment setup, and how the model works.

experiment_slides

Html slides and figures for Experiments 1-3 which were run via qualtrics.

plots

Results plots as shown in the paper.

videos

Video clips from each experiment in .swf and .mov format.

cicl-stanford/omission