This software package is a companion to our upcoming paper for the Precision Nudging project. The scientific aim of this project is to test multiple Machine Learning algorithms to find the treatment effect for a nudge for people, given their personal circumstances. For this, we have created an involved simulation setup that can compare different methods, knowing the true treatment effect for each individual in the dataset (something that is not possible with real data).
Thus, we are interested in the heterogeneous treatment effect of nudges. Conventionally, heterogeneous treatment effects are found by dividing the study data into subgroups (e.g., men and women, or by age) and comparing the conditional average treatment effect (CATE) between subgroups. A challenge with this approach is that each subgroup may have substantially less data than the study as a whole, and there may not be enough data to accurately estimate the effects on subgroups. Recently and increasingly, however, machine learning is used to estimate heterogeneous treatment effects, even to the level of individuals, see e.g. Künzel et al. 2019. In this study, we apply different machine learning models to determine the heterogenity of treatment effects.
The project can be split in several steps:
- We investigate different methods of determining the heterogeneity of treatment effects;
- We create realistic synthetic data to compare the performance of the different methods;
- We investigate the validity of the methods on open data from published studies.
We have a quick online tutorial.
Install a version of Python>=3.7 for this project.
Open a command line terminal (not a python interpreter!).
Clone the repository with git clone https://github.com/UtrechtUniversity/nudging.git.
Go into the newly created directory cd nudging (Linux/MacOS).
Install the package with pip install .
The tutorial is available under examples/tutorial.ipynb.
Contributions are what make the open source community an amazing place to learn, inspire, and create. Any contributions you make are greatly appreciated.
To contribute:
- Fork the Project
- Create your Feature Branch (
git checkout -b feature/AmazingFeature) - Commit your Changes (
git commit -m 'Add some AmazingFeature') - Push to the Branch (
git push origin feature/AmazingFeature) - Open a Pull Request