Methodology

Continue reading to learn more about our approach and techniques to each aspect of the project.

Data Collection

The vast majority of our data was collected from the openFEC API, which is free and open to the public. To aid in our process of gathering the data, we authored the now open-sourced opynfec package: a Python wrapper for the openFEC API. Feel free to use this for your own use cases, although not all endpoints have been implemented. Pull requests are always welcome!

Node Graph

The node graph was implemented primarily using D3.js (specifically a force simulation. Although we have a lot more data collected than we allow a user to view, it is just not practical to draw that hundreds of thousands of nodes and still learn anything from it. This is why there are user-configurable controls that limit the amount of data that is shown. The data is limited using a standard tree search algorithm where the smaller quantity transactions are pruned off. It can also be truncated by limiting the total number of nodes per candidate.

Map

The map was also implemented with D3.js, primarily a projection.

Clustering

Clustering on the committees and organizations combined data from the FEC as well as some industry data from SOURCE??. Principal Component Analysis was done on the data to reduce the dimensionality, and then it was clustered with a standard $k$-means algorithm.

The results of the clustering are displayed when you hover over a node in the node graph. Displayed with other information is all of the similar committees to the one you have selected.