PI: Adam Sheingate (Krieger School of Arts & Sciences)
Efforts to weaken or undermine the electoral system are one of several existential threats to democracies around the world. Effective electoral oversight is therefore essential if citizens are to perceive elections as both free and fair and is key to the survival of the democratic project. The application of AI and data science tools can help deliver effective and trusted elections, advance social science research, and inform the public about campaigns and elections. For example, natural language processing can automate the classification of millions of campaign records in ways that facilitate election monitoring and scholarly analysis. The development of multi-lingual NLP models makes it possible to extend these tools to the analysis of global election spending and text-based descriptions of expenditures in different languages.
To that end, we are using our IDIES seed grant to explore how data science tools can assist in the collection and dissemination of campaign finance and political spending data from around the world. In June, we organized a workshop on “The Global Elections Dashboard: Harnessing Data Science to Enhance Electoral Oversight” at the JHU Bloomberg Center in DC. The meeting brought together faculty from both the Krieger and Whiting schools along with data journalists and transparency advocates from the US, UK, and Brazil. We discussed a variety of approaches and tools that could assist with the analysis of campaign data such as large language models, machine-learning classifiers, optical character recognition, and entity disambiguation among others. We also discussed how a global elections dashboard that enables users to visualize and analyze campaign data could serve various stakeholders including election regulators, transparency advocates, and academic researchers.
Currently, a team of research assistants is assessing the availability of campaign finance data in consolidated democracies around the world. Researchers will determine whether the election management body in each country has bulk data available for download or through an API, whether the data is in a language other than English, the type of data available (contributions or spending), and for which years. From this survey, we hope to identify a subset of candidate countries with sufficient data available to serve as the basis for a second meeting planned for early 2025. At this next meeting, we will focus on identifying specific tools and tasks needed to produce a unified dataset to populate the dashboard such as dataset linkages, data dictionaries, and multi-language models.
Working with colleagues in the UK and the International Institute of Democracy and Electoral Assistance (International IDEA), we are also exploring ways to develop a global standard for the collection and reporting of campaign finance data. We hope to identify a set of best practices for data on political contributions, categories of political spending, and vendors who provide products and services to parties and candidates. Standardized reporting will improve the transparency and accountability of electoral processes while also making it easier for social scientists, journalists, transparency advocates, and citizens to track the flow of money in politics.
Finally, the 2024 U.S. Election affords us an opportunity to test some of our ideas about the analysis and visualization of campaign spending data. Using a supervised machine-learning model, we are able to classify millions of individual expenditure records into categories such as media, fundraising, or digital advertising. We can then compare these categories of spending over time and across election cycles as well examine differences in spending by political party or political office. Figure 1 shows a screenshot comparing media spending by party in campaigns for the U.S. House of Representatives in 2020 and 2022. We will continue to update the dashboard as more data becomes available for the 2024 election cycle. We will also add additional features to the dashboard that will enable users to examine spending by individual candidates, specific campaign committees, or particular states or congressional districts.
You can explore the U.S. dashboard for yourself at https://election-spending-data. shinyapps.io/dashboard/.
