Inspiration
Across our team, we've worked in political consulting, canvassing, and political news writing. We've found ourselves wanting better tools for smaller political groups to access voter information and find people to reach out to and message, because most of this information is buried behind large corporations and paywalls. We hope that democratizing voter information can make it easier for people to get involved with the democratic process.
What it does
There were 2 goals: make it easy for people to get names and addresses for different uses, but also to help them know which of those people are worth reaching out to. We used a machine learning model to predict what people will do in the 2024 election: 1) how likely they are to vote, and 2) what party they are likely to vote for. We made maps and infographics to demonstrate the data we collected, and built 3 methods of distribution to help different political organizations access that information on a per-voter basis.
- Campaign treasurers can find people who strongly support their party and may be willing to donate.
- Canvassers can find people on the fence about parties, so they can visit and talk to them.
- Voting registrars can target people who have historically been voters but now aren't to help them get back involved. Each of these lets you pick specific counties and ages, then automatically download a list of people that match the criteria for you to go visit or send a flyer to.
How we built it
The core was a Machine Learning model that used voter data. Using their street address and positional coordinates, we could add in additional voter information from Census data. We ran 2 models: one predicted whether or not someone would vote in the next election, and the other predicted which party that person would vote for.
People on our website use the results of that model, but don't have to understand how it works. Our main website is built in Quarto and rendered on GoDaddy. We used R Shiny to build the interactive dashboards for the 3 types of political groups, with our data hosted on Github.
We also used those positional coordinates to build interactive maps in Leaflet.
Challenges we ran into
Our voter data contains 8 million voters, and hundreds of columns of census data and individual voting habits spanning 20 years. When we ran our machine learning models to predict people's future voting patterns and whether or not they should be contacted, our computers often ran out of memory and crashed! After jumping platforms and adjusting memory usage, we were able to run it through Google Colab.
We had a similar problem getting coordinates for the millions of addresses so we could map it: by the time we got them organized our computer would crash again! We had to manipulate our packages
Accomplishments that we're proud of
We've never run machine learning models with this kind of scope before, and we were really pleased with the accuracy we got. When we predicted 2020's voting as a measure of our 2024 voting, we were with .5 percentage points of the true 2020 voter turnout, which gave us a lot of confidence in the results we got.
This is by far the biggest project any of us have led, and doing it in a single 24-hour period with a high degree of precision and overall success is something we're all happy with.
What we learned
We learned how to blend lots of different tools together. We displayed Python ML results in R mixed with interactive Javascript, broadcast through mapping systems into HTML. The classes we've taken use each of these programs in isolation, but this opportunity to work the system together has broadened our understanding of how tech systems work across the board.
What's next for Make Ohio Voter Data Accessible
We want to use the process we've developed to make other states files more available. We also plan to reach out to Ohio groups and let them know about the platform we've developed and how it can help them.
Log in or sign up for Devpost to join the conversation.