Replication + Datasets

From early September through Election Day, the MIT Election Data + Science Lab, as part of our contributions to the Stanford-MIT Elections Performance Central, has been documenting core election performance statistics in a number of states of interest, and posting them to our social media channels and to the Elections Performance Central website.

These plots cover the number of registered voters, the number of mail-in ballots issued to and received from voters, and the number of early in-person ballots cast. We update plots of these data as often as a new dataset is released — some are updated less than once a month, others are updated every day — across nine key states: Arizona, Florida, Georgia, Michigan, Nevada, North Carolina, Pennsylvania, Texas, and Wisconsin (and we pepper in data from other states as the opportunity arises). Wherever possible, we break these numbers down by party, and compare them to data collected the same number of days before the 2020 election. 

In the last few days before the election we updated and posted about 10 plots per day. Our workflow is:

  1. Automatically go to the URL for any dataset that is regularly updated and save it
  2. Run a script that updates a plot of the feature of interest
  3. Post that plot

So far we have only released the plots. Now, we are releasing the data behind the plots, and all the code that we use to collect those data, and all the code we use to turn those data into plots. You can find all of that on GitHub here: github.com/MEDSL/2024-EPC

This repository more or less contains almost all the information you would need to replicate our work. There are only two caveats. First, we have not retained our actual file structure, which is complicated by many factors that don’t need to be reproduced here, so the code do not directly run; we have assembled them into a simple folder structure but some setup would be needed for you to have the same automatic process running on your machine that we have set up on ours. Second, some files are either too big to share, or require purchasing from a state source; in those cases, we direct you to the original file source rather than posting it ourselves.