VA Election Night: Data Reliability Issues
EPEC Team offers early analysis about VA's performance of vote-data feeds after Tuesday's historic national election swept in a Republican White House, Senate majority and likely hold of the House.
By Jon Lareau, EPEC Team volunteer Chief Technology Officer
Our team at EPEC monitored the Election Night Reporting (ENR) data feed published by the Virginia Department of Elections (ELECT) during the 2024 General election.
While we have not finished collecting or examining the election results, we have observed a few issues already that are presented below.
There were three specific issues that we have noticed so far in our analysis of the ENR from ELECT. None of these issues look to have impacted the count, as far as we can tell, but these are system or procedural issues that should be documented, addressed and corrected going forward.
We will continue to update on these issues as we find out more; with much more data analysis to conduct, here are some preliminary observations.
1. We observed a distinct jump, and then immediate reversion, in 11 localities at 12 am on Nov. 6th in the ENR data feed.
Bland, Campbell, Carrol, Cumberland, Dinwiddie, Floyd, Nottaway, Orange and Rappahannock counties, as well as Franklin City and Norfolk City, all show the same “glitch.”
The images of data plots of a few of the counties (see below) show this “glitch” in the data feed, all occurring at midnight. We have not (yet) reached out directly to ELECT or any of these counties to inquire as to the reason. We will update if we find out anything as to the cause of this “glitch.” (Mobile readers can “pinch up” the images.)
2. The second issue we noticed was that Orange County had the first report of their totals at 10:10 PM pm Nov. 5th, and initially reported the election day count for Donald Trump (7,891) and Kamala Harris (3,852).
Then the data feed immediately removed the election day count information and began reporting just the Early Voting and Mailed Absentee numbers (5,593 Trump, 4,259 Harris), only to add the same Election Day totals back in to the data feed at 9:06 AM on Nov. 6th.
3. Significant issues with the reliability of the public feed provided by VA ELECT over the course of the vote-counting proved interesting.
Multiple team members were monitoring the feed from different locations around the state, and therefore had different endpoint internet connection configurations. We all noticed that ELECT’s JSON data feed (JavaScript Object Notation, a lightweight data-interchange format) would routinely produce incorrectly formatted data that could not be parsed by standard tools.
The errors seemed random and did not have a specific repeatable pattern when we tried to look at the data being returned. Sometimes, it was simply a missing bracket, or quotation mark. Other times it appeared to be missing data or had malformed sections of the data. Tools such as Python, MATLAB, Tableau, Visualizer and other standard JSON parser libraries were unable to parse these errant data files.
Might this be an issue with IT infrastructure, such as bandwidth issues at ELECT causing dropped data packets? Or possibly an error in the server-side systems that respond to “GET” requests for the data?
We do know that within the last few years, ELECT has partnered with “Enhanced Voting” to supply the ENR data feed. We hope this feedback is useful to help ELECT improve their ability to supply the public with reliable Election Night Reporting.
EPEC is happy to work with ELECT to help identify and correct these issues. ##
Jon Lareau is EPEC’s volunteer Executive Director and CTO. He publishes regular analysis on DigitalPollwatchers.org, where this article post first ran. You can follow his “X” feed here. #