On January 6, 2021, The Reverend Raphael Warnock, in a stunning upset, won the U.S. Senate runoff election in Georgia. His win makes him the first Black man from Georgia to hold a seat in a traditionally majority-white, American institution.
Warnock, and his co-senator, John Ossoff, emerged victorious, with 51 and 50.6 percent of the vote respectively against their opponents, Kelly Loeffler and David Purdue.
In the final polls, compiled by FiveThirtyEight on January 4 and 5, three firms predicted Warnock would lose the election to Kelly Loeffler by 1 point. In comparison, just one polling company predicted a loss for Ossoff.
Considering that Warnock had a larger margin of victory when all the votes were counted than Ossoff, how can analysts understand the lower levels of confidence in Warnock’s success?
The answer is in the data; polls offer just one data source in a complex web of intersectional considerations. In an ex-post analysis of AI generated financial data collected by Rossum, an OCR-based AI company, researchers can begin to piece together a new story about financial power and changing demographics in Georgia’s election.
Setting the scene: the Georgia runoff.
In the Georgia Senate races for the November 4 election, no candidate received more than 50 percent of the vote, necessitating, by Georgia law, that the top candidates automatically enter a runoff election on January 5.
In the runoff, four candidates faced off, Warnock, a Reverend, and Ossoff, the CEO of an impact-driven journalist group, Insight TWI, in the Democrat camp. For the Republicans, Loeffler, a business leader, and Purdue, the incumbent who assumed office in 2015.
The contentious race directed the attention and wallets of Americans toward Georgia in the weeks before the runoff. Donations from across the country to all candidates made this Senate race the most expensive in history. The outcome would determine which party controlled the Senate.
AI-generated finance data adds another data perspective
Finances are not the ultimate predictor of election results. While it is true that the candidate who spends the most money often wins the election, there is not necessarily causation between spending and winning. How that money is spent can transform a race.
Spending early, shying away from attack ads, targeting specific areas and groups are all tactics to spend money more efficiently.
Reviewing each candidate’s county spending data, before and after November 4, Purdue spent more in voting districts with major cities. Ossoff, on the other hand, increased his spending in rural communities. Ossoff’s efforts gained him a few points in Northeast Georgia.
Rossum tracked Senate spending for each week of 2020. Between late August and early November, Warnock outspent Loeffler in every week, except one. Only after the runoff was announced did Loeffler’s campaign truly increase their spending.
Earlier spending helps with name recognition. The money that Loeffler’s campaign did spend, if we’re following recommendations on when and how to spend money in an election, came too late.
Where AI falls short in interpreting elections
The rise of data can provide more accurate knowledge about the world but it is insufficient. Large quantities of data need to be fed into a computer for it to learn; not all firms have access to that volume of data. In addition, not all important information can be tracked using automation, and AI-generated data can be biased, especially against minority groups.
Exclusively using AI to understand the election would have left out important qualitative data. According to Pew, white voters in Georgia are 59 percent likely to be Republicans, whereas Black voters are 73 percent likely to be Democrats. Despite historically lower voter turnout for minority voters, Black and Latinx-led community organizing groups enfranchised non-white voters and directly led predominantly Black counties to high voter turnout for the runoff.
AI, for now, can’t see and learn everything about short term human-to-human experiences.
Where AI is helpful in understanding election data
The Georgia Senate runoff was the most expensive in history. Digging deeper into who spent money, where, and when adds context and depth to election predictions that polling alone, with its inherent uncertainties, might miss.
In a data-driven world, information needs to be collected constantly and interpreted quickly, an ideal job for AI. Rossum’s AI program was able to segment data from before the November 4 election and after, creating the opportunity to quickly compare and contrast financial data in the lead up to each vote.
While it doesn’t tell the whole story, Rossum’s election data demonstrates that public affairs can use AI to build additional insights.