Political advertising has surged in the US ahead of the country’s November 8 congressional election. According to campaign funding tracker OpenSecrets, over $9 billion will be spent on this year’s political races; $50 million was spent in September alone.
Much of this money goes to political advertising on social media. A newly improved online tool, the Ad Observatory, can help journalists analyze the nature of this ad spend.
In a recent ICFJ Global Crisis Reporting Forum webinar, Nancy Watzmana strategic advisor for NYU Cybersecurity for Democracy and founder of Lynx LLC, spoke about how journalists can use the Ad Observatory to study political advertising on social media.
How it works
Developed by NYU’s Tandon School of Engineering, the Ad Observatory allows users to search data related to political ads on social media sites owned by Meta, the parent company of Facebook and Instagram. Search parameters can include keywords, subject, sponsor, or region. Users can also use the tool to perform basic Boolean searches and examine keywords in relation to each other.
The tool uses data from the publicly available Meta Ad Library, which is an archive of all ads that run on meta properties. It combines this data with available election information to create easy-to-understand graphs. These visualizations can show trends such as spend over time, by top sponsor, by ad type and partisan lean, or by audience demographics. Data can also be date-restricted, for example, if a journalist is looking for ad spend for a specific time period.
This additional layer of processing makes NYU’s tool a better outlet for journalists than the Meta Ad Library, Watzman explained. “We don’t clean [the data] as much as adding layers of analysis on top of that. These topics – you won’t find them on Facebook. We also have ad classifications: buy versus persuade versus connect. That’s not something Facebook offers; we’re adding that value,” she said.
However, the Ad Observatory doesn’t include individual examples of posts flagged as political — only aggregated data, Watzman noted. Users need to go to the Meta Ad Library to see these posts themselves.
“What you can find in the Meta Ad Library are the trees, and we show you the whole forest [the Ad Observatory]. If you want to go back and find a specific tree, the display library is very useful.”
how to use it
Journalists have used Ad Observatory data to inspire and enhance their stories about politics. Some examples are: learning which candidates use social media advertising more often; comparing a single candidate’s spending over different time periods to draw conclusions about political aspirations; understanding of current political issues such as immigration and abortion; and examining the history of political sponsors.
The Ad Observatory is best used in conjunction with the archived ads at the Meta Ad Library and other sites such as OpenSecrets, a research group that provides data on money in American politics, Watzman said. She warned journalists not to jump to conclusions about the forces behind political winners or losers after the upcoming elections based on information in the Ad Observatory.
“Expenditure is just a metric,” she said. “It’s true that historically, the biggest spender in a race usually wins. But it’s not just Facebook spending, and there are always exceptions to that rule.”
Instead, it’s best to use the Ad Observatory’s information as a supplement to understand larger stories about trends in politics. These stories can also be used to highlight the defect of the information available, as there are no rules dictating what ad information Meta and other social media sites must publish.
“There is so much we don’t know, and we don’t know because it’s not necessary. We just have to trust Meta to identify those political ads,” she said.
International use
Currently, the tool is most effective at analyzing English-language ads from the United States. One reason is that the Meta Ad Library, the source of the tool’s raw data, underperforms at categorizing political versus non-political ads from other countries. Although you can use the Ad Observatory in Spanish and search for Spanish language ads, the results provided are not as comprehensive as the English language results.
“We use a fairly conservative method to determine whether an ad is in Spanish or not. We’re still optimizing this model,” said Watzman. “We hope that this expertise we’re gaining on language and advertising recognition will be really useful in global elections.”
Watzman emphasized that the Ad Observatory was built with a unique, portable infrastructure that the NYU team plans to spread globally. In addition to creating ad observatories in other languages, Watzman says the team also hopes to incorporate the nuances of regional or country-specific jargon and idioms into the tool’s algorithms.
“Sometimes adverts about immigration [in the U.S.] Never mention the word immigration. They use these code words recognized as such by various political extremes, like “the caravan”. [by right-leaning organizations]’ Watzman said. “I think in each country it will be very, very specific what the terms used are and the topic modeling will have to be different. One of the research questions the team is investigating is how to create these flexible topic models to reflect these different languages and different modes of expression.”
midterms and beyond
As the U.S. interim period approaches, Watzman said political advertising on social media will only get more aggressive.
“It’s going to be hard to fathom. It’s going to be hard to figure out what’s going on. It’s just going to be a rush,” she said. “It’s a very important election in the United States and there’s a lot at stake for a lot of people. We need journalists to watch.”
Watzman encourages those who have questions about the tool or would like to help improve the Ad Observatory to contact the NYU Cybersecurity for Democracy team.
“After we are done with this choice, we will find out what our next projects are. So we’re always open to discussions,” she said.
Photo by Georgia de Lotz on Unsplash.
