Influence agents were responsible for roughly 25% of political support spread via Twitter for candidates in the Arizona and Florida midterm elections, researchers report.

A new body of research by Morpheus Cybersecurity and APCO Worldwide, entitled "Impact of Influence Operations Targeting Midterm Elections," explores the effects of disinformation campaigns. They analyzed hundreds of thousands of retweets from thousands of accounts, looking for non-organic behavior – for example, high numbers of daily tweets for a long time frame.

The researchers' goal was to include all types of influence agents and explore the myriad ways in which bots and humans effectively swayed politicians and journalists with disinformation. 

Influence agents span a broad range of actors, including fully automated bots, semi-automated bots partially operated by humans, people who leverage software to generate traffic, political volunteers working together, and paid influencers employed by a central organization. Actors helped candidates appear to be more popular and generate organic support they didn't have.

The first phase of this study (June 2018 to August 2018) found an average of 27% of support for each political candidate in Arizona and 24% for each candidate in Florida appeared to come from non-organic accounts. Those numbers remained consistent in phase 2 (September 2018), when 26% of support for Arizona candidates and 28% of support for Florida candidates came from non-organic accounts.

Phase 3 consisted of collecting proof of influence. Researchers analyzed thousands of conversations between influence agents and politicians, journalists, and thought leaders. Their findings included a candidate agreeing with statements provided by influence agents, another engaging in a Q&A session with an influence agent, and a journalist discussing his work with an influence agent who was continually threatening him.

Read more details here.