New Research Points to Possible Algorithmic Bias on X

Elon Musk formally endorsed Donald Trump in the race for President on July 13, 2024, and spent more than $200 million on a bid to help re-elect him. After his victory, the President-elect has involved the billionaire tech in his transition planning, with Musk making recommendations on key appointments and joining calls with international leaders. President-elect Trump’s recent announcement that Musk will lead a new “Department of Government Efficiency” (DOGE), an initiative to advise the White House on ways to reduce the size of the federal government and its budget, indicates that his influence only stands to grow in a Trump presidency.

Tech Policy Press is tracking Musk’s involvement in politics in the US and around the world. One key question is how much Musk is willing to use X, the social media platform he acquired in 2022, to achieve his political goals. While Musk has taken various actions to shape the platform around his interests, from welcoming far-right figures back to the platform to promoting and encouraging various issues and ideas (as well as conspiracy theories and falsehoods), the extent to which he may have used X to shape online discourse leading up to the election remains an open question.

Researchers are looking for clues. A new pre-print research paper by Timothy Graham at Queensland University of Technology in Brisbane, Australia, and Mark Andrejevic of Monash University in Melbourne suggests possible algorithmic bias in engagement metrics favoring Musk’s personal X account and that of some Republican-leaning accounts. While the findings are preliminary, they suggest “important questions about the potential impact of algorithmic adjustments on public discourse and the ‘neutrality’ of social media platforms as information carriers.”

The study was conducted in two phases: Phase 1 focused on comparing engagement metrics (views, retweets, and followers) for Elon Musk's personal X account and other prominent accounts, while Phase 2 compared Republican-leaning and Democratic-leaning accounts.

To gauge whether Musk’s account was intentionally amplified in the lead-up to the 2024 US election, the researchers employed a statistical method called CUSUM (Cumulative Sum) to each engagement metric on X and captured the point where Musk’s engagement metrics showed a sudden, dramatic increase compared to historical averages. This “change point” happened to coincide with Musk’s endorsement of Donald Trump.

To evaluate if Musk’s engagement gains were unique, they used a Difference-in-Differences (DiD) approach. DiD is a statistical method that compares shifts in engagement before and after a certain point in time; in this case, the point was identified to be July 13, 2024. They found that “Elon Musk’s account had substantially higher baseline view counts compared to other accounts, with an average difference of about 4.43 million views prior to the change point.” Post July 13, 2024, the study found that Musk's posts received a boost of approximately 6.4 million views, independent of the uptick on the other accounts. In other words, there was a 138.27% increase in views for Musk's account post endorsement.

These results show that “Elon Musk’s account not only started from a higher baseline of views but also benefited from an enhanced increase in visibility post-change relative to other accounts.” A similar trend was also observed for retweets and user interaction, indicating that Musk’s account saw a sudden increase in engagement after July 13, 2024. The authors conclude that this may hint at the “possibility of algorithmic prioritization or bias, positioning Musk's content favorably in terms of platform visibility and user engagement.”

In Phase 2, the researchers shifted their focus from Elon Musk’s account to see if engagement boosts were distributed unevenly between Republicans and Democrats. The researchers collected data for five popular X accounts from both parties using an open-source tool called Zeesschuimer. They looked for posts from the “target accounts” and included all posts shared by these accounts between January 1, 2024, to October 25th, 2024, producing a dataset of 56,184 posts. Similar to Phase 1, the authors employed a statistical tool to detect deviations in engagement from the average over time and found the “change point” to be July 14, 2024, suggesting that this date “represents a platform-wide alteration affecting all engagement types, rather than isolated shifts within specific metrics.”

Interestingly, the authors do not find compelling evidence of algorithmic bias as they found when comparing Elon Musk’s personal X account with other prominent users. However, the findings show that Republican accounts had higher baseline views compared to Democrat accounts, highlighting a pro-Republican bias well before Musk endorsed Trump. Post endorsement, Republican accounts in the sample garnered 952,300 additional views. Other engagement metrics like retweets saw no significant partisan differences before or after Musk’s endorsement. It is possible that these findings were influenced by Musk’s boost in engagement, which may have led to greater visibility for Republican accounts that shared similar views.

The findings must be interpreted in the context of the study’s limitations. The authors acknowledge the difficulty in collecting data from X after the company restricted access to its Academic API. While efforts were made to ensure all posts were collected, there is no guarantee that all posts from these accounts were included, potentially biasing the findings.

“The authors highlight a compelling pattern and an intriguing explanation for why it occurred. As the authors are the first to acknowledge, given data limitations, it’s hard to discern for sure whether this was intentional or a quirk of a warped algorithm,” said Dr. Joe Bak-Coleman, a researcher affiliated with the University of Konstanz Center for the Advanced Study of Collective Behavior who read the preprint paper. He also noted that it should be “obvious that X is not ideologically neutral in platform design, moderation, and affordances.”

Concerns over Musk’s handling of X and allegations of political bias have led to more than 700,000 users migrating to Bluesky in recent weeks. This study indicates that these fears may have an empirical basis and raises important questions about the impact of algorithmic design on political content visibility. While we cannot definitively ascertain whether engineers at X intentionally amplified one political side over the other, the observed patterns suggest that X’s content algorithm elevated posts from Elon Musk and those he seeks to support, further warping the platform around his worldview and politics.