A new study published in the journal PNAS Nexus has shed light on how social media algorithms favor politically sponsored content from certain parties, even when the same budget is applied. This research, conducted by the Politecnico di Milano, LMU – Ludwig Maximilians Universität of Munich, and the CENTAI institute of Turin, analyzed over 80,000 political ads on Facebook and Instagram leading up to the 2021 German federal elections.
These ads, representing parties across the political spectrum, generated over 1.1 billion impressions among more than 60 million eligible voters. The findings reveal significant discrepancies in the effectiveness of these ads, with a bias towards more extremist groups.
Social media platforms like Facebook and Instagram have become essential tools for political campaigns due to their vast user base and sophisticated targeting capabilities. However, there are growing concerns about the fairness, accountability, and transparency of these platforms’ proprietary algorithms. These algorithms determine which users see specific ads, potentially introducing biases that favor certain political messages and groups.
To conduct their study, the researchers gathered data from the Meta Ad Library API, focusing on political ads from the 2021 German federal elections. The data included detailed information about ad content, spending, start and stop dates, and the number of impressions distributed across different demographics (gender and age). The team analyzed ads from six major political parties in Germany: Linke, Grüne, SPD, FDP, Union, and AfD.
Die Linke, or The Left, is a democratic socialist party advocating for social justice, anti-capitalism, and policies that support the working class and disadvantaged groups.
Bündnis 90/Die Grünen, or The Greens, focuses on environmental protection, sustainability, and social justice, promoting policies to combat climate change and support human rights.
The Social Democratic Party of Germany (SPD) is a center-left party advocating for social democracy, workers’ rights, and a strong welfare state to ensure social equity.
The Free Democratic Party (FDP) is a libertarian party that promotes individual freedom, free-market economic policies, and reduced government intervention in the economy.
The Union, comprising the Christian Democratic Union (CDU) and its Bavarian sister party, the Christian Social Union (CSU), is a center-right political alliance supporting conservative values, a social market economy, and European integration.
Alternative für Deutschland, or Alternative for Germany (AfD), is a right-wing populist party known for its anti-immigration stance, Euroscepticism, and conservative social policies.
The researchers used statistical and machine learning methods to analyze the data. They computed the discrepancies between the intended (targeted) and actual audiences reached by the ads. The team also evaluated the efficiency of ads in terms of impressions-per-EUR (a measure of how many impressions an ad generates for each euro spent). They employed regression analysis to identify key factors influencing ad reach and used a random forest model to predict ad performance based on the available data.
The study revealed that 72.3% of all political ads used targeting strategies, accounting for 72.6% of the total ad spending. Parties used a wide range of targeting categories, with a preference for exclusion criteria to narrow down audiences. This approach allowed the social media platform’s algorithms to optimize ad delivery among broad audiences.
The analysis showed significant disparities in the efficiency of ads across different political parties. On average, political ads generated 126.71 impressions per euro spent. However, ads from the Grüne party achieved only 36.18 impressions per euro, while those from the FDP and AfD were much more efficient, with 181.53 and 203.49 impressions per euro, respectively. This suggests that certain parties, particularly more extremist groups, benefited more from the platform’s algorithmic delivery.
“The greater success of their advertising could be explained by the fact that the incendiary political issues promoted by populist parties tend to attract a lot of attention on social media. Consequently, algorithms would favor campaign ads with such content,” explained Francesco Pierri, a researcher from the Data Science research group of the Department of Electronics, Information, and Bioengineering at the Politecnico di Milano, who co-led the study.
The researchers also found discrepancies between the targeted and actual audiences. Ads generally reached a younger audience than intended, except for the far-right AfD, which reached an older audience. Gender-wise, most ads were shown to fewer female users than targeted, except for those from the Grüne party, which reached more female users than intended. This indicates a potential algorithmic bias in ad delivery that could reinforce existing political and social biases.
The regression analysis revealed that more granular targeting criteria often resulted in lower impressions-per-EUR, especially when using exclusion criteria. Targeting single-gender audiences was associated with higher ad efficiency. Ads with positive sentiment and those published earlier in the week or for longer durations also tended to perform better.
The machine learning model’s low prediction performance suggested that the available data was insufficient to fully explain the variance in ad performance. This highlights the need for greater transparency from social media platforms regarding their ad delivery algorithms and pricing mechanisms.
“We see a systematic bias in how the political ads of different parties are distributed. If they aim at a specific audience or send contradictory messages on political issues to different groups, this can limit the political participation of disadvantaged groups,” Pierri said. “Even worse, the algorithms used by the platforms do not allow verification if they involve biases in ad distribution. If, for example, some parties systematically pay higher prices than others for similar ads, this damages political competition. We need greater transparency from the platforms regarding political advertising to ensure fair and uncompromised elections.”
The study, “Systematic discrepancies in the delivery of political ads on Facebook and Instagram,” was authored by Dominik Bär, Francesco Pierri, Gianmarco De Francisci Morales, and Stefan Feuerriegel.
