Abstract
Social media has transformed political communication, yet its direct impact on election results remains debated. This study analyzes 21,641 political ads from the 2021 German federal election to estimate the association between online advertising and candidate success. Using a combination of discrete choice modeling and causal sensitivity analysis, we find that social media ads have a significant positive effect on vote shares. Specifically, an additional 200,000 impressions are associated with a 2.1% increase in votes, at an estimated cost of just €4 per vote. These findings suggest that social media advertising is a powerful, cost-effective tool capable of swaying outcomes in close races.
The Challenge of Measuring Digital Influence
While political scientists have long studied the impact of traditional media and general campaign spending, social media platforms have historically remained "black boxes." Linking specific ad "impressions" (views) to actual election results across an entire country is a complex task.
A primary hurdle in this research is endogeneity: the difficulty of determining whether a candidate receives more votes because they purchased ads, or if they purchased more ads because they were already popular and well-funded. This paper seeks to move beyond simple correlations to estimate the causal impact of social media advertising within the multi-party landscape of Germany.
Methodology and Data Integration
The study integrates three distinct data sources to isolate the effect of digital advertising:
- Meta Ad Library: Data on 21,641 ads published on Facebook and Instagram, totaling approximately 126 million impressions.
- Official Election Results: Data from the German Federal Returning Officer for all 299 constituencies.
- GLES Candidate Study: A survey providing "hidden" variables such as a candidate’s specific campaign budget, team size, and time investment, which serve as vital controls for candidate quality and effort.
To analyze this data, the researchers employed a multi-pronged statistical approach. They used a Conditional Logit Model to account for the interdependence of vote shares in a multi-party system. To ensure robustness, they applied the sensemakr framework for causal sensitivity analysis and Instrumental Variable (IV) Regression, using ad behavior from neighboring constituencies as an instrument. Additionally, Llama-3-70B was utilized to categorize the intent of the ads (e.g., persuasion, mobilization, or information), achieving an 83% agreement rate with human annotators.
Key Insights
The Power of Reach
Interestingly, the specific goal of an ad—whether it was intended to mobilize the base or persuade swing voters—did not significantly alter the outcome. The primary driver of success was the sheer volume of impressions. While Instagram is growing in popularity, Facebook remained the dominant platform for political reach during the 2021 cycle.
Low Barriers and High Efficiency
The study found that the average price per impression was roughly €0.01. This low cost allows candidates with smaller budgets to compete, potentially "democratizing" electoral competition, even though larger parties still maintain a higher total "share of voice."
Causal Robustness
The causal sensitivity analysis provided strong evidence for a true causal link. It revealed that unobserved confounding factors (such as a candidate's personal charisma) would need to be three times as strong as the effect of being an incumbent—a massive advantage in German politics—to explain away the estimated impact of the ads.
Quantitative Results
- Vote Increase: A one-standard-deviation increase in ad impressions (approx. 200,000 additional views) predicts a 2.1% increase in a candidate’s votes.
- Cost per Vote: The authors estimate the average price per vote at only €4.
- Electoral Sway: In the context of the 2021 German election, a 2.1% shift in votes for the second-place candidate would have been enough to change the outcome in 12 out of 299 constituencies.
Limitations and Future Directions
While these findings highlight the significance of social media, the study is limited to the Meta ecosystem (Facebook and Instagram). Future research is required to evaluate the impact of newer platforms like TikTok, which utilize different algorithmic delivery mechanisms. Furthermore, while impressions measure the opportunity to see an ad, they do not guarantee active engagement or persuasion. As voters become more "digitally fatigued," the long-term effectiveness of these strategies remains an open question for future election cycles.
Conclusion
This research provides empirical evidence that social media advertising is a highly effective and relatively inexpensive tool for influencing election outcomes. By establishing a causal link between impressions and vote shares, the study underscores the necessity of transparency and rigorous analysis in the digital political landscape.
Summary updated