From Correlation to Causation: Proof That Algorithms Drive Polarization

Establishing causation rather than mere correlation represents a crucial scientific achievement. Previous research showed that heavy social media use correlates with political polarization, but this study proves that algorithmic choices cause polarization through experimental manipulation that isolates platform effects from other variables.
The distinction matters enormously for policy and intervention. Correlational evidence leaves room for alternative explanations—perhaps polarized people simply use social media more rather than social media causing polarization. But experimental evidence from over 1,000 users randomly assigned to receive different feed compositions removes such ambiguity.
Participants who received slightly more divisive content became measurably more polarized than those who received less, with the random assignment ensuring these differences resulted from algorithmic manipulation rather than pre-existing user characteristics. This establishes clear causal relationships between specific algorithmic choices and psychological outcomes.
The causal evidence strengthens arguments for platform accountability. Companies can no longer plausibly claim they merely reflect existing divisions without contributing to polarization. The research proves that algorithmic choices actively increase or decrease political animosity depending on what content is amplified or suppressed.
This causation also points toward solutions. Since algorithms cause polarization, changing algorithms can reduce it. The research demonstrated that down-ranking divisive content decreased animosity by amounts matching the increases seen when such content was boosted. Platforms possess the technical capability to reduce division—the question is whether they’ll voluntarily exercise that capability or whether external pressure will be necessary.