Structure and Dynamics of Online Information Ecosystems: Concentration, Segregation, and Governance

Who shapes the online information environment, along which demographic lines, and under what conditions that environment changes are questions that prior research has addressed primarily through aggregate behavioral data. This dissertation argues that answering them requires individual-level data linked to ground-truth demographics, and that doing so produces answers that differ substantially from what aggregate measures suggest. Three empirical studies use a panel of registered U.S. voters matched to social media activity over a multi-year observation window to examine concentration, segregation, and the dynamics of information quality. The first study shows that online news sharing is extremely concentrated and that the people responsible are not a representative cross-section of the public. The top 1% of news sharers account for 51.5% of all news shares (Gini = 0.911), a figure that prior studies have estimated with a factor-of-three range due to unreported methodological choices, primarily observation window length. The sharing elite is eighteen years older than the panel mean and substantially more Democratic. Activity concentration and reach concentration identify largely different people, overlapping by only 11.1% at the top 1%. A gender gap in reach that is absent at the individual level emerges at the joint extreme of the activity and follower distributions, an instance of aggregate-level inequality arising from individual-level equality. The second study shows that standard measures of online political segregation are biased by the same concentration the first study documents. Visit-weighted segregation indices estimate the segregation of the average share rather than the average person, and the direction of the bias depends on the level of aggregation. Corrected person-weighted measures show that partisanship is the least segregating of three demographic dimensions, falling below both gender and age. The strongest lines of segregation in shared information run along demographic rather than ideological dimensions, a finding that challenges the dominant framing of online filter bubbles as primarily partisan phenomena. The third study shows that platform governance is an active determinant of information quality, not a background condition. Following a major ownership transition, the platform exhibited a statistically significant decline in information quality, driven by an increase in content from low-quality sources and a decrease in content from high-quality sources, with measurable consequences for the information ecosystem of registered voters. What online information ecosystems look like depends on how they are measured, who is counted, and under what governance conditions they operate. These are not merely methodological choices. They determine whether concentration appears extreme or moderate, whether segregation runs along partisan or demographic lines, and whether platform transitions leave a trace. The dissertation makes those choices visible and shows that the answers change when they do.