
A small fraction of users produces most of what circulates online, and the expectation that falling publication costs would flatten this gradient has not been borne out. Two questions about that concentration remain open. Published estimates of how concentrated sharing is vary by a factor of three at comparable thresholds, with no account of why, and the demographic composition of heavy sharers has received little systematic attention outside the study of misinformation, where existing work does not separate the users who share the most from those whose shares reach the most people. Using 38 million news shares by 493,237 registered US voters matched to their Twitter accounts across 57 months, we address both. Observation-window length alone reproduces the full factor-of-three range within a single dataset, holding population and content fixed. The top one percent accounts for 22 percent of shares at a one-week window and 52 percent at 57 months. Two tests establish that the rise is compositional and not behavioral, since holding sample size fixed does not flatten the curve and a cohort of the same users followed across all windows stays near 21 percent throughout. Matching observation windows to prior studies that overlap our own panel narrows the gap for comparable content almost entirely, while a residual gap against misinformation-specific estimates reflects a difference in content class rather than window choice. Ranking the same users by sharing volume and by audience reach identifies substantially overlapping but still distinct populations, overlapping by 49.9 percent at the top one percent against one percent expected by chance. Both elites are older and more Democratic than the panel, with mean age rising from 40.5 years to 57.0 under activity ranking. Gender behaves differently. Male and female users share at statistically indistinguishable rates throughout the activity distribution, and a thirteen-point gap appears only at the extreme of the follower tail, where female share falls to 34.8 percent. Elite position is durable at the head of the distribution and close to memoryless in the middle. The users who share the most only partly overlap with the users whose shares travel furthest, and both groups are unrepresentative of the voting public in overlapping but distinct ways.
citation
Ozturan, B., & Lazer, D. (2026). Concentration Without Representation [Preprint]. SocArXiv. https://doi.org/10.31235/osf.io/dhq73_v1
topics
Online Behavior
Artificial Intelligence
Artificial Intelligence
Artificial Intelligence
related work

Measuring Agency-Amplifying AI Literacy: The HuskyAI Coaching Platform for Workforce Readiness
July 2026
HuskyAI is a web platform on which students develop AI knowledge by working through authentic
Artificial Intelligence
Artificial Intelligence
Working Paper


Measuring Agency-Amplifying AI Literacy: The HuskyAI Coaching Platform for Workforce Readiness
July 2026
HuskyAI is a web platform on which students develop AI knowledge by working through authentic
Artificial Intelligence
Artificial Intelligence
Working Paper


Measuring Agency-Amplifying AI Literacy: The HuskyAI Coaching Platform for Workforce Readiness
July 2026
HuskyAI is a web platform on which students develop AI knowledge by working through authentic
Artificial Intelligence
Artificial Intelligence
Working Paper

.png)