Simple contagion drives population-scale platform migration

TL;DR

This work links 276,431 scholars on Twitter/X to their respective new profiles among the universe of all 16.7 million Bluesky accounts, tracked from January 2023 to December 2024, using a scalable, high-precision cross-platform matching pipeline.

Abstract

Social media platforms mediate professional communication, political expression, and community formation, making the rare instances when users collectively abandon an incumbent platform particularly consequential. Strong network effects raise switching costs and strengthen incumbents' positions, making coordinated exit difficult. Here we link 276,431 scholars on Twitter/X to their respective new profiles among the universe of all 16.7 million Bluesky accounts, tracked from January 2023 to December 2024, using a scalable, high-precision cross-platform matching pipeline. Exploiting exogenous variation from Brazil's court-ordered suspension of Twitter/X and a dynamic matching design, we show that adoption is peer-driven, treatment effects are short-lived and dose-dependent, and contagion is simple, not complex. Three patterns characterize adoption and retention. Adoption concentrates among users deeply embedded in Twitter's social graph. Public political expression predicts migration, consistent with homophilous inflows into a largely left-of-center Bluesky information space. Early reconnection with prior contacts predicts longer tenure and engagement. Our findings provide the first population-scale causal evidence of peer influence in a social media platform migration by exploiting exogenous exposure variation in a natural experiment and using daily dynamic matching. Rather than the complex contagion mechanism often emphasized in the literature, contagion is predominantly simple. Our findings recast migration as a multi-homing strategy that insures against governance uncertainty and show that users who quickly reconnect with prior contacts remain active longer on Bluesky.

Figures (7)

• Figure 1: Descriptive patterns of the academic migration from Twitter/X to Bluesky.a Daily number of academic transitions over time, highlighting five major shock periods that drove concentrated migration waves, with the November 2024 US election producing the largest spike (see Table S10 for detailed discussion). b Transition rates by academic discipline, revealing substantial variation from 13.3% in Medicine & Health to 31.3% in Arts & Humanities, with sample sizes shown for each field. c Transition rates by user characteristics, comparing top tercile (blue arrows, right) and bottom tercile (red arrows, left) users across academic and social media metrics. Twitter-specific indicators show stronger associations with transition behavior than traditional academic metrics. PageRank is used to compute the centrality of academics in the Twitter network. d Ratios of user characteristics during invite-only versus public release periods, showing early adopters had higher Twitter engagement but lower academic productivity (99% bootstrapped CIs, 1,000 replicates). e Share of total sign-ups on Bluesky during major shock events, comparing academics (triangles) versus the general public (circles). The US presidential election (red star) generated the largest migration surge
• Figure 1: Feature importance for contagion mechanism classification.a GAIN scores from XGBoost, measuring the average reduction in loss from splits using each feature. b LIME coefficients showing direction and magnitude of local linear effects for each mechanism. Simple contagion (blue) depends on exposure duration and active influences. Complex contagion (yellow) relies on peer saturation. Spontaneous adoption (green) shows relatively uniform importance. Shock-driven transitions (red) concentrate on shock intensity and recency.
• Figure 2: Political expression and transition to Bluesky. Linear Probability Model (LPM) coefficients showing associations between expressing opinions on political topics and Twitter-to-Bluesky transition probability ($N = 206{,}898$; $\text{df} = 206{,}819$). Points represent joint-model estimates that include all topic indicators alongside the same baseline controls. Boxplots on the right summarize the distribution of coefficients for political topics (red) versus non-political controls (gray). Error bars indicate 95 confidence intervals.
• Figure 2: Degree-adoption order correlation test fromcencettiDistinguishingSimpleComplex2023cencettiDistinguishingSimpleComplex2023. The observed negative correlation ($\rho = -0.21$, $p < 0.001$, $N = 49{,}533$) is consistent with simple contagion as the dominant mechanism.
• Figure 3: Brazil's suspension of Twitter/X as a natural experiment.a Absolute transitions in 24-hour bins from day $-14$ to day 30, split into Brazilian (red) and others (gray). b Country-level 30-day transition rates among sizeable groups (top five shown) plus the network average (black). c,d Event-study estimates showing the relative probability of transitioning around day 0 for non-Brazilian scholars, grouped by number of Brazilian ties. We compare treated users against users with no Brazilian connections ($N = 8{,}645{,}021$ user-days, df = $192{,}508$). Panel c uses Brazilian followees and panel d uses Brazilian followers. Shaded areas denote 95% confidence intervals. Sample sizes for Followees: 0 ties = $202{,}729$, 1--3 ties = $31{,}592$, 4+ ties = $5{,}596$ and for Followers: 0 ties = $187{,}789$, 1--3 ties = $42{,}618$, 4+ ties = $9{,}510$.