Measuring online grassroots action 

Learning the recipe for successful grassroots action is one of the holy grails of Computational Social Science. Understanding the mechanics of the processes that motivate people to take action on common causes could provide us with a blueprint on how to make best use of the power of digital, planetary-scale communication to enable mass cooperation and tackle some pressing global issues like climate change. However, studying collective coordination occurring in the wild is hard, primarily because these events are rarely documented with hard data.

The Twitter Migration as a giant collective behavior change experiment

The user migration from Twitter to Mastodon that followed Elon Musk's purchase of the Blue Bird site represents one of the largest digital migrations in the history of the Social Web and it is one rare examples of collective behavioral change that is documented through large-scale digital traces. This migration exhibits two uncommon properties that render it especially interesting from the perspective of behavioral change studies. First, the migration unfolded organically within Twitter, with users engaging in discussions and potentially influencing their peers by signaling their intention to migrate. Second, transitioning to a different social platform entails practical and psychological costs associated with changing habits, as well as a social cost associated with adopting a behavior that deviates from mainstream norms. These characteristics are shared with other grassroots processes of behavioral change that are generally desirable for human societies, like climate action.

Follow the Migration

To study how the migration unfolded, we collected 1.3M tweets marked with  the #TwitterMigration hashtag, posted by 0.5M users. To determine which of these authors had migrated to Mastodon, we identified the users who advertised their new Mastodon handles on Twitter. We were able to accurately link 75K Twitter users to their newly created Mastodon accounts.

The widespread practice of using Twitter to announce one’s decision to migrate to Mastodon raises the question of whether social influence played a role in Twitter users’ migration choices. To investigate this, we characterize the “infectiousness” of migration decisions usin the SIRS compartmental epidemic model —yes, the one that's been talked about a lot during the dreaded COVID times. Epidemic models have been extensively used to simulate information diffusion within social systems, under the assumption that the process of influence spreads through social interactions, akin to the transmission of communicable diseases. We found that a SIRS model can reproduce quite faithfully the temporal trace of the number of migrations to Mastodon (Figure 1).

The SIRS model is characterized by the reproduction number R0. When R0>1 the diffusion process grows as individuals become infected at a higher rate than they recover.  Our model estimated an R0 value of 4.57, indicating a highly infectious process of influence. Most interestingly, the R0 varies considerably (from 1 to 11.82)  across different social communities that compose the Twitter social network, which begs the question of what factors made some communities more effective than others in the process of behavior change.

The drivers of migration

To gain insights into the factors that contribute to the acceleration of the influence process, we explored the correlation between the R0 parameter and three families of factors that have previously been associated with the adoption of new opinions and behaviors in social groups:

• Network topology. Features describing the shape of the social network of the community in which the process of diffusion takes place.
• Reiterated commitment. The extent to which users in the community re-iterate a call to action. In our case, we considered the average frequency of posting the #TwitterMigration hashtag as an indication of commitment.
• Language use. We focused on the social pragmatics of language, namely the intended social function of an utterance (e.g., expressing trust, or conveying social support). Recent research has identified dimensions of social pragmatics commonly observed in everyday language, and used Machine Learning models to quantify them from language.

The results of a regression analysis (Table 1) showed that communities that migrated more rapidly were those where:

1. Social connections are relatively sparse. The incentive to migrate seems to be proportional to the fraction of a user’s friends who have already migrated, and such fraction increases more rapidly in networks with fewer social connections.
2. Members repeteadly signaled their commitment to migrate. We interpret this result in the light of theories linking the rapid emergence of consensus to the influence exerted by committed individuals, even when they constitute a small minority.
3. The social discourse emphasizes shared identity and engages in exchanges of factual knowledge. Extensive research in social psychology has established a connection between psycho-linguistic aspects of social interaction and successful, spontaneous coordination. In particular, the Identity Theory posits that cooperation can be facilitated through cognitive mechanisms that foster a sense of belonging to the same social group, suggesting that identity may be pivotal in overcoming social dilemmas involving coordinated behavior that entail inherent risks or a non-zero cost of action. Moreover, the exchange of truthful, factual information has been identified as a prerequisite for constructive debates and, ultimately, persuasion. 

Interestingly, the combined influence of identity and knowledge explains more variance in our data than the combined influence of density and commitment, highlighting the significant role of psycho-linguistic aspects as key drivers of behavioral change.

Is language the main key to unlock social cooperation?