Periodic Reporting for period 3 - DAFINET (Dynamic Attitude Fixing: A novel theory of opinion dynamics in social networks and its implications for computational propaganda in hybrid social networks (containing humans and bots))
Okres sprawozdawczy: 2021-12-01 do 2023-05-31
Understanding the coordination of attitudes in societies is important for many disciplines and global social challenges, to understand the social processes by which people adopt new attitudes (e.g. in health behaviour interventions) or reject them (e.g. in radicalization resistance). This is a foundational problem across the social sciences and sciences. Improved understanding of the coordination of attitudes in groups therefore will potentially impact a broad spectrum of disciplines and global social challenges, including climate change, threats to democratic processes, etc. A stubborn problem in social influence research is the paradox that attitudes are sometimes deeply resistant to change and sometimes propagate rapidly. This problem is becoming more pressing as evidence mounts that public opinions are being powerfully influenced by bots (autonomous agents) in social media networks, for example in the Trump/Clinton election or the ‘Brexit’ referendum.
Dynamic fixing is proposed as a novel model for social network opinion dynamics with promise for understanding the mechanisms of computational propaganda in hybrid social networks (consisting of human and bot agents). Dynamic fixing potentially explains how attitudes are can be both highly resistant to influence (e.g. in relation to health behaviour campaigns) and highly changeable (e.g. during revolutionary political events). The model synergistically integrates attitudes and identity with multilayer network modelling to propose an inter-disciplinary theory of opinion dynamics. The theory of dynamic fixing is premised on recent research linking attitudes with identity in ethnomethodology and social psychology, namely that people: (1) simultaneously hold a repertoire of multiple (often ambivalent) attitudes on issues, (2) use attitude expressions to produce social identity; and (3) continuously manage social accountability in interaction by initiating and reacting to affiliative and disaffiliative responses. It is proposed that when people interact in groups the emergent accountability networks progressively fix the range of attitudes associated with particular identities. Attitude expressions are highly sensitive to the immediate social context (i.e. network state), and changes in network state can therefore lead to apparently dramatic flips in attitudes. The dynamic fixing model holds that these shifts are related to the immediate accessibility of specific attitudes in agents’ existing attitudinal repertoires rather than (or in addition to) actual attitude change. This explains how dramatic collective attitude shifts (e.g. in the run up to an election) may be observed without any permanent shifts in underlying individual attitudes; and also how attitude shifts can be so quickly and easily unwound (e.g. on the morning after an election).
The DAFINET project aims to: (1) theoretically validate the dynamic fixing model of opinion dynamics using mathematical models and agent-based modelling; (2) empirically validate the model with network experiments with human participants using the Virtual Interaction APPLication and other platforms for social experiments with interaction ; (3) experimentally test model hypotheses in hybrid networks with bots and humans; and (4) experimentally explore the utility of the dynamic fixing model for predicting network vulnerability and resistance to social influence from bots in hybrid networks.
Dynamic fixing is proposed as a novel model for social network opinion dynamics with promise for understanding the mechanisms of computational propaganda in hybrid social networks (consisting of human and bot agents). Dynamic fixing potentially explains how attitudes are can be both highly resistant to influence (e.g. in relation to health behaviour campaigns) and highly changeable (e.g. during revolutionary political events). The model synergistically integrates attitudes and identity with multilayer network modelling to propose an inter-disciplinary theory of opinion dynamics. The theory of dynamic fixing is premised on recent research linking attitudes with identity in ethnomethodology and social psychology, namely that people: (1) simultaneously hold a repertoire of multiple (often ambivalent) attitudes on issues, (2) use attitude expressions to produce social identity; and (3) continuously manage social accountability in interaction by initiating and reacting to affiliative and disaffiliative responses. It is proposed that when people interact in groups the emergent accountability networks progressively fix the range of attitudes associated with particular identities. Attitude expressions are highly sensitive to the immediate social context (i.e. network state), and changes in network state can therefore lead to apparently dramatic flips in attitudes. The dynamic fixing model holds that these shifts are related to the immediate accessibility of specific attitudes in agents’ existing attitudinal repertoires rather than (or in addition to) actual attitude change. This explains how dramatic collective attitude shifts (e.g. in the run up to an election) may be observed without any permanent shifts in underlying individual attitudes; and also how attitude shifts can be so quickly and easily unwound (e.g. on the morning after an election).
The DAFINET project aims to: (1) theoretically validate the dynamic fixing model of opinion dynamics using mathematical models and agent-based modelling; (2) empirically validate the model with network experiments with human participants using the Virtual Interaction APPLication and other platforms for social experiments with interaction ; (3) experimentally test model hypotheses in hybrid networks with bots and humans; and (4) experimentally explore the utility of the dynamic fixing model for predicting network vulnerability and resistance to social influence from bots in hybrid networks.
The focus at the start of the project has been on evaluating the applicability of the proposed network models to the phenomenon (theory development), and experimentally confirming the proposed psychological mechanisms linking attitudes and social identity (basic human experiments).
We developed methods to constructs bipartite graphs from survey-based data to visualize how people are linked by shared attitudes, and also between attitudes (by being jointly held by people).The latter produces a network-view of opinions similar belief-network analysis, but with fewer statistical assumptions (arXiv:2012.11392). We have applied this technique to empirical attitude data collected in the first wave of the COVID pandemic in the UK, and have demonstrated that the dynamic opinion-based groups so identified predict group relevant behaviours (doi:10.1111/bjso.12396).
We have connected with work in sociology and social psychology modelling social belief systems. We have identified also discovered threads of literature where bipartite network structures have been used to model attitudes, most notably by sociologists Ronald Breiger and Bonnie Erikson (1988). We have worked to situate our theory in relation to these existing, but little-known, approaches (under review; preprint doi:10.31234/osf.io/mh4z8). This theory-building work has helped us realize the mathematical relationship between our bipartite network and existing modelling techniques, such as cluster analysis. We have established that our approach is similarly effective as cluster-analysis in identifying cohesive opinion-based groups, but with the advantage that we can locate individuals precisely in the structural opinion-space. We have demonstrated that this bipartite network approach to opinion-modelling allows us to sensitively detect polarization, even without extremism (ie. when group opinions are tightly synchronized on several moderate opinions; see preprint arXiv:2104.14427).
We have realized that Axelrod’s classic model of cultural diffusion natively relies on a bipartite network structure of people connected by attitudes. We have extended this model, introducing a multidimensional equivalent of an agreement threshold, and shown that the model generates plausible clusters similar to those observed in real data (doi:10.1371/journal.pone.0233995) and is relatively impervious to underlying social network topology (e.g. friendship links), allowing us to apply it in realistic social systems (doi:10.1016/j.physa.2021.126086).
We have developed a method for generating novel attitude statements (attitudes to which people have not yet been exposed), and now have a battery to deploy in our experiments.
Despite Covid-related delays to our virtual interaction experiments (which previously relied on people participating together in a lab), we have run three experiments with human participants. These demonstrate that:
(1) Sharing novel opinions (and more specifically, expressing agreement on such opinions) results in people experiencing a sense of shared group identity.
(2) The sense of identification produced in opinion-based groups is stronger than that produced classical “minimal group” conditions where groups are differentiated on arbitrary dimensions.
(3) People come to have more certainty about attitudes that are associated with an emerging group identity.
The VIAPPL software platform has been updated to allow a novel network game where participants exchange opinions.
We developed methods to constructs bipartite graphs from survey-based data to visualize how people are linked by shared attitudes, and also between attitudes (by being jointly held by people).The latter produces a network-view of opinions similar belief-network analysis, but with fewer statistical assumptions (arXiv:2012.11392). We have applied this technique to empirical attitude data collected in the first wave of the COVID pandemic in the UK, and have demonstrated that the dynamic opinion-based groups so identified predict group relevant behaviours (doi:10.1111/bjso.12396).
We have connected with work in sociology and social psychology modelling social belief systems. We have identified also discovered threads of literature where bipartite network structures have been used to model attitudes, most notably by sociologists Ronald Breiger and Bonnie Erikson (1988). We have worked to situate our theory in relation to these existing, but little-known, approaches (under review; preprint doi:10.31234/osf.io/mh4z8). This theory-building work has helped us realize the mathematical relationship between our bipartite network and existing modelling techniques, such as cluster analysis. We have established that our approach is similarly effective as cluster-analysis in identifying cohesive opinion-based groups, but with the advantage that we can locate individuals precisely in the structural opinion-space. We have demonstrated that this bipartite network approach to opinion-modelling allows us to sensitively detect polarization, even without extremism (ie. when group opinions are tightly synchronized on several moderate opinions; see preprint arXiv:2104.14427).
We have realized that Axelrod’s classic model of cultural diffusion natively relies on a bipartite network structure of people connected by attitudes. We have extended this model, introducing a multidimensional equivalent of an agreement threshold, and shown that the model generates plausible clusters similar to those observed in real data (doi:10.1371/journal.pone.0233995) and is relatively impervious to underlying social network topology (e.g. friendship links), allowing us to apply it in realistic social systems (doi:10.1016/j.physa.2021.126086).
We have developed a method for generating novel attitude statements (attitudes to which people have not yet been exposed), and now have a battery to deploy in our experiments.
Despite Covid-related delays to our virtual interaction experiments (which previously relied on people participating together in a lab), we have run three experiments with human participants. These demonstrate that:
(1) Sharing novel opinions (and more specifically, expressing agreement on such opinions) results in people experiencing a sense of shared group identity.
(2) The sense of identification produced in opinion-based groups is stronger than that produced classical “minimal group” conditions where groups are differentiated on arbitrary dimensions.
(3) People come to have more certainty about attitudes that are associated with an emerging group identity.
The VIAPPL software platform has been updated to allow a novel network game where participants exchange opinions.
We have made substantial progress in testing the groundbreaking theory that opinion-based groups can be represented as bipartite networks, modelling it mathematically, and demonstrating its value for detecting and understanding opinion-based groups and associated social phenomena, like polarisation. We have encouraging preliminary results confirming the basic psychological processes underlying this structure. We expect to deliver the results outlined in the DOA by the end of the project.