2. Application of the framework to our empirical investigations
Our experimental paradigm (Dezecache, Allen, von Zimmerman, & Richardson, Reference Dezecache, Allen, von Zimmerman and Richardson2021) tests conflict between groups with unequal resources, and asks how relative deprivation and social identity interact to generate conflict in the laboratory. Players are assigned to two teams in a computer game played in person or online. They tap on their devices to build features in a park, such as a flower bed. However, players also have the option to vandalise the other team's park, trampling the flower beds or damaging the playground, and assign a disadvantage to one of the teams – they have to work harder to build their park features. We find that this structural inequality in the game produces a higher rate of aggressive behaviour towards the advantaged team, a conclusion supported through our agent-based models.
Within our paradigm, groups can also be a mix of preexisting identities and minimal group assignments, allocated in different proportions across conditions. Therefore, although players are assigned at random to the advantaged or disadvantaged park, they are also informed, for example, that their team mates share their political beliefs, or have varying political opinions. An open question is then whether individuals consider others within their empirically assigned team to be part of their group, or if it is preexisting identities driving any observed intergroup conflict.
Considering the target article's computational approach, we find that insight is gained by assuming that participants compute groups through assigning roles in triads, and particularly through considering the frequency of observed behaviours between different groups. By observing the behaviour of the bystander, and taking the assumption that participants compute group membership through the observation of roles and behaviour, we could infer which groups participant's consider themselves to be part of through observed triad frequencies. As we observe changing proportions of vandalism over time, the frequency of different triads may also inform which group (e.g., team or wider identity) is salient at any one time. By observing the behaviour of different bystanders towards the relative success of others, we may be able to better understand how individuals are making these comparisons, and whether groups or individual comparisons are key. Therefore, depending on the level of observed conflict between groups, we can infer if participants are considering individual or group relative deprivation.
However, the fuzzy nature of real group dynamics means there are a number of things we cannot infer by assuming that individuals use this computational method. Our own modelling points towards coordination and conflict coexisting within and between groups. By defining groups in a concrete manner, the opportunity for a wider number of roles is lost. If an individual performs a different task to the rest of the group, does that mean they are not a member of the group, or does it just mean that they are a part of the group, but coordinating behaviour?
Finally, to infer feelings of relative deprivation, and also any other drivers of intergroup conflict, we feel that more attention must be paid to an agent's internal state. Although assigning roles based on behaviour works in areas such as conflict, which is an overt behaviour, this is not true of affective states such as frustration, or even the act of making a comparison. Therefore, how is it that an observer could then classify others into groups?
We also consider how this computational approach contributes to our understanding of real-world riots. One explanation from social psychology is that riots emerge through aggressive action perceived through salient identities (e.g., Drury et al., Reference Drury, Stott, Ball, Reicher, Neville, Bell and Ryan2020; Reicher, Reference Reicher1984; Stott & Drury, Reference Stott and Drury2017; Stott et al., Reference Stott, Ball, Drury, Neville, Reicher, Boardman and Choudhury2018). This explanation aligns with the target article's framework. Rather than defining groups through a number of context-dependent external markers, it is the role the individual takes within this scenario that is important. By understanding conflict through the allocation of roles, this framework also introduces a flexibility to help us understand this phenomenon across scales. What begins as a local phenomenon can soon spread, as an individual bystander observing aggression between two individuals or two groups is spurred into action in a similar way.
3. Conclusion
We find the idea of asking what is the minimum information required to understand a “group” intriguing. When applied to the real-world example of riots, this framework provides insight through its flexibility to describe conflict across scales, and its ability to marry contagion models with social psychology. Within our empirical paradigm, by assuming that participants do indeed compute groups in this way, we may gain information on exactly who they consider to be part of their group.
More usefully, this framework adds to the continuing discussion on how best to model a riot (e.g., Baudains, Braithwaite, & Johnson, Reference Baudains, Braithwaite and Johnson2013a; Baudains, Johnson, & Braithwaite, Reference Baudains, Johnson and Braithwaite2013b; Bonnasse-Gahot et al., Reference Bonnasse-Gahot, Berestycki, Depuiset, Gordon, Roché, Rodriguez and Nadal2018; Davies, Fry, Wilson, & Bishop, Reference Davies, Fry, Wilson and Bishop2013). Currently, most models consider contagion to be the central mechanism of riot spreading. However, what is lacking from these models is the inclusion of the importance of the identity salience to who imitates who, how this changes over time, and how this structures their social interactions. However, by including bystanders, and considering how they are likely to act next, the computational framework enables us to understand why these contagion models are successful at matching the data, even before many of the factors discussed in social psychology are considered.
However, despite the simplicity and flexibility of this framework, it does have a number of issues. Specifically, we find that the myriad of possible behaviours required to explain more complicated real-world scenarios, such as we find in the laboratory, cannot be understood solely through this computational model.
You have
Access
1. Introduction
We intend to “stress-test” the author's computational approach to groups using concrete examples of intergroup conflict – specifically, riots. Drawing on psychological and sociological accounts of the London riots of 2011, we previously developed a series of lab-based games of collective action where we experimentally manipulate psychology factors and use agent-based modelling to understand our participants' behaviour. We consider the idea proposed in the article that individuals perceive a group through roles and behaviour, and examine whether this gives insight into our findings.
2. Application of the framework to our empirical investigations
Our experimental paradigm (Dezecache, Allen, von Zimmerman, & Richardson, Reference Dezecache, Allen, von Zimmerman and Richardson2021) tests conflict between groups with unequal resources, and asks how relative deprivation and social identity interact to generate conflict in the laboratory. Players are assigned to two teams in a computer game played in person or online. They tap on their devices to build features in a park, such as a flower bed. However, players also have the option to vandalise the other team's park, trampling the flower beds or damaging the playground, and assign a disadvantage to one of the teams – they have to work harder to build their park features. We find that this structural inequality in the game produces a higher rate of aggressive behaviour towards the advantaged team, a conclusion supported through our agent-based models.
Within our paradigm, groups can also be a mix of preexisting identities and minimal group assignments, allocated in different proportions across conditions. Therefore, although players are assigned at random to the advantaged or disadvantaged park, they are also informed, for example, that their team mates share their political beliefs, or have varying political opinions. An open question is then whether individuals consider others within their empirically assigned team to be part of their group, or if it is preexisting identities driving any observed intergroup conflict.
Considering the target article's computational approach, we find that insight is gained by assuming that participants compute groups through assigning roles in triads, and particularly through considering the frequency of observed behaviours between different groups. By observing the behaviour of the bystander, and taking the assumption that participants compute group membership through the observation of roles and behaviour, we could infer which groups participant's consider themselves to be part of through observed triad frequencies. As we observe changing proportions of vandalism over time, the frequency of different triads may also inform which group (e.g., team or wider identity) is salient at any one time. By observing the behaviour of different bystanders towards the relative success of others, we may be able to better understand how individuals are making these comparisons, and whether groups or individual comparisons are key. Therefore, depending on the level of observed conflict between groups, we can infer if participants are considering individual or group relative deprivation.
However, the fuzzy nature of real group dynamics means there are a number of things we cannot infer by assuming that individuals use this computational method. Our own modelling points towards coordination and conflict coexisting within and between groups. By defining groups in a concrete manner, the opportunity for a wider number of roles is lost. If an individual performs a different task to the rest of the group, does that mean they are not a member of the group, or does it just mean that they are a part of the group, but coordinating behaviour?
Finally, to infer feelings of relative deprivation, and also any other drivers of intergroup conflict, we feel that more attention must be paid to an agent's internal state. Although assigning roles based on behaviour works in areas such as conflict, which is an overt behaviour, this is not true of affective states such as frustration, or even the act of making a comparison. Therefore, how is it that an observer could then classify others into groups?
We also consider how this computational approach contributes to our understanding of real-world riots. One explanation from social psychology is that riots emerge through aggressive action perceived through salient identities (e.g., Drury et al., Reference Drury, Stott, Ball, Reicher, Neville, Bell and Ryan2020; Reicher, Reference Reicher1984; Stott & Drury, Reference Stott and Drury2017; Stott et al., Reference Stott, Ball, Drury, Neville, Reicher, Boardman and Choudhury2018). This explanation aligns with the target article's framework. Rather than defining groups through a number of context-dependent external markers, it is the role the individual takes within this scenario that is important. By understanding conflict through the allocation of roles, this framework also introduces a flexibility to help us understand this phenomenon across scales. What begins as a local phenomenon can soon spread, as an individual bystander observing aggression between two individuals or two groups is spurred into action in a similar way.
3. Conclusion
We find the idea of asking what is the minimum information required to understand a “group” intriguing. When applied to the real-world example of riots, this framework provides insight through its flexibility to describe conflict across scales, and its ability to marry contagion models with social psychology. Within our empirical paradigm, by assuming that participants do indeed compute groups in this way, we may gain information on exactly who they consider to be part of their group.
More usefully, this framework adds to the continuing discussion on how best to model a riot (e.g., Baudains, Braithwaite, & Johnson, Reference Baudains, Braithwaite and Johnson2013a; Baudains, Johnson, & Braithwaite, Reference Baudains, Johnson and Braithwaite2013b; Bonnasse-Gahot et al., Reference Bonnasse-Gahot, Berestycki, Depuiset, Gordon, Roché, Rodriguez and Nadal2018; Davies, Fry, Wilson, & Bishop, Reference Davies, Fry, Wilson and Bishop2013). Currently, most models consider contagion to be the central mechanism of riot spreading. However, what is lacking from these models is the inclusion of the importance of the identity salience to who imitates who, how this changes over time, and how this structures their social interactions. However, by including bystanders, and considering how they are likely to act next, the computational framework enables us to understand why these contagion models are successful at matching the data, even before many of the factors discussed in social psychology are considered.
However, despite the simplicity and flexibility of this framework, it does have a number of issues. Specifically, we find that the myriad of possible behaviours required to explain more complicated real-world scenarios, such as we find in the laboratory, cannot be understood solely through this computational model.
Financial support
JMA and DCR are funded by a grant from the Nuffield Foundation (The Psychological Roots of Societal Self Harm, 42868) awarded to DCR.
Conflict of interest
None.