Dr Corinna Elsenbroich

This is an Evaluation Policy and Practice Note that explores the application of Agent-Based Modelling (ABM) for complex policy evaluation.

In this paper we assess the construct validity and theoretical emdeddedness of agent-based models of normative behaviour drawing on experimental social psychology. We contend that social psychology and agent-based modelling share the focus of 'observing' the processes and outcomes of the interaction of individual agents. The paper focuses on two from a taxonomy of agent-based models of normative behaviour. This enables the identification of the assumptions the models are built on and in turn, reflection on the assumptions themselves from a socio-psychological perspective.

As policy makers require more rigorous assessments for the strength of evidence in Theory-Based evaluations, Bayesian logic is attracting increasing interest; however, the estimation of probabilities that this logic (almost) inevitably requires presents challenges. Probabilities can be estimated on the basis of empirical frequencies, but such data are often unavailable for most mechanisms that are objects of evaluation. Subjective probability elicitation techniques are well established in other fields and potentially applicable, but they present potential challenges and might not always be feasible. We introduce the community to a third way: simulated probabilities. We provide proof of concept that simulation can be used to estimate probabilities in diagnostic evaluation and illustrate our case with an application to health policy.

We present an analysis for modelling social norms. In social psychology three different normative behaviours have been identified: obedience, conformity and compliance. We show that this triad is a useful conceptualisation of normative behaviour and that current models only ever deal with conformity and obedience two, neglecting compliance. We argue that this is a result from modelling having so far focussed too much on agent behaviour rather than agent knowledge and that cognitive models of normative behaviour are needed to capture this third and arguably most interesting normative behaviour.

In this paper we assess the construct validity and theoretical emdeddedness of agent-based models of normative behaviour drawing on experimental social psychology. We contend that social psychology and agent-based modelling share the focus of ‘observing’ the processes and outcomes of the interaction of individual agents. The paper focuses on two from a taxonomy of agent-based models of normative behaviour. This enables the identification of the assumptions the models are built on and in turn, reflection on the assumptions themselves from a socio-psychological perspective.

We present DIAL, a model of group dynamics and opinion dynamics. It features dialogues, in which agents gamble about reputation points. Intra-group radicalisation of opinions appears to be an emergent phenomenon. We position this model within the theoretical literature on opinion dynamics and social influence. Moreover, we investigate the effect of argumentation on group structure by simulation experiments. We compare runs of the model with varying influence of the outcome of debates on the reputation of the agents. © JASSS.

Abstract Extortion racketeering is a crime that blights the lives of everyone in societies where it takes hold. Whilst most European countries have some form of extortion racketeering, in most countries it is isolated to some ethnic communities. In Southern Italy and Sicily, extortion racketeering is still a feature of overall society. This paper attempts to look at the phenomenon from the angle of collectives, of resistance building through civic organisations such as Addiopizzo. For this investigation a computational model is presented to analyse the effect of team-reasoning on levels of resistance in systemic extortion rackets. An agent-based model is presented that implements the interaction of different kinds of decision-making of extortion victims with law enforcement deterrence. The results show that established extortion rackets are hard to undermine unless bottom-up civic engagement and law enforcement go hand in hand.

This book presents a multi-disciplinary investigation into extortion rackets with a particular focus on the structures of criminal organisations and their collapse, societal processes in which extortion rackets strive and fail and the impacts of bottom-up and top-down ways of fighting extortion racketeering. Through integrating a range of disciplines and methods the book provides an extensive case study of empirically based computational social science. It is based on a wealth of qualitative data regarding multiple extortion rackets, such as the Sicilian Mafia, an international money laundering organisation and a predatory extortion case in Germany. Computational methods are used for data analysis, to help in operationalising data for use in agent-based models and to explore structures and dynamics of extortion racketeering through simulations. In addition to textual data sources, stakeholders and experts are extensively involved, providing narratives for analysis and qualitative validation of models. The book presents a systematic application of computational social science methods to the substantive area of extortion racketeering. The reader will gain a deep understanding of extortion rackets, in particular their entrenchment in society and processes supporting and undermining extortion rackets. Also covered are computational social science methods, in particular computationally assisted text analysis and agent-based modelling, and the integration of empirical, theoretical and computational social science.

The concept of self-organization in social science is reviewed. In the first two sections, some basic features of self-organizing dynamical systems in general science are presented and the origin of the concept is reconstructed, paying special attention to social science accounts of self-organization. Then, theoretical and methodological considerations regarding the current application of the concept and prospective challenges are examined.

Systematic extortion involves a long term parasitic relationship between the criminal and the victim. Game theory analysis has provided insight into the choices of individual hypothetical criminal and victim pairs. In this paper we present an agent-based model so as to extend the analysis to the relationship between extorters and other potential victims. The model is developed in two stages, the first to be closest to game theory, the second one making the decision informed by the social environment of the victim. The agent-based model shows the importance of social aspects for the functioning of extortion rackets.

Collective dilemmas have attracted widespread interest in several social sciences and the humanities including economics, sociology and philosophy. Since Hardin’s intuitive example of the Tragedy of the Commons, many real-world public goods dilemmas have been analysed with a wide ranging set of possible and actual solutions. The plethora of solutions to these dilemmas suggests that people make different kinds of decision in different situations. Rather than trying to find a unifying kind of reasoning to capture all situations, as the paradigm of rationality has done, this article develops a framework of agent decision-making for social simulation, that takes seriously both different kinds of decision making as well as different interpretations of situations. The Contextual Action Framework for Computational Agents allows for the modelling of complex social phenomena, like dilemma situations, with relatively simple agents by shifting complexity from an agent’s cognition to an agent’s context.

The book focusses on questions of individual and collective action, the emergence and dynamics of social norms and the feedback between individual behaviour and social phenomena. It discusses traditional modelling approaches to social norms and shows the usefulness of agent-based modelling for the study of these micro-macro interactions. Existing agent-based models of social norms are discussed and it is shown that so far too much priority has been given to parsimonious models and questions of the emergence of norms, with many aspects of social norms, such as norm-change, not being modelled. Juvenile delinquency, group radicalisation and moral decision making are used as case studies for agent-based models of collective action extending existing models by providing an embedding into social networks, social influence via argumentation and a causal action theory of moral decision making. The major contribution of the book is to highlight the multifaceted nature of the dynamics of social norms, consisting not only of emergence, and the importance of embedding of agent-based models into existing theory.

In the ourishing research area of agent-based social simulation, the focus is on the emergence of social phenomena from the interactions of individual autonomous agents. There is, however, a relative underexposure of the cognitive properties of agents, as the existing agent architectures often focus on behaviour alone. Cognition becomes particularly salient when the subject under investigation concerns social phenomena where agents need to reason about other agents' beliefs. We see this as a requirement for any communication with some degree of intelligence. In this paper we use concepts and methods from dynamic epistemic logic to build agents capable of reasoning about other agents' beliefs and their own. In dynamic epistemic logic, agents are assumed to be perfect rational reasoners. We break with this unrealistic assumption in order to bridge the gap between the sociological and the logical approach. Our model is based on a minimal set of assumptions representing cognitive processes relevant to modelling the macro-phenomena of group formation and radicalisation.

What kind of knowledge can we obtain from agent-based models? The claim that they help us to study the social world needs unpacking. I will defend agent-based modelling against a recent criticism that undermines its potential as a method to investigate underlying mechanisms and provide explanations of social phenomena. I show that the criticism is unwarranted and the problem can be resolved with an account of explanation that is associated with the social sciences anyway, the mechanism account of explanation developed in Machamer et al. (2000). I finish off discussing the mechanism account with relation to prediction in agent-based modelling. © Copyright JASSS.

Theoretical game theory has been a successful theory in economics and other social sciences. Experimental game theory, on the other hand, seems to open more problems than it solves. Almost every experimental setup results in much higher levels of cooperative behaviour than rationality allows. This paper presents an agent-based model to generate the population level outcomes of some prominent explanations of human behaviour by implementing alternatives to perfect rationality.

This paper presents an agent-based model of team reasoning in a social dilemma game. Starting from the conundrum of empirically high levels of cooperation in dilemma games, contradicting traditional utility maximisation assumptions of game theory, Bacharach (1999, 2006) developed a theory of team reasoning. The idea behind team reasoning is that agents do not try to maximise their own utility but make choices as part of a team. This paper presents a model of preference convergence, mirroring adaptation dynamics of team reasoning. It describes an agent-based model simulating a repeated public goods game between a designated set of agents, a team. In the model agents have a probability to choose cooperation or defection, adjusting this preferences in the face of the revealed preferences of other players. The model is a classic binary choice model mapping an individual's preference for cooperation onto the binary behaviour choice of cooperation and defection. Preferences are updated in reaction to the behaviour choices of the team. Starting from simple stated preferences, the model implements a reframing of utility maximisation as applying to a group rather than an individual, modelling the importance of social interaction for individual preferences and the dependency of choice on social context. Results show that team reasoning, as implemented here, can explain high levels of cooperation found in the real world resulting from a wide range of settings. It also shows that team reasoning, as implemented here, is not a ‘sucker’ strategy except when adaptation rates are very slow. This paper demonstrates how agent-based models can be used to examine the role of social contexts for individual decision making.