Agent Based Modelling Support System in Crime Science

Making realistic predictions about the occurrence of crime is a challenging research area. City-
wide crime patterns depend on the behaviour and interactions of a huge number of people
(including victims, offenders, and passers-by) as well as a multitude of environmental factors.
Modern criminology theory has highlighted the individual-level nature of crime—whereby
overall crime rates emerge from individual crimes that are committed by individual people in
individual places—but traditional modelling methodologies struggle to capture the complex
dynamics of the system. The decision whether or not to commit a burglary, for example, is based
on a person's unique behavioural circumstances and the immediate surrounding environment. To
address these problems, individual-level simulation techniques such as agent-based modelling
have begun to spread to the field of criminology. These models simulate the behaviour of
individual people and objects directly; virtual ‘agents' are placed in an environment that allows
them to travel through space and time, behaving as they would do in the real world. We outline
an advanced agent-based model that can be used to simulate occurrences of residential burglary
at an individual level.

With respect to modeling crime, an agent based model can be build directly simulate the
behavior of offenders/victims/guardians as they travel around their environment on typical
routine activities and predict, as a result these individual behavior patterns, when a crime
occurrence is likely to occur.

Crime occurrences are driven by a complicated mix of distinct influences, including those of the
environment, the surrounding social context, and personal behavior/psychology of the people
who could influence a crime event. Agent-based modelling is a methodology used in computer
simulation that concentrates on individual-level behaviors and is ideally suited to modelling
crime. This is particularly true of crimes such as burglary or street crime, which are heavily
influenced by environmental factors and by the behavior of individual people. In an agent-based
crime model, virtual “agents” are placed in an environment that allows them to travel through
space and time, behaving as they would do in the real world. This entry will discuss why the
crime system is such an ideal candidate for agent-based models and will review a number of
crime models that have recently arisen.
Criminologists focus on explaining crime and criminal behavior. This necessarily requires an
examination of individual decision-making within the context of social processes that occur over
time. To complicate matters, social processes are multifaceted and include spatial, temporal, and
cultural dimensions. Collecting data on and modeling these processes is difficult using traditional
empirical approaches.

Crime events are the output of a complex system. Where and when crime does and does not
occur, by whom it is carried out, and against whom/what it is perpetrated, are all the result of
interconnected, interdependent, and situated interactions, amongst multiple, heterogeneous,
adaptive actors and the environment within which they are situated.

These interactions of crime commission (and control), and the mechanisms that underlie them
are consistently difficult to observe or control in service of the scientific method. Academically,
this limits theoretical advancement. But more importantly, we would argue, it constrains efforts
to respond to crime, disrupt ongoing events, and better understand those offences that have
occurred previously to facilitate future prevention.

In this lecture we discuss computational analytic methods that embrace this complexity and aim
to explore it, with the hope of increasing our understanding of crime, and informing what we
might do about it.

Drawing on research from the field of distributed computing, and emboldened by a range of
other social sciences with compatible goals, in the last decade crime scientists have begun to
apply computational models to better understand how the actions and interactions of individuals
influence complex social systems, the crime opportunities and events they give rise to, and the
effectiveness of potential responses to them. Computational models allow researchers to
construct simplified versions of real-world systems by encoding them as computer programs.

One particular approach, the individual- or agent-based model (ABM), is seeing a considerable
up swell in application. ABMs allow researchers to construct synthetic environments and
populate them with virtual decision makers (referred to as agents) designed to represent key
societal actors. These agents are imbued with behaviours that reflect theoretical proposition
and/or previous empirical insight. Once a model is constructed it can be used to conduct
simulated experiments whereby populations of agents are Agent-based decision-support systems
instantiated in some particular configuration, and the aggregate outcome patterns (e.g. patterns of
crime) that result from their repeated interactions observed.

Within this burgeoning field of computational criminology, ABMs have been productively
constructed to study a diverse range of crime problems and responses to them. This approach to
understanding crime events and patterns departs from a range of other quantitative modelling
techniques that typically operate in a ‘top-down’ analytic fashion, inferring individual-level
explanations from observed aggregate associations (e.g. affluence and crime).

By contrast, ABMs allow researchers to explore crime events from the bottom-up, casting crime
as the emergent outcome of a series of lower-order interactions. By doing so, individual crime-
event actors, their characteristics and calculi are formalised, and by running the model one is
able to assess if proposed individual-level mechanisms are capable of generating plausible
outcome patterns.

Modelling the connection between individual behaviour and crime events confers several
inherent analytic advantages in examining the complex social systems that produce and shape
crime problems. First, it permits the explicit modelling of heterogonous populations. Agents
within an ABM are represented as instantiations of an agent class or template that contains
individual characteristics and behaviours that can be set on a per-agent basis.

Thus, realistically diverse populations of actors can be modelled. This approach overcomes
traditional analytical assumptions that often enforce population homogeneity in order to manage
analytical tractability. Second, by modelling the repeated interaction of agents over time, ABMs
permit the exploration of longitudinal processes.

The agent based modeling a system is modeled as a collection of autonomous decision making
entities called agents. Each agent individual assessed its situation and makes decisions on the
basis of a set of rules. Agents may execute various behaviours appropriate for the system they
represents for example producing , consuming, or selling.

Finally Agent Based Modeling consists of a system of agents and the relationship between them
(social, environment and individual complex behavior). Even a simple agent based model can
exhibit complex behavior patterns.

Why this model?

2. Provide a natural description of a system

3. Flexible in nature; agent interactions are heterogeneous and can generate network effects.

ABM simulate urban crime patterns quantify the following information.

1. Characteristics of the population, model and agents

2. Crime types investigated

3. Interrogation of the model via sensitivity testing and validation

By constructing agents that represent potential offenders, victims and crime controllers, and
examining their interaction, ABMs allow researchers to understand what configurations of
behavioural and environmental conditions are capable of generating particular patterns of crime,
and, in turn, to construct and estimate the effectiveness of crime-control strategies that might be
implemented to respond to them. ABMs thus provide computational laboratories in which
researchers can prototype, test, and refine both criminological theory, and proposed intervention
– free from ethical and logistical constraints and prior to the carrying out of necessary, but likely
costly, empirical investigation.

In this discussion we focus on the application of ABMs for the latter purpose – as a means to
investigate the likely impacts of differing crime-control strategies. In doing so, we aim to
demonstrate how ABM can be productively used in the field of crime science as a means to
support decision-making in complex systems. Following this rationale, the remainder of the
discussion begins by briefly describing the notion of ABMs as a method for decision support.
Subsequently, we describe an example ABM of police resource deployment and present a series
of illustrative experiments run using it. We conclude by discussing some of the insights gleaned
from this simple model and highlighting several potential fruitful areas of research where we
believe ABM may be productively incorporated into key crime science problem areas

As we see it, policymakers have three primary objectives: (i) identifying substantive social
problems, (ii) understanding the underlying causes of said problems, and (iii) activating the most
effective means for addressing them. In the interests of space, we assume the first two of these
tasks are appropriately conducted (our policymaker has been able to locate a substantive social
problem and that research has identified some underlying cause for the problem). The challenge
now is to identify the most effective way of tackling the underlying cause(s).

The second constraint is scale. Some social problems can be tackled in a demonstration or pilot
project. But many cannot. Social marketing, for instance, relies on wide-scale network effects,
which will be outside the financial reach of most pilot schemes. An action research paradigm
might provide some effective evidence, but even the most promising intervention can become
diluted when mainstreamed.

The problems-----the models------- the model environment------call for services

We demonstrated from these discussions how ABM will be productively used in supporting the
decisions of crime scientists and practitioners who seek to devise responses to crime problems
that are the product of a complex system. While ABMs do not seek to replace existing analytical
methods, they do provide new means to support theoretical and empirical efforts, especially in
areas of research where a more ‘traditional’ approach is hampered by logistical and ethical
constraints.