Computational simulation is useful when the phenomenon to be studied is not directly
accessible or is difficult to observe (Davidsson, 2002). Simulation helps to study the
emergence of macro-level phenomena driven by individual actions, the influence of macro structure on individual behaviours, the consequences of social networks and interactions,
and the implications of learning and adaptation. It allows the development of tools that
can inform practical actions and contribute to scientific theory (Gilbert,1999).
2.2.1 Simulation with Theory and Reality
Computational simulation is a mediator between theory and data to help to explain
social phenomena (Sawyer,2004). Unlike mathematical modeling, computational mod-
eling is not limited by traditional mathematical tools. It enjoys expressive power while
remaining precise (Sun and Naveh,2007). Agent-based modeling is appropriate for the
domains characterized by a high degree of localisation and distribution and dominated by discrete decisions. Equation-based modeling is naturally applied to systems that can be modeled centrally, and in which the dynamics are dominated by physical laws rather
than information processing (Davidsson,2002). Both equation-based and agent-based
modelling have empirical connections and have value in understanding complex societies
(Byrne and Callaghan,2013). Computational models score well on internal validity, sta-
tistical conclusion validity, external validity and construct validity when compared with field experiments and laboratory experiments, except for a subcategory of correspon- dence of the model where field experiments are better. This is the concern for realism
and why realism must be addressed in computational modeling (Burton and Obel,1995).
Computational simulation is connected with the real world by processing data collected from society and the environment. A data-driven simulation is the interaction between multi-agent simulations and real world. The frequency of interaction increases as new sources of data become available. The simulation models such as agent-based models can be constructed to support persistent run-time interactions between computer agents and real-world entities via general types of input-output data streams. In this way, agent- based models become data-driven dynamic application systems entailing the capability of incorporating additional data into an executable application and, vice versa, the ability
of applications to dynamically steer the measurement process (Conte et al.,2012).
2.2.2 Process of Computational Simulation
The process of developments unfolding through time is an essential basis for understand- ing complex systems. Process-tracing is a methodology well-suited for testing theories in a world marked by multiple interaction effects, where it is difficult to explain outcomes
in terms of a small number of independent variables (Byrne and Callaghan,2013). The
model statement of variables, parameters, relations and the computational process need to be matched. A experimental design is to try different initial settings of simulations to find the best strategy to satisfy purpose while the data analysis is the analysis of
1995). The process of computational simulation started from initial settings and ended by data analysis mainly follows bottom-up and top-down directions.
A complex system may be simulated to emerge from a collection of interacting objects. The process of simulating complex systems is describing, exploring, modelling, and es- tablishing causality. In a complex system, a structure is the result of actions by objects within the simulation rather than something pre-existing. But a structure may have a
causal potential in relation to the possibility of future actions (Byrne and Callaghan,
2013). Complex social systems are characterised by multiple ontological levels with
multi-directional connections. The processes are not only from the micro to the macro-
scopic levels but also back from the macro to the micro-levels (Conte et al.,2012). The
Micro-Macro link is the loop process by which behaviour at the individual level gener- ates higher-level structures, i.e., bottom-up process, which feedback to the lower level,
i.e., top-down process (Conte et al.,2012).
A probability distribution exists on the upper level while the realisation of this probabil- ity distribution is determined on the lower level. The attributes of the population and of
the individuals are interdependent (Gilbert and Troitzsch,2005). In a macro simulation,
the set of individuals is viewed as a structure that can be characterized by a number of variables, whereas in a micro simulation, the structure is viewed as emergence from
the interactions between the individuals (Davidsson, 2002). These simulations need to
be accompanied by the micro-macro-loop theories, i.e., the theories of mechanisms at the individual level that affect the global behaviour, and the theories of loop-closing
downward effects or second-order emergence (Conte et al.,2012).
The study of emergent social behaviour has benefited from computational simulation. Hierarchical and multilevel cultural models are needed to take into account the interde- pendence of cultural features and the interconnection of cultural dynamics with other
social processes (Conte et al.,2012).
2.2.3 Types of Computational Simulation
Computational Simulations may not only be used to simulate the reality to predict the future trend or explain existing phenomena, but also be utilised to build up abstract models to explore the principles of social emergence. The former can be named “thick”
simulations while the latter can be called “thin” simulations (Kliemt,1996). Regarding
to different causes, the computational simulations can be used to explore the sequence of causes and results to explain existing phenomena. In terms of different purposes, computational simulations may be used to test hypotheses, train managers, understand decision making design organisations etc.
2.2.3.1 “Thick” and “Thin” Simulations
Kliemt (1996) put forward that computational simulations could be “thick” or “thin”,
where “thick” simulations are detailed, draw on abundant empirical data, and tell the investigator a lot about a specific question. Such simulations are useful in domains which employed case studies traditionally. “Thin” simulations are the tools for “con- trolled speculation”, useful in disciplining theory formation, simplifying and distorting
assumptions (Gotts et al.,2003).
In order to clarify which aspects of a specific situation to be modelled are likely to be most important in understanding its dynamics, the use of “thin” simulations may help to prevent the over-interpretation of results from “thick” ones. Because understanding how such systems are distinctive requires understanding of how complex effects can result
from interactions between much simpler elements via “thin” simulations (Gotts et al.,
2003).
Burton and Obel (1995) discussed the balance of reality and simplicity between “thick”
simulations and “thin” simulations. The reality of the model is a central issue. But how close must a model be to ”reality” is relative. The balance between “thick” simulations and “thin” simulations should be considered according the purposes and issues which
need to be addressed (Burton and Obel,1995).
2.2.3.2 Cause-Driven Simulations
The causes of phenomena should be identified to clarify the questions: what might cause changes in the future state of the system and what changes would result from such causes. And the causes should be used by agents to take action, which will result in some particular results. The adjustment of parameters has profound causal power in relation to the outcome states, which emerge from interactive agents with the capability
of organisation and decision making (Byrne and Callaghan,2013).
The evaluation of causes is about what worked as a basis of saying what would work, and realist evaluation is always framed contextually and makes claims constructed in
applicable conditions (Byrne and Callaghan, 2013). A great number of quantitative
variations may result in qualitative changes, i.e., the changes of kind which are more
significant than the changes of degree (Byrne and Callaghan,2013).
Selection pressures cause evolutionary social change, which is a process of increasing structural differentiation to upgrade its longer-term flexibility and adaptability. Social systems are open-ended dynamic wholes interacting with their environments, including
other systems (Byrne and Callaghan,2013). A new adoption is a response to the current
relative adoption success to extrinsic, environmental factors applying selection pressures
(Watts and Gilbert,2014).
2.2.3.3 Purpose-Driven Simulations
A purpose could be to formulate theories which explain why existing oragnisations be- haved in particular ways, to test these theories by comparing the observed past be- haviours with the simulated behaviours generated by the models, and to predict how
these organisations would behave in the future (Burton and Obel, 1995). The models
used for predictive purposes are related with certain mechanisms. They are output- oriented since, for a given set of initial conditions, to show the state of the system
evolving in time (Conte et al., 2012). In addition, abstract approaches could be im-
proved using real data as the basis for specifying an agent-based model (Byrne and
Callaghan,2013).
Another purpose might be to explore the implications of reasonable assumptions about organisational behaviours, in order to determine what the world is like when these
assumptions are true (Burton and Obel, 1995). For example, if the prevalence of two
sex organisms needs to be explained, what the biological world would be like if there
were three sexes can be studied (Conte et al.,2012).
Some other purposes could be to determine which organisation forms are suited to particular goals, or to train people to function better in organisational settings and
operations (Burton and Obel,1995).
2.2.4 The Application of Computational Simulation
The identification of a plausible generative mechanism through simulation is not the
end of research, but rather a step to further researches and developments (Watts and
Gilbert,2014). The application of simulation in education and engineering is an open-
ended process of modification, updating and improvement.
In the 1980s, computational simulation was applied in the development of education tools to help students complete their courses such as physics, chemistry and architecture. In the early stages, the simulation tools used in physics and chemistry reduced the opportunities of understanding the real process of physics and chemistry phenomena for students because early simulation software packages were developed as “black boxes”. Students were only allowed to input data, and the final result would be output directly without any explanation and illustration. Thus they did not know how the result was generated. Over time the simulation software improved. Students could experience each step of virtual experiment and be asked to find errors in the simulations. Similar improvement occurred in the development of simulations for computer aided design tools.
The early architecture software made students feel like engineers rather than designers when using these tools. But the tools were continually updated and new functions
embedded into them to enhance the user experience. Consequently, the simulation
used in architecture and civil engineering improved the efficiency and possibilities of
generating more interesting concepts and structures (Turkle et al.,2009).
Computational simulations have also been applied in the evolution of artificial languages to explore the origin of language and related mechanisms. Computational linguistics, which develops concrete operational models of the processing, can sustain language as a
complex adaptive system (Steels,2012). The language games as a type of computational
simulation for social science are mainly used to explore the relationship between the
evolution of language and social creativity (Saunders and Bown,2015). To improve social
creativity, the analysis of conversations are important for understanding the functions well described on a specific body of knowledge accumulated through communications. Conversations produce ordered accounts of particular phenomena instead of rules. Thus conversation analysis should be treated as useful data rather than as a prescription for
designers (Luff et al.,2014).
Although computational simulation is widely used in a number of fields, it has some limits. The iterative calculations involved in numerical simulations often produce accu- mulation errors. Such errors have the potential of misleading the evaluation of results
and the validity of the model. In addition, simulation is limited with the memory
space and computational performance that cause optimising initial settings and pro-
cesses (Cangelosi and Parisi,2012). Thus only partial realities are likely to be reflected.