3 Research Methodology
3.3 Research methods and tools
3.3.2 Modelling as a research method
A model is defined as "an external and explicit representation of parts of reality as seen by the people who wish to use that model to understand, to change, to manage and to control that part of reality" (Pidd, 1996: p15). In the management science, models are used for understanding, changing, managing, and controlling reality. Pidd emphasised on simplification in modelling, due the fact that it is not possible or feasible to model an operational system in its full details. Models, as simplified versions of reality, are used extensively in decision making due to their lower costs, timely delivery of needed information, ability in examining subjects which would be experimentally impossible, and providing insights and understandings about the decision problems of interest (Ragsdale, 2008). Figure 14 represents the general process of problem solving using a model as opposed to the experiment in the real world. As shown in this Figure, modelling starts with mapping the problem from the real world into the virtual world (abstraction), analysis and optimization of the model, and mapping back the solution into the real world. Human cognition is usually erroneous and results in inaccurate judgments due to factors such as irrationality in decision making (Ragsdale, 2008). Consequently, the main reason of using models is improving the process of problem solving and decision making by discovering and eliminating misunderstood elements of a problem during building the model, or gaining insights needed to make a decision through careful analysis of a completed model. Problem solving starts with the identification of the problem as its key step. Based on the nature of the problem, the simplest type of models which fits the problem and accurately reflects its characteristics is created and formulated. The selected model then is used to analyse the problem by generating and evaluating alternative scenarios which could lead to a solution.
The feasibility and quality of each potential solution needs to be tested, before implementation.
Figure 14: Modelling vs real‐world experiment (Borshchev and Filippov, 2004)
Compared to the real‐world experiments, modelling has lower cost, and is faster, safer, and more legally compatible while it is also possible to replicate the same conditions in order to repeat the simulation with any combination of decisions. In addition, it is rarely feasible or even possible to conduct a real‐world experiment, so it is inevitable to use a model. Based on their level of abstraction models could be classified into four categories as shown in Figure 15. In general, more accurate results are achieved when the level of abstraction in the model is low, while the modelling cost and difficulty increase with decreasing the level of abstraction. The Operation Exercise, as the first category with lowest level of abstraction, directly operates with the real environment in which the problem under study exists.
External abstractions and oversimplifications are narrowly introduced in this approach, so it has the highest level of reality among other modelling approaches. It involves designing and conducting a set of experiments in the real environment, and analysing their results.
Experiments should be designed carefully, consider errors resulting from measurement inaccuracies while evaluating results, and make inferences based upon the performed observations. Similar to the empirical research in the natural sciences, the operation exercise has an inductive approach in essence, allowing generalization of results drawn from observations of a given phenomenon. The drawbacks of operation exercise approach, similar to the real‐world experiment, are its high implementation cost, and difficulty in thoroughly analysing available alternatives, which lead to suboptimal conclusions. Due to these problems, although this approach has a significant contribution in improving the managerial decision making, still its usage is limited (Bradley et al., 1977).
The second approach to modelling, according to Figure 15, is Gaming, which provides decision makers with a responsive mechanism to check the performance of available
alternatives. In some cases, games are identified as a type of simulation modelling. It is a useful tool for helping managers to cope with the inherent complexities in a decision‐making process. Human interactions affecting the decision environment are active participants in gaming. Compared with the operation exercise approach, some degree of realism is lost in gaming, due to operating in an abstract environment, although a part of human interactions in the real environment are maintained. However, as the result of the higher level of abstraction, the cost of processing different alternatives would be lower, and the performance of these alternatives would be measured with higher speeds, compared with the previous approach (Bradley et al., 1977).
While in mentioned modelling approaches, human interactions are parts of the modelling process, in two other approaches human is external to the modelling process. As the third approach to modelling, Simulation provides a tool for evaluating the performance of existing alternatives. The main difference of the simulation with the gaming is removing human interactions from the modelling process. Similar to previous approaches, simulation is inductive in essence, and does not generate new alternatives or an optimum answer for the problem under study, but just evaluate previously identified alternatives. In simulation, exclusive definition of the problem using analytic terms is not necessary, giving this approach the flexibility in model formulation which is useful in the presence of uncertainties in decisions (Bradley et al., 1977).
The last category of modelling approaches with the highest level of abstraction is Analytical Models. In this approach, using an objective function, the problem is completely represented in mathematical terms. Decision conditions are portrayed as a set of mathematical constraints, and the objective function is sought to be maximized or minimized in a way to satisfy all constraints to reach to an optimum solution for the model. This type of modelling approach has the lowest cost and is the easiest model to develop, while having the least accuracy in the results due to the high degree of assumptions simplification (Bradley et al., 1977).
Figure 15: Model classification based on the level of abstraction (Bradley et al., 1977) In a different approach to the research, Harrison, et al. (2007) stated that theoretical and empirical analyses are two methods which scientific progress has historically relied on. The former is based on the formulation of a set of assumptions as mathematical relationships and deducing the consequences of those assumptions through mathematical proves.
According to Meredith et al. (1989), in the theoretical method a complex phenomenon is simplified and primarily addressed with mathematical models. The researcher looks at the problem through mathematical models and tries to find solutions within the defined model and to make sure that these solutions provide insights to the structure of the problems as defined in the model. Empirical research, as the second type, is driven mainly by empirical results and measurements (Burton and Obel, 1995). It starts with the observation of variables in real life and then analysing them to find the patterns of relationships, and is more concerned with the fit between the model and the observation in real‐life situations (Bertrand and Fransoo, 2002). Both approaches have some shortcomings, for example, in the theoretical analysis approach, especially in social science, due to the complexity and stochastic nature of social phenomena, analytically determination of results of assumptions using mathematical techniques is very difficult. This difficulty leads to choosing the assumptions mostly based on their usefulness for driving the desired consequences, not their correspondence to reality. In empirical approach, the main problem is the unavailability of data and difficulty to measure them, in addition to the need for comparable measures across a sample or an extended time period (Harrison et al., 2007).
Harrison et al. (2007) considered the introduction of simulation models as the third method of research which allows researchers to handle complex mathematical relationships using computer‐based numerical methods. This makes them able to use more realistic assumptions instead of compromising them with analytically convenient assumptions. In addition, using simulation, researchers could produce their own virtual data, thus overcoming the data availability problem in empirical approach, to some extent.