2.1 Introduction

In this chapter, research methods are studied with respect to their adequacy and appropriateness for the detection, explanation, prediction, and combating of clandestine and dark economic activities. The main result is that there are two pitfalls of such studies, namely “the fallacy of misplaced concreteness” (A.N. Whitehead) and “the fallacy of disregarded abstractness” (M. Schramm). These potential fallacies point to the issue of the adequacy and the appropriateness of research methods and research goals. The more abstract the research objective, the more general its propositions; the more specific the study intentions, the more specific the various considerations. As a consequence, there can be no single method of research for all objectives. At this point, the availability of research methods becomes a major issue. For instance, in the more distant past, the tools consisted of theoretical models and a certain number of empirical methods. Neither laboratory and field experiments nor computer simulation models were on hand. Newly available methods now allow for an expansion of research in two directions toward a deeper understanding of economic and social phenomena, and to different levels of analysis.

Take tax evasion as an example. Theoretically, there are a number of general theories (expected utility theory, psychological theories, etc.) that are useful in formulating hypotheses. With experimental methods and computer simulation models, these theories can now be tested in detail. For instance, one can consider what role social norms may play and how the interaction of people changes their tax-paying behavior. However, if the research objective is to find effective measures to combat, say, VAT fraud, these theories are of limited use. In the latter case, the availability of electronic cash registers, for instance, might be superior to knowing the degree of risk aversion that plays an important role in the relevant economic theory. Moreover, the more realistic and policy relevant a research question becomes, the larger will be the number of intervening variables that are not included in the theory. In the VAT fraud example, the size of the firm, its owners, as also the nature of its business are important variables to consider. Hence, the evaluation of research methods may only provide conditional recommendations as to their applicability and effectiveness, by pondering their advantages and shortcomings, given the research objective.

Another focus of the paper is the complexity of clandestine economic and social interactions. For instance, as is well known, tax evasion, money laundering, and tax havens are interrelated (Tavares, 2013). Consider a firm evading the corporate income tax in its home country via business operations in a tax-haven country. To avoid the immediate detection of this tax fraud, various hiding operations are necessary. In order to spend the money earned in this way on official transactions, laundering activities are required. A large number of transactions and persons are involved in such tax evasion. In addition, the social network structure of communities of people and societies renders clandestine activities particularly complex. As a consequence, research methods must be suitable, appropriate, and applicable concerning the complexity of the different layers of analysis.

The rest of this chapter is organized as follows. The next section deals with why it may be useful to study clandestine (economic) behavior. In Section 2.3, the available tools for such studies are presented and briefly discussed. In Section 2.4, the focus is on the complexity of interactions in clandestine activities. In this context, network effects and the dynamics of interactions are considered. Section 2.5 takes account of the layers of analysis, from the individual, to groups, to the macro level. The adequate application of the available research tools is discussed in Section 2.6. Section 2.7 concludes.

2.2 Why Study Clandestine Behavior At All?

Clandestine behavior results mainly from three very different motivations. Firstly, it provides an area of privacy; secondly, to hide an activity because it is expected to be socially (i.e., by significant others) unacceptable; and thirdly, to conceal illegal behavior. Of course, to prevent and to combat crimes, it is necessary to discover their underlying mechanisms. Nevertheless, clandestine activities, whether legal or illegal, are interesting from a social science perspective. As demonstrated by John List (2006) and Levitt and List (2007), people behave quite differently when they know that they are observed, in comparison to behavior in similar circumstances when they assume they are not observed. This is a general problem for research in social sciences; if people behave differently when observed, how can one study authentic behavior? One possibility is field experiments in which participants do not know that they participate in an experiment, but this is only feasible if participants did not have to give their informed consent (List, 2008). Another possibility is to use social media data, that is, data provided unintentionally by a very large number of users of these media and collected there (Rudder, 2014). This data demonstrates clearly that unobserved behavior varies greatly from that which is observed and pretended (see Rudder, 2014, for instructive examples). Hence, if authentic behavior is to be studied, clandestine behavior may be the key. In effect, social science cannot ignore such behavior if it wants to study genuine human activity.

However, social science is not entirely an end in itself. Although pure science also plays a role, social science results are necessary to provide a basis for better policies. A trial-and-error approach to social problems might be very expensive or even completely ineffective and futile. Knowledge about the mechanisms of crimes, drug trafficking, tax evasion, money laundering, corruption, and other activities may help develop policies and approaches to counter such activities effectively and efficiently. The social costs of illegal clandestine activities are certainly high, but the costs and damages of misconceived policies could be even higher.

In addition to the dependency of human behavior on simply feeling or actually being observed, a crucial question in social science is that of causality. Behavior occurs in a social environment in which action and reaction depend on and cause each other. Consequently, almost all human behavior is determined by the relevant social (sub)system. Therefore, behavior will genuinely suffer from circular causality (see Thomas, 2006). Wherever the circle of behavior is cut, the “causal” starting point is fixed. This means that the concept of causality loses some of its power. However, it may be that not all elements in the social system are of equal importance. Hence, the essence of research is to find those elements that are crucial for the behavior of the social subsystem under study. This requires a very careful investigation of the behavior in question. Without such studies, the chances of finding the drivers of clandestine activities such as crimes are very small. To go beyond symptomatic policies (that are not well-suited to solving social problems), research in the area of clandestine behavior is required. Since there are good reasons to study clandestine behavior and activities, the next question is how to do it.

2.3 Tools for Studying Clandestine Activities

In order to study human behavior in general, the usual tools are theory, empirical data analysis, questionnaires, laboratory and field experiments, and computer simulations. Recently, the huge amounts of data produced in social media and telecommunication networks have also be applied to study human behavior and clandestine activities.

The intended research method depends first on the (scientific and academic) goal of the analysis (Figure 2.1). Generally, three goals are relevant: description, explanation, and prediction (see Bunge, 1967b). This is particularly true for clandestine activities such as crimes and tax evasion. A description is required to separate the particular activity under scrutiny from all others; moreover, it also should provide a method for measuring the dimension of the respective activity. This is the first serious problem with clandestine activities. The next step of explanation consists of reduction and abstraction; an isolated phenomenon is embedded in an abstract and reduced frame in which (causal) relationships can be modeled. The last step of prediction requires a lower level of abstraction in order to anticipate the (potential) results of activities in a real-world scenario.

These scientific objectives require, first of all, theories and models. Theories are highly abstract mental constructs that are based on a few basic axioms that are employed to derive the logic consequences of these axioms. However, theories as such are not testable; they are metaphysical objects (Duhem, 1908/1978, Chapter 1). To become both testable and applicable, specifications are necessary. In this way, a theory is divided into many models, in which “a model is an idealized representation of a class of real objects” (Bunge, 1967a, p. 386). Moreover, a single model can never represent the entire theory (Leijonhufvud, 1997, p. 193). As depicted in Figure 2.1, at least three different kinds of models can be identified: descriptive, explanatory, and predictive.

To start with the first kind of model, even descriptions and measurement are not possible without a descriptive model as a basis. Although the notions of “crime,” “tax evasion,” and so on are legally defined, the specification as well as the quantitative measurement of the dimensions of these concepts are not so straightforward because the respective actions are clandestine. Although hidden actions are by nature not directly observable, their consequences become evident sooner or later. A descriptive model has to combine observable outcomes with unobservable activities. Hence, measurement without such a model seems either impossible or merely ad hoc.

To understand a real-world phenomenon in general, an explanatory theory and many explanatory models are not only possible but also necessary. Theories are of necessity highly abstract. Therefore, theories cannot really be considered as “realistic,” a failure that Alfred Whitehead called “the fallacy of misplaced concreteness” (Whitehead, 1925/1967, p. 51). Moreover, theories may imply absurd conclusions, but, “I doubt there is any set of assumptions that does not produce absurd conclusions when applied to circumstances far removed from the context in which they were conceived” (Rubinstein, 2006, p. 871). Therefore, whether one likes it or not, theories and clearly specified models are required to explain and understand economic (and noneconomic) phenomena, irrespective of whether they are based on legal or illegal behavior.

Predictive models presuppose even more empirical information that is not necessarily incorporated in explanatory models. As already recognized and analyzed by Duhem (1908/1978) and Quine (1951), testable and applicable models need a number of additional assumptions and specifications that are not strictly part of the theory. Therefore, theories cannot be regarded as true or false, independent of facts. Hence, the commonly supposed dichotomy between analytic and synthetic truth does not exist Quine (1951, p. 20). Put differently, no explanatory or predictive model is useful without the incorporation of empirical facts.

This leads to empirical studies being used as research tools in the area of clandestine activities; with respect to the shadow economy and tax compliance, see Slemrod and Weber (2012) and Feld and Larsen (2012). However, these activities, and sometimes even their results and outcomes, remain undetected by the public and the authorities. Although detected cases of fraud and crimes can be studied empirically, the number of undetected cases is inevitably high. This restricts empirical research approaches. Moreover, questionnaire research is also very limited, since the willingness to admit to socially outlawed behavior such as crimes and fraud, even if not sanctioned, might be rather low. Experimental studies, in the laboratory, as well as in the field, are both possible and useful. As demonstrated by List (2006) and Levitt and List (2007), there is a great difference in the behavior of people when they are observed and when they think that they are not observed. Field experiments are expensive and pose ethical questions, because the condition of “informed consent” would destroy participant perceptions of being unobserved (List, 2008). However, as, for instance, experimental laboratory studies with tax compliance show, such experiments provide some insights into the tax compliance behavior of participants; see Alm and Jacobson (2007), Alm (2012), and Pickhardt and Prinz (2014). Nevertheless, there are clear limitations to the generalizability of these results (Levitt and List, 2007). In this respect, field experiments are of greater value (Kleven et al., 2011), but are not always possible, as, for example, with serious crimes.

Social media and telecommunication data, as indicated above, have become new sources for research. Data sets are also available for crimes, tax fraud, money laundering, and so on. Each activity leaves traces in electronic media and since these traces are the result of unobserved, authentic behavior, they are unbiased. Nowadays, authorities use this data, sometimes by intruding into the private sphere of innocent people, which indicates the ethical limits for this kind of research. Last, but not least, computational studies are both feasible and useful. The advantage of simulations is that the connectedness and social embeddedness of clandestine behavior can be accounted for (see Korobow et al., 2007), which does not violate individual privacy. Moreover, not only the social environment and networks of persons can be incorporated but also the dynamics of behavior and of networks can be studied; for an overview of such studies regarding tax compliance, see Pickhardt and Prinz (2014). Such studies may provide the basis for predictive analyses of clandestine activities.

To sum up, all research tools available can, in principle, be applied to clandestine activities. However, the limited availability of empirical data cannot be overcome completely with the methods at hand, although social media and telecommunication data sources are useful. Nonetheless, the lack of reliable empirical data is one of the major remaining issues of hidden activities.

2.4 Networks and the Complexity of Clandestine Interactions

Although many crimes are committed individually, without interaction with others, in economic crimes and fraud a number of people are usually involved. Moreover, crimes and fraud take place in a social environment that might encourage or discourage criminal and fraudulent activities. This implies that clandestine activities exhibit a network structure. The motivation for all kind of activities (not only illegal ones) depends also on the personal social environment, as do the potential activities, as well as their aggregate outcomes.

This network structure makes it difficult to isolate individual actions from the network structure. The theoretical reason is that networks may create nonlinear feedback loops (Heylighen, 2001) between persons and actions, which are not under individual control. According to the sociologist Niklas Luhmann, social systems create and reproduce themselves (Luhmann, 1984, 1996, Chapter 1), and are barely controllable from outside. Although social subsystems reduce complexity, they nevertheless create their own internal complexity (Luhmann, 1984, 1996, Chapter 5). This means that once a social subsystem has created itself, its further development is indeterminate and unpredictable. Self-organization due to voluntary cooperation – known as “catallaxy” in the terminology of Hayek (1976) – is the key notion for such systems and subsystems. Not only is the official economy a “self-organizing system” (Krugman, 1996), but also the shadow economy, including drug trafficking, human trafficking, and so on.

A case in point is tax evasion. In a Networked Agent-Based Compliance Model, Korobow et al. (2007) demonstrated the network effects of neighbors at the aggregate level of tax compliance. Given a certain tax law enforcement strategy, the aggregate level of tax compliance was higher when agents knew little of their neighbor's pay-off, and vice versa.

An important characteristic of self-organizing systems is their level of adaptivity (Heylighen, 2001), that is, their ability to adapt themselves to (adverse) changes in the environment; for more on the origin of the adaptation principle in evolutionary biology and psychology, see Buss et al. (1998). The level of adaptivity is a central issue in combating illegal activities. For instance, highly adaptable crime networks (systems) are nearly impossible to eradicate or even to restrict. Moreover, the adaptivity of self-organizing systems is also crucial for the study of such systems. What research tools are required and which are available? What can(not) be said of these systems?

It seems that the network structure of clandestine activities – the “dark internet” – is a case in point Bartlett (2014); its self-organizing character, adaptivity, and complexity make it very difficult to analyze and even more difficult to counter.

A more detailed economic analysis of clandestine activities may shed some light on the question of how to detect it. The above-mentioned network structure and self-organizing property of clandestine activities is nevertheless governed by economic incentives; moreover, it has an economic production structure (Figure 2.2). The “black box” of clandestine (economic) activities to a large extent resembles legal and observable economic activities. This means that an input–throughput–output process is at work. Moreover, even the activity of individual firms in open markets can be described and analyzed economically without looking directly into the “black box” of internal production processes. Firms can be defined as legal structures for organizing cooperative production processes, as famously stated by Fama (Fama, 1980, p. 290): “The firm is just the set of contracts covering the way inputs are joined to create outputs and the way receipts from outputs are shared among inputs.” For instance, observing inputs and outputs is sufficient to determine a firm's productivity and efficiency. In the context of clandestine activities, irrespective of whether they are carried out at the individual or the firm level, observing inputs and outputs may also reveal much about the “black box” of clandestine activities.

First of all, at least some output of clandestine activities will always be observable. For instance, crime victims will press charges, drugs are discovered and addictions become evident, the shadow economy leaves its marks in the demand for cash transactions, and so on. Additionally, incomes are observable to a certain degree if they are consumed and money laundering can in principle be detected. Although there are also many strategies to cover the tracks (i.e., by combining illegal activities with legal ones), some part of the outcomes of clandestine activities are observable. If the output is detected, inputs may also be tracked to a certain extent. In this way, observations of outputs and inputs can be employed to measure clandestine activities. If they are measurable, it is possible to “guess” the business model, as well as the network structure of the business. As a case in point, the connection of input and output is applied to measure the size of the shadow economy; the electricity consumption of an economy can be used to estimate the output of unofficial production, given the output of official firms; see Schneider and Enste (2000), on different methods for estimating the size of the shadow economy.

Measuring clandestine activities and re-engineering the network of clandestine activities may not be sufficient to counter it effectively. As indicated above, clandestine activities are self-organizing systems that are (more or less) adaptable to new circumstances as, for instance, police raids. Highly adaptable clandestine activities will reorganize their business model; they may even completely change their network structure and survive the raids. The economic nature of clandestine activities, their illegality, and the pressure of police raid will have an impact. However, whether these activities can be fought effectively is an open question since the development of self-organizing systems is indeterminate. Adaptivity might prove to be the crucial question in this respect. As suggested by the so-called Red Queen Model (Van Valen, 1973, pp. 17 ff.) of evolution (see Roberts and Newman, 1996, for comments, literature and “a model of evolution and extinction”), permanent adaptation to a changing environment is crucial for survival, whatever the species (Carroll, 1872/1993): “Now here, you see, it takes all the running you can do, to keep in the same place”; (Van Valen, 1973, p. 25, endnote 32). This applies to all kinds of legal as well as illegal behavior. In terms of evolutionary game theory, behavior that provides a pay-off at least slightly higher than the average will survive and even expand. To reduce or to eradicate such behavior, its relative pay-off must become (much) lower than the average (Van Valen, 1973).

2.5 Layers of Analysis

The analysis of real-world phenomena requires a number of layers, simply because these phenomena are also layered and have different dimensions. It does not matter much in this respect whether the phenomenon is legal or illegal. The only difference in illegal and clandestine activities is the additional dimension of obscurity and opacity. This should be borne in mind when looking at the levels of activity and analysis depicted in Figure 2.3.

Starting at the individual level, the next layer is a group of individuals who share a common interest and activity. Tax evasion and other (economic) crimes may be committed by individuals (who are themselves usually included in informal and formal networks that are not necessarily criminal), but also by groups that can be characterized as informal firms in the sense of Fama (1980). In the latter case, crimes are committed by such firms, irrespective of whether they have a formal organizational structure. A case in point is the Mafia with its hierarchical, but informal structure. Another case is a firm that evades a tax; it too has a formal organizational structure. The internal organization is relevant for the criminal activity as well as for the feasibility of police raids and law enforcement. However, it should be clear that the organizational structure may change very quickly under police raids, if it saves the whole business.

Consider again the connections between tax evasion, tax havens, and money laundering. Imagine an individual establishing a dummy firm in a tax haven country with the sole intention of evading the income tax of the home country. Although the intention and the activities are purely individual, they cannot take place without an adequate and conducive environment. First of all, a tax haven country is required that allows for establishing an offshore company and that does not transmit data to the home country. Moreover, successful tax evasion creates additional income; in order to spend it safely in the home country, the illegally earned money must be legalized through laundering. This also requires an infrastructure that provides the necessary services.

More generally, individuals, as well as groups, spread themselves in and over regions. The regional distribution depends, for example, on the level of law enforcement, availability of input resources, and proximity of customers (as in the case of drug trafficking, for instance). Last, but not least, the respective activity may be at a global level, as in the above example of tax evasion. Tax evasion and money laundering on a large scale are global businesses nowadays.

In order to better understand the relationship between theories, experiments, empirical investigations, and simulations, some general remarks seem necessary. It should be emphasized that these remarks are relevant for all of these analytic layers and approaches.

The analytical layer starts with theory. As stated, without a theory, real-world phenomena cannot be investigated scientifically. Descriptive models are required, in order to render a phenomenon measurable, explanatory models combine ideas on the determinants of a certain phenomenon to form a coherent explanation, and predictive models are specifications for forecasting the respective phenomenon. Theories and models are necessary to inform both laboratory and field experiments.

Experiments can be designed according to theories and models; such a design is a precondition for the interpretation of experimental results. As stated by the philosopher Willard Van Orman Quine (Quine, 1951, p. 42), science “ is similarly but more extremely underdetermined by experience”. Experiments do not tell an interpretable story, unless their design is specified on the basis of theoretical considerations. According to the philosopher Nicholas Rescher (Rescher, 2006, p. 82), the main reason is that “Facts are infinite in number. The domain of fact is inexhaustible: there is no limit to facts about the real.”. This means that limiting the meaning of an unlimited number of facts, theories, and models is the prerequisite. Moreover, theories and the models derived from them are also required – for the same reason – to conduct and interpret empirical studies, as well as for running meaningful simulations. Nevertheless, experiments and empirical investigations are indispensable for analysis, because they produce results that enable testing models and even theories, in a kind of feedback-loop. Put differently, without experiments and empirical studies, theories and models would remain in their metaphysical realm; that is, they would not reveal anything about real-world phenomena. In contrast to empirical investigations, experiments allow for a design that makes it feasible to study causal links between certain external changes and human behavior. Although this does not imply that experiments represent the real world in its entirety, in well-designed experiments, certain supposed causal links may be tested.

However, experiments face a scalability limit. Although large-scale laboratory and field experiments can be imagined, their costs are usually unjustifiably high. This is one of the best reasons to use large-scale simulation studies. Their very substantial advantage is that the calculation power of today's computers allow for very large numbers of such simulations at very low cost. Large-scale simulations in fact solve the scalability problem that cannot otherwise be fixed at reasonable and justifiable cost as, for example, by using laboratory or field experiments. Unfortunately, virtual agents in computer simulations differ from human beings, owing to their low degrees of behavioral freedom. However, institutions, norms, and social control of all kinds are among those factors that make human behavior predictable to a certain degree. In addition, “behavioral rules” (Heiner, 1983, p. 561) reduce the flexibility of potential behavior, which then suffers from uncertainty of outcome. After all, human behavior does not seem to be all that unpredictable. It might even be that the network structure of human interactions – in legal and illegal, as well as in observable and clandestine cases – contributes more to the stochasticity of aggregate behavior than underdetermined individual actions. In the latter case, computer simulation can be used in two different ways, firstly, to identify the network contribution to behavioral and outcome fluctuations, and secondly, to forecast aggregate behavior and its outcome.

It should be clear that all layers of analysis, with all available research tools, can contribute quite substantially even to the knowledge on and understanding of clandestine activities. Consider again the example of tax evasion. Theoretical approaches such as the Allingham and Sandmo (1972) and Yitzhaki (1974) model (cf. Chapter 1) try to explain individual tax-evading behavior. The embedding of individual taxpayer behavior in a social environment has also been analyzed theoretically, for instance, by Myles and Naylor (1996), as well as Prinz et al. (2014). Many laboratory experimental studies are available (see Kastlunger et al., 2011), as well as a few field experiments; see Cummings et al. (2009) and Alm (2012). Additionally, a growing number of simulations has been published (see Pickhardt and Prinz, 2014, for a survey), some of them based on econophysics approaches, as, for example, Hokamp and Pickhardt (2010) and Pickhardt and Seibold (2014).

The next question is: How can one apply research tools appropriately to the analysis of clandestine activities? What might reasonably be expected, and what not?

2.6 Research Tools and Clandestine Activities

As indicated above, the available research tools for studying clandestine activities are theories, laboratory and field experiments, empirical investigations, as well as computer simulations. Although all these research tools are indispensable, their potential contribution to the analysis itself is quite different. As a consequence, the question is: what are the specific advantages of the respective method for analyzing clandestine activities?

To start with, some general issues should be discussed. There are several trade-offs in scientific research concerning real-world phenomena. According to Rescher (2006), there is a trade-off between “security and confidence,” denoted by , on the one hand and “definiteness and detail,” denoted by , on the other hand, so that (Rescher, 2006, p. 1). Rescher calls this “Duhem's law of cognitive complementarity,” according to his interpretation of Duhem (1908/1978). According to Rescher, “Duhem's law” implies that science can only produce “secure” results at a rather abstract level of reasoning. The more details that are taken into account, the less “secure” the results. Economically, this means that there is a trade-off between the generality of theoretical results and their applicability to real-world problems. In a similar manner, (Roth, 2002) describes the role of the economist as an engineer with respect to designing market mechanisms for solving real problems, in contrast to merely analyzing markets. In order to design market mechanisms, all available tools must play a role. To solve real-world problems, very specific situations and variable constellations have to be accounted for. In contrast, a theoretical analysis attempts to find a few general variables that drive behavior and activities. As Ariel Rubinstein put it: “ economic theory is an abstract investigation of the concepts and considerations involved in real life economic decision-making rather than a tool for predicting or describing real behavior” (Rubinstein, 2001, Abstract). Hence, the role of theory in policy making is to provide an intuitive understanding of the problem and its feasible solution (McAfee and McMillan, 1996, p. 172).

How can one apply research methods adequately to the analysis of (clandestine) activities? There are at least two fallacies in this respect, namely, the “fallacy of misplaced concreteness” (Whitehead, 1925/1967, p. 51) and the “fallacy of disregarded abstractness” (Schramm, 2015, p. 29). Obviously, “science” has other goals than “policy,” so that it should not come as a surprise that one gets into trouble if the one is mistaken for the other (Figure 2.4).

As “science” and “abstraction” are very closely related to each other, so too are “policy” (i.e., the intervention in real-world affairs) and “concreteness.” As a consequence, general scientific results cannot be expected to be directly applicable. To try and do so may result in Schramm's “fallacy of disregarded abstractness.” By contrast, attempting to make science specific may lead to Whitehead's “fallacy of misplaced concreteness.” These fallacies are all the more important, as there is no direct and immediate outcome of “disregarded abstractness” that could demonstrate the fallacy in the case of clandestine activities. Although theories and models are indispensable as analytical research tools, their abstractness renders them poorly suited for a direct application to policy issues of clandestine activities. For instance, countering tax evasion with models could imply an assumption that penalties and audits are perfect substitutes for each other. In practice, they are not. In addition, quantifying detection probabilities and penalties does not seem possible with theoretical models alone.

So far, only unspecific remarks on the applications of research methods have been provided in this section. In the next step, the peculiarities of clandestine activities are considered with respect to research methods. Put differently, the applicability and usefulness of the methods under the specific conditions of secrecy and complexity are accounted for. Secrecy is a crucial factor, since it only rarely allows for a direct test of hypotheses on “dark” behavior. Moreover, the multilayer network structure of secret activities makes them very complex. Hence, a useful application of research methods to clandestine activities must account for the complexity of concreteness, as demonstrated in Figure 2.5. According to this figure, theories (and models) of clandestine activities must provide concepts for their measurement. Moreover, the models must also determine the most important determinants of these activities, as well as concepts concerning the interactions of individuals, groups, and other individuals in these respects. Alternatively expressed, in addition to a static theory of the determinants of (legal and illegal) clandestine activities, theories and models for the dynamics and their network structures are required. For instance, Prinz et al. (2014) provide just such a dynamic model for tax compliance with two population groups. One group is compliance minded and the other evasion minded; that is, the latter group will evade taxes, whenever possible and profitable. The compliance-minded group will not evade taxes, unless the gain from (undetected and unpunished) tax evasion is so high that even compliance-minded people become evasion minded. An interesting result of this dynamic model is that the punishment over and above a certain level may increase rather than decrease tax evasion.

Recently, Perc et al. (2013) studied the evolutionary dynamic development of crime with the so-called “inspection game” – see, for instance, Avenhaus et al. (2002) and Avenhaus (2004) – that is, a game between criminals, inspectors, and ordinary people, using a spatial variant. Based on simulations, they find a cyclical domination of “criminals,” “inspectors,” and “ordinary people” in the case of low or moderate inspection costs. They conclude that crime may be evolutionarily recurrent and that it requires “very much counter-intuitive and complex” strategies to contain such crime (Perc et al., 2013). This result is generated by applying evolutionary game theory and computer simulation, which shows how to study the evolutionary dynamics of clandestine (in this case criminal) behavior. Although the direct usability for law enforcement and crime containing is rather limited, the study is nonetheless valuable for the insights it provides.

As explained above, not only theoretical research, but all kinds of experiments as well as computer simulations may prove very helpful, even on the theoretical level; see Rubinstein (2001), on studying the relationship between theory and experiments in economics, and Card et al. (2011), on the use of theory in field experiments.

One of the main difficulties for policy-oriented research is, however, the existence of a (probably) high number of hidden variables that interact with the known variables and structures in the real-world of these activities. Hence, without further empirical observations and data, the theoretical analysis is of very limited use in finding effective policies.

Duijn et al. (2014) analyze raids on criminal networks (organized cannabis cultivation) based on network and network resilience theory; additionally, they apply data sets from police investigations to test the efficacy of network disruption raids. Although network resilience theory describes potentially useful approaches – see, for example, Sparrow (1991), Klerks (2001), Prinz (2005), and Schwartz and Rouselle (2009) – the study found that police interventions tended to increase the efficiency and efficacy of the criminal network. Obviously, such a criminal network demonstrates a very high degree of raid resilience, whereby policy attacks even increase its efficiency. From an evolutionary point of view, the result is not so surprising after all. This study is important though, as it shows that the results of theoretical and simulation studies cannot be converted immediately into successful application. Moreover, it also demonstrates the importance of collecting and analyzing empirical data. This brings us back to the secrecy problem of “dark” (economic) behavior. How can data collection on this clandestine behavior and its results be improved and extended?

Agresten et al. (2015) use data sets on telephone contacts among suspects and data on the relationships between individuals involved in criminal offences perpetrated by the Sicilian Mafia. They emphasize that the very high resilience of criminal networks to raids in fact seems to be vulnerable through their phone contacts. Moreover, Tseng et al. (2012) describe text-mining methods, in combination with network analysis methods, to detect criminal structures and networks. In this way, social media data may be combined with network analysis to detect criminal networks and activities. Even hidden ties in criminal networks may be detected via network theory, when relevant data sets are available (Isah et al., 2015). Consequently, social media and telecommunication activities provide new and very large data sets that may be used to detect and to measure clandestine activities. Nevertheless, there are two limitations in this respect. Firstly, there are legal and ethical restrictions that protect the privacy of these activities. Secondly, the availability of huge data sets requires concepts and algorithms to decide what to look for. Although such concepts and algorithms are available, the risk of error is high; see Lazer et al. (2014), for errors in flu prediction on the basis of Google data.

Another data source is that of solved criminal cases. Tax evasion, money laundering, and so on are detected to a certain extent. These cases automatically produce large data sets through police, court, and administrative activities. As shown by Duijn et al. (2014), such data sets can be used successfully for policy-oriented studies. They reveal some secret activities, and this knowledge can be employed to calibrate network and simulation models.

As a research tool, computer simulations offer a wide range of applications for studying clandestine activities. Although the external validity of such simulation cannot be determined without reliable empirical data (see Prinz, 2016, for a detailed analysis of the usability of simulations in tax compliance research), they nevertheless enhance the potential for policy-oriented research. Whether they can be reasonably employed for forecasting the effects of anti-crime policies, for instance, depends on the calibration of the simulation models. However, calibration itself is restricted by the availability of sound empirical data. As pointed out by Dirk Helbing, “ computer simulation can be seen as an experimental technique for hypothesis testing and scenario analysis which can be used complementarily and in combination with experiments in real-life, the lab or the Web” (Helbing, 2012, p. 25). A further step in the development of dynamical and interactive modeling for social and economic processes, whether observable or clandestine, legal or illegal, is “experimental econophysics,” as described in detail by Huang (2015). In this approach, laboratory experiments with humans are combined with agent-based simulations in order to study the emergent properties of real-world phenomena (Huang, 2015). Self-organization, as well as complexity, plays a crucial role in this method.

A method that combines the behavioral approach to tax compliance with a social network model and agent-based modeling, called “predictive analytics,” has been introduced by Hashimzade et al. (2016), in order to investigate tax evasion. Taxpayers form subjective beliefs on tax audit probability in their social interactions with fellow taxpayers and choose their employment (i.e., as an employee or in the form of self-employment) according to their level of risk aversion. The tax authority chooses a nonrandom audit strategy that has an impact on subjectively perceived audit probabilities. In this way, taxpayers are self-selected into groups with different attitudes to tax evasion. This can be used by the tax authority to raise revenue.

Experimental econophysics and predictive analytics are examples of a recent class of hybrid research methods. These methods combine several analytical and computational strategies to develop more specific real-world models and obtain the associated results. Although they are “only” models, their capacity to analyze real-world phenomena in a much more complex social, but still controllable, environment is larger than with single “pure” methods. However, hybrid models too come with a “price tag”; it is no longer possible to say exactly which effect was caused by what. This is less satisfactory from a scientific point of view, but is much more relevant to and suitable for the characteristics of self-organizing systems.

A bundle of further research methods can be found in artificial intelligence. In this respect, machine learning (see Barber, 2012, Chapter 13 ff.) may play a major role. If a huge database is available, computers can be trained to analyze the data thoroughly and automatically. Of course, models are required to train machines in such a manner as to detect the respective structures of clandestine activities. However, it is an open question whether and when machine learning tools reach a level of perfection so that the results are valid and reliable.

Obviously, each research method has its merits and demerits, particularly in the study of clandestine activities. In Table 2.1, an overview of research methods is provided concerning the areas of application with respect to clandestine activities, their advantages, and their shortcomings.

First of all, the methods mentioned in this chapter may all contribute substantially to the study of clandestine activities. The external validity of research (i.e., the meaning of the research results for the respective real-world problem) is not so easy to establish. However, the availability of sound empirical data may improve the situation considerably. Nonetheless, until recently the direct applicability of research results for law enforcement, for instance, seemed rather restricted. The availability of social media and telecommunication data, when combined with other methods, may improve the situation drastically. But even then, it should be borne in mind that the evolutionary stability and robustness of crime and illicit activities may limit the success of enforcement policies.

Having said this, in the second column of Table 2.1, the areas of application for the respective research method are indicated. More or less all kinds of clandestine activities may ultimately be accessible for the various research methods. In the third column, the advantages of the respective research method are briefly described. Theories, models, and laboratory experiments can be put into one class, as they study clandestine behavior in a mainly scientific way, which means that an immediate application of their results may neither be intended nor very successful. The remaining methods, although based on results from the first class of methods, seem to be more appropriate for policy research and application. This is particularly true for hybrid methods. The fourth and last column points to the main shortcomings of the respective method. For the first class of methods, unknown external validity is the most serious problem. The other methods may solve this problem to the extent that valid and reliable data in sufficient amounts is available. Moreover, although social media and telecommunication data may alleviate the data availability problem, they create their own problems, since the quality of the data (validity and reliability) is not as good as expected and the algorithms of data analysis may lead to serious errors. Nevertheless, their advantage is that the data is based on authentic behavior.

Last, but not least, the procedure of scientific studies concerning clandestine activities can be summarized as follows (see Gale and Slemrod, 2001, in the context of the US debate on the estate tax). Theoretical analysis and investigation, in combination with empirical facts, should be used in simulation and hybrid models to obtain at least informed guesses on the mechanics of and policies on clandestine behavior.

2.7 Conclusion

This chapter contains an analysis of how clandestine activities such as crimes, tax fraud, and other “dark” activities can be studied scientifically. The goals of such scientific analysis have been identified as description, explanation, and prediction of the relevant behavior. Moreover, the usefulness of scientific studies for policy interventions is also considered.

It is argued that even the measurement of clandestine activities requires a theory and probably more than one model (in the sense of Mario Bunge's theory of scientific research). That is, without any theory, a scientific study of clandestine behavior – and all other behavior too – is not feasible. However, descriptive models alone are insufficient, as they do not provide rationales for the respective activities; hence, explicative theories and models are required. Such theories, as well as models, must of necessity be parsimonious to be of value. This implies that they are not directly applicable for policy analysis. Moreover, in order to forecast real-world phenomena, predictive models are necessary that encompass all kinds of variables and many specifications that are omitted from explicative theories and models. Explanation requires a high degree of generality, and prediction presupposes a large number of additional variables and facts. As a consequence, there is a generality – specificity gap between explanation and prediction. Policy analysis is all about specificity and, therefore, different and much more specific approaches are essential. The consequence is two potential pitfalls, namely the “fallacy of misplaced concreteness” (Whitehead) for science, and the “fallacy of disregarded abstractness” (Schramm) for policy.

For policy research and evaluations, empirical data are indispensable. This is particularly true for clandestine activities. Recently, social media and telecommunication sources have become progressively more available. This could lead to a major leap in our knowledge about clandestine activities, from drug and human trafficking to tax evasion and other crimes. Although there are ethical concerns involved, communication and trading activities leave traces that can be used even without identifying individuals or firms. This might reduce ethical reservations in using these data empirically. Nevertheless, the validity and reliability of data, as well as of the analytical algorithms employed for their analysis, are far from faultless.

Finally, it should be emphasized that simulations may play a crucial role in theoretical studies and, even more importantly, in policy studies. The reason is that simulations can differentiate between agents as well as between social networks and their interactions. Hence, simulations will be much more complicated than models. Based on theories and models, simulations are theoretical research tools; calibrated with empirical data, they become policy research instruments. The more data on clandestine activities is available, the more attractive and useful simulations will become as policy research tools.

Acknowledgment

I would like to thank two anonymous reviewers for helpful comments on an earlier version of the chapter and Brian Bloch for extensive text editing. However, all remaining errors are mine.

References

1. Agreste, S., Catanese, S., De Meo, P., Ferrara, E., and Fiumara, G. (2015) Network structure and resilience of Mafia syndicates. arXiv: 1509.01608v1 [cs.SI] 4 September 2015.
2. Allingham, M.G. and Sandmo, A. (1972) Income tax evasion: a theoretical analysis. Journal of Public Economics, 1, 323–338.
3. Alm, J. (2012) Measuring, explaining, and controlling tax evasion: lessons from theory, experiments, and field studies. International Tax and Public Finance, 19 (1), 54–77.
4. Alm, J. and Jacobson, S. (2007) Using laboratory experiments in public economics. National Tax Journal, 60, 129–152.
5. Avenhaus, R. (2004) Applications of inspection games. Mathematical Modelling and Analysis, 9 (3), 179–192.
6. Avenhaus, R., Von Stengel, B., and Zamir, S. (2002) Inspection games, in Handbook of Game Theory with Economic Applications, vol. 3 (eds R.A. Aumann and S. Hart), North Holland, pp. 1947–1987.
7. Barber, D. (2012) Bayesian Reasoning and Machine Learning, Cambridge University Press.
8. Bartlett, J. (2014) The Dark Net. Inside the Digital Underworld, William Heinemann.
9. Bunge, M. (1967a) Scientific Research I: The Search for System, Springer-Verlag, Berlin, Heidelberg, New York.
10. Bunge, M. (1967b) Scientific Research II: The Search for Truth, Springer-Verlag, Berlin, Heidelberg, New York.
11. Buss, D.M., Haselton, M.G., Shackelford, T.K., Bleske, A.L., and Wakefield, J.C. (1998) Adaptations, exaptations, and spandrels. American Psychologist, 53 (5), 533–548.
12. Card, D., DellaVigna, S., and Malmendier, U. (2011) The role of theory in field experiments. Journal of Economic Perspectives, 25 (3), 39–62.
13. Carroll, L. (1872/1993) Through the Looking-Glass: And What Alice Found There, HarperCollins, New York.
14. Cummings, R.G., Martinez-Vazquez, J., McKee, M., and Torgler, B. (2009) Tax morale affects tax compliance: evidence from surveys and an artefactual field experiment. Journal of Economic Behavior & Organization, 70, 447–457.
15. Duhem, P. (1908/1978) Ziel und Struktur der physikalischen Theorien [La théorie physique, son objet et sa structure], Felix Meiner Verlag, Hamburg.
16. Duijn, P.A.C., Kashirin, V., and Sloot, P.M.A. (2014) The relative ineffectiveness of criminal network disruption. Nature Scientific Reports, 4 (4238), 1–15.
17. Fama, E.F. (1980) Agency problems and the theory of the firm. Journal of Political Economy, 88 (2), 288–307.
18. Feld, L.P. and Larsen, C. (2012) Self-perceptions, government policies and tax compliance in Germany. International Tax and Public Finance, 19 (1), 78–103.
19. Gale, W.G. and Slemrod, J. (2001) Rhetoric and economics in the estate tax debate. National Tax Journal, 54 (3), 613–627.
20. Hashimzade, N., Myles, G.D., and Rablen, M.D. (2016) Predictive analytics and the targeting of audits. Journal of Economic Behavior & Organization, 124, 130–145.
21. Hayek, F.A. (1976) Law, Legislation and Liberty: The Mirage of Social Justice, vol. 2, University of Chicago Press.
22. Heiner, R.A. (1983) The origin of predictable behavior. American Economic Review, 73 (4), 560–595.
23. Helbing, D. (2012) Social Self-Organization. Understanding Complex Systems, Springer-Verlag, Berlin, Heidelberg.
24. Heylighen, F. (2001) The science of self-organization and adaptivity. The Encyclopedia of Life Support Systems, 5 (3), 253–280.
25. Hokamp, S. and Pickhardt, M. (2010) Income tax evasion in a society of heterogeneous agents. Evidence from an agent-based model. International Economic Journal, 24, 541–553.
26. Huang, J.P. (2015) Experimental econophysics: complexity, self-organization, and emergent properties. Physics Reports, 564, 1–55.
27. Isah, H., Neagu, D., and Trundle, P. (2015) Bipartite network model for inferring hidden ties in crime data. arXiv: 1510.02343v1 [cs.SI].
28. Kastlunger, B., Muehlbacher, S., Kirchler, E., and Mittone, L. (2011) What goes around comes around? Experimental evidence of the effects of rewards on tax compliance. Public Finance Review, 39, 150–167.
29. Klerks, P. (2001) The network paradigm applied to criminal organizations: theoretical nitpicking or a relevant doctrine for investigators? Recent developments in the Netherlands. Connections, 24 (3), 53–65.
30. Kleven, H.J., Knudsen, M.B., Kreiner, C.T., Pedersen, S., and Saez, E. (2011) Unwilling or unable to cheat? Evidence from a tax audit experiment in Denmark. Econometrica, 79, 651–692.
31. Korobow, A., Johnson, C., and Axtell, R. (2007) An agent-based model of tax compliance with social networks. National Tax Journal, 60 (3), 589–610.
32. Krugman, P. (1996) The Self Organizing Economy, Wiley-Blackwell, Cambridge, MA and Oxford.
33. Lazer, D., Kennedy, R., King, G., and Vespignani, A. (2014) The parable of Google Flu: traps in big data analysis. Science, 343, 1203–1205.
34. Leijonhufvud, A. (1997) Models and theories. Journal of Economic Methodology, 4 (2), 193–198.
35. Levitt, S.D. and List, J.A. (2007) Viewpoint: on the generalizability of lab behaviour to the field. Canadian Journal of Economics, 40 (2), 347–370.
36. List, J.A. (2006) The behavioralist meets the market: measuring social preferences and reputation effects in actual transactions. Journal of Political Economy, 114 (1), 1–37.
37. List, J.A. (2008) Informed consent in social science. Science, 322 (5902), 672.
38. Luhmann, N. (1984) Soziale Systeme. Grundriss einer allgemeinen Theorie, Suhrkamp, Frankfurt am Main.
39. Luhmann, N. (1996) Social Systems (translated by John Bednarz), Stanford University Press.
40. McAfee, R.P. and McMillan, J. (1996) Analyzing the airwaves auction. Journal of Economic Perspectives, 10 (1), 159–175.
41. Myles, G.D. and Naylor, R.A. (1996) A model of tax evasion with group conformity and social customs. European Journal of Political Economy, 12, 49–66.
42. Perc, M., Donnay, K., and Helbing, D. (2013) Understanding recurrent crime as system-immanent collective behavior. PLoS ONE, 8 (10), e76063.
43. Pickhardt, M. and Prinz, A. (2014) Behavioral dynamics of tax evasion. Journal of Economic Psychology, 40 (1), 1–19.
44. Pickhardt, M. and Seibold, G. (2014) Income tax evasion dynamics: evidence from an agent-based econophysics model. Journal of Economic Psychology, 40 (1), 147–160.
45. Prinz, A. (2005) The network structure of illegal firms, in Entscheidungsorientierte Volkswirtschaftslehre (eds M. Göcke and S. Kooths), Peter Lang, Frankfurt am Main, pp. 465–480.
46. Prinz, A. (2016) Simulations in tax compliance research: what are they good for? Review of Behavioral Economics, 3 (1), 20–46.
47. Prinz, A., Muehlbacher, S., and Kirchler, E. (2014) The slippery slope framework on tax compliance: an attempt to formalization. Journal of Economic Psychology, 40 (1), 20–34.
48. Quine, W.V. (1951) Main trends in recent philosophy: two dogmas of empiricism. Philosophical Review, 60 (1), 20–43.
49. Rescher, N. (2006) Epistemetrics. Chapter 1: Asking for More Than Truth: Duhem's Law of Cognitive Complementarity, pp. 1–7;Chapter 3: Spencer's Law of Cognitive Development, pp. 15–28;Chapter 7: How Much Can Be Known? pp. 73–94, Cambridge University Press.
50. Roberts, B.W. and Newman, M.E.J. (1996) A model for evolution and extinction. Journal of Theoretical Biology, 180, 39–54.
51. Roth, A. (2002) The economist as engineer: game theory, experimentation, and computation as tools for design economics. Econometrica, 70 (4), 1341–1378.
52. Rubinstein, A. (2001) A theorist's view of experiments. European Economic Review, 45, 615–628.
53. Rubinstein, A. (2006) Dilemmas of an economic theorist. Econometrica, 74 (4), 865–883.
54. Rudder, C. (2014) Dataclysm: Who We are (When We Think No One's Looking), Crown Publishing, New York.
55. Schneider, F. and Enste, D. (2000) Shadow economies: size, causes, and consequences. Journal of Economic Literature, 38 (1), 77–114.
56. Schramm, M. (2015) How the (Business) World Works. Business Metaphysics & “Creating Shared Value”. Annual Meeting of the Working Group on “‘Wirtschaftsphilosophie und Ethik”’ 2015: “‘Creating Shared Value -Concepts, Experience, Criticism”’, Zeppelin University (Friedrichshafen).
57. Schwartz, D.M. and Rouselle, T.D.A. (2009) Using social network analysis to target criminal networks. Trends in Organised Crime, 12, 188–207.
58. Slemrod, J. and Weber, C. (2012) Evidence of the invisible: toward a credibility revolution in the empirical analysis of tax evasion and the informal economy. International Tax and Public Finance, 19 (1), 25–53.
59. Sparrow, M.K. (1991) The application of network analysis to criminal intelligence: an assessment of the prospects. Social Networks, 13, 251–274.
60. Tavares, R. (2013) Relationship Between Money Laundering, Tax Evasion and Tax Havens. Special Committee on Organized Crime, Corruption and Money Laundering (CRIM) 2012–2013, European Parliament.
61. Thomas, R. (2006) Circular causality. IEE Proceedings - Systems Biology, 153 (4), 140–153.
62. Tseng, Y.-H., Ho, Z.-P., Yang, K.-S., and Chen, C.-C. (2012) Mining term networks from text collections for crime investigations. Expert Systems with Applications, 39, 10082–10090.
63. Van Valen, L. (1973) A new evolutionary law. Evolutionary Theory, 1 (1), 1–30.
64. Whitehead, A.N. (1925/1967) Science and the Modern World, The Free Press, New York.
65. Yitzhaki, S. (1974) A note on income tax evasion: a theoretical analysis. Journal of Public Economics, 3, 201–202.