An engineering perspective for policy design: self-organizing crime as an evolutionary social system

Abstract

Here we introduce methodological guidelines for designing policies against organized crime. We employ the evolutionary ontology proposed by Kurt Dopfer for conceiving organized crime as the outcome of social, intelligent agents whose strategies evolve through time. To illustrate the use of this ontology we explore the case of corruption in public procurement processes in Colombia in which criminal organizations—groups of corrupt agents—converge spontaneously. The ontology leads to conceive corruption as a knowledge process that adapts according to the evolution of problem-solving rules that are created, used and discarded by agents that seek to attain personal gains by means of public resources. We also use an engineering perspective that favors model-aided design. We built a simulation model that illustrates how the dynamics of such evolving-rules systems can be conceptualized for exploring potential policies. The application of the evolutionary ontology shows why corruption exhibits self-organization: system-level patterns develop from spontaneous interactions that use only local information. Rule dynamics form a changing structure of rule-populations that adapt to novel environmental conditions and generate meta-stable adaptions that explain why corruption persists despite continuous challenges from the environment. This engineering approach forms the ground for proposing policies that instead of addressing the operant level of a social system (according to observed operations and data), should meet the dynamics of rules that govern those operations. Hence, the role of regulators shifts from “controllers” to inventors of selectionist environments that facilitate suitable change through the introduction or promotion of counter-crime rules, the design of selective pressures that favor the evolution of desirable rules and the attention to coordination gaps at the macro-structure. The recognition of organized crime as the outcome of an evolving-rules system changes the questions that orient policy-making and focuses on the redesign of evolving knowledge. Accordingly, our methodological guidelines address such evolutionary dynamics and can be applied to several forms of organized crime.

Annex 1 – model parameters and tests

Table 9 Description of key parameters of the model

Sensitivity analysis

Sensitivity analysis asks whether results change in ways important to the purpose as assumptions and parameters change; this type of analysis should focus on examining those relationships and parameters that are uncertain and likely to be influential (Sterman 2000). Windrum et al. (2007) also suggest that sensitivity analysis should explore how results depend on i) micro and macro parameters, ii) initial conditions, and iii) across run variability induced by stochastic elements. The results from these tests can be categorized in: a) numerical sensitivity, b) behavior mode sensitivity, and c) policy sensitivity (Sterman, 2000). Based on this categorization, we performed sensitivity tests with parameters upon which we are uncertain of their estimation; these are: corruption threshold, denounce rate of contractors and minimum accumulated allegations by citizens required for the local government to present allegations to controllers. We varied these parameters and focused on the behavior rule #4 use since it is the corrupt rule that entails higher complexity (association among groups of corrupt contractor and groups of corrupt public officers). In the following table we brief the specifications and conclusions of these tests; we also present the simulation results of the tests in the subsequent graphs. Because there were no observed changes in behavioral patterns, the uncertainty of the estimation for these parameters is not of concern in terms of the conclusions that can derive from this model, particularly considering that the outcomes of this model are merely illustrative and aim to shed light towards public policy broad guidelines.

Table 10 Sensitivity analysis - test description and results

Figure 5 shows that use of rule #4 has numeric sensitivity to variations in propensity for corruption parameters. However, the behavior mode through the simulation does not change. For lower values of propensity for corruption, the use of rule #4 decreases with regard to those simulations with higher levels of propensity for corruption.

Fig. 5

Sensitivity Analysis: Variations in Corruption Threshold

Figure 6 shows that the use of rule #4 decreases when the denounce rate increases and vice versa. However there is no change in the behavior mode throughout the simulations.

Fig. 6

Sensitivity Analysis: Variations in Denounce Rate of Contractors

In the model, allegations from local governments are a response to an accumulation of citizen allegations. Figure 7 shows that the use of rule #4 does not seem to change when this parameter changes. Significant increases in the use of rule #4 are observed at the last part of the simulation when this parameter has values of 700 and 750, however there is no significant change in the behavior mode.

Fig. 7

Sensitivity Analysis: Variations in Minimum Accumulated Allegations by Citizens Required for Local Governments to Make Allegations to Controllers

Extreme conditions

With the Extreme Conditions test, we examined if the model behaves in a realistic fashion when extreme inputs are explored (Sterman, 2000) such as no propensity for corruption, total propensity for corruption, no public officers involved in procurement processes, or large numbers of public officers involved in procurement processes. The description and results of this test are summarized in Table 10; we also show the simulation results of the tests in the subsequent graphs. From the results we can conclude that the rule dynamics of the model is consistent with what is expected when we assume extreme values for critical parameters.

Table 11 Extreme conditions analysis - test description and results

Figure 8 shows three simulations: 1. No corruption (Corruption Threshold = 100), 2. Normal Corruption Threshold (Corruption Threshold = 36) and 3. Absolut corruption (Corruption Threshold = 0). The simulations show that in the case of No corruption, rule #4 is never used while in the case of Absolut corruption the utilization of rule#4 is largely increased with respect to the Normal Corruption Threshold used in the model. These results show that the model responds accordingly to extreme conditions.

Fig. 8

Extreme Conditions Analysis: Use of Rule #4 under Extreme Conditions of Corruption Threshold

Figure 9 shows three simulations: 1. No public officers in process (Public officers = 0), 2. Normal number of Public Officers in process (Public officers = 3, value used in the model), and 3. Large number of public officers in process (Public officers = 9). For the first case, No public officers in process, rule#4 is not used because there are no public officers with whom to make corrupt alliances with. Conversely, in the case of Large numbers of public officers in the process, the chance of making a corrupt alliance with several public officers is higher since there are more public officers; consequently, the use of rule#4 increases significantly. These results are coherent with the expected behavior.

Fig. 9

Extreme Conditions Analysis: Use of Rule #4 under Extreme Conditions of the Number of Public Officers

Further analysis could be extended in terms of how the model responds to extreme conditions. Analysis could focus not only in the dynamics of the use of rule#4 but as well in the use of the other three rules. This test, as well as the sensitivity analysis serve as illustrative examples of how this type of models could be tested with respect to its purpose from an engineering perspective.