The Use of Actor-Level Attributes and Centrality Measures to Identify Key Actors

Abstract

Social network analysis (SNA) conducted on criminal networks can identify key players and shed light on important patterns of connectivity. This information can be used to develop interventions to dismantle or disrupt criminal networks. Drawing upon the network capital construct, this study demonstrates that integrating centrality measures (such as degree or betweenness centrality) with other individual attributes related to functional roles and access to tangible and intangible resources will enhance efforts to identify critical actors. Using a drug trafficking network that operated in Australia in the 1990s, we identify actors who are key to the network by virtue of their position in the network, their attributes, and combinations of these factors. Implications for law enforcement practice are discussed.

Keywords

drug trafficking centrality of actors roles social network analysis network capital

Introduction

Much of the previous research using social network analysis (SNA) to study criminal networks has focused on identifying “key actors” in networks, usually based on the analysis of centrality scores¹ (e.g., Baker & Faulkner, 1993; Krebs, 2002; Morselli, 2009a; Xu & Chen, 2009). Key actors may be important for reasons unrelated to centrality. Robins (2009) argues that the sole use of SNA centrality metrics leads to the omission of important information about actor-level factors such as demographic characteristics, capacities such as skills and expertise, and psychological characteristics such as personality traits. Applying this argument to criminal networks, greater utility may be extracted from SNA when actor attributes are considered (see, for example, Calderoni, 2014). Actors may be more critical if they possess key resources required for the crime commission process (e.g., precursor chemicals, bomb-making skills, knowledge of law enforcement strategies). The present article extends previous research by examining the relationship between centrality and actor attributes, and by identifying “key” actors in a criminal network using both centrality scores and actor attributes (i.e., possession of particular resources).

Actor Attributes in Criminal Networks

Previous studies examining trafficking networks have investigated whether actor attributes offer strategic direction for law enforcement (e.g., Bright, Hughes, & Chalmers, 2012; Bright, Greenhill, & Levenkova, 2014; Calderoni, 2014; Malm & Bichler, 2011; Morselli & Roy, 2008; Natarajan, 2000, 2006). Each of these studies used the activity that an actor engages in to generate a functional or role-based attribute. However, few of these studies evaluated whether targeting key actors would disrupt trafficking.

Natarajan (2000) used task analysis to identify three major roles in a cocaine trafficking group: (a) bosses, (b) assistant managers, and (c) field workers. Although the network was described using both SNA metrics (centrality) and the roles performed by actors, there was no attempt to evaluate the relative “importance” of centrality scores compared with actor attributes. Bright et al. (2012) also classified actors in a methamphetamine manufacture and trafficking network into seven roles: managers, clandestine laboratory managers, wholesale level dealers, resource providers, specialists, workers, and corrupt officials. Although the roles were used to describe the network and to identify key actors, no formal modeling was conducted to determine the relative “value” of centrality scores compared with roles. Similarly, Malm and Bichler (2011) examined roles (or niches) in drug trafficking (e.g., production, transport). Some actors were involved in only one niche, whereas others were involved in two related niches (e.g., smuggling and retail). Degree centrality scores were found to be highest for those involved in complex transport and complex supply, whereas betweenness centrality scores were highest for actors in complex transport and supply niches and in financial niches. Results indicated that examining network positioning alongside information about roles or niches may facilitate a more nuanced understanding and more specific interventions against particular market niches. Calderoni (2014) examined actor status and found that “specific activities within the drug trafficking chain account for the position of individuals within the network” (p. 170). Traffickers, member of the ’Ndrangheta and bosses had higher degree and betweenness centrality scores than other actors. Furthermore, actors with high status were less centrally positioned, whereas actors with moderate status were more central and therefore more vulnerable to law enforcement.

One way to evaluate the relative importance of actors is to examine the impact of actor removal on the network. Natarajan (2006) categorized actors in a heroin trafficking group into four main roles: sellers (wholesale), retailers, brokers, and secretaries. Natarajan tested the impact of arrest by simulating the removal of high centrality actors. Removal of actors with high centrality scores did not lead to network fragmentation. However, the study did not examine the impact of the removal of actors who played particular roles, and no formal measurement of network fragmentation was used. In a subsequent study, Bright et al. (2014) examined the impact of four law enforcement interventions aimed at dismantling and disrupting the network: (a) removal of actors with the highest degree centrality, (b) removal of actors who play the most important roles, (c) removal of actors based on a combination of centrality scores and role, and (d) random removal of actors. The most effective strategy was to remove nodes based on degree centrality scores or to use a mix of degree centrality and role weighting. Removing nodes based on role was found to be relatively ineffective. Morselli and Roy (2008) examined the importance of particular roles within a crime commission process (stolen vehicle exportation), in concert with the importance of brokers. They found that the impact of the removal of brokers from the network depended on which crime “scene” they were part of, and what role they played in the crime commission process. For example, brokers who disposed of vehicles were found to be more important than brokers who concealed vehicles. The analysis and results illustrate the utility of a method that complements SNA with information about actor-level attributes such as roles. While the research reviewed here advances how we identify critical actors in an illicit network, an alternative method is to consider actor resources.

Network Capital

Some scholars argue that the best way to identify critical actors is to combine SNA with information about the resources an actor can access and bring to the network. Schwartz and Rouselle (2009) advocate a “network capital” framework in which the tangible and intangible attributes possessed by network members is considered in addition to, or alongside, centrality scores. Network actors who possess a larger number of resources and who share these resources with other actors in the network, contribute more significantly to network capital. Actors who contribute more to the network capital of a criminal network would be considered more important to the crime commission process. These actors possess and share resources that are needed for the sequence of actions required for the crime commission process to be successfully implemented. Accordingly, it can be hypothesized that the removal of “high network capital actors” will exert a negative impact on the network. The negative impact on the network will be both structural due to the actors’ positional importance and functional by virtue of the key resources possessed by these actors.

The number of resources possessed by a network actor can be quantified in “attribute weightings” (Schwartz & Rouselle, 2009). Schwartz and Rouselle (2009) recommend the use of two types of weights, one for nodes and one for edges (or links). Actor attribute weightings can be combined with actor centrality scores, or can be considered separately. The larger the number of tangible and/or intangible resources an actor possesses, and the more the actor shares the resources with other network actors, the greater the weightings applied to the actor. In this conceptualization, even actors with low connectivity (low degree or betweenness centrality) may add significantly to network capital if they obtain high attribute weights. That is, actors may be critical to the network for reasons unrelated to centrality. The combination of centrality scores with node attribute weightings may provide a more powerful method for identifying key actors, and compared with the use of centrality measures or attribute weightings used in isolation.

The present study seeks to extend the results of previous work on actor attributes in criminal networks by calculating a weighting for every actor in the network based on the set of resources possessed by each. The study seeks to answer the following questions: (a) What resources are possessed by actors in the criminal network? (b) What is the relationship between network position (centrality) and actor-level attributes? (c) Which actors are key based on centrality scores, attribute weightings, and a combination of these two factors?

The hypotheses were as follows:

Hypothesis 1: A combined scoring system (centrality and attributes) will generate a list of key players that includes actors not included in a list based on centrality scores alone.

Hypothesis 2: Actors who possess both tangible and intangible resources will be more active in the network (higher degree centrality) and will be positioned more strategically (higher betweenness and centrality) than other actors.

Method

Data Source

Data were extracted from file records of the Office of the Director of Public Prosecutions in New South Wales (NSW), Australia. We were provided access to the files of two individuals who were investigated and prosecuted for their part in the manufacture and trafficking of methamphetamine in NSW in the 1990s. The two actors (seeds) were selected after being identified as key players in the network based on evidence and police reports referenced in judicial sentencing comments. A range of documents, filling 24 boxes, were examined, including transcripts from listening devices and telephone intercepts, telephone call records, police surveillance reports, witness statements, sections of trial transcripts, and judges’ sentencing comments.

Network Generator

Each document was assessed as to whether it contained relevant information (i.e., information on the actors, relational data on links between actors, and information about resources possessed by actors). Actors were included if they contributed to the illicit activities being conducted by the group. Individuals were excluded if they had ties with actors, but there was no evidence that they participated in illicit behavior (e.g., family members who did not engage in illicit activities).

Variables

Information was initially recorded in narrative form and subsequently recoded into an attribute file and association matrix. For each actor, data were recorded regarding the presence or absence of each type of resource (1 = resource present, 0 = resource absent). The final set of resources is displayed in Table 1, including examples of coding for each.

Table 1.

List of Actor Attributes (Resources) With Example Coding.

Attribute	Examples
Drugs	N61 was supplying large quantities of drugs; N127 supplied others with ecstasy.
Money	N8 met with N23 and N66 to discuss supplies and to pay outstanding monies; N19 had a safe house where he stored firearms, drugs and money.
Premises	The cook started in the bathroom of a unit owned by N87; N95 asked N99 to rent a house with his girlfriend so they could cook there.
Equipment	N19 was supplying N95 with precursor chemicals and equipment for the manufacture of prohibited drugs; glassware, equipment and chemicals were given to N95 from N23.
Precursors	N66 asked N15 to get him some red phosphorous; N52 said he has 990 boxes of Sudafed.
Information	N23 called N19 to tell him some details; while in prison, N95 heard from N53 that N13 has robbed a chemical factory.
Skills/knowledge	N95 was recruited to manufacture amphetamines; N66 was engaging N67 and N38 for his cooks.
Labor	N63 was hired to babysit the cook; N110 picked up chemicals and delivered them to N66.

After initial data collection was completed by two researchers, a third researcher checked the data to identify and correct any errors in data extraction, transfer, and coding. Once all data were collected, the network had 128 actors and 741 (directed) links. The spreadsheet was anonymized by allocating an alphanumeric identifier to each actor in the network (from N1 to N128).

The SNAs were conducted using the UCINET software package (Borgatti, Everett, & Freeman, 2002). SNA metrics (centrality, betweenness, outdegree centrality) were calculated for the network. Degree centrality is a measure of connectedness, and is sometimes considered to be an indicator of the power or influence of actors in the network. Betweenness centrality measures the extent to which actors are on the shortest paths (geodesics) between other pairs of actors in the network, and is considered a measure of brokerage. Outdegree centrality is a measure of the outward flowing connections from each actor. In this study, outdegree centrality was used to indicate the extent to which actors had the capacity to share the resources they possessed with other network actors.

Attribute data for each actor were stored in a separate file (connected by the unique alphanumeric identifier). Each actor in the network was assigned an attribute weighting, calculated as follows. The attributes information and labor were assigned weight 1, and all other attributes (money, drugs, precursors, premises, equipment, skills/knowledge) were assigned weight 5. These latter attributes were determined to be more critical as they are required for the crime commission process; without this set of resources, the manufacture and trafficking of methamphetamine could not be successfully completed. On the other hand, information and labor were determined to be more peripheral; these resources can improve security and efficiency of the process, but were not determined to be critical resources. Note that the ratio of 5:1 used here is fairly arbitrary.² The total attribute weight of a given node is the sum of the weights of the attributes assigned to it. For example, an actor with the attributes drugs, money, and information receives an attribute weighting of 11.

Attributes Possessed by Actors: Descriptive Results

Table 2 shows the number of actors who possessed each of the eight attributes. The majority of actors possessed tangible resources, whereas fewer actors possessed intangible resources. Of the tangible resources, the majority of actors possessed at least one of money and drugs, whereas fewer possessed premises and precursors. Of the intangible resources, more actors possessed skills/knowledge and labor compared with information. Out of 128 actors, 34 did not possess any resources.

Table 2.

Number of Actors With Attributes (Some Actors Have Multiple Attributes).

Attributes	No. of actors
Tangible
Money	45
Drugs	45
Premises	24
Equipment	3
Precursors	26
Intangible
Information	9
Skills/knowledge	17
Labor	16
Total actors with at least one attribute	94
Actors who possess no attributes	34

Table 3 reports the number of attributes possessed by network actors. Among those actors who possessed only a single resource, the resource most frequently possessed was premises. It is interesting to note that no actors possessed equipment only. The most common combination of resources possessed by actors was money and drugs. Five actors possessed the combination of money, drugs and precursors, and all other combinations of resource types were uncommon (possessed by at most two actors).

Table 3.

Combinations of Attributes Possessed by Actors in the Network.

No. of attributes	No. of different combinations	% of actors with attributes (n)
One type	8	48.91 (45)
Two types	7	27.17 (25)
Three types	7	14.13 (13)
Four types	4	4.35 (4)
Five types	4	4.35 (4)
Six types	1	1.09 (1)

Results

Attribute Weightings

Scatterplots provide a bivariate analysis comparing the attribute weighting with the SNA centrality scores of each actor in the network (see Figures 1 and 2). The figures plot each actor’s strategic position in the network against actor’s functional importance. Means and standard deviation lines are displayed on each graph. Each graph displays one vertical and one horizontal solid line (one for each axis) showing the mean for each of the two measures. The broken lines (vertical and horizontal) represent one standard deviation above or below the mean.

Figure 1.

Scatterplot of degree centrality and total attribute weighting.

Figure 2.

Scatterplot of betweenness centrality and total attribute weighting.

Figure 1 shows that some actors were high on both degree centrality and total attribute weighting (e.g., N95 and N66). Other actors have low degree centrality scores, combined with high attribute weightings. For example, actors N77 and N108 have degree centrality scores below the mean, but have attribute weightings that are greater than one standard deviation above the mean.

Figure 2 shows that four actors (N19, N23, N66, and N95) have scores greater than one standard deviation above the mean for both betweenness centrality scores and attribute weighting scores. This group of actors appears to occupy brokerage positions for the exchange of some critical resources. One actor (i.e., N125) had a betweenness centrality score above the mean, combined with a low attribute weighting score (below the mean). This actor may occupy a brokerage position, facilitating the exchange of resources, without actually being in possession of the resources.

Combination of Centrality Scores and Weighting

Figure 3 shows a three-dimensional graph on which actors are plotted across three values—degree centrality (x-axis), betweenness centrality (y-axis), and attribute weights (z-axis). To allow us to make meaningful comparisons across measures, all values have been rescaled so that each axis ranges between 0 and 1 inclusive, by dividing by the maximum values. For example, the maximum degree in the network is 64, so all degree centrality scores were divided by 64 to give a value between 0 and 1. Because all values were scaled, the data points can be displayed in a one unit cube with one corner at the origin (0, 0, 0) and the opposite corner at the point (1, 1, 1).

Figure 3.

Three-dimensional plot of (scaled) degree centrality, betweenness centrality, and attribute weight.

The three measures (degree, betweenness, and weight) can now be combined to give one numerical quantity, the Euclidean norm of the corresponding data point. The Euclidean norm (distance to the origin) of each point was calculated using the formula $‖ (x, y, z) ‖ = \sqrt{(x^{2} + y^{2} + z^{2})}$ . The attribute weight was calculated as described earlier, with the attributes drugs, money, precursors, premises, skills contributing five times more to the attribute weight than information and labor. This distance gives a combined measure of the contribution of that node to the network taking into account a combination of two centrality scores (e.g., degree and betweenness) and attribute weight. This equals the straight-line distance from the origin (0, 0, 0) to the data point in three-dimensional space.

The Euclidean norm (Euclidean distance A) of the data point corresponding to an actor is therefore a combined measure of their contribution to the network. The top 10 actors corresponding to the nodes with the greatest Euclidean norm are shown in Table 4 (Euclidean A). As can be seen from the three-dimensional plots, for all actors except N66 and N95, the bulk of the distance to the origin comes from high degree centrality, high weight, or both.

Table 4.

Top 10 Network Actors: Centrality Scores vs. Euclidean Distance.

Position in top 10	Degree	Betweenness	Euclidean A	Euclidean B
1	N95	N95	N66	N66
2	N66	N66	N95	N67
3	N19	N19	N23	N36
4	N23	N23	N19	N114
5	N126	N61	N67	N23
6	N42	N126	N99	N95
7	N61	N42	N69	N77
8	N8	N8	N114	N99
9	N67	N125	N36	N108
10	N29	N67	N61	N19

Note. Elements in bold highlight actors who do not appear in the top 10 list by centrality alone.

In criminal networks, the trade-off between degree centrality and betweenness centrality can be used to identify actors who are less visible (and therefore less vulnerable), and who occupy strategic brokerage positions (see Morselli, 2010). Actors can obtain high scores on Euclidean distance A if they have high degree centrality and high betweenness scores. To investigate strategic positioning (i.e., to identify actors with low degree and high betweenness scores), we also calculated the distance of each data point from the corner of the cube (1, 0, 0: we call this Euclidean distance B). Data points that are far from the corner (1, 0, 0) will include actors with relatively low degree centrality, such as N67, N69, and N99. This was used to identify actors who are strategically positioned with low centrality but high betweenness. Nonetheless, actors with high degree centrality can obtain large relative distance from (1, 0, 0) if they also score highly on betweenness centrality, attribute weight, or a combination of these (e.g., N66). The top 10 actors in the network with the greatest distance from the point (1, 0, 0) are shown in Table 4 (Euclidean distance B).

Centrality Scores and Attributes

To determine whether there was a significant relationship between the type of attribute possessed by actors and actor centrality scores, network actors were categorized into one of four groups: (a) possessed no resources, (b) possessed only tangible resources, (c) possessed only intangible resources, and (d) possessed both tangible and intangible resources. Table 5 shows the means for degree and betweenness centrality across the four groups.

Table 5.

Means for Degree and Betweenness Across Attribute Groups.

Group	n	Degree	Betweenness
None	45	3.27	1.44
Tangible only	46	4.48	25.30
Intangible only	16	3.56	10.23
Both	21	16.67	510.84

Analyses³ yielded significant main effects for degree, Welch (3, 47.34) = 10.43, p < .001 and betweenness, Welch (3, 40.79) = 11.98 (3, 40.79), p < .001. Temhanes post hoc test indicated that actors who possessed both tangible and intangible resources were higher on degree centrality than actors who possessed no resources, tangible resources only, or intangible resources only (all p values < .001). Actors who possessed both tangible and intangible resources were also higher on betweenness centrality compared with actors who possessed no resources, tangible resources only, or intangible resources only (both > none, p < .001; both > intangible, p < .01; both > tangible, p < .01). In addition, actors who possessed tangible resources only had higher betweenness scores than those who possessed no resources (p < .05).

Discussion

Resources Possessed by Network Actors

The combinations of resources possessed by network actors showed considerable variation. For example, two actors possessed drugs, money, and labor; one actor possessed drugs, precursors, and premises; whereas, another possessed money, precursors, and skills. About half of all network actors did not specialize in the supply of any one particular resource, and instead supplied multiple resources, possibly in response to exigencies of the manufacture and trafficking process. While the majority of actors possessed at least one resource, 34 out of 128 (27%) of actors did not possess any resources. It is not clear whether this is a true reflection of the network, or an artifact of the data.

More actors in the network possessed tangible than intangible resources, which may be an artifact of the data (that is, police may be less likely to record possession of intangible resources). Money and drugs were the resources most frequently possessed by actors in the network, and indeed were the most common resources possessed in combination. This is not too surprising, given that the main motivation for involvement in drug trafficking is the profit from selling illicit drugs (Desroches, 2005). No actors possessed equipment only; equipment was possessed by actors only in combination with other resources. One possible explanation for this is that many different pieces of equipment (e.g., laboratory equipment) are required to set up a clandestine laboratory (see Ritter, Bright, & Gong, 2012). Existing members of the network (who may have been providing other resources) may have been called upon to obtain particular items of equipment when necessary. Of those who possessed only one type of resource, the most common was premises. The finding suggests that some actors supply their residential premises only, but are otherwise uninvolved in other parts of the manufacture and trafficking process. This is consistent with the results of Malm and Bichler (2011) who found that some actors were involved in only one niche, whereas others were involved in multiple niches. This finding also underscores the importance of residential premises for methamphetamine manufacture and trafficking. Methamphetamine cannot be manufactured without a relatively secure location for clandestine laboratories. At the time the network was operating, the most typical location for clandestine laboratories were private residences (Bright et al., 2012; Ritter et al., 2012), which offered some security against the laboratory sites being discovered by law enforcement.

Tangible/Intangible Resources and Centrality Scores

Actors who possessed diverse types of resources (i.e., tangible and intangible) had more connections with other actors and were more likely to be in brokerage⁴ positions in the network. Once again, this finding is consistent with Malm and Bichler (2011) who found that degree centrality was higher for actors involved in multiple niches compared with those involved in only a single niche. We also found that actors who possessed tangible resources were higher on brokerage than those who did not possess any resources. The same result was not found for intangible resources. This suggests that actors are more likely to fill structural holes for the exchange of tangible resources, but not for intangible resources. Tangible resources (like money, drugs, and precursor chemicals) are critical in a criminal network and are the main drivers for profit. Given that the motivation for involvement in drug trafficking is usually profit (e.g., Desroches, 2005), actors may be more likely to be involved in the exchange of such resources either directly or as brokers.

Actor Attributes: Total Weighting Score and Centrality Scores

Actors with high scores on attribute weightings are those who have access to a combination of a number of important resources, especially those considered “critical” to the crime commission process (i.e., drugs, money, premises, precursors, skills). These critical resources were weighted on a 5:1 ratio with labor and information which were considered less important to the crime commission process. While degree centrality scores have been used to indicate an actor’s power influence, or vulnerability within a criminal network (Morselli, 2009b), node-level attributes may provide an alternate measure of an actor’s status within a criminal network.

For both degree and betweenness centrality, the results demonstrate that analysis of both network position and actor-level attributes can generate important insights regarding criminal networks, especially with respect to the identification of key actors within such networks. For example, actors exhibiting high scores (i.e., greater than one standard deviation above the mean) for both degree centrality and attribute weighting were critical to the network both in terms of connectedness (possibly influential hubs), and because they possessed important resources required for the crime commission process.

Other actors who had low scores on degree centrality (i.e., below the mean) also possessed important resources (e.g., N77 and N108; see Figure 1). For this set of actors, an analysis based on degree centrality alone would not have identified them as key or high value actors in the network. The findings support the contention that actors can be important in criminal networks for reasons unrelated to centrality and that an analysis of both strategic positioning and actor-level attributes can reveal insights about the network that analysis based on strategic positioning alone cannot (e.g., Robins, 2009; Schwartz & Rouselle, 2009). Another distinct group of actors had high scores (i.e., greater than one standard deviation above the mean) on both betweenness centrality and attribute weighting (e.g., N19, N23, N66, and N95). These actors may be important brokers in the network, and they also possess critical resources that are necessary for the crime commission process to be successfully completed. It is not clear, based on the results, whether these actors actually possess and supply the resources, or are facilitators of the exchanges. Nonetheless, removal of these actors is likely to have both a structural and functional impact on the network. Removal of these actors is likely to create “structural holes” (Burt, 1992) in the network, and constrain the flow of important resources through the network, impairing the capacity of the network to successfully complete the crime commission process. In contrast to these findings, one actor (N125) showed a unique pattern of results on betweenness centrality and attribute weighting: a low attribute weighting score (below the mean), and a relatively high betweenness centrality score (above the mean). This pattern of results suggests that N125 may have occupied a brokerage role, and facilitated the exchange of important resources between other actors, but was never actually in possession of the resources.

In our analyses, the combination of degree centrality and attribute weighting scores most closely approximates the network capital construct as described by Schwartz and Rouselle (2009). Some actors were found to have high attribute weighting scores in combination with low centrality scores (e.g., N14, N56). These actors possess important resources, but have relatively few ties to other actors. While they may be suppliers of key resources (e.g., precursor chemicals), they appear to supply such resources to only a select few actors. Other actors had high scores (greater than one standard deviation above the mean) on degree centrality and on attribute weighting scores (e.g., N8, N19, N23, N99, N126). These actors make significant contributions to network capital. Such actors may indeed be worthy of law enforcement attention—their removal is likely to have a significant detrimental impact on network structure and function. It is important to note that an analysis of degree centrality alone would not have identified all members of this group of actors as key, and as important contributors to network capital.

Implications

Our findings suggest that law enforcement strategies should focus not only on actors who are in strategic positions in criminal networks but also on actors who possess important resources. Law enforcement agencies interested in disrupting and dismantling criminal networks should collect data not only on relationships between actors in criminal networks but also on actor-level attributes (such as resources). This information can be used separately and in combination with centrality scores to identify targets for law enforcement interventions (e.g., surveillance and arrest). Results suggest that law enforcement interventions will have maximum impact when both actor centrality and actor-level attributes are considered. Furthermore, when law enforcement target particular areas in a criminal network (e.g., precursor chemical supply, financial transactions), they may be best to target actors who are both well connected and possess a wide range of resources including the resource of interest. Such a strategy is likely to exert the greatest impact on both network structure and function. Of course, law enforcement agencies may be constrained in their capacity to gather detailed information about both actor connectedness and the range of resources they have access to. Therefore, this recommendation is limited by the quality and quantity of information available to law enforcement agencies at the time of any planned intervention.

Limitations

There are a number of limitations to this study. First, criminal justice data may include error. The sources of this error include intentional misinformation provided by actors (e.g., in recorded telephone conversations) and errors of transcription. Second, some data are likely to have been missing. For example, there may have been actors in the network who were not identified through law enforcement surveillance and investigation, there may have been connections between actors that were not documented, and law enforcement may be less likely to collect information on intangible resources compared with tangible resources (drugs, money). Third, as with all SNA research using criminal justice data, there is the potential for biased centrality scores based on the focus of the police investigation and the files used. However, the files used were for actors N66 and N126; two actors who were under heavy law enforcement surveillance. A review of the results reveals that these are not always the actors who feature as being central in the analyses, suggesting that the bias does not account for all results. Furthermore, a recent study (Berlusconi, 2013) concluded that centrality measures are robust even under conditions of data limitations related to the use of criminal justice file information (e.g., surveillance transcripts). Fourth, it is possible that more data may have been collected on some actors, leading to an artificial inflation of both centrality and attribute weights. Finally, as the study focused on a case study of a methamphetamine trafficking network operating in a single country in the 1990s, the generalizability of the results may be limited. The results mirror similar results found in different illicit contexts (e.g., Malm & Bichler, 2011); however, further research is needed using similar methodologies across different contexts (e.g., other drug trafficking, terrorism) to ascertain the extent to which our results will generalize to other criminal networks.

Conclusion

Despite the limitations, this study supports the collection and analysis of actor-level attributes by researchers and law enforcement agencies, as a complement to SNA. Our results suggest that actors in criminal networks can be important for reasons unrelated to centrality. For example, an actor may be low on centrality, but possess important resources such as precursor chemicals or the finances needed to purchase important equipment. Consideration of actor-level attributes in concert with centrality scores can be used to shed new light on the structure and function of criminal networks, and may be useful for identifying “key actors” beyond the usual focus on connectivity.

Footnotes

Acknowledgements

The authors would like to thank the Office of the Director of Public Prosecutions (ODPP; NSW) for providing access to data. Without the support of the ODPP (NSW), the project could not have been completed. The authors also acknowledge the contributions of Mr. Jordan Delaney and Dr. Helen Finn who collected some of the data for the project, and Dr. Thomas Britz who produced Figures 1 and .

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The project was funded by an Australian Research Council Discovery Project grant (DP120101744). Alison Ritter is the recipient of an NHMRC Senior Research Fellowship (APP1021988).

Notes

Author Biographies

David A. Bright is a forensic psychologist and senior lecturer in criminology with the School of Social Sciences, UNSW, Australia. His research focuses on criminal networks, organised crime, drugs and crime, and offender rehabilitation.

Catherine Greenhill is an associate professor in the UNSW School of Mathematics and Statistics. Her research interests range from asymptotic, probabilistic and algorithmic combinatorics through to the use of combinatorics to study real-world networks.

Michael Reynolds is a mathematician and programmer, with experience in data modelling, quantitative analysis, and using visualisation as a tool for understanding and exploring mathematical proofs.

Alison Ritter, is a professor, drug policy scholar and Director of the Drug Policy Modelling Program (DPMP) at the National Drug and Alcohol Research Centre (NDARC) at the University of New South Wales. She is an NHMRC Senior Research Fellow (2012 to 2016). She is the current President of the International Society for the Study of Drug Policy.

Carlo Morselli is a professor at the École de criminologie and member of the International Centre for Comparative Criminology, Université de Montréal. He is also editor-in-chief for the journal Global Crime. His current scholarship examines illegal firearm markets, synthetic drug markets, and collusion in the construction industry. Recent publications include two books (Crime and Networks and Inside Criminal Networks), in addition to articles appearing in Crime, Law and Social Change, Criminology, the Journal of Research in Crime and Delinquency, Critical Criminology, and Social Networks.

References

Baker

W. E.

Faulkner

R. R.

(1993). The social organization of conspiracy: Illegal networks in the heavy electrical equipment industry. American Sociological Review, 58, 837-860.

Berlusconi

(2013). Do all the pieces matter? Assessing the reliability of law enforcement data sources for the network analysis of wire taps. Global Crime, 14, 61-81.

Borgatti

S. P.

Everett

M. G.

Freeman

L. C.

(2002). UCINET 6 for Windows. Columbia, SC: Analytic Technologies.

Bright

D. A.

Greenhill

Levenkova

(2014). Dismantling criminal networks: Can node attributes play a role. In Morselli

(Ed.), Crime and Networks (pp. 148-162). New York, NY: Routledge.

Bright

D. A.

Hughes

C. E.

Chalmers

(2012). Illuminating dark networks: A social network analysis of an Australian drug trafficking syndicate. Crime, Law, and Social Change, 57, 151-176.

Burt

R. S.

(1992). Structural holes. Cambridge, MA: Harvard University Press.

Calderoni

(2014). Strategic positioning in mafia networks. In Morselli

(Ed.), Crime and Networks (pp. 163-181). New York, NY: Routledge.

Desroches

F. J.

(2005). The crime that pays: Drug trafficking and organized crime in Canada. Toronto, Ontario: Canadian Scholars’ Press.

Krebs

V. E.

(2002). Mapping networks of terrorist cells. Connections, 24(3), 43-52.

10.

Malm

Bichler

(2011). Networks of collaborating criminal: Assessing the structural vulnerability of drug markets. Journal of Research in Crime & Delinquency, 48, 271-297.

11.

Morselli

(2009a). Hells Angels in springtime. Trends in Organized Crime, 12, 145-158.

12.

Morselli

(2009b). Inside criminal networks. New York, NY: Springer.

13.

Morselli

(2010). Assessing vulnerbale and strategic positions in a criminal network. Journal of Contemporary Criminal Justice, 26, 382-392.

14.

Morselli

Roy

(2008). Brokerage qualifications in ringing operations. Criminology, 46, 28, 71-98.

15.

Natarajan

(2000). Understanding the structure of a drug trafficking organisation: A conversational analysis. In Natarajan

Hough

(Eds.), Illegal drug markets: From research to prevention policy (pp. 273-298). Monsey, NY: Criminal Justice Press.

16.

Natarajan

(2006). Understanding the structure of a large heroin distribution network: A quantitative analysis of qualitative data. Journal of Quantitative Criminology, 22, 171-192.

17.

Ritter

Bright

D. A.

Gong

(2012). Evaluating drug law enforcement interventions directed towards methamphetamine in Australia. Canberra, Australia: National Drug Law Enforcement Research Fund.

18.

Robins

(2009). Understanding individual behaviors within covert networks: The interplay of individual qualities, psychological predispositions, and network effects. Trends in Organized Crime, 12, 166-187. doi:10.1007/s12117-008-9059-4

19.

Schwartz

D. M.

Rouselle

D. A.

(2009). Using social network analysis to target criminal networks. Trends in Organized Crime, 12, 188-207.

20.

Chen

(2009). Untangling criminal networks: A case study. Intelligence and Security Informatics, 2665, 232-248.