This is the author’s accepted manuscript of Calderoni, Francesco, and Elisa Superchi. 2019. “The Nature of

Organized Crime Leadership: Criminal Leaders in Meeting and Wiretap Networks.” Crime, Law and Social

Change, March. https://doi.org/10.1007/s10611-019-09829-6. The final publication is available at https://link.s-

pringer.com/article/10.1007/s10611-019-09829-6

The Nature of Organized Crime Leadership: Criminal Leaders in Meeting

and Wiretap Networks

Francesco Calderonia and Elisa Superchib

a Transcrime, Università Cattolica del Sacro Cuore, Milan, Italy

b Dipartimento di sociologia e ricerca sociale, Università degli Studi di Milano Bicocca,

Milan, Italy

Abstract: Criminal leaders enhance their social capital by strategically brokering infor-

mation among associates. To balance security and efficiency, leaders may favor meet-

ings instead of telephones, potentially affecting analyses relying solely on wiretap data.

Yet, few studies explored criminal leaders’ use of meetings in the management of crim-

inal groups.

We analyze criminal leaders’ participation in meetings and telephone calls in four dis-

tinct investigations. For each case, we extracted meetings and wiretap networks, ana-

lyzed leaders’ network positioning and identified leadership roles through logistic re-

gressions relying on network centrality.

Results show that leaders minimize telephone use (20% missing in wiretap net-works),

and act as brokers, particularly in meeting networks (betweenness 18 times higher than

non-leaders). Regressions on meeting networks identify leaders more effectively than

wiretap networks, with betweenness centrality as the strongest predictor of leadership.

Leaders’ centrality in meetings shows their strategic brokering position and the social

embeddedness of criminal groups. While meeting participation is a sign of power, it is

also a social obligation that leaders can hardly minimize. This makes them more visible,

with possible benefits to investigations and intelligence.

Keywords: network analysis, organized crime, criminal networks, dark networks, lead-

ership

1

Acknowledgments We wish to thank two anonymous reviewers, Scott Decker, David

Krackhardt, Giulia Berlusconi, Elisa Bellotti, and Tomáš Diviak for the useful com-

ments to our manuscript. This work is the result of the joint efforts by both authors. F.C.

and E.S. jointly contributed to the design of the study to the data collection and to the

analysis of the results. F.C. wrote the introduction and discussion, and reviewed the fi-

nal manuscript.

1. Introduction

Leaders are fundamental for organizational success. Since the 1970s, research on leadership

has re-focused from human capital, i.e. individuals’ traits, to social capital, i.e. resources em-

bedded in social relations (Coleman, 1988). Studies analyzed the connections between indi-

viduals’ status and their role within their social networks, pointing up how relations play a

major role in the achievement of leadership (Burt, 1980; Granovetter, 1973). Access to new

information generates relational asymmetries that may be crucial in competitive settings, and

individuals exploit connections that are sources of novel information (Burt, 1992; Granovet-

ter, 1983). People in such privileged positions are brokers, effectively exploiting the flows of

information, connecting disconnected others, and bridging across structural holes, the “empty

spaces in social structure” (Burt, 2007, p. 16).

Brokers are even more crucial within criminal networks, which operate in stateless en-

vironments without standard mechanisms regulating transactions and disputes (Kleemans &

de Poot, 2008; Kleemans & Van Koppen, 2014; Paoli, 2002; Reuter, 1983). Social ties giving

access to profitable opportunities determine success within criminal organizations (Kleemans,

2014; Kleemans & Van Koppen, 2014). Contemporarily, criminals have to balance the effect-

ive management of illicit activities with the security of the group (Morselli, Giguère, & Petit,

2007).

Since the 2000s, studies have empirically examined criminal networks (Mancuso,

2014; D. A. Bright, Hughes, & Chalmers, 2012; Malm, Bichler, & Van De Walle, 2010;

2

Morselli, 2009b; Natarajan, 2006; Varese, 2006b; Morselli & Tremblay, 2004).1 Criminal

leaders are strategic brokers within their networks, often reporting higher scores of between-

ness centrality, a measure capturing how frequently an individual lies on the shortest path be-

tween any other two individuals in the network (Freeman, 1979). In organized crime groups,

embedded in the surrounding social environment, leaders bridge among other criminals but

also among people from legitimate businesses and politics, exploiting ties providing access to

criminal opportunities (Kleemans & Van De Bunt, 1999; Papachristos & Smith, 2014;

Kleemans & Van Koppen, 2014; Vargas, 2014).

Most studies identified criminal leaders through wiretap data. However, many crimi-

nals are wary of telephone interceptions and may minimize telephone use (Agreste, Catanese,

De Meo, Ferrara, & Fiumara, 2016; Calderoni, 2014; Campana & Varese, 2011; Morselli,

2009b). Notoriously, criminals face a efficiency vs. security trade-off, limiting information

sharing so as to prevent detection at the cost of lower efficiency (Morselli et al., 2007). Crimi-

nal leaders often avoided telecommunications, limiting it to forced or unnecessary involve-

ment (Kleemans, Brienen, & Van De Bunt, 2002). Furthermore, leaders may delegate to mid-

dle-level criminals the operational management of risky activities. In some cases, they com-

pensate direct (degree centrality) with indirect connections (betweenness centrality) (Baker &

Faulkner, 1993; Morselli, 2009b). In other cases, centrality fails to capture the status within a

network, as middle-level criminals are the most central individuals (Agreste et al., 2016; D. A.

Bright et al., 2012; Calderoni, 2012; Carley, Krackhardt, & Lee, 2002).

While criminals’ strategies may bias wiretap data, little research has explored the ef-

fectiveness of alternative sources for the identification of leaders (Bouchard & Konarski,

2014; D. A. Bright et al., 2012). Different data may yield different results, particularly when

1 Social network analysis is a statistical method allowing the mapping and measuring of relations among individuals constituting networks. Within a network, the members are the actors and can represent people or other entities; whereas the links show relations between nodes (e.g. telephone calls, friendship, etc.) (Wasserman & Faust, 1994).

3

individuals avoid specific channels of communications. Comparison of intelligence, surveil-

lance, and co-offending networks of the same street gang showed that the data provided “fun-

damentally different pictures” (Rostami & Mondani, 2015, p. 16). Research may need to ex-

plore alternative, hardly concealable, behaviors to better identify leaders. The analysis of

meetings may be a promising step in this direction.

Leaders may be unable to minimize meeting participation since meetings are essential

to manage the criminal activities. Two studies showed that mafia leaders took part in more

meetings and had higher values of betweenness centrality than others participants (Calderoni,

2014, 2015). Logistic regressions on leadership successfully classified 92.6% of the individu-

als, with betweenness as the main predictor of leadership. Furthermore, the models were ro-

bust even at early stages of the investigation (Calderoni, 2015). Given their high betweenness,

mafia bosses participated in meetings where most other individuals were absent, a pattern typ-

ical of brokers. However, the results were based on one case study and lacked a direct com-

parison with the corresponding wiretap network.

This study aims at improving the knowledge on criminal leadership by analyzing

meeting and telephone call participation across four distinct organized crime investigations.

The objectives of this study are (1) to analyze criminal leaders’ participation in meetings and

telephone calls; (2) to compare meeting and wiretap networks in identifying leadership roles.

The novel contribution of this work consists in its focus on criminal leadership, a surprisingly

under researched field in criminology, despite leaders being among the few high-achievers in

crime. Our new study provides a novel examination of the relations among meeting participa-

tion, leadership, and brokerage, and a systematic comparison between wiretap and meeting

networks on the four cases. The results show that leaders minimize telephone use and favor

indirect connectivity, particularly in meetings, revealing a brokering position which gives

them strategic advantages. Criminal leadership is thus the result of specific social opportunity

4

structures, where leaders can hardly avoid participating in certain meetings. This makes them

vulnerable to identification through logistic regressions.

The rest of the article is structured as follows: the second section presents the case

studies and the analytical approach; the third section describes the results; the fourth one dis-

cusses the findings as well as their limitations. The final section concludes with some consid-

erations for future research.

2. Methodology

2.1. The case studies and the organization of the ‘Ndrangheta

This analysis focused on four large investigations tackling groups belonging to the

‘Ndrangheta, an Italian mafia from Calabria (a southern Italian region). Examination of the

‘Ndrangheta is important beyond the borders of Calabria because it has expanded to several

Italian regions (Transcrime, 2013; Varese, 2006a), and countries (Germany, Switzerland,

Canada, and Australia) (Calderoni, Berlusconi, Garofalo, Giommoni, & Sarno, 2016;

Campana, 2013; KLPD, 2011; Sciarrone & Storti, 2014; Sergi, 2015). The investigations

tackled groups in Calabria (Operation Crimine, 2010), Lombardy (Operation Infinito, 2010),

Piedmont (Operation Minotauro, 2011), and Emilia-Romagna (Operation Aemilia, 2015).

They lasted for several years and were conducted by the specialized anti-mafia prosecutor’s

offices and anti-mafia police units of the Carabinieri.2

In all operations, the main charge was the offence of mafia association (Art. 416-bis of

the Italian Criminal Code), with a focus on establishing the existence and internal structure of

the groups including their leadership. In addition, each investigation concerned dozens of mi-

nor charges for e.g. violence, extortion, loan sharking, corruption, possession of weapons and

2 The cases were the most exceptional actions in recent years against the ‘Ndrangheta in Italy. After the arrest of nearly six hundred people, the courts confirmed the charges and convicted most defendants for Infinito, Crimine, and Minotauro. In April 2016, the first grade trial for Aemilia also confirmed the charges.

5

drug trafficking. The cases are thus particularly useful for assessing the participation of crimi-

nals in wiretaps and meetings, especially because of the internal structure of the ‘Ndrangheta,

based on blood ties and formal hierarchy.

The organization of the ‘Ndrangheta primarily revolves around blood kinship (Paoli,

2003; Varese, 2006a). Affiliates linked by blood ties or marriages form a ‘ndrina; several

‘ndrine in the same geographic area form a locale, an autonomous group which controls a

specific territory (Sergi & Lavorgna, 2016; Serenata, 2014; Paoli, 2003). Even far from the

home territory, the ‘Ndrangheta replicates its distinctive structure through affiliation cere-

monies and a well-defined internal hierarchy (KLPD, 2011; Transcrime, 2013; Campana,

2013; Sciarrone & Storti, 2014; Sergi, 2015; Calderoni et al., 2016). The hierarchy comprises

offices (cariche) and ranks (doti) (Paoli, 2003, p. 46). Each locale has offices charged with

specific duties and powers. The most important include capolocale (the boss of the locale),

caposocietà (the deputy boss), mastro di giornata (“day master”, ensuring the flow of infor-

mation), contabile (“accountant”, managing of the common funds of the locale). The offices

are elective, although often monopolized by powerful individuals or families, and only high-

rank affiliates can compete for high-level offices (Paoli, 2003, pp. 46–51). The ranks (doti)

correspond to the individual’s status within the organization, like the steps in the ‘Ndrangheta

career. The award of a higher rank requires the agreement of the most important Calabrian af-

filiates, and only top-ranked individuals can aspire to high offices (Paoli, 2003, 1994).

Given the objective of proving the structure of mafia groups, the four investigations

carefully charted the internal hierarchy of the ‘Ndrangheta. The investigative files were useful

sources for the gathering of data on the criminal networks and their leadership. The sources

for the case studies were four pre-trial court orders, judicial documents similar to arrest war-

rants (Procura della Repubblica presso il Tribunale di Reggio Calabria, 2010; Tribunale di

6

Bologna, 2015; Tribunale di Milano, 2010; Tribunale di Torino, 2011).3 The data gathering

consisted in the manual extraction and coding of every individual and event (telephone call or

meeting) mentioned in the documents. Different researchers independently collected and vali-

dated the data to ensure consistency.4 The analysis focused on time periods with constant law

enforcement surveillance.5

The source generated three datasets per case study: one with individual attributes, and

two with relational data on meetings and telephone calls, respectively. The individual at-

tributes comprised name, year of birth, locale of affiliation, office, the number of events at-

tended, mafia charge (whether the suspect was charged for mafia association), leadership

roles, and additional information (e.g. arrested, murdered). The datasets on meetings and tele-

phone calls recorded the date, place, and hour of the event, and the number and identity of

participants.

2.2. The classification of leaders

The identification of leaders relied on direct evidence from the criminals (e.g. content of wire-

taps, audio or video surveillance) and was thus independent from the prosecutor’s judgement

or network position. During several conversations the suspects reviewed in detail the formal

hierarchy of the different ‘Ndrangheta locali. Remarkably, the classification was independent

from the actual participation in meetings or telephone calls, since a court would dismiss any

such argument for lack of evidence.

3 In Italy, specialized judges in every court issue such orders upon request of the prosecution. The orders aim to impose pre-trial restrictions against the suspects by motivated decision (challengeable before other courts). Especially for large investigations these documents amount to hundreds of pages and include detailed information on all suspects’ activities. The literature has often used these sources for analyzing criminal networks (Varese, 2006b; Campana & Varese, 2011; Scaglione, 2011; Berlusconi, 2013).4 Consistency checks included removal of duplicate events, synthesis of incomplete events, exclusion of uncorroborated events.5 Selected time spans were mid-2011 to 2012 for Aemilia; 2007 to mid-2010 for Crimine; 2008 to early 2010 for Infinito; 2007 to early 2011 for Minotauro. The selection entailed the exclusion of only few events and marginal individuals.

7

The variable “leaders” identifies only the most important criminals in the four case

studies (leader=1; non-leader =0). Operations Infinito, Crimine, and Minotauro tackled more

than one locale, so that “leaders” includes only the two highest offices of each locale – the

boss (capolocale) and the deputy boss (caposocietà) – because they may sometimes overlap

(Paoli, 2003, p. 49).6 Conversely, Operation Aemilia focused only on one ‘ndrina, and “lead-

ers” included six senior suspects considered the bosses by the court based on direct evidence

of their roles within the group.7 The precautionary selection resulted in only a few “leaders” in

each case, from 6 in Aemilia to 25 in Infinito (5%-14% of the individuals in the networks)

(Table A in the Annex).8

To ensure the robustness and reliability of the results, the analyses also considered an

alternative classification comprising “leaders & organizers”. These included all individuals

classified as “leaders”, but also other individuals with high ranks. In Infinito, Crimine, and

Minotauro “leaders & organizers” included the bosses, the deputy boss, and the main charges

for each locale (i.e. contabile, and mastro di giornata);9 in Aemilia, they comprised five sus-

pects labelled “organizers” by the court. As a result, “leaders & organizers” were 11 in Ae-

milia, 31 in Crimine, 34 in Infinito, and 33 in Minotauro.

6 In Infinito, “leaders” includes also two individuals responsible for a special body (“la Lombardia”) tasked with the coordination between Calabria and Lombardy.7 At the end of 2018, a final judgement confirmed the charges and convicted the five out of six leaders. A separate, first grade judgement convicted the last leader.8 Given the complexity of the investigations, some leaders made only sporadic appearances. To ensure that the analyses focused on the leaders actually managing their own criminal groups, a few leaders were classified as non-leaders for two reasons which may have biased the analyses. First, some were leaders affiliated to locali outside the scope of the investigation: for example, ‘Ndrangheta leaders from other areas (Calabria, but also other Italian regions and countries of the world) who seldom interacted with the groups under investigation. Second, other leaders were incapacitated to participate in events, e.g. due to imprisonment or death at an early stage of the investigations.9 In Minotauro, “leaders & organizers” also comprised former holders of the offices (only this investigation identified them), and the members of the “Crimine”, a special unit responsible for violent actions.

8

2.3. Network measures and logistic regression models

Relational data generated two, two-mode affiliation matrixes recording individuals’ participa-

tion in meetings and wiretaps. In turn, two-mode matrixes generated one-mode and undirected

adjacency matrixes reporting individuals’ co-participations in events and allowing assessment

of networks’ topology (Table A and Figure A in the Annex) (Wasserman & Faust, 1994).10

These also enabled computation of actors’ centrality measures, including: degree centrality,

measuring an individual’s number of direct connections; valued degree centrality, showing

the number of co-participations in events; betweenness centrality, measuring the extent to

which a node is in the shortest path between a pair of individuals (Wasserman & Faust,

1994).11 All measures were computed on binary networks, then normalized (max

score=1,000) to improve readability, except valued degree (valued networks and raw scores)

(Table 1). A preliminary exploration of centrality calculated the ratios between leaders’ and

non-leaders’ scores.

The identification of leaders relied on Firth’s logistic regressions to assess the proba-

bility of being a leader given the values of some predictors. Firth’s method fits data better

than traditional logistic regressions, where maximum likelihood estimations suffer from a

small number of cases in one of the two categories (i.e. only few leaders in networks), as it re-

duces such bias through penalized likelihood and overcomes the problem of separation (Al-

lison, 2013; Firth, 1993).12

10 All network analysis operations were performed using UCINET 6 (Borgatti, Everett, & Freeman, 2002). 11 Preliminary explorations included also closeness and eigenvector centrality, but these measures reported limited differences between leaders and non-leaders.12 Complete or quasi-complete separation occurs when the outcome variable separates a predictor variable. For instance, in most observations leader=1 corresponds to Mafia charge=1.

9

Table 1. Descriptive Statistics of Variables by Network

Network Leadership roles

Events Mafia charge Degree Valued Degree Betweenness

Mean St. dev. Mean St. dev. Mean St. dev. Mean St. dev. Mean St. dev.Aemilia meeting

Leaders 29.7 11.2 1.0 0.0 240.3 43.6 105.2 17.7 71.2 21.5Others 3.3 5.9 0.2 0.4 45.6 55.4 14.0 20.3 5.5 21.2

Aemilia wiretap

Leaders 64.8 50.5 1.0 0.0 95.9 53.3 42.8 29.7 152.4 132.5Others 5.0 7.7 0.3 0.4 10.3 11.4 7.2 18.1 4.4 15.9

Crimine meeting

Leaders 18.8 31.5 1.0 0.2 78.0 89.4 59.4 78.5 48.1 18.8Others 2.6 4.7 0.2 0.4 37.1 50.2 19.4 24.7 1.7 2.6

Crimine wiretap

Leaders 33.4 71.1 1.0 0.0 45.9 84.5 33.8 72.2 112.8 33.4Others 5.1 7.6 0.2 0.4 9.5 10.0 5.2 7.8 16.8 5.1

Infinito meeting

Leaders 28.0 33.8 1.0 0.2 174.9 78.2 107.3 76.5 44.0 28.0Others 3.9 5.5 0.5 0.5 52.8 49.9 20.4 26.4 3.0 3.9

Infinito wiretap Leaders 9.8 9.8 1.0 0.2 33.8 25.4 9.8 9.8 102.1 9.8Others 3.3 3.5 0.6 0.5 12.4 9.6 3.3 3.5 19.6 3.3

Minotauro meeting

Leaders 39.1 42.2 0.9 0.3 383.0 127.8 376.4 202.6 30.5 31.0Others 4.0 6.2 0.3 0.5 105.1 112.1 57.1 77.6 2.0 7.2

Minotauro wiretap

Leaders 40.1 81.9 0.9 0.3 40.5 51.4 40.5 82.6 84.6 130.9Others 8.7 24.3 0.2 0.4 8.2 13.1 8.9 24.5 6.0 29.5

Firth’s regressions explored the capacity of meeting and wiretap networks to identify

leaders. The dependent variable for each network was “leaders” or “leaders & organizers”

(where leaders=1; non-leaders=0). The main predictors included the centrality measures better

discriminating leaders based on the preliminary exploration of the leaders/ non-leaders ratios

and often used in the literature (Morselli, 2009b): in meetings these were betweenness and

valued degree, in wiretaps betweenness and degree centrality.13 The regressions controlled for

possible investigation biases due to the non-random sample. “Mafia charge” considered that

investigators deliberately focused the surveillance on suspect mafia members. The number of

events controlled for possible additional biases, e.g. that individuals merely participating in

more events were mistaken for leaders, or that the law enforcement likely prioritized alleged

leaders, thus resulting in higher frequencies.

3. Results

3.1. Network topology and leaders’ behavior

The ‘Ndrangheta leaders minimized involvement in wiretaps, either completely avoiding tele-

phones or using them in a manner hardly distinguishable from non-leaders. They generally fa-13 In wiretap networks, an individual’s participations in telephone calls often corresponds to the valued degree (number of co-participations) as most calls involve only two people.

10

vored indirect connectivity, particularly in meeting networks, with betweenness scores nearly

eighteen times higher than non-leaders.

All the networks were loosely connected (Table A in the Annex, density ranging be-

tween 1.2% and 11.6%), with meeting networks more cohesive than wiretap networks.14 The

participation of mafia leaders to the networks reveals three specific patterns. First, mafia lead-

ers were generally more present in meetings than in wiretaps, notwithstanding the higher

number of wiretaps. While always present in meeting networks, some leaders were absent

from three out of four wiretap networks. Specifically, only 12 out of 24 leaders participated in

telephone calls in Crimine, 23 out of 25 in Infinito, and 13 out of 15 in Minotauro. Overall,

23% of leaders were missing in the wiretap networks.

Second, even when participating in telephone calls, leaders showed some preference

for meetings. Table 2 and Figure 1 report the leaders’/non-leaders’ ratios in the average num-

ber of events and centrality scores. Differences were assessed through Student’s independent

samples, two-tail, t-tests performed through UCINET on 10,000 permutations to account for

the dependency among network data. Unsurprisingly, leaders were always more active than

non-leaders, e.g. participating in more events (on average about 6.6 times more across the

eight networks, first column in Table 2), and having a larger number of contacts (on average

four times higher, as expressed by the degree centrality, second column in Table 2). However,

those leaders present in wiretaps used telephones five times more than non-leaders, whereas

all leaders attended 8.3 times more meetings than non-leaders.

14 Across the four investigations, 48.6% and 36.1% of individuals in meetings and wiretap networks, respectively, took part in one event. Participants in 10+ events amounted to 9.5% and 15.7%, respectively.

11

Table 2. Ratios between Leaders and Non-Leaders in the Mean Number of Events and the Centrality Measures by Network

Network N. of events Degree Valued Degree BetweennessAemilia meeting 8.9*** 5.3*** 7.5*** 13.0***Aemilia wiretap 5.9** 5.6*** 5.9** 12.1**Crimine meeting 7.2*** 2.1*** 3.1*** 28.9***Crimine wiretap 6.5*** 4.8*** 6.4*** 6.7**Infinito meeting 7.1*** 3.3*** 5.3*** 14.7***Infinito wiretap 3.0*** 2.7*** 3*** 5.2***Minotauro meeting 9.8*** 3.6*** 6.6*** 14.9***Minotauro wiretap 4.6** 4.9*** 4.5** 14.1***Avg. all networks 6.6 4.0 5.3 13.7Avg. meeting networks 8.3 3.6 5.6 17.9Avg. wiretap networks 5.0 4.5 5.0 9.5+ p≤0,1, * p≤0,05, ** p≤0,01, *** p≤0,001

Third, leaders preferred indirect over direct connections, with remarkably high be-

tweenness centrality scores (on average nearly fourteen times those of non-leaders, vs. four

times for degree centrality). The prevalence of indirect connectivity was stronger for meeting

than for wiretap networks: while leaders’ betweenness centrality was 17.9 and 9.5 times that

of non-leaders across meeting and wiretaps networks, degree centrality was instead only 3.6

and 4.5 times higher, respectively (Table 2).

Figure 1. Ratios between Leaders and Non-Leaders in the Mean Number of Events and the Centrality Measures by Network

AemiliaMeetings

AemiliaWiretaps

CrimineMeetings

CrimineWiretaps

InfinitoMeetings

InfinitoWiretaps

MinotauroMeetings

MinotauroWiretaps

-5

5

15

25Events

Degree

Valued Degree

Betweenness

12

3.2. Comparison between meeting and wiretap networks in identifying leaders

Across all the four case studies, models based on meetings identified mafia leaders better than

models on wiretaps. The regressions with “leaders” as a dependent variable relied on six spec-

ifications (Table 3).15 Models on meetings provided better fitting models and higher success in

the classification of leaders than models on wiretaps. Tjur’s R2 revealed an acceptable or good

fit of the models for meeting networks, and a poorer fit for wiretaps (Table 3).16 On average

Tjur’s R2 for meetings was almost double than for wiretaps. Meetings also provided better

classifications of leaders than wiretaps (average successful leader identification of 26.74% vs.

9.13%, respectively, Table 3). Comparing the best models among all specifications, meetings

identified almost three times more leaders for Aemilia and Crimine, and more than twice for

Infinito and Minotauro (Table 4).

15 Specification I only included betweenness centrality, given its importance in the literature and in the analysis of the ratios between leaders and non-leaders. Specification II included only control variables to assess the impact of the mere participation in events and the possible biases by investigation. Specification III comprised betweenness and mafia charge, while specification IV betweenness and both controls. Specification V relied on betweenness, mafia charge, and valued degree for meetings and degree for wiretaps. Specification VI presented the complete models. Exploration of additional specifications yielded consistent results.16 Tjur’s R2, or Tjur’s coefficient of discrimination, varies from 0 to 1 as the prediction improves. It calculates the mean of the predicted probabilities of an event for each category of the dependent variable (i.e. whether or not a member is a leader) and then it takes the difference between the two (Tjur, 2009).

13

Table 3. Firth’s Logistic Regression Results for “Leaders” for Meeting and Wiretap Networks, Odds Ratios

Network Variables I II III IV V VI

Aemilia Meeting

Betweenness 1.032** 1.019* 0.973 0.995 0.977Valued Degree 1.050** 1.036+Mafia charge 11.45 21.49* 10.37 4.266 4.007

Meetings attended 1.101** 1.204** 1.108Tjur’s R2 0.114 0.327 0.169 0.424 0.444 0.508

Predicted leaders (%) 0.0 16.7 0.0 33.3 16.7 50.0

Aemilia Wiretap

Betweenness 1.011* 1.006+ 1.042 0.99 1.016Degree 1.037 1.098*

Mafia charge 26.58* 24.78* 23.61* 18.20+ 9.441Tel. calls 1.014+ 0.911 0.871*Tjur’s R2 0.068 0.11 0.132 0.232 0.15 0.318

Predicted leaders (%) 0 0 0 16.667 0 0

Crimine Meeting

Betweenness 1.076***   1.048* 1.056+ 1.060* 1.058+Valued Degree 0.993 0.997Mafia charge 52.29*** 45.89*** 48.15*** 45.04*** 46.51***

Meetings attended 1.044* 0.967 0.969Tjur’s R2 0.224 0.263 0.324 0.317 0.329 0.318

Predicted leaders (%) 20.8 16.7 20.8 20.8 25.0 25.0

Crimine Wiretap

Betweenness 1.007***   1.004 1.001 1.002 1.002Degree 1.007 0.979

Mafia charge 113.1** 105.0** 107.1** 104.6** 108.3** Tel. calls 1.012 1.009 1.034Tjur’s R2 0.087 0.284 0.278 0.289 0.287 0.296

Predicted leaders (%) 8.3 8.3 8.3 8.3 8.3 8.3

Infinito Meeting

Betweenness 1.099***   1.079*** 1.038* 1.045** 1.042*Valued Degree 1.022** 1.020+Mafia charge 5.947* 6.231* 4.737+ 3.659 3.7

Meetings attended 1.157*** 1.096* 1.011Tjur’s R2 0.417 0.405 0.421 0.453 0.473 0.471

Predicted leaders (%) 40 44 40 44 44 40

Infinito Wiretap

Betweenness 1.014***   1.012*** 1.008* 1.005 1.004Degree 1.053* 1.083+

Mafia charge 4.763+ 4.761+ 3.923 3.41 3.459Tel. calls 1.166*** 1.100+ 0.915Tjur’s R2 0.181 0.2 0.196 0.233 0.263 0.264

Predicted leaders (%) 17.4 13.0 17.4 13.0 17.4 17.4

Minotauro Meeting

Betweenness 1.077***   1.060*** 0.996 1.015 1.001Valued Degree 1.010*** 1.008*Mafia charge 24.23+ 9.936* 26.16+ 4.525 7.451

Meetings attended 1.096*** 1.099** 1.031Tjur’s R2 0.247 0.326 0.247 0.325 0.392 0.392

Predicted lea (%) 20.0 20.0 20.0 20.0 26.7 33.3

Minotauro Wiretap

Betweenness 1.012***   1.011** 1.024 1.002 0.996Degree 1.023 1.090*

Mafia charge 25.46*** 23.98*** 18.22** 25.28*** 15.50** Tel. calls 1.011+ 0.981 0.969* Tjur’s R2 0.122 0.128 0.206 0.238 0.201 0.262

Predicted leaders (%) 15.4 0.0 7.7 15.4 0.0 15.4+ p≤0.1 level; * p≤0.05 level; ** p≤0.01 level; *** p≤0.001 level.

Given that the number of leaders varies across networks, Table 4 and Figure 2 com-

pare the correct classifications with the random probability of identifying a leader (bench-

mark) across the eight networks. Meetings identified leaders between four and nineteen times

better than the benchmark in the best models. Conversely, wiretaps ranged between 1.2 (for

Infinito, thus only slightly better than random selection) and 5.6 (for Aemilia, although all

other specifications failed to identify any leader).

14

Table 4. Ratios between the Share of Leaders Correctly Predicted by Firth’s Logistic Regressions and the Share of Leaders Correctly Predicted by Random Probability (Benchmark) by NetworkNetwork N.

of leadersBenchmark

(%)Regressions’ specifications

I II III IV V VIAemilia Meeting 6 2.6 Correct identifications (%) 0.0 16.7 0.0 33.3 16.7 50

Ratio correct identifications/benchmark 0.0 6.3 0.0 12.6 6.3 18.9Aemilia Wiretap 6 3.0 Correct identifications (%) 0 0 0 16.67 0 0

Ratio correct identifications/benchmark 0.0 0.0 0.0 5.6 0.0 0.0Crimine Meeting 24 5.1 Correct identifications (%) 20.8 16.7 20.8 20.8 25.0 25.0

Ratio correct identifications/benchmark 4.1 3.2 4.1 4.1 4.9 4.9Aemilia Wiretap 12 5.8 Correct identifications (%) 8.3 8.3 8.3 8.3 8.3 8.3

Ratio correct identifications/benchmark 1.4 1.4 1.4 1.4 1.4 1.4Infinito Meeting 25 9.9 Correct identifications (%) 40.0 44.0 40.0 44.0 44.0 40.0

Ratio correct identifications/benchmark 4.0 4.4 4.0 4.4 4.4 4.0Infinito Wiretap 23 14.1 Correct identifications (%) 17.4 13.0 17.4 13.0 17.4 17.4

Ratio correct identifications/benchmark 1.2 0.9 1.2 0.9 1.2 1.2Crimine Meeting 15 4.0 Correct identifications (%) 20.0 20.0 20.0 20.0 26.7 33.3

Ratio correct identifications/benchmark 5.0 5.0 5.0 5.0 6.7 8.4Minotauro Wiretap 13 4.8 Correct identifications (%) 15.4 0.0 7.7 15.4 0.0 15.4

Ratio correct identifications/benchmark 3.2 0.0 1.6 3.2 0.0 3.2Best identifications in bold

Figure 2. Ratios between the Share of Leaders Correctly Predicted by Firth’s Logistic Regressions and the Share of Leaders Correctly Predicted by Random Probability (Benchmark) by Network

I II III IV V VI02468

101214161820

Aemilia MeetingCrimine MeetingInfinito MeetingMinotauro MeetingAemilia WiretapCrimine WiretapInfinito WiretapMinotauro Wiretap

The predictors of leadership showed different patterns between meeting and wiretap

networks. In meetings, centrality measures were statistically significant and positively associ-

ated with the probability of being a leader, particularly for betweenness. For Crimine and In-

finito, betweenness increased the probability of being a leader in all specifications, all vari-

ables constant (between +4.8% and +7.6% in Crimine and between +3.8% and +9.9% in In-

finito). In Aemilia and Minotauro, betweenness was statistically significant only for specifica-

tions I and III, possibly due to the small numbers of leaders and the collinearity among inde-

15

pendent variables.17 Also valued degree increased the probability of being a leader, except for

Crimine (specifications V and VI).

In wiretaps, centrality showed an unclear pattern. Betweenness often lost its predictive

capacity as soon as other variables were included. Only in half of the cases did degree central-

ity contribute to the identification of leaders. In fact, in most wiretap networks centrality mea-

sures and controls were strongly correlated (Pearson’s R often above a 0.850), resulting in

lowered reliability whenever multiple predictors entered the models.

3.3. Identifying “leaders” and “leaders and organizers”

Additional regressions tested the robustness of the results by identifying “leaders & organiz-

ers” to mitigate the problem of small number of “leaders”, especially in Aemilia.18 Table 5

compares specifications III to VI from Table 4 with four new specifications (VII to X) with

the same setup but “leaders & organizers” as the dependent variable. The new specifications

fit the data better than those for “leaders”, with improved Tjur’s’ R2 values. Except for Crim-

ine, also the share of correctly identified leaders increased, peaking at 63.6% for Aemilia,

58.5% for Infinito and 42.4% for Minotauro (Table 5). As the share of correct identifications

increased, so did the benchmark (i.e. probability of randomly identifying “leaders & organiz-

ers”). The higher benchmark offset the improved identification successes, and the ratios be-

tween correct identifications and the benchmark increased only for Aemilia, decreased for

Crimine, and Minotauro, and were stable for Infinito.

17 Aemilia comprised only six leaders, making it hard to achieve reliable estimations. Moreover, betweenness strongly correlated with the number of meetings (Pearson’s R 0.914), causing problems for collinearity in the specifications. Similarly, in Minotauro betweenness and meetings were closely associated (Pearson’s R 0.839) and the inclusion of meetings in specifications IV and VI made betweenness not statistically significant.18 This subsection presents results for meeting networks for reasons of simplicity. Exploration for wiretap networks yielded results similar to those presented in the previous subsection, with wiretaps models performing significantly worse than meetings.

16

Table 5. Firth’s Logistic Regression Results for “Leaders” and “Leaders & Organizers” for Meeting Networks Only, Odds RatiosNetwork Variables

Leaders Leaders & organizers

III IV V VI VII VIII IX X

Aemilia meetings

Betweenness 1.019* 0.973 0.995 0.977 1.062*** 1.061* 1.050* 1.063*Valued Degree 1.050** 1.036+ 1.01 1.018Mafia charge 21.49* 10.37 4.266 4.007 16.27+ 16.06+ 11.76 10.64

Meetings attended 1.204** 1.108 0.999 0.94Tjur’s R2 0.169 0.424 0.444 0.508 0.638 0.63 0.628 0.613

Predicted leaders (%) 0.0 33.3 16.7 50.0 63.6 63.6 54.5 54.5Predicted/Benchmark 0.0 12.6 6.3 18.9 13.1 13.1 11.3 11.3

Crimine meetings

Betweenness 1.048* 1.056+ 1.060* 1.058+ 1.055* 1.078* 1.058+ 1.070*Valued Degree 0.993 0.997 0.998 1.006Mafia charge 45.89*** 48.15*** 45.04*** 46.51*** 68.46*** 76.50*** 66.47*** 74.47***

Meetings attended 0.967 0.969 0.934 0.931Tjur’s R2 0.324 0.317 0.329 0.318 0.347 0.348 0.348 0.349

Predicted leaders (%) 20.8 20.8 25.0 25.0 22.6 22.6 22.6 22.6Predicted/Benchmark 4.1 4.1 4.9 4.9 3.4 3.4 3.4 3.4

Infinito meetings

Betweenness 1.079*** 1.038* 1.045** 1.042* 1.121*** 1.047+ 1.050** 1.050*Valued Degree 1.022** 1.020+ 1.035*** 1.036**Mafia charge 6.231* 4.737+ 3.659 3.7 8.356* 7.193* 4.399 4.326

Meetings attended 1.096* 1.011 1.141** 0.994Tjur’s R2 0.421 0.453 0.473 0.471 0.481 0.526 0.573 0.570

Predicted leaders (%) 40 44.0 44.0 40.0 47.1 55.9 58.8 58.8Predicted/Benchmark 3.5 4.4 4.4 4.0 3.5 4.1 4.4 4.4

Minotauro meetings

Betweenness 1.060*** 0.996 1.015 1.001 1.110*** 1.011 1.041* 1.019Valued Degree 1.010*** 1.008* 1.008*** 1.005+Mafia charge 9.936* 26.16+ 4.525 7.451 38.01*** 51.49* 24.84** 52.98*

Meetings attended 1.099** 1.031 1.118** 1.05Tjur’s R2 0.247 0.325 0.392 0.392 0.377 0.435 0.451 0.452

Predicted leaders (%) 20.0 20.0 26.7 33.3 21.2 33.3 42.4 39.4Predicted/Benchmark 5.0 5.0 6.7 8.4 2.4 3.8 4.8 4.5

+ The correlation is statistically significant at p≤0.1 level.* The correlation is statistically significant at p≤0.05 level.** The correlation is statistically significant at p≤0.01 level.*** The correlation is statistically significant at p≤0.001 level

Betweenness centrality remained statistically significant and positively associated with

the leadership. In particular, it improved in Aemilia, where leaders passed from 6 to 11, possi-

bly stabilizing the estimations and improving their reliability (betweenness always statistically

significant and increasing the probability of identifying a leader by 5%-6.3%, all other vari-

ables constant, specification X in Table 5). In Minotauro, betweenness was statistically signif-

icant only in specifications VII and IX. Whenever the model included the number of meet-

ings, betweenness lost statistical significance due to the already mentioned correlation with

the meetings. Valued degree centrality confirmed its inconsistent pattern: not statistically sig-

nificant in Aemilia and Crimine, and positive and statistically significant in Infinito and Mino-

tauro.

17

Overall, the models on “leaders” and “leaders & organizers” provided consistent re-

sults. In meetings, betweenness positively influenced the probability of being a leader even

when controlling for other variables. Contrarily, in wiretaps, betweenness often provided re-

dundant information. When focusing on “leaders & organizers”, the larger number of leaders

improved the predictive power of the models, enabling correct identification of between one

third and nearly two thirds of leaders and confirming the positive impact of betweenness.

4. Discussion

4.1. Criminal leaders’ brokerage in meetings

The ‘Ndrangheta leaders seem to prioritize security over efficiency by avoiding or minimizing

telephone calls and prioritizing indirect connectivity (Morselli et al., 2007). Indeed, direct

connections often involve a higher risk of detection and conviction (Baker & Faulkner, 1993;

Morselli, 2009a; Morselli, Masias, Crespo, & Laengle, 2013). Studies have shown that crimi-

nal leaders often avoid telephones for fear of detection (Kleemans et al., 2002; Agreste et al.,

2016). Leaders may often let middle-level members directly manage riskier tasks (Hofmann

& Gallupe, 2015). The more skilled and high-ranked individuals are indirectly involved in

criminal activities and that also non-leaders record high degree centrality (Baker & Faulkner,

1993; Calderoni, 2012; Morselli, 2009b, Chapter 8 and 9, 2009a).

However, criminals can hardly balance efficiency and security, as “covert networks

are not generally capable of trading efficiency for security” (Morselli et al., 2007, p. 151).

Leaders cannot entirely delegate criminal activities, owing to limited possibilities to control

subordinates (Paoli, 2002; Reuter & Haaga, 1989; Reuter & Rubinstein, 1982). Contrarily,

they often have to take direct action, especially in complex criminal activities requiring more

intense interaction (“high information-processing needs” Baker & Faulkner, 1993, p. 854;

“less time-to-task” Morselli et al., 2007, pp. 151–152). Consequently, direct and indirect con-

18

nectivity are not mutually exclusive; in fact, they are often correlated (Calderoni, 2012, p.

342; Morselli, 2009b, p. 18). Leaders often balance the high number of direct contacts with

high indirect connectivity (Morselli, 2010). This is the case also for the ‘Ndrangheta net-

works, where leaders reported higher degree and betweenness than non-leaders. Nevertheless,

the absence of one in five leaders from wiretaps and the lower ratios in the participation in

telephone calls avoidance suggest that they may resort to other, less visible, channels to con-

trol the criminal organizations. Results from the meeting networks support this argument.

Meetings may provide leaders with a way to balance security and efficiency. First, in

the four investigations, meetings often occurred in secure environment such as open-air spa-

ces or separate rooms in restaurants and bars. Leaders often prefer meetings in person to tele-

phone use due to lower risks of interception (Agreste et al., 2016; Calderoni, 2012; Kleemans

et al., 2002). In the ‘Ndrangheta networks, leaders participated in meetings 8 time more than

non-leaders (vs. only 5 times, and 23% of leaders were completely absent, in wiretaps). Sec-

ond, meetings enable direct interaction, more efficient than telephones in managing delicate

situations (e.g. murders, competition for offices and ranks).19 The mafias have a tradition of

formal summits for strategic planning and dispute settlement (Paoli, 2003, p. 59).20 Third,

mafia meetings often have social and cultural functions where leaders play an important role

(affiliation rituals, ceremonies and important celebrations) (Gambetta, 1993, Chapter 6; Paoli,

2003, Chapter 2). In the ‘Ndrangheta, these events strengthen the internal trust and the mem-

19 In Infinito the investigators filmed a ‘Ndrangheta meeting in a town nearby Milan, which took place on 31 October 2009 in a social club for elderly people named after Giovanni Falcone and Paolo Borsellino, two judges killed by Cosa Nostra in 1992. The meeting aimed at appointing the new Mastro generale. 23 men sat at a horseshoe-shaped table, eating dinner and discussing the appointment. All the leaders of the locali approved the candidate and subsequently celebrated with a toast.20 For example, a major event in the ‘Ndrangheta is the annual meeting at the Sanctuary of Santa Maria di Polsi, a mountain town in Calabria (Paoli, 2003, p. 53). In Operation Crimine, the investigation recorded for first time the discussions at the Polsi meeting of September 2010, where more than 300 affiliates were present. Only highest-ranked individuals from locali around the world participate discussing internal organization, activities, and awarding new high ranks. Crimine was the first time the Italian authorities has managed to monitor the Polsi meeting, showing how carefully managed meetings may provide better protection from investigations than telephone conversations.

19

bers’ sense of belonging (Sergi & Lavorgna, 2016; Paoli, 2003; Simmel, 1950, Chapter 4).21

Last, criminal leaders do not attend just more meetings than non-leaders: their outstanding be-

tweenness scores highlight a pattern typical of brokers.

The ‘Ndrangheta leaders can effectively broker across otherwise disconnected people.

Studies on co-offending networks have shown that non-redundant contacts provide higher

earnings (Morselli & Tremblay, 2004) and more offending versatility (McGloin & Piquero,

2010). Brokering skill is a key factor in the success of career criminals in the drug trade

(Morselli, 2001), in the American Mafia (Morselli, 2003), and in Dutch organized crime

(Kleemans & de Poot, 2008, pp. 82–92). Leaders’ high brokerage yields not only security but

also closer control over the criminal activities (Morselli, 2009b, Chapter 8 and 9). The evi-

dence on the benefits of brokerage for criminals supports the general claim by social sciences

that “brokers do better” (Burt, 2007, p. 7; Morselli, 2009a, p. 16): brokers are higher-achiev-

ers and well-respected (Boissevain, 1974), able to exploit new opportunities (Granovetter,

1973), and generally advantaged “in terms of breadth of knowledge, early knowledge, and op-

portunities for strategically coordinating across groups” (Burt, 2007, p. 55).

The brokerage in meetings also enables leaders to fulfil social obligations deriving

from their role. In the ‘Ndrangheta, participation in specific meetings is a prerogative of lead-

ers, who are both willing and obliged to participate (Simmel, 1950, Chapter 4). This concerns

both high-level criminal meetings and social occasions. Presence at these events is a sign of

high status and prestige, and intrinsically associated with leadership (Kleemans et al., 2002;

21 In Infinito leaders of the locali in Lombardy were invited to the wedding between the children of two powerful ‘Ndrangheta dynasties, the Pelle and the Barbaro on 19 August 2009. This was a major event for the ‘Ndrangheta, and Infinito leaders discussed participation and presents at length. Invitations reflected the status and power of a locale, whereas non-invitation showed its weakness.

Furthermore, the largest recorded event during Operation Minotauro was a funeral held on 3 January 2009 in Settimo Torinese. The service concerned the caposocietà (deputy boss) of the locale of Natile di Careri, near Turin, murdered in an attack a few days earlier because of disputes internal to the ‘Ndrangheta. The ceremony was attended by dozens of criminals, who gathered to pay their respects to the deceased boss. Participation in the funeral represented an opportunity to demonstrate criminal standing and reinforce a group cohesion in difficult times.

20

Paoli, 2002). Failure to participate would miss important decisions, spread gossip, affect the

reputation, and show disrespect. Therefore, meeting participation may often represent a neces-

sity due to the social embeddedness of the ‘Ndrangheta (Kleemans & Van De Bunt, 1999).

Organized criminals are embedded in a network of personal relations providing unequal ac-

cess to information and people. Successful organized crime careers benefit from the “social

opportunity structure”, i.e. “social ties providing access to profitable criminal opportunities”

(Kleemans & de Poot, 2008, p. 75).

These considerations may be extended to other criminal groups. The four ‘Ndrangheta

groups revealed remarkably similar patterns, despite being based in different regions and in-

vestigated by separate prosecutor’s offices. The internal functioning of the ‘Ndrangheta may

be similar in countries like Australia, Germany, Canada, and the Netherlands (Calderoni et al.,

2016; KLPD, 2011; Sciarrone & Storti, 2014; Sergi, 2015, 2018). Furthermore, these patterns

may be common to the leaders of different groups, ranging from other mafias to drug traffick-

ing organizations, organized street gangs and terrorist groups (Decker & Pyrooz, 2015). First,

most criminal groups face the efficiency/security trade-off (D. Bright, Koskinen, & Malm,

2018; Morselli et al., 2007; Simmel, 1950, Chapter 4; Waniek, Michalak, Rahwan, &

Wooldridge, 2017). Security may discourage telephone use and favor more secure communi-

cations strategies. Meetings may provide more secure communication within most criminal

groups. It is surprising that the literature has rarely analyzed meetings, and has never com-

pared meeting with wiretap or other networks. To the best of our knowledge, this is the first

study systematically comparing meetings and wiretaps across four cases of the same criminal

organization. Second, brokerage and social embeddedness provide advantages to most social

groups (Granovetter, 1973; Boissevain, 1974; Burt, 2007). Brokerage benefits criminal lead-

ers in the mafias (DellaPosta, 2017; Lo, 2010; Papachristos & Smith, 2014; Wang, 2017),

drug trafficking organizations (Dorn, Murji, & South, 1992; Pearson & Hobbs, 2001; Zaitch,

21

2002; Decker & Townsend Chapman, 2008), terrorist organizations (Demiroz & Kapucu,

2012; Krebs, 2002), and street gangs (Benson & Decker, 2010; Morselli, 2009b, Chapter 9).

4.2. The identification of leaders

While indicative of the advantages of brokerage, leaders’ participation in meetings can make

them vulnerable to identification. The regression on meetings identify leaders more effec-

tively than wiretaps across all four networks (Table 4 and Figure 2), corroborating previous

works (Calderoni, 2014, 2015).

The findings are also robust to alternative definitions of leaders (Table 5). Logistic re-

gressions on “leaders & organizers” perform similarly to, or even better than, those on “lead-

ers”. These results are due to both statistical and organizational reasons. First, regressions on

“leaders & organizers” enlarge the number of leaders in the logistic regressions, notoriously

affected by few cases in one category of the dependent variable. While Firth’s method par-

tially addresses this problem, the share of “leaders” is still particularly low (around 6% of the

sample), making the estimations difficult.22 “Leaders & organizers” increase the numbers in

the category to about 11%, thus improving the performance of the regressions.23 Second, in

the organizational structure of the ‘Ndrangheta, the power of brokerage is shared by different

office-holders. High betweenness centrality is typical not only of the boss and the deputy boss

of a locale (the only individuals in “leaders”), but also of other criminals holding important

offices (in “leaders & organizers”). Thus betweenness captures the social capital of the most

senior and powerful affiliates (Paoli, 2003, pp. 49–51).

While consistent across the four networks and robust to the variation of the leadership

classification, the regressions correctly classify only some leaders, with some false positives

22 In particular, “leaders” numbered six (about 3% of the total) in Aemilia. Also in Minotauro, leaders numbered fifteen in the meeting network (4% of the total) and thirteen in the wiretap network (4.8%), respectively.23 Unsurprisingly, Aemilia reports the best improvement, as the number of “leaders & organizers” (n=11) nearly doubles the number of “leaders”.

22

(non-leaders wrongly classified as leaders) and negatives (leaders mistaken for non-leaders).

This may be due to the secrecy of organized crime activities and the consequent incomplete-

ness of investigative sources. At the same time, it may also reveal the actual distribution of

brokerage capacities across criminals. As in most social groups, formal positions and actual

skills only partially overlap. Few leaders may hold an office despite limited brokerage in the

meeting networks, due to personal conditions (e.g. incapacitated, temporarily arrested) or in-

ternal dynamics (diminished prestige, disputes). These situations may signal the crisis of the

leadership within a criminal network (Morselli & Petit, 2007). Conversely, criminals with no

office may record high brokerage, e.g. younger, ambitious criminals making their way into

the higher offices, or former leaders dismissed from due to internal dynamics (Morselli,

2009b, p. 158).

These considerations are also relevant when considering the risk of reverse causality

in our models, i.e. that leadership determines network position. While the cross-sectional ap-

proach of this study cannot rule out reverse causality by design, there are reasons to contend

that mafia leaders’ social connections enabled them to secure a ‘Ndrangheta office rather than

this driving their links. The literature on brokerage in social networks often points out that so-

cial relations drive success and leadership (Watts, 2004; Burt, 2007). Criminal leadership po-

sitions are due not only to individuals’ characteristics but also to the structure of their per-

sonal networks (Kleemans & de Poot, 2008, p. 91; Kleemans & Van Koppen, 2014, p. 291),

confirming that brokering roles often “describe the consequences of position in social struc-

ture, not the goals that led incumbents to the position” (Burt, 2007, pp. 241, footnote 13) and

that, for criminals, “rank […] is but a prize for knowing how to fit in amongst others”

(Morselli, 2009b, p. 138).24 Lastly, a reverse interpretation could hardly explain the hetero-

geneity in leaders’ centrality observed across the four networks.

24 Morselli described how repeated drug seizures affected N1, the group leader’s position (2009b, Chapter 6). N1’s centrality scores decreased as the law enforcement agencies intercepted more drug consignments, while other criminals organized new shipments, resulting in increased centrality.

23

The identification of criminal leaders through meeting networks contributes to the

growing literature on the identification of important actors and network disruption strategies

(Bouchard & Konarski, 2014; D. A. Bright, Greenhill, & Levenkova, 2014; D. A. Bright,

Greenhill, Reynolds, Ritter, & Morselli, 2015; Calderoni, 2012; Carley et al., 2002; Duijn,

Kashirin, & Sloot, 2014; Morselli, 2010; Morselli & Roy, 2008; Robins, 2009; Schwartz &

Rouselle, 2009). Some scholars warn that an excessive focus on network topology (e.g. cen-

trality measures) may be misleading: actors in dark networks are actively trying to hide their

activities (Carley et al., 2002; Robins, 2009); centrality may omit actor-level attributes, like

specific skills, expertise, and demographic factors (D. A. Bright et al., 2015; Schwartz & Rou-

selle, 2009). These works propose integration of network measures with measures of individ-

uals’ attributes. Nevertheless, while offering valuable approaches to the identification of lead-

ers, such methods inevitably imply some degree of discretion in the operationalization of indi-

vidual characteristics; scores and weights are inevitably network-specific and may hardly ap-

ply to other criminal groups without substantial adaptations. Lastly, a study comparing differ-

ent disruption methods (prioritizing both social and human capital) concluded that targeting

based on betweenness was the most efficient strategy (D. Bright, Greenhill, Britz, Ritter, &

Morselli, 2017).

Our study provides a different approach, taking into account the risks of overreliance

on network topology and still avoiding excessively discretionary manipulations. On the one

hand, the results confirmed that overreliance on wiretap data may indeed jeopardize the effec-

tive identification of leaders. Contrarily, meeting data reveal the importance of meeting par-

ticipation for the brokering and social functions of criminal leaders. High indirect connectiv-

ity in meetings may synthesize individuals’ social capital, thus making “the people bridging

structural holes a productive target for destabilizing a criminal network” (Burt, 2007, p. 87).

On the other hand, leader identification through logistic regressions and betweenness central-

24

ity is a parsimonious method, without discretion about the scoring and weighting of individu-

als’ attributes. As such, it could be easily replicated to identify leaders within other criminal

groups.

The capacity to effectively identify leaders from meeting networks may have impor-

tant policy implications for law enforcement agencies and prosecutors, who are increasingly

integrating social network analysis into strategic and tactical intelligence (Duijn & Klerks,

2014). Network analysis is often used to identify possible additional investigative paths and

persons of interest (Burcher & Whelan, 2017). First, the analysis of meeting networks may

yield useful insights during the early stages of an investigation, enabling better focusing of

limited surveillance resources against potential leaders. The content of telecommunications

provides detailed information on the networks, but wiretapping is expensive and follows spe-

cific requirements by courts. Law enforcement agencies can more easily control individuals’

movements than their communications, making data on meetings is more easily accessible;

the analysis of meetings may uncover leaders avoiding participation in telephone calls and

thus uncover patterns unknown to the authorities. Second, when available in national legisla-

tion, preventive or administrative orders may prohibit leaders from meeting each other, which

may ultimately jeopardize the internal organization of criminal networks (Calderoni, 2015).

Third, leaders with high-brokering characteristics may be the target for arrests, although re-

search showed these operations may impact differently on the levels of violence depending on

the criminal group embeddedness in the social relations with other groups (Vargas, 2014;

Calderón, Robles, Díaz-Cayeros, & Magaloni, 2015; Vargas, 2016).

4.3. Limitations

In addition to the limitations discussed in the previous subsections, the study may suf-

fer from network incompleteness and non-random sampling. First, network completeness and

boundary definition may be relevant for studies relying on judicial sources, as the investiga-

25

tions may have missed important nodes or links biasing the boundaries of the observed net-

work (Sparrow, 1991; Morselli & Petit, 2007; Morselli, 2009b; Campana & Varese, 2011;

Faust & Tita, 2019). However, the investigations analyzed here were long and carefully built

cases, specifically intended to reconstruct the internal organizations of the ‘Ndrangheta. Trials

for Infinito, Crimine, and Minotauro have confirmed the charges, finally convicting nearly all

defendants. In October 2018, the third grade trial for Aemilia also confirmed the charges

against a group of defendants, while a first grade trial upheld the charges against the others.

Consequently, the risk of omissions that may have radically biased the networks appears low.

Furthermore, several studies reject the concerns about network completeness, arguing that,

under conditions, judicial sources are reliable for the analysis of criminal groups (Campana &

Varese, 2011; Morselli, 2009b, Chapter 5); that network measures are reliable even if ran-

domly tested for missing data (Berlusconi, Calderoni, Parolini, Verani, & Piccardi, 2016; Xu

& Chen, 2003); and that network analysis may even contribute to the identification of missing

links (Berlusconi et al., 2016; Chen, Gao, Karampelas, Alhajj, & Rokne, 2011; Clauset,

Moore, & Newman, 2008; Rhodes & Jones, 2008).

Second, investigative data provide non-random samples possibly biased by the investi-

gators’ strategies, who may concentrate on a few individuals, e.g. those initially believed to be

high-profile criminals and/or the more visible targets, who may result as the leaders of the

criminal networks. The research design carefully addressed these possible biases, by identify-

ing leaders relying only through information independent from either investigators’ judge-

ments or network topology. The majority of leaders were identified from direct evidence pro-

vided by the suspects during intercepted conversations. Furthermore, logistic regressions in-

cluded events participation and criminal charges for the main offence, thus controlling for

possible biases against high-profile targets. Lastly, results of logistic regressions are consis-

tent across the four criminal networks and robust to changes in the leader classification.

26

5. Conclusions

Previous research on organized crime networks has frequently relied on data based on tele-

phone communications. However, wiretaps may bias the analysis of criminal groups as crimi-

nal leaders often minimize telephone use in favor of other communication channels. Among

these, meetings may provide a valuable alternative to assess the nature of criminal leadership,

offering insight into the security strategies and the brokering roles of the leaders and the so-

cial embeddedness of organized crime.

This study examined the nature of criminal leadership examining leaders’ participation

in meeting and telephone communications, and comparing meeting and wiretap networks in

identifying leaders. The results showed that leaders minimize telephone use and prefer indi-

rect connectivity, particularly in meeting networks. Through meetings, not only do leaders en-

sure more secure management of the criminal activities, but they also achieve strategic bro-

kering positions, bridging across unconnected individuals. Besides these advantages, meetings

are also social and cultural obligations, hardly avoidable for security concerns. Indeed, logis-

tic regressions on meeting networks identified a relevant share of criminal leaders by relying

solely on centrality measures. Particularly, high betweenness centrality, a measure of an indi-

vidual’s brokerage, substantially increases the probability of being a criminal leader.

The research can be extended in many directions. First, future studies might replicate

the analyses to test the external validity of the results further, by focusing on other groups,

such as street gangs, drug trafficking organizations, and terrorist groups. The extent to which

these replications will be possible will depend largely on the quality of the data on meeting

participation (Campana & Varese, 2011). Second, further analyses could focus on meeting at-

tendance patterns within groups spread across different regions or countries, and see whether

this constraint obliges leaders to resort to telephone calls because meetings may be more diffi-

cult to arrange. Third, more operational applications might test whether the identification of

27

leaders in several networks is reliable at early stages of the investigation (Calderoni, 2015),

paving the way for the support of law enforcement agencies in the timely identification of the

best target for network disruption (Bichler & Malm, 2015).

In conclusion, the findings complement the previous literature in demonstrating crimi-

nals’ constant struggle to balance security and efficiency, and the capacity of betweenness

centrality to capture leaders’ brokering roles in meetings. By focusing on meetings, the study

indicates a new source of information which can improve knowledge on the behavior of crim-

inal leaders and on the social structure of criminal groups. Furthermore, the identification of

criminal leaders through network analysis of their meeting participations may yield remark-

able benefits to investigations and intelligence.

28

References

Agreste, S., Catanese, S., De Meo, P., Ferrara, E., & Fiumara, G. (2016). Network Structure

and Resilience of Mafia Syndicates. Information Sciences, 351(C), 30–47.

Allison, P. (2013). What’s the Best R-Squared for Logistic Regression? Retrieved May 31,

2016, from http://statisticalhorizons.com/r2logistic

Baker, W. E., & Faulkner, R. R. (1993). The Social Organization of Conspiracy: Illegal Net-

works in the Heavy Electrical Equipment Industry. American Sociological Review,

58(6), 837–860. https://doi.org/10.2307/2095954

Benson, J. S., & Decker, S. H. (2010). The organizational structure of international drug

smuggling. Journal of Criminal Justice, 38(2), 130–138.

Berlusconi, G. (2013). Do all the pieces matter? Assessing the reliability of law enforcement

data sources for the network analysis of wire taps. Global Crime, 14(1), 61–81.

https://doi.org/10.1080/17440572.2012.746940

Berlusconi, G., Calderoni, F., Parolini, N., Verani, M., & Piccardi, C. (2016). Link Prediction

in Criminal Networks: A Tool for Criminal Intelligence Analysis. PLOS ONE, 11(4),

e0154244. https://doi.org/10.1371/journal.pone.0154244

Bichler, G., & Malm, A. E. (Eds.). (2015). Disrupting Criminal Networks: Network Analysis

in Crime Prevention. Boulder: Lynne Rienner Publishers.

Boissevain, J. (1974). Friends of Friends: Networks, Manipulators and Coalitions. Oxford:

Basil Blackwell.

Borgatti, S. P., Everett, M. G., & Freeman, L. C. (2002). Ucinet 6 for Windows: Software for

Social Network Analysis. Harvard: Analytic Technologies.

Bouchard, M., & Konarski, R. (2014). Assessing the core membership of a youth gang from

its co-offending network. In C. Morselli (Ed.), Crime and Networks (pp. 81–93). New

York - London: Routledge.

29

Bright, D. A., Greenhill, C., & Levenkova, N. (2014). Dismantling criminal networks: Can

node attributes play a role? In C. Morselli (Ed.), Crime and Networks (pp. 148–162).

New York - London: Routledge.

Bright, D. A., Greenhill, C., Reynolds, M., Ritter, A., & Morselli, C. (2015). The Use of Ac-

tor-Level Attributes and Centrality Measures to Identify Key Actors: A Case Study of

an Australian Drug Trafficking Network. Journal of Contemporary Criminal Justice,

31(3), 262–278. https://doi.org/10.1177/1043986214553378

Bright, D. A., Hughes, C. E., & Chalmers, J. (2012). Illuminating dark networks: a social net-

work analysis of an Australian drug trafficking syndicate. Crime, Law and Social

Change, 57(2), 151–176. https://doi.org/10.1007/s10611-011-9336-z

Bright, D., Greenhill, C., Britz, T., Ritter, A., & Morselli, C. (2017). Criminal network vul-

nerabilities and adaptations. Global Crime, 18(4), 424–441. https://doi.org/

10.1080/17440572.2017.1377614

Bright, D., Koskinen, J., & Malm, A. (2018). Illicit Network Dynamics: The Formation and

Evolution of a Drug Trafficking Network. Journal of Quantitative Criminology, 1–22.

https://doi.org/10.1007/s10940-018-9379-8

Burcher, M., & Whelan, C. (2017). Social network analysis as a tool for criminal intelligence:

understanding its potential from the perspectives of intelligence analysts. Trends in

Organized Crime, 1–17. https://doi.org/10.1007/s12117-017-9313-8

Burt, R. S. (1980). Models of Network Structure. Annual Review of Sociology, 6(1), 79–141.

Burt, R. S. (1992). Structural Holes: The social structure of competition. Cambridge, MA:

Harvard University Press.

Burt, R. S. (2007). Brokerage and Closure: An Introduction to Social Capital. Oxford: Ox-

ford University Press.

30

Calderón, G., Robles, G., Díaz-Cayeros, A., & Magaloni, B. (2015). The Beheading of Crimi-

nal Organizations and the Dynamics of Violence in Mexico. Journal of Conflict Reso-

lution, 59(8), 1455–1485. https://doi.org/10.1177/0022002715587053

Calderoni, F. (2012). The structure of drug trafficking mafias: the ‘Ndrangheta and cocaine.

Crime, Law and Social Change, 58(3), 321–349. https://doi.org/10.1007/s10611-012-

9387-9

Calderoni, F. (2014). Identifying Mafia Bosses from Meeting Attendance. In A. J. Masys

(Ed.), Networks and Network Analysis for Defence and Security (pp. 27–48). Cham:

Springer International Publishing Switzerland.

Calderoni, F. (2015). Predicting Organized Crime Leaders. In G. Bichler & A. E. Malm

(Eds.), Disrupting Criminal Networks: Network Analysis in Crime Prevention (pp. 89–

110). Boulder, CO: Lynne Rienner Publishers.

Calderoni, F., Berlusconi, G., Garofalo, L., Giommoni, L., & Sarno, F. (2016). The Italian

mafias in the world: A systematic assessment of the mobility of criminal groups. Eu-

ropean Journal of Criminology, 13(4), 413–433. https://doi.org/

10.1177/1477370815623570

Campana, P. (2013). Understanding Then Responding to Italian Organized Crime Operations

across Territories. Policing, 7(3), 316–325. https://doi.org/10.1093/police/pat012

Campana, P., & Varese, F. (2011). Listening to the wire: criteria and techniques for the quan-

titative analysis of phone intercepts. Trends in Organized Crime, 15(1), 13–30.

Carley, K. M., Krackhardt, D., & Lee, J.-S. (2002). Destabilizing networks. Connections,

24(3), 79–92.

Chen, A., Gao, S., Karampelas, P., Alhajj, R., & Rokne, J. (2011). Finding Hidden Links in

Terrorist Networks by Checking Indirect Links of Different Sub-Networks. In U. K.

31

Wiil (Ed.), Counterterrorism and Open Source Intelligence (pp. 143–158). Springer

Vienna.

Clauset, A., Moore, C., & Newman, M. E. J. (2008). Hierarchical structure and the prediction

of missing links in networks. Nature, 453(7191), 98–101. https://doi.org/10.1038/na-

ture06830

Coleman, J. S. (1988). Social capital in the creation of human capital. American Journal of

Sociology, 94, 95–120.

Decker, S. H., & Pyrooz, D. C. (2015). Street Gangs, Terrorists, Drug Smugglers, and Orga-

nized Crime: What’s the Difference? In S. H. Decker & D. C. Pyrooz (Eds.), The

Handbook of Gangs (pp. 294–308). Chichester: Wiley.

Decker, S. H., & Townsend Chapman, M. (2008). Drug Smugglers on Drug Smuggling:

Lessons from the Inside. Philadelphia: Temple University Press.

DellaPosta, D. (2017). Network closure and integration in the mid-20th century American

mafia. Social Networks, 51, 148–157. https://doi.org/10.1016/j.socnet.2016.11.005

Demiroz, F., & Kapucu, N. (2012). Anatomy of a dark network: the case of the Turkish Er-

genekon terrorist organization. Trends in Organized Crime, 15(4), 271–295. https://

doi.org/10.1007/s12117-012-9151-7

Dorn, N., Murji, K., & South, N. (Eds.). (1992). Traffickers: Drug Markets and Law Enforce-

ment. London: Routledge.

Duijn, P. A. C., Kashirin, V., & Sloot, P. M. A. (2014). The Relative Ineffectiveness of Crim-

inal Network Disruption. Scientific Reports, 4. https://doi.org/10.1038/srep04238

Duijn, P. A. C., & Klerks, P. (2014). Social network analysis applied to criminal networks:

Recent developments in Dutch law enforcement. In A. Masys (Ed.), Networks and

Network Analysis for Defence and Security. Cham: Springer International Publishing.

32

Faust, K., & Tita, G. E. (2019). Social Networks and Crime: Pitfalls and Promises for Ad-

vancing the Field. Annual Review of Criminology, 2(1), 99–122. https://doi.org/

10.1146/annurev-criminol-011518-024701

Firth, D. (1993). Bias reduction of maximum likelihood estimates. Biometrika, 80(1), 27–38.

https://doi.org/10.1093/biomet/80.1.27

Freeman, L. C. (1979). Centrality in social networks conceptual clarification. Social Net-

works, 1(3), 215–239.

Gambetta, D. (1993). The Sicilian Mafia: the business of private protection. Cambridge: Har-

vard University Press.

Granovetter, M. S. (1973). The Strength of Weak Ties. American Journal of Sociology, 78(6),

1360–1380.

Granovetter, M. S. (1983). The Strength of Weak Ties: A Network Theory Revisited. Socio-

logical Theory, 1(1), 201–233.

Hofmann, D. C., & Gallupe, O. (2015). Leadership protection in drug-trafficking networks.

Global Crime, 16(2), 123–138. https://doi.org/10.1080/17440572.2015.1008627

Kleemans, E. R. (2014). Theoretical perspectives on organized crime. In L. Paoli (Ed.), The

Oxford Handbook of Organized Crime (pp. 32–52). London: Oxford University Press.

Retrieved from http://www.oxfordhandbooks.com/view/10.1093/oxfordhb/

9780199730445.001.0001/oxfordhb-9780199730445-e-025

Kleemans, E. R., Brienen, M. E. I., & Van De Bunt, H. (2002). Dutch Organized Crime Mon-

itor: Georganiseerde Criminaliteit in Nederland. Tweede Reportage Op Basis van de

WODC-Monitor. Den Haag: WODC.

Kleemans, E. R., & de Poot, C. J. (2008). Criminal Careers in Organized Crime and Social

Opportunity Structure. European Journal of Criminology, 5(1), 69–98.

33

Kleemans, E. R., & Van De Bunt, H. (1999). The social embeddedness of organized crime.

Transnational Organized Crime, 5(1), 19–36.

Kleemans, E. R., & Van Koppen, M. V. (2014). Careers in Organized Crime. In G. Bruinsma

& D. Weisburd (Eds.), Encyclopedia of Criminology and Criminal Justice (pp. 285–

295). Springer New York. https://doi.org/10.1007/978-1-4614-5690-2_321

KLPD. (2011). The ’Ndrangheta in the Netherlands: The nature, criminal activities and modi

operandi on Dutch territory. Amsterdam: National Crime Squad National - Police

Agency of the Netherlands.

Krebs, V. E. (2002). Mapping networks of terrorist cells. Connections, 24(3), 43–52.

Lo, T. W. (2010). Beyond Social Capital: Triad Organized Crime in Hong Kong and China.

The British Journal of Criminology, 50(5), 851–872. https://doi.org/10.1093/bjc/

azq022

Malm, A. E., Bichler, G., & Van De Walle, S. (2010). Comparing the ties that bind criminal

networks: Is blood thicker than water? Security Journal, 23(1), 52–74.

Mancuso, M. (2014). Not all madams have a central role: analysis of a Nigerian sex traffick-

ing network. Trends in Organized Crime, 17(1–2), 66–88. https://doi.org/10.1007/

s12117-013-9199-z

McGloin, J. M., & Piquero, A. R. (2010). On the Relationship between Co-Offending Net-

work Redundancy and Offending Versatility. Journal of Research in Crime and

Delinquency, 47(1), 63–90. https://doi.org/10.1177/0022427809348905

Morselli, C. (2001). Structuring Mr. Nice: entrepreneurial opportunities and brokerage posi-

tioning in the cannabis trade. Crime, Law and Social Change, 35(3), 203–244.

Morselli, C. (2003). Career opportunities and network-based privileges in the Cosa Nostra.

Crime, Law and Social Change, 39(4), 383–418.

34

Morselli, C. (2009a). Hells Angels in Springtime. Trends in Organized Crime, 12(2), 145–

158.

Morselli, C. (2009b). Inside Criminal Networks. New York: Springer.

Morselli, C. (2010). Assessing Vulnerable and Strategic Positions in a Criminal Network.

Journal of Contemporary Criminal Justice, 26(4), 382–392. https://doi.org/

10.1177/1043986210377105

Morselli, C., Giguère, C., & Petit, K. (2007). The efficiency/security trade-off in criminal net-

works. Social Networks, 29(1), 143–153. https://doi.org/10.1016/j.socnet.2006.05.001

Morselli, C., Masias, V. H., Crespo, F., & Laengle, S. (2013). Predicting sentencing outcomes

with centrality measures. Security Informatics, 2(1), 4. https://doi.org/10.1186/2190-

8532-2-4

Morselli, C., & Petit, K. (2007). Law-Enforcement Disruption of a Drug Importation Net-

work. Global Crime, 8(2), 109–130. https://doi.org/10.1080/17440570701362208

Morselli, C., & Roy, J. (2008). Brokerage Qualifications in Ringing Operations. Criminology,

46(1), 71–98.

Morselli, C., & Tremblay, P. (2004). Criminal Achievement, Offender Networks and the Ben-

efits of Low Self-Control. Criminology, 42(3), 773–804. https://doi.org/10.1111/

j.1745-9125.2004.tb00536.x

Natarajan, M. (2006). Understanding the Structure of a Large Heroin Distribution Network: A

Quantitative Analysis of Qualitative Data. Journal of Quantitative Criminology, 22(2),

171–192. https://doi.org/10.1007/s10940-006-9007-x

Paoli, L. (1994). An Underestimated Criminal Phenomenon: The Calabrian ’Ndrangheta. Eu-

ropean Journal of Crime, Criminal Law and Criminal Justice, 2(3), 212–238.

Paoli, L. (2002). The paradoxes of organized crime. Crime, Law and Social Change, 37(1),

51–97.

35

Paoli, L. (2003). Mafia brotherhoods: Organized crime, Italian style. Oxford: Oxford Univer-

sity Press.

Papachristos, A. V., & Smith, C. M. (2014). The Embedded and Multiplex Nature of Al

Capone. In C. Morselli (Ed.), Crime and Networks (pp. 97–115). New York: Rout-

ledge.

Pearson, G., & Hobbs, D. (2001). Middle Market Drug Distribution (Home Office Research

Study No. 227). London: Home Office Research, Development and Statistics

Directorate.

Procura della Repubblica presso il Tribunale di Reggio Calabria. Decreto di fermo di indiziato

di delitto - artt. 384 e ss. c.p.p. (1389/2008 R.G.N.R. D.D.A., Operazione Crimine),

No. 1389/2008 R.G.N.R. D.D.A. (Procura della Repubblica Presso il Tribunale di

Reggio Calabria luglio 2010).

Reuter, P. (1983). Disorganized crime: the economics of the visible hand. Cambridge: MIT

Press.

Reuter, P., & Haaga, J. (1989). The Organization of High-Level Drug Markets: An Ex-

ploratory Study. Santa Monica: Rand.

Reuter, P., & Rubinstein, J. (1982). Illegal Gambling in New York: a case study in the opera-

tion, structure, and regulation of an illegal market. Washington D.C.: U.S. Depart-

ment of Justice, National Institute of Justice.

Rhodes, C. J., & Jones, P. (2008). Inferring missing links in partially observed social net-

works. Journal of the Operational Research Society, 60(10), 1373–1383. https://

doi.org/10.1057/jors.2008.110

Robins, G. (2009). Understanding individual behaviors within covert networks: the interplay

of individual qualities, psychological predispositions, and network effects. Trends in

Organized Crime, 12(2), 166–187.

36

Rostami, A., & Mondani, H. (2015). The Complexity of Crime Network Data: A Case Study

of Its Consequences for Crime Control and the Study of Networks. PLOS ONE, 10(3),

e0119309. https://doi.org/10.1371/journal.pone.0119309

Scaglione, A. (2011). Reti Mafiose: Cosa Nostra e Camorra: organizzazioni criminali a

confronto. Milano: FrancoAngeli.

Schwartz, D. M., & Rouselle, T. (2009). Using social network analysis to target criminal net-

works. Trends in Organized Crime, 12(2), 188–207.

Sciarrone, R., & Storti, L. (2014). The territorial expansion of mafia-type organized crime.

The case of the Italian mafia in Germany. Crime, Law and Social Change, 61(1), 37–

60. https://doi.org/10.1007/s10611-013-9473-7

Serenata, N. (Ed.). (2014). The ’Ndrangheta and Sacra Corona Unita - The History, |

Nicoletta Serenata | Springer. New York: Springer.

Sergi, A. (2015). The Evolution of the Australian ‘Ndrangheta. An Historical Perspective.

Australian & New Zealand Journal of Criminology, 48(2), 155–174. https://doi.org/

doi:10.1177/0004865814554305

Sergi, A. (2018). Polycephalous ‘ndrangheta: Crimes, behaviours and organisation of the Cal-

abrian mafia in Australia. Australian & New Zealand Journal of Criminology. https://

doi.org/10.1177/0004865818782573

Sergi, A., & Lavorgna, A. (2016). ‘Ndrangheta: The Glocal Dimensions of the Most Powerful

Italian Mafia. Cham: Springer International Publishing.

Simmel, G. (1950). The Sociology of Georg Simmel. (K. H. Wolf, Trans.). New York: The

Free Press.

Sparrow, M. K. (1991). Network vulnerabilities and strategic intelligence in law enforcement.

International Journal of Intelligence and Counter Intelligence, 5(3), 255–274.

37

Tjur, T. (2009). Coefficients of Determination in Logistic Regression Models—A New Pro-

posal: The Coefficient of Discrimination. The American Statistician, 63(4), 366–372.

https://doi.org/10.1198/tast.2009.08210

Transcrime. (2013). Progetto PON Sicurezza 2007-2013: Gli investimenti delle mafie.

Rapporto Linea 1. Milano: Ministero dell’Interno. Retrieved from

www.investimentioc.it

Tribunale di Bologna. Ordinanza di applicazione di misure cautelari coercitive - artt. 272 e ss.

284, 285 c.p.p. (N.20604-10 RNR DDA N. 17375-11 R GIP, Operazione Aemilia),

No. N.20604-10 RNR DDA N. 17375-11 R GIP (Sezione dei giudici per le indagini

preliminari e l’udienza preliminare gennaio 2015).

Tribunale di Milano. Ordinanza di applicazione di misura coercitiva con mandato di cattura -

art. 292 c.p.p. (N.43733/06 R.G.N.R. N. 8265/06 R.G.G.I.P., Operazione Infinito), No.

N.43733/06 R.G.N.R. N. 8265/06 R.G.G.I.P. (Ufficio del giudice per le indagini

preliminari luglio 2010).

Tribunale di Torino. Ordinanza di applicazione di misura coercitiva con mandato di cattura -

art. 292 c.p.p. (N.6191/07 + 9689/08 R.G. notizie di reato N. 5418/07 + 4775/09 R.G.

G.I.P., Operazione Minotauro), No. N.6191/07 + 9689/08 R.G. notizie di reato N.

5418/07 + 4775/09 R.G. G.I.P. (Sezione dei giudici per le indagini preliminari maggio

2011).

Varese, F. (2006a). How Mafias Migrate: The Case of the ’Ndrangheta in Northern Italy. Law

& Society Review, 40(2), 411–444.

Varese, F. (2006b). The Structure of a Criminal Network Examined: The Russian Mafia in

Rome. Oxford Legal Studies Research Paper, 21.

38

Vargas, R. (2014). Criminal Group Embeddedness and the Adverse Effects of Arresting a

Gang’s Leader: A Comparative Case Study. Criminology, 52(2), 143–168. https://

doi.org/10.1111/1745-9125.12033

Vargas, R. (2016). Wounded City. Violent turf wars in a Chicago barrio. Oxford: Oxford Uni-

versity Press.

Wang, P. (2017). The Chinese Mafia: Organized Crime, Corruption, and Extra-Legal Protec-

tion. Oxford: Oxford University Press.

Waniek, M., Michalak, T. P., Rahwan, T., & Wooldridge, M. (2017). On the Construction of

Covert Networks. In Proceedings of the 16th Conference on Autonomous Agents and

MultiAgent Systems (AAMAS 2017) (pp. 1341–1349). Richland, SC: International

Foundation for Autonomous Agents and Multiagent Systems. Retrieved from http://

dl.acm.org/citation.cfm?id=3091282.3091310

Wasserman, S., & Faust, K. (1994). Social Network Analysis: Methods and Applications.

Cambridge: Cambridge University Press.

Watts, D. J. (2004). Six Degrees: The Science of a Connected Age. New York: W. W. Norton

& Company.

Xu, J., & Chen, H. (2003). Untangling Criminal Networks: A Case Study. In H. Chen, R. Mi-

randa, D. D. Zeng, C. Demchak, J. Schroeder, & T. Madhusudan (Eds.), Intelligence

and Security Informatics (Vol. 2665, pp. 232–248). Berlin, Heidelberg: Springer.

Zaitch, D. (2002). Bosses, Brokers and Helpers. Labour and business relations amongst

Colombian cocaine traffickers. Amsterdams Sociologisch Tijdschrift, 29(4), 502–529.

39

Annex

Table A. Networks’ Topology and Connectivity*

Network N. of events

Leadership roles

N. of Nodes

Mafia charge Density N. of

linksAverage Degree

Degree centralization

N. of components

N. of nodes in secondary components

Average Geodesic Distance

Aemilia meeting 234

Leaders 6 65.1% 2,606 11.5 23.9% 5 11 2.8Others 221 54

Total 227 60

Aemilia wiretap 802

Leaders 6 61.5% 608 3.0 23.0% 2 3 3.6Others 197 53

Total 203 59

Crimine meeting 540

Leaders 24 233.9% 8,540 18.3 27.1% 5 13 3.0Others 443 82

Total 467 105

Crimine wiretap 700

Leaders 12 121.2% 496 2.4 29.1% 6 13 6.2Others 196 31

Total 208 43

Infinito meeting 483

Leaders 25 246.5% 4,100 16.3 25.2% 2 2 2.8Others 227 105

Total 252 129

Infinito wiretap 342

Leaders 23 221.6% 410 2.5 8.4% 3 4 6.3Others 140 87

Total 163 109

Minotauro meeting 437

Leaders 15 1411.6% 16,466 43.7 47.1% 3 10 2.3Others 362 99

Total 377 133

Minotauro wiretap 1,385

Leaders 13 121.8% 708 2.6 17.6% 5 12 5.1Others 259 40

Total 272 69*The density and the average degree reveal a network’s connectivity by calculating the ratio between the total observed direct relations and the total possible ones and the average number of nodes’ direct connections, respectively. The average geodesic distance generally describes a network’s information efficiency by averaging the shortest (geodesic) distance for all possible pairs of network nodes.

Figure A. Operation Minotauro, Two-Mode and One-Mode Meeting Networks

On the left pane, red nodes are meetings and black nodes individuals. On the right pane, red nodes are leaders, black nodes are non-leaders.

Nodes sized by betweenness centrality.

40