Come at the king, you best not miss: criminal network adaptation after law enforcement targeting of key players

Figures & data

Table 1. Descriptive statistics

		Phase 1	Phase 2		Phase 3
Network size		63	68		55
Edge count		146	110		91
Density		0.07	0.05		0.06
Mean degree		4.63	3.24		3.31
Maximum degree		26	19		18
Degree centralisation		0.36	0.24		0.28
Clustering coefficient		0.35	0.26		0.23
Nationality
Italian		49	53		34
Moroccan		9	12		21
Romanian		3	0		0
Other/unknown		2	3		0
Task
Supplier		6	9		8
Trafficker		11	13		10
Courier		6	5		7
Buyer		14	19		11
Support		18	8		11
Unknown		8	14		8
Role in the drug supply chain
Supply		8	10		9
Importation		32	22		25
Distribution		15	22		13
Unknown		8	14		8
Status
High		6	6		6
Medium		34	43		28
Low		23	19		21

Table 2. Changes across investigative phases

	Phase 1 to 2	Phase 2 to 3
No. observed actors
Actors at t	63	68
Actors at t + 1	68	55
Size difference	5	−13
Joint	27	25
Combined	104	98
Jaccard index	26.0	25.5
No. observed ties
Ties at t	146	110
Ties at t + 1	110	91
Jaccard index	16.9	15.5

Figure 1. Criminal collaboration networks across investigative phases

Note: recurring nodes maintain the same coordinates across networks, N3’s position is highlighted in red in the first investigative phase.

Table 3. ERGM parameters⁸²

Parameter	statnet name	Interpretation
Edge	edges	Baseline probability for tie formation. Negative for sparse networks (security)
Preferential attachment	gwdeg	Negative for decentralised networks (security); positive for centralised networks (efficiency)
Activity spread for high status actors, ‘Ndrangheta members, and traffickers	nodefactor	Positive when high status actors, traffickers, and ‘Ndrangheta members are directly involved in day-to-day operations (efficiency)
Homophily by task and role	nodematch	Negative when actors tend to collaborate more with those with different tasks or in different roles (efficiency)
Multiplexity of ties (kinship and formal organisation)	edgecov	Positive when actors tend to collaborate based on pre-existing relations to reduce uncertainty and increase trust (security)
Indirect connections	gwdsp	Positive for networks with chain-like patterns of connection (security); negative for networks with short geodesic distances (efficiency)
Triadic closure	gwesp	Positive when actors tend to collaborate in dense, local groups (efficiency), but also when collaboration is based on kinship and/or shared criminal affiliation (security)

Table 4. Estimates and standard errors from ERGMs

	Phase 1			Phase 2			Phase 3
	Estimate		SE	Estimate		SE	Estimate		SE
Edges	−7.53	***	0.69	−6.96	***	0.56	−7.97	***	0.77
Activity spread (high status)^a	0.51	**	0.19	0.40		0.22	0.89	**	0.31
Activity spread (trafficker)	−0.07		0.15	0.53	*	0.21	0.31		0.20
Activity spread (‘Ndrangheta member)	0.25		0.16	0.25		0.18	−0.79	***	0.23
Homophily by task	0.08		0.24	0.00		0.38	0.26		0.41
Homophily by role	0.59	**	0.22	0.03		0.33	−0.31		0.37
Multiplex ties (kinship)	2.58	***	0.55	1.93	**	0.66	3.32	***	0.88
Multiplex ties (formal org.)	0.98	*	0.39	1.42	***	0.36	1.27	*	0.50
Preferential attachment	2.87	***	0.79	3.10	***	0.77	3.93	**	1.21
Triadic closure	1.81	***	0.28	1.19	***	0.21	1.16	***	0.23
Indirect connections	0.00		0.03	0.03		0.03	0.07	*	0.03
Homophily by nationality	0.77	***	0.23	0.66	*	0.28	1.77	***	0.35
Activity spread (targeted)	0.99	***	0.28	0.78	**	0.26	1.14	***	0.33

*** p < 0.001; ** p < 0.01; * p < 0.05.

^aReference category: low status.

Figure A1. Goodness-of-fit diagnostics for Phase 1 model

Figure A2. Goodness-of-fit plot for Phase 2 model

Figure A3. Goodness-of-fit plot for Phase 3 model