Cocaine and Heroin Trafficking Networks
This study conceives of cocaine and heroin trafficking as two weighted and directed networks, in which nodes represent countries and edges identify drug trafficking between any two countries. We used the methodology developed by the UNODC (2015b) to estimate the presence and weight of the edges in the cocaine and heroin networks. This methodology is based on five main steps (Fig. 1): (1) identification of connections between any two countries; (2) estimation of the relative weight of each connection with respect to each country’s other incoming flows; (3) estimation of the quantity of drug consumed and seized in each country; (4) estimation of the total quantity of drug imported by each country; (5) estimation of the quantity of drug exchanged between any two countries. We produced year-by-year estimates of both cocaine and heroin trafficking flows (2011–2016) and calculated their mean over the entire period.
Step 1 identified connections—i.e. drug trafficking flows—between any two countries. To do this, we drew on the UNODC Individual Drug Seizure (IDS) dataset for the period 2011–2016. Seizure data often includes information on the quantity of drugs intercepted by law enforcement agencies, the country where the seizure occurred, and the origin, transit, and—occasionally—destination country. For the years 2011–2016, the IDS dataset contains 13,021 records of seizures of coca derivatives and 8052 records of seizures of opium and its derivatives providing indications of 19,155 and 11,652 connections in the cocaine and heroin trafficking networks, respectively. The resulting cocaine network comprised 147 countries and 817 edges while the heroin network included 137 countries and 437 edges.
Step 2 involved estimating the relative weight (%) of each connection with respect to each country’s other incoming flows. We used information on the quantity of drug intercepted along the identified connections to obtain these estimates. We adjusted each seizure case for its estimated purity, which was calculated as the average purity of the drug at wholesale level in the importing and exporting countries.Footnote 1 Finally, we estimated the relevance of each edge with respect to the total seizures performed by importing countries:
$$ {F}_{d,t,j\leftarrow k}={S}_{d,t,j\leftarrow k}/{\sum}_{k=1}^K{S}_{d,t,j\leftarrow k} $$
(1)
where Sdt is the quantity of drug d seized in year t, j is the importing country, and k is the exporting country. Figure 1 shows an example where 60% of the drugs seized by country A were smuggled from country D, while 40% was coming from country C. See Appendix 2 for further details.
A first limitation of this approach is that out of scale seizures may distort the estimates as particularly large seizures may reflect either an increase in shipments or luckier law enforcement operations (MacCoun and Reuter 2001; Kilmer and Hoorens 2010; Aziani 2018).Footnote 2 This is rarely the case for land and air routes because loads tend to be limited in scale; it may however be more likely in the case of sea connections, whose shipments are possibly larger. We used three-year moving averages to mitigate possible distortions driven by bulky seizures or limited data reporting in any given year.Footnote 3
Step 3 comprised estimating the total quantity of drug consumed and seized in each country, i.e. the quantity of drug that a country needs to import in order to satisfy its internal demand. To estimate the quantity of drug consumed in each country, we used a combination of demand- and supply-based approaches. First, we estimated the total quantities of cocaine and heroin available for consumption, starting from the UNODC (2019b) estimates of the global production of the two drugs. Second, we divided these quantities by the global estimates of the number of users of the two substances, thus obtaining estimates of the annual consumption per user. Third, for each country, we multiplied the estimated number of users by purity-adjusted estimated quantities of cocaine and heroin consumed by each user (Kilmer et al. 2013, 2011; Legleye, Lakhdar, and Spilka 2008; Wilkins, Bhatta, and Casswell 2002). Finally, we adjusted seizures for purity to make them comparable across countries before adding them to the total quantity of cocaine and heroin consumed (Paoli, Greenfield, and Reuter 2009). See Appendix 3 for further details.
Step 4 estimated the total quantity of drug imported by each country. For non-producing countries, this was equivalent to the sum of the quantity of drug consumed and seized (Step 3) and the quantity of drug imported and then exported to other countries. In the toy model proposed in Fig. 1, country C imported one ton of drug to meet its internal demand and an additional 40% to export to country A, for a total of 1.4 tons.
Step 5 involved estimating the quantity of drug exchanged by any two countries. As Fig. 1 shows, country B required a total of two tons to satisfy its internal demand. Given that 100% of the seized drugs came from country D, we assumed that country B imported the two tons from country D. 60% of country A’s imports came from country D, while the remaining 40% came from country C, which led to an estimate of 0.6 tons having been exchanged between country D and country A, and 0.4 tons between country C and country A. See Appendix 4 for further details on the algebra behind step 5. Finally, we calculated the average of all annual estimates (2011–2016) to alleviate possible distortions due to exceptional events and episodic issues in data reporting.
Producing, Transit, and Consumer Countries
We classified the countries in the cocaine and heroin networks into three groups—‘producer’, ‘transit’, and ‘consumer’ countries—according to the role that they play in international drug trafficking. These labels identify the main role of countries, as only a few are exclusively consumers (i.e. they do not export drugs), while none are exclusively a transit country (i.e. they do not have any internal consumption). This classification will be used subsequently to identify the most effective interception strategies.
‘Producer’ countries were identified based on data on coca and poppy cultivation published by the UNODC (2019a). The main cocaine producers were Bolivia, Colombia, Ecuador, and Peru, while the main heroin producers included Afghanistan, Colombia, Laos, and Myanmar. We used information on drug imports and exports to classify countries as ‘transit’ or ‘consumer’. The natural logarithm of the ratio between total exports and consumption in each country was taken as a measure of countries’ main role in transnational drug trafficking. That is to say, the higher the ratio, the more pronounced the transiting role played by a country; the lower the ratio, the more a country is a destination market. We then classified the countries by partitioning them into two non-overlapping clusters through the use of a k-medians method, as any threshold to discriminate between ‘transit and ‘consumer’ countries would introduce an element of arbitrariness in the absence of further information. A list of the countries that make up the cocaine and heroin networks and their cluster is available on the authors’ Open Science Framework project profile, while Table 1 provides a summary of the clusters.Footnote 4
Table 1 Countries’ clusters Network Statistics and Simulations
The first part of our analysis focused on providing descriptive statistics of the two networks.Footnote 5 We quantified network parameters by using social network analysis measures. More specifically, we used density (i.e. the proportion of edges in the network) and mean degree (i.e. the average number of edges for the countries in the network) to assess the degree of network cohesion (Wasserman and Faust 1994). We also used centralisation measures to assess whether the networks were characterised by the presence of a few trafficking hubs, or whether most of the countries had similar levels of involvement in trafficking activities (Giommoni, Aziani, and Berlusconi 2017).
We also calculated statistics for every country within the two networks, namely, degree centrality, betweenness centrality, and Gould and Fernandez’s (1989) brokerage measures. Degree centrality measures the number of countries that a country imports drugs from (in-degree) or exports drugs to (out-degree). Betweenness centrality measures the number of times that a country is located along the shortest path between any other two countries in the network (Wasserman and Faust 1994).
Gould and Fernandez’s (1989) brokerage measures afford the identification of the different brokerage roles that countries perform in the cocaine and heroin networks, respectively. They are based on the idea that ‘flows within groups should in general be distinguished from flows between groups’ (1989, 91). While betweenness centrality is expedient for identifying the countries that are located along drug trafficking routes, Gould and Fernandez’s brokerage roles are valuable in terms of identifying the countries that liaise between ‘producer’, ‘transit’, and ‘consumer’ countries.
Gould and Fernandez (1989) identified five brokerage roles (Fig. 2):
-
Coordinators mediate between two countries in the same cluster, such as in the case of a ‘transit’ country which lies along a drug trafficking route between two countries that are also primarily involved in transit activities (Fig. 2A).
-
Itinerants mediate between two countries from the same cluster, while belonging to a different cluster (Fig. 2B). A ‘transit’ country connecting two ‘producer’ countries is an example of this type of mediating role.
-
Representatives and gatekeepers mediate between two countries from different clusters. However, representatives share the same cluster as the country exporting drugs, whereas gatekeepers share the same cluster as the country importing drugs (Fig. 2 C and D, respectively).
-
Liaisons mediate between two countries in two different clusters, while, simultaneously, belonging to a third cluster. This relates to the case of a ‘transit’ country that lies on a drug trafficking route between a ‘producer’ and ‘consumer’ country.
The inclusion of Gould and Fernandez’s brokerage roles allowed us to consider not only the position of countries within the entire network—as indicated by betweenness centrality—but also their capacity to connect different clusters. This is valuable information for developing targeted interdiction strategies, given the different clusters that countries belong to in the international trafficking of cocaine and heroin (i.e. producer, transit, and consumer). For instance, intervening in a transit country that plays a coordinating role may produce a very different impact on drug trafficking flows than intervening in the same country if it acts as a liaison between ‘producer’ and ‘consumer’ countries, even if the betweenness centrality score for that country is the same.
The second part of the analysis tested the impact of different interdiction strategies on the level of connectivity within the networks, and on countries’ closeness to all other countries in the network, in order to identify those strategies that were more likely to have a disruptive effect on cocaine and heroin flows between countries. Specifically, we measured network connectivity by producing the reachability matrix—i.e. the matrix that identifies whether any two nodes are connected by one or more directed paths—and calculating its density to obtain the fraction of all dyads (i.e. pairs of countries) that are connected through a directed path (Wasserman and Faust 1994). We measured countries’ closeness by calculating the average closeness centrality score for all countries in the network using Opsahl et al.’s (2010) approach, which can be used for fully connected networks and networks with disconnected components alike. We simulated two types of interdiction strategies.
The first strategy focused on countries. It was based on the removal of nodes (i.e. countries) from the networks, and tested the disruptive effect of concentrating interdiction resources within specific countries. Vertices were removed: (1) randomly; based on (2) unweighted and (3) weighted in-degree centrality scores; based on (4) unweighted and (5) weighted out-degree centrality scores; (6) based on betweenness centrality scores; and (7) based on the five brokerage roles identified by Gould and Fernandez (1989). With respect to both random removal and those cases in which more than one country had the same score, we simulated node removal 100 times, and reported the average connectivity and closeness scores for the 100 replications.
The second strategy focused on drug routes. It was based on the removal of edges from the networks, and tested the effect of concentrating interdiction resources on specific trafficking routes between countries. Edges were removed: (1) randomly; (2) based on their weight, i.e. the estimated quantity of drugs exchanged between any two countries; (3) based on their edge betweenness, i.e. the number of times an edge occurred on the shortest path between any two countries (Freeman 1979); (4) based on whether they connected two countries in the same cluster (e.g. two ‘transit’ countries); (5) and based on whether they connected two countries in different clusters (e.g. a ‘transit’ and a ‘consumer’ country). Similar to above, for the purposes of both random removal and those cases in which more than one edge had the same score or fell within the same category, we simulated edge removal 100 times, and reported the average connectivity and closeness scores for the 100 replications.
Supply-side reduction policies in producing countries have thus far been proven to have only a marginal impact on the availability of drugs, if not actually being wholly detrimental (Angrist and Kugler 2008; Collins 2014; Mejía and Restrepo 2016; Veillette 2005). In consideration of previous literature which has already addressed the limitations of supply-side interventions in producer countries, we excluded these countries from node removal, and instead solely focused on ‘transit’ and ‘consumer’ countries. It should also be acknowledged that there is a cost factor associated with removing nodes or edges that must be considered in the development of drug interdictions strategies. For instance, although enforcing the border between Mexico and the United States is likely to be more disruptive than interdiction strategies targeting the connection between Belize and Mexico, the former also involves higher costs. This paper focuses exclusively on the benefits that targeted interventions may produce but overlooks their costs. This kind of cost-benefit analysis is outside the scope of this paper but should be further investigated by future research.
Finally, it is worth emphasising here that the aim of our simulations is not to literally remove a country or a trafficking route, as sealing borders and impeding illicit drugs from penetrating a country has been found to not be practically achievable. Rather, our analysis aims to inform drug law enforcement by showing where interdiction efforts are more likely to have the most disruptive effects on the entire drug trafficking network. At a historical juncture characterised by reduced public expenditure, our analysis can both inform how scarce resources are allocated and increase the effectiveness of drug law enforcement.
Limitations
Our approach is not without its drawbacks. Specifically, seizures might not be fully representative of actual drug flows. First, it is reasonable to expect a degree of heterogeneity in the effectiveness of different agencies’ enforcement activities, as drug policies vary across countries, and air and sea routes are easier to enforce than land routes (Reuter, 2014). Therefore, seizure data are likely to over-estimate the relevance of drug trafficking within more aggressive countries, and along sea and air routes.
Second, while seizures account for a large share of inflows in several countries, they are limited and only account for a small fraction of total inflows in other countries—see Appendix 5 for the share of cocaine and heroin seized, consumed and exported by each country. Moreover, some countries do not report information on drug seizures to the UNODC every year, while others report only certain cases or avoid sharing any information at all (UNODC 2019b). Certain trafficking connections thus emerged from limited evidence. Finally, as already discussed, particularly large seizures might introduce an upward bias in the estimate of the relative relevance of a specific connection between two countries.
The combined use of information from multiple countries in multiple years helps mitigate the consequences stemming from countries’ differences in counter-drug efforts and seizure reporting, changes in countries’ border control activities, and exceptionally large seizures that might skew our estimates (Berlusconi, Aziani, and Giommoni 2017). Moreover, while working on the 1998–2014 UNODC cocaine IDS, Aziani (2018) noted that importing countries provide the vast majority of information actually usable to build a network. Since the relevance of each connection is estimated with respect to the importing country—i.e. through a comparison of incoming flows—this fact further reduces the biases due to different law enforcement and reporting systems, as these differences are likely to be more severe between countries than within countries. Nonetheless, the extent to which seizures represent actual flows or enforcement capacity remains unknown (MacCoun and Reuter 2001; Kilmer and Hoorens 2010); therefore, the resulting networks are an approximation of the drug trafficking activities taking place worldwide rather than their faithful reconstruction.
There are further uncertainties pertaining to our assumptions about users and their consumption, along with drug purity. The heterogeneity of data available across countries forces us to rely, from time to time, on linear interpolations and eventually on regional values. The lack of estimates is particularly severe with respect to quantities of drug consumed by the average user, which has thus been estimated starting from drug production data (see Appendix 3 for details). A series of inconsistent reports by government agencies and international organizations, together with the impossibility to state the exact time lag between drug production and consumption, led several authors to be sceptical about supply-side estimations (Reuter and Greenfield 2001; Kilmer, Reuter, and Giommoni 2015). While we do agree with this position, in the context of the current study, the limits of this strategy primarily concern the lack of adjustments for national consumption values per user and eventually a bias in the relative weight of consumption and seizures. Uncertainties about the time lag between production and consumption are instead likely to affect only marginally our analyses because we combine estimates referring to multiple years, and because the analysis of law enforcement interventions is not performed on a dynamic network.
Despite these aforesaid limitations, the approach adopted in this study has been extensively applied in other studies because of its unique capacity to produce new information on the existence of trafficking activities from available data (Dugato and Aziani 2020). Paoli, Greenfield, and Reuter (2009) were the first to conceive of international drug trafficking as a series of trading relations across countries. Boivin (2013, 2014a, 2014b) utilised information from seizures to study the network structure of international drug trafficking, while Chandra et al. (2011, 2014, 2015) used a similar approach to analyse drug trafficking flows across countries, albeit they constructed their networks with price-based data rather than seizure data. The UNODC (2015b) developed a model to estimate heroin trafficking flows along the Balkan route, with the same methodology subsequently being used in academic articles on cocaine and heroin trafficking (Aziani, Berlusconi, and Giommoni 2019; Berlusconi, Aziani, and Giommoni 2017; Giommoni, Aziani, and Berlusconi 2017). Hence, this is an established method through which to estimate drug trafficking across countries.