Researching a Local Heroin Market as a Complex Adaptive System

ORIGINAL PAPER

Researching a Local Heroin Market as a Complex AdaptiveSystem

Lee D. Hoffer • Georgiy Bobashev•

Robert J. Morris

Published online: 17 October 2009� Society for Community Research and Action 2009

Abstract This project applies agent-based modeling(ABM) techniques to better understand the operation,organization, and structure of a local heroin market. Thesimulation detailed was developed using data from an 18-month ethnographic case study. The original research,collected in Denver, CO during the 1990s, represents thehistoric account of users and dealers who operated in theLarimer area heroin market. Working together, the authorsstudied the behaviors of customers, private dealers, street-sellers, brokers, and the police, reßecting the core elementspertaining to how the market operated. After evaluating thelogical consistency between the data and agent behaviors,simulations scaled-up interactions to observe their aggre-gated outcomes. While the concept and Þndings from thisstudy remain experimental, these methods represent anovel way in which to understand illicit drug markets andthe dynamic adaptations and outcomes they generate.Extensions of this research perspective, as well as itsstrengths and limitations, are discussed.

Keywords Illegal drug markets� Ethnographic research�Complex adaptive systems� Applied social simulationmodeling

Introduction

This paper describes an experimental study using agent-based modeling (ABM) techniques to computationallyreproduce and investigate the operation of a local heroinmarket in Denver, CO, circa 1992Ð1996. The objective ofthe illicit drug market simulation (IDMS) project was tocomputationally simulate what occurred in DenverÕs Lari-mer area open-air heroin market, using ethnographic data.Simulation experiments evaluate and extend our under-standing of two research questions: (1) How important wasthe role of ÔÔbrokers,ÕÕ heroin users who facilitated trans-actions between customers and dealers,1 to the marketÕsoverall operation? (2) To what extent did police activityinßuence the market and its eventual transformation?

It is important to note, the model described in thismanuscript is not intended as a forecasting tool. Rather, itwas created as an exploratory laboratory to expand eth-nographic Þndings, and to identify essential parameters tothe marketÕs operation and transformation. In developingan ABM logically consistent with ethnographic research,the IDMS project was a proof of concept study. Fewexamples of this type of synthesis exist (Agar2001, 2005),but applying this approach in prospective research couldmake it easier to interpret conventional market outcomes,as well as expand research on local illicit drug marketoperations in novel ways.

This paper highlights utilizing ABM as a research tool toexploit the unique strengths of ethnography. Namely, eth-nographic research occurs within the participantÕs natural

This manuscript was prepared when Dr. Hoffer was a researchinstructor in the Department of Psychiatry at Washington UniversitySchool of Medicine, St. Louis, MO. He is currently an AssistantProfessor of Anthropology at Case Western Reserve University.

L. D. Hoffer (&)Department of Anthropology, Case Western Reserve University,11220 Bellßower Road #205, Cleveland, OH 44106, USAe-mail: [email protected]: http://www.case.edu/artsci/anth/

G. Bobashev� R. J. MorrisRTI International, Research Triangle Park, NC, USA

1 The term ÔÔcoppingÕÕ is an indigenous term often used in theliterature to describe when one user buys drugs for another. Theauthor uses ÔÔbrokerÕÕ and ÔÔbrokeringÕÕ to better specify theseactivities. These more widely known terms also show their crucialrole to the marketÕs operation and adaptation.

123

Am J Community Psychol (2009) 44:273Ð286DOI 10.1007/s10464-009-9268-2

environment. This way, researchers cannot help but addresssocial and environmental interactions. The ethnographermust confront, and attempt to explain, nuance and change.The challenge of the IDMS project involved reformulatingÞndings of an ethnography using ABM. Could we repro-duce the essential behaviors of market participants withinan ABM to reßect (and quantify) what was discovered inthe ethnography? What would ÔÔscaling-upÕÕ results dem-onstrate, and what could we learn from this new data?

Future applications of the hybrid methodology describedin this paper are numerous and could make substantialcontributions to research in new drug prevention andtreatment interventions and in impact evaluations of drugpolicies. Community outreach, needle-exchange, treatmentprograms, and other services aimed at reducing the nega-tive health consequences of drug use could be designed,integrated, and tested using such simulation techniques.Additionally, taking advantage of geographic informationsystems (GIS) and other geographical data sources, futureABMs of drug markets could observe illicit drug activity inreal time. However, to apply ABMs to inform policy,researchers must be cautious. Unlike Systems Dynamicsand other more mature simulation and modeling techniques(Forrester1994; Homer and Hirsch2006; Mabry et al.2008), guidelines for integrating data into ABMs andprocedures for validating such models need further devel-opment. In fact, theoretically oriented ABMs do not requiredata to be useful. Because of ABMÕs micro-simulationcharacteristics, they provide a unique opportunity toexamine rare and ÔÔunusualÕÕ scenarios. Statistical andsystem dynamics models focus on describing the means,but ABMs allow researchers to examine trajectoriesstrongly deviating from the mean trend and Þnd out whysuch factors occur. For research on illegal drug markets,this paper hopes to advance methods for designing ABMsusing ethnographic data, better integrating behaviors withintheir local environments.

Perspectives on Heroin Markets

The global economy for heroin is comprised of countlessnumbers of heterogeneous agents who produce, distribute,sell, and consume the drug (Eck and Gersh2000). As withother drugs not produced in the US, it is recognized that theheroin economy has a hierarchical structure, with proÞtsand resources coalescing exponentially at its highest levels(Curtis and Wendel2000). At the top of this structure sitlarge-scale, well organized, well Þnanced, and politicallyentrenched international cartels.

Understanding heroin markets in local communitiesindirectly involves what cartels do. In Denver, the mar-ket was connected to a global enterprise. Heroin sellerswho were researched on the streets of Denver for this

project sold heroin for Mexican Drug TrafÞcking Orga-nizations (DTOs). Sellers went to the market, workedshifts, and received pay on commission based on theirsales. The trafÞckers recruited sellers from villages inCentral America, arranged for their transportation, andprovided them with housing. Expenses incurred, in turn,justiÞed the sellerÕs servitude. The Mexican DTOÕscontrolled heroin distribution in Denver.

This perspective highlights the topÐdown structure inillicit drug markets, implying what happens locally isdirected by forces acting higher up the economic hierarchy.The arrangement is ÔÔsupply-centric.ÕÕ It emphasized howproducers inßuence the behavior of distributors. Distribu-tors then inßuence the behaviors of dealers. Dealers theninßuence the behaviors of customers. In this ÔÔcommandeconomyÕÕ system, higher echelons are presumed to controlsubordinates, promoting the ideology that interventionsshould be targeted upward. This, after all, is wheredynamics are set in motion. This topÐdown approach iscommensurate with a ÔÔsupply-reductionÕÕ policy, and isoften used to justify the ÔÔwar on drugs.ÕÕ However, coor-dination and independence coexist in illicit drug markets,and this admixture confounds such simpliÞcation (Johnson2003).

A ÔÔBottomÐUpÕÕ Perspective on Heroin Markets

Considering the economic structure, hierarchy, and verticalintegration of heroin distribution, changes in organizationcan reverberate ÔÔdownward,ÕÕ inßuencing subordinateoperations and end-users, i.e. customers. However, a sig-niÞcant portion of activity in local drug markets occursoutside such ÔÔorganizedÕÕ frameworks. People and groupsare often operationally independent from, or in directcompetition with, one another (Bourgois1997). Forexample, a distributor interacts with a dealer to provide thedealer with heroin; however, the distributor and the dealerare independent businesspeople. In other words, it is up tothe dealer to make sales, enforce rules, and managetransactions with customers, and the distributor has tomanage suppliers bringing them product. As a result, basedon the primary authorÕs ethnographic research, we couldnot understand DenverÕs open-air heroin market only byunderstanding the activities of the DTO that supplied theheroin. Appearances suggested a functional monopoly.However, this facüade only partially explained how heroinwas sold.

Like most street-based markets, the Larimer area wasnot DenverÕs only venue for heroin sales. A considerablenumber of independent (i.e., private) dealers sold heroinfrom their homes. Our ethnographic case study examinedone such organization. Like street-based sellers, the dealersresearched purchased the half-ounce supply of heroin for

274 Am J Community Psychol (2009) 44:273Ð286

123

$750 from a supplier afÞliated with a Mexican DTO. Theywere then free to sell it with no operational support,guidance, or oversight from the DTO.

Illicit drug markets involve hundreds, if not thousands,of loosely afÞliated organizations, gangs, groups, andindividuals, all in business to make proÞts. How the actionsof these heterogeneous agents are coordinated, or syn-chronized, is important. This statement does not diminishthe importance of the global or organized nature of the drugeconomy. Nonetheless, many sales operations are onlyloosely, if at all, afÞliated with larger groups and operatealmost entirely outside any control. In the primary authorÕsethnography, an important observation was how littleinßuence the dealerÕs supplier had on day-to-day opera-tions. Maintaining autonomy, the dealer was only depen-dent on his supplier for material, i.e., heroin. Similar to theorganization of a large intelligence agency, or terroristnetwork, operational activity in illicit drug markets iscompartmentalized.

TopÐdown management inßuences local markets, butthis momentum also works in reverse. Local marketdynamics often shape (or reshape) the larger economy ofan illicit drug. Local markets respond and adapt to whatlocal dealers do, to local demand, and to changes thathappen locally. Unlike organizational directives thatÔÔtrickle down,ÕÕ this type of endogenous bottomÐupactivity evolves in seemingly spontaneous ways, making itexceedingly challenging to understand or anticipate. Thismixture of organization and independence promotes non-linearity and results in complexity (Caulkins and Reuter2006; Hough and Natarajan2000). Crack cocaine andmethamphetamine use are two modern examples of trendsthat emerged in this manner.

Cocaine cartels did not invent crack. An anonymousgroup of entrepreneurial mid-level drug dealers inventedcrack to increase proÞts from the product they were sellinglocally. This innovation completely reorganized cocainesales, along with HIV and other associated health risks(Singer2006). Outlaw motorcycle gangs were the Þrst touse, produce and distribute methamphetamines. However,based on a steady increase in popularity (i.e., localdemand), large-scale Mexican DTOs began producing,importing, and selling methamphetamines.

The history of crack and methamphetamines is a dra-matic example of the power of local market dynamics. Inboth instances, local demand set in motion macro-scalerestructuring. Simply stated, in many instances, whathappens in drug economies emerges from low-level inter-actions and feedback, reconstituting the nature, function,and structure of the larger system.

A Þrst step in realizing the potential of this localdynamic perspective was to use this ÔÔbottomÐupÕÕ

approach to understand DenverÕs heroin market, the con-ceptual objective of the IDMS project.

The methods of this project involve the descriptiveÞndings from the background ethnography. This data isthen contextualized through an understanding of complexadaptive systems, and how such systems are researched.Once the data and conceptual framework are outlined,details are provided on the model building processes weused and how data were extracted from the ethnography toinform ABM development. Finally, the model, initialconditions, and experimental results are discussed.

Methods

It is important to recognize that ethnographic researchdescribing local factors associated with the operation andadaptation of local illicit drug markets is not new. Since the1960s, researchers have described the roles (Adler1985;Agar 1973; Johnson et al.1985; Preble and Casey1969;Reuter et al.1990) and interactive factors (Agar et al.1998;Bourgois1997; Caulkins1997; Hamid 1991a, b; Johnsonet al. 2000; Koester 1994) that shape behavior in drugmarkets. However, no one has explored how these micro-interactions scale-up to produce macro-level patterns, orhow such interactions constitute the larger system. Thisproject integrates ABM to address this speciÞc limitation.

In the mid-1990s, one of us (LH) conducted ethnographicresearch in an open-air drug market in Denver, CO through anumber of research projects funded by the National Institutesof Health, National Institute on Drug Abuse. This work,summarized in the ethnographyJunkie Business: the Evo-

lution and Operation of a Heroin Dealing Network (Hoffer2006), describes how two homeless heroin users developedand managed a full-time business selling heroin.

The data for the ABM presented here were extracted fromJunkie Business. The portion of the ethnography re-analyzedrelates to the operation of the Larimer area open-air heroinmarket, from which the dealersÕ business emerged. Duringthe ethnography, a natural experiment occurred in which thepolice, in combined effort with other city groups and orga-nizations, effectively ÔÔclosedÕÕ the Larimer area drug mar-ket. To understand the marketÕs operation before, during,and after this intervention, the authors present it as a com-plex adaptive system (CAS) (Holland1995, 1998; Waldrop1992). Before addressing what affected this perspective, wepresent details of the case.

DenverÕs Heroin Market, Circa 1990Ð1993

The operation of the Larimer area heroin market involvedthree primary groups of actors: immigrant sellers, local

Am J Community Psychol (2009) 44:273Ð286 275

123

junkie brokers, and customers. Before 1993, this marketoperated in a state of ÔÔdynamic stability,ÕÕ with agentsmutually beneÞting from their relationships with the otheragents. The Larimer market was an ÔÔopen-airÕÕ drug mar-ket, meaning customers did not need an introduction topurchase drugs and could walk or drive into the area andbuy heroin from one of two sources: directly from a selleror indirectly through a local junkie broker. Even thoughmost people who concerned themselves with these matters(e.g., the police, social service and health care providers)knew drugs were sold openly in Larimer, the area was alsohome to DenverÕs homeless and transient populations. Theconsiderable number of people on the street disguised theseactivities. Private heroin dealers, who were not part of thisLarimer ÔÔscene,ÕÕ also had an integral role in the marketÕsoperation and eventual transformation. We will discussthese dealers in greater detail later.

The most visible heroin sellers in Larimer came fromCentral America. They did not use heroin. Mexican maÞaorganizations recruited sellers and exploited them as labor.These organizations controlled the cityÕs heroin supply.Typically, sellers stayed ÔÔin businessÕÕ for short-termproÞts and for a limited time. Working as part of anorganized syndicate, sellers made money based on a dailysales commission. Immigrant sellers dominated the market,but they had unsophisticated sales strategies; they stood onthe street and asked people who passed by if they wanted tobuy drugs. Based on these public and primitive methods,sellers often made themselves targets of arrest by thepolice.2

Beyond selling heroin, sellers had little involvement orinterest in the lives of customers, or in developing rela-tionships with them. Few sellers spoke English beyond theminimum needed to sell heroin or other drugs. This was aclosed community. Most sellers did not use drugs andlooked down on, depersonalized, and mistreated customers,viewing them as an economic means to an end. As a result,sellers often cheated customers by ripping them off tomake a little extra money. Anyone could buy heroin from aseller in Larimer, and if a customer had no other options ordesired a quick, no-nonsense, anonymous sale, they went toa seller. But Larimer had another popular option for heroinsales.

Among the homeless loitering in the streets of Larimerexisted a relatively small but active group of local junkies.3

As heroin users who lived on the streets, junkies knew the

detailed infrastructure of the market. They knew who thesellers were, where to Þnd them, and when sellers were onthe streets. If a customer wanted to purchase heroin using abroker, transactions were indirect. Customers would givetheir money to the broker, the broker would purchaseheroin from a seller, and then the broker would return tothe customer with the heroin purchased. Brokers requiredÔÔpaymentÕÕ for their services and, because customers andjunkies were heroin users, customers often paid junkies byinjecting drugs together. In other words, junkie brokerswould receive a small portion of the liqueÞed drug from acustomer during the process of injecting (Koester andHoffer 1994). This tax, referred to locally as a ÔÔkick-down,ÕÕ was how the local junkies maintained theiraddictions (Koester1994). Brokers knew customers andhad no language or other cultural barriers interacting withthem. Before living on the streets, brokers were ordinarycustomers. In moving to the streets they did not severrelations with their peers. In fact, as brokers, they couldimprove these relations by acting as third parties.

Brokers stood less of a chance of immigrant sellerscheating them because the drug market was their ÔÔhome.ÕÕSellers understood that brokers could bring them businessor, if mistreated, steer customers away from them anddamage their reputations.

Ironically, despite the fact that brokers could easilycheat customers, they were less likely than sellers to do so.Local junkie brokers were a relatively small and highlyconnected group of users. They all knew and relied on oneanother to meet their basic needs, both for heroin but alsofor more instrumental resources like food and shelter.Although this group had no formal leadership, they occu-pied a small geographical area, and communication trav-eled rapidly within the group. If a junkie consistentlycheated a customer, their fellow junkies suffered becausenext time the customer came to purchase heroin they wouldbe less likely to use a fellow broker to purchase drugs. As aresult, the group enforced fairness. Customers also avoidedcheating a junkie. If they did so, it jeopardized their chanceof using another junkie next time they came to Larimer.The most important beneÞt to customers in using a brokerwas safety.

Using a broker effectively insulated both sellers andcustomers from arrest because the broker made the herointransaction. Although such instances were rare, brokersarrested typically did not carry large enough quantities ofdrugs4 or money to receive long, if any, jail time. Arresting

2 It should be noted that arrest tactics had little long-term success indisrupting market activities, because after sellers were arrested theywere deported. Once back in Mexico, or their home country, theywould make the trip back to Denver and resume their job sellingheroin.3 The authors do not use the term ÔÔjunkieÕÕ in a derogatory way.Rather this was the term the group self-reported.

4 Because brokers used drugs immediately after obtaining them intransactions, they rarely possessed drugs. If they were caught withdrugs it was usually an amount intended for personal consumption,not distribution.


123

brokers also had no long-term inßuence on the marketbecause police only removed indirect participants in thedistribution process. Police rarely arrested brokers;5 byspending all day on the streets, brokers became familiar withpolice tactics and they knew who the real sellers were.Customers coming to the area quickly learned who thebrokers were. As a result, police had a difÞcult time inÞl-trating the system by posing as either customers solicitingsales from brokers, or as brokers soliciting sales from sellers.

Over time, interactions between sellers, brokers, andcustomers developed into an arrangement of mutual depen-dence and cooperation. Customers who did not wish to paythe heroin tax could purchase heroin directly from a seller, aquick and anonymous transaction some customers preferred.Immigrant sellers clearly beneÞted from direct transactionswith customers. Sellers also beneÞted from brokers bringingthem business, as an enterprising broker could bring them aconsiderable number of customers. Sellers could also accesscustomers through brokers who might otherwise be cautiousof going directly to a seller. Because many customers did nottrust sellers, they could use a broker, thereby decreasing theirchance of being arrested and getting ripped off, bothimportant considerations. Finally, brokers beneÞted fromtheir unique position in the market and in developing goodreputations with both customers and sellers. Brokers alsoclearly beneÞted from open communications and maintain-ing strong ties with other brokers.

It is important to acknowledge that the relationshipsdescribed had no basis in mutual trust, communal feelings,or empathy, although some relations in time became per-sonal. Interactions that occurred between sellers, brokers,and customers were fundamentally pragmatic and moti-vated by individual economic self-interest. Over time,intra-group norms emerged, as did punishment for non-compliance (i.e., a local junkie who consistently ripped offcustomers would eventually be socially ostracized by otherjunkies). Nonetheless, all the players in the market oper-ated in autonomous ways, motivated independently. Thesymbiotic connection was a byproduct of mutual economicself-preservation, not afÞnity.

Secondly, the authorsÕ research delineates brokers andcustomers as distinct social groups, but this distinction wassomewhat artiÞcial. Customers commonly recognize bro-kering as a strategy to reduce the cost of their own druguse. Brokers in Larimer were distinct only because: (1)their primary access to the drug came from brokering, (2)they spent most of their time with other homeless heroinusers who engaged in similar activities, and (3) they spentmost of their time in the Larimer area. But this ÔÔcommu-nityÕÕ had porous boundaries, meaning that if a heroin user

with a home decided to broker heroin sales to reduce thecost of addiction, they were welcome to do so. Junkies withhomes maintained their membership in the community ofbrokers in several ways. These junkies used heroin withothers in the networks through which the brokers survivedand also participated in the informal exchange relationshipsthat made up broker networks.

Police saw ÔÔMexicansÕÕ (i.e., the immigrant sellers whooperated on the streets) as the only relevant group associ-ated with the distribution of heroin (and other illegal drugs)in the city, but this inaccuracy had important consequences.The organization of the Larimer area heroin market, andsubsequent homeostatic relations between actors, was not afunction of central organization. This observation wascritical in casting the Larimer market in terms of a complexadaptive system. In other words, elements imposed by boththe vertical integration of the Mexican DTOs and con-ventional forces of supply and demand contributed to howthe market operated, and yet neither exclusively justiÞedthe markets sustainability or stability. Juxtaposing formaland mechanistic understandings of market behaviors,activities in the Larimer market were organic, meaningthey were self-organizing, self-correcting and solidiÞedthrough the social ties of the actors.

Closing the Market

In 1993, Larimer abruptly changed. As a continuation ofurban renewal efforts already ßourishing in lower down-town, the city decided to build Coors Field, home of theColorado Rockies major league baseball team, on theborder of the Larimer area drug market. As one of the fewareas of undervalued and undeveloped real-estate adjacentto the downtown business district, Larimer was ideal for re-development. With popular support for the new stadiumfrom the city, state and region, the homeless (and drugmarket) did not stand a chance. It is important to recognizethat the Larimer area has no ofÞcial residents. Other thanthe homeless, no one lived in this area of town. As a result,the media largely ignored its transformation. Within just afew years (1993Ð1995), the Larimer area was ÔÔcleaned-upÕÕ and transformed, at least superÞcially.

During this time frame, through the uncoordinated yetdedicated efforts of a number of different groups (i.e., thepolice, parks and recreation department, and homelessshelters), the numbers of street-people were drasticallydiminished in Larimer, permanently. Without the home-less, only immigrant sellers and brokers remained on thestreets, primarily because neither had resources to goelsewhere. As a result, police efforts started to have moreimpact on open-air sales.

Immigrant sellers responded in two ways. First, theyvaried their locations on the streets. In other words, instead

5 The police rarely arrested brokers because they did not meet theproÞle of drug dealers in the area.


123

of standing on the same corner every day they movedlocations to make it more difÞcult for the police to Þndthem. Second, some immigrants left the area altogether.Some went to other areas of the city where their presenceon the streets was less noticeable while others migrated tomore private venues, switching to a beeper system to Þllorders and make sales.

Unlike immigrant sellers who were required to changetheir tactics, brokers did not have to signiÞcantly alter theiractivities. In other words, customers still came to Larimerto purchase heroin and were still able to Þnd brokers.Brokers maintained contacts with immigrant sellers duringthis time, although these connections were more difÞcult tomaintain as the immigrants moved and/or relocated.However, during this interim period brokers had anotherready access point to Þll orders for heroin. As formercustomers, in their interaction with current customers andthrough communication networks with other brokers, manybrokers had and maintained contacts with private dealerswho resided outside the Larimer area.

In general, private dealers were a valuable resourcebecause they usually made better deals and sold higherpotency heroin than could be purchased on the streets.Residing in neighborhoods unassociated with open-air drugsales, most private dealersÕ businesses were invisible to thepolice. In fact, occasionally brokers used private dealersinstead of immigrant sellers when brokering sales evenbefore Larimer was under pressure. However, most brokersdid not take full advantage of such opportunities for oneimportant reason: it jeopardized their continued livelihood.If either a customer or private dealer became aware that abroker was making consistent sales for a customer, it wasin the interest of both the customer and dealer to, in effect,cut out the middleman, i.e., the broker. Inevitably, brokeredsales using private dealers would ultimately result in thebroker losing the customer to the dealer. Nonetheless, asLarimer was under pressure this was an important short-term strategy brokers employed to Þll customer orders.This option for brokers was a vital component to the heroinmarketÕs successful and largely uneventful transition dur-ing the Larimer area clean-up effort. Once brokers rees-tablished connections with immigrant sellers, and thingssettled down in the area, business returned to normal,although the area was forever changed after the clean-up.Without the cover of large populations of homeless, after1995 sellers and brokers needed to be more cautious.

Paradoxically, in a review of the available drug trendindicator data for Colorado, the transformation of the Lari-mer market did not register via traditional economic indi-cators. In other words, based on the primary authorÕsresearch with customers and dealers in the market at the time,heroin price, purity, and sales unit sizes remained constant.With more brokers operating during the clean-up and

changing a ÔÔkick down,ÕÕ one could argue that the ÔÔcostÕÕ ofheroin increased slightly for customers over this time period.However, the aggregated effect of this change was negligiblebecause the marketÕs adaptation was so rapid.

Drug Markets as Complex Adaptive Systems (CAS)

This paper suggests Larimer and other local illicit drugmarkets operate as self-organizingcomplex adaptive sys-

tems (CAS). As a science, Complexity theory is concernedwith understanding precisely these systems throughanswering the fundamental question, ÔÔHow could thedecentralized local interactions of heterogeneous autono-mous agents generate the given regularity [in the system]?ÕÕ(Epstein2006: 5). This shift in perspective has importantimplications for how markets are researched, as well as theanalytic tools required to understand them.

A self-organizing CAS is a type of social systemdirected by the micro-level individual interactive behaviorsof ÔÔactorsÕÕ occurring within a dynamic environment(Conte and Gilbert1995; Conte et al.1997; Doran andGilbert 1994; Epstein and Axtell1996; Gilbert and Conte1995; Gilbert and Toritzsch1999; Holland 1995, 1998;Marney and Tarbert2000; Sawyer2003; Schelling1978;Waldrop1992). Interactions between actors, or actors andthe environment, are non-linear and non-reducible viaaggregation (Agar1999). The environment is proactive in aCAS, both shaping and being shaped by how agentsbehave. The self-organizing property of a CAS acknowl-edges that there is no centralized pre-existing infrastructureor organizational framework that directs it. Rather, out-comes are endogenous and based exclusively upon thesystemÕs internal momentum. Borrowing from Non-linearDynamics and Complexity theory, a CAS is ÔÔemergent,ÕÕoperating solely from its members interacting with theenvironment (Epstein2006).

As with natural ecosystems, not all interactions betweenagents in a CAS are meaningful. Some activity has little or noaffect relative to the dynamics of the system or its output.Other times, small changes to the system result in majorrepercussions. A related feature of a CAS is that it is dynamic,meaning constantly active. Similar to when an ecosystem isdescribed as ÔÔin balance,ÕÕ ÔÔstable,ÕÕ or ÔÔin equilibrium,ÕÕ suchstates are not necessarily Þxed or perdurable.

Finally, based on its lineage from Complexity theory, aCAS borrows elements from Chaos theory (Sawyer2003,2006). However, this does not imply disorganization. Onthe contrary, it suggests spontaneous organization (Wal-drop 1992). As a result, within the context of local heroinmarkets, hierarchy and formal structure can be incorpo-rated to varying degrees within a CAS. As a caveat, this isnow one of many features contributing to the systems


123

overall operation. Conventional market assumptions oflinearity and rationality are relaxed. Using a CAS model,the realities of a local drug market are decoupled froma priori economic assumptions, allowing more equitableinclusion of behaviors, factors, and contextualized inter-actions (see Fig.1). Recognizing this ßexibility, researchon emergence using ABM is rapidly becoming a scientiÞcÞeld unto itself (Bankes2002; Bonabeau2002; Carley2002; Conte2002; Holland1998; Waldrop1992; Wolfram1986). Research in public health is also beginning to adoptthese tools to inform our understanding of interventionsand their unintended consequences (Green2006; Leischowand Milstein 2006; Lempert 2002; Levy et al. 2006;McLeroy 2006; McManus2002; Sterman2006).

Modeling a CAS

To evaluate a CAS, researchers simulate its operationthrough time. This requires the researchers to construct thesystem using the available data. In this project an agent-based (a.k.a multi-agent) modeling (ABM) approach wasused. Agents are virtual objects reßecting speciÞc statesand behaving according to pre-speciÞed rules. Technically,an agent is a computer program (Gilbert2007). In thisproject agents were developed to represent actors in theLarimer heroin market. For example, ÔÔcustomerÕÕ agents(see below) were designed to represent heroin users. Cus-tomers existed in satiated, craving, or withdrawal states andcould use, search for, and purchase heroin. Transitions instates and behaviors were patterned, i.e., programmed, byconverting ethnographic narratives into decision trees andcomputer algorithms.

Ethnographic data were well suited to ABM program-ming because it focused on the micro-interactive behaviorsof dealers, customers, and other groups operating withinthe market. This level of granularity was important becauseit included individual decisions, heuristics, behaviors, and

memory, as well as changes in decisions based on inter-actions. In ABM research it is important to emphasize thatthe agent is programmed with the rules that generate itsbehavior when the simulation is run. Outcome data expressthe aggregated behaviors of agents. The environment inwhich agents are placed is also programmed. Ultimately,ÔÔThe aim of agent design is to create programs whichinteract ÔintelligentlyÕ with their environmentÕÕ (Gilbert andToritzsch 1999, p. 158). In this way, ABMs build thesystem from the ground up and system outcomes are ulti-mately the product of agents.

Although we selected ABM to maximize the utility ofour ethnographic data, it is important to recognize othersimulation techniques have a longer tradition of use insocial science applications. For example, system dynamics(SD) is a particularly powerful modeling tool that relies onconceptualizing an entire system, incorporating stocks andßows of system variables, and understanding systemfeedback loops. However, a fundamental weakness in usingSD for modeling local drug markets is that it relies onaggregate data of the total system. Because heroin usersand dealers are hidden populations, valid data concerningstocks and ßows of local illicit drug markets simply do notexist. Proxy data such as treatment entry statistics, drugprice, or emergency department mentions are difÞcult totrust and challenging to interpret in this context. The onlyprevious attempt to use SD modeling of a heroin market(Levin et al.1975) was highly criticized for these reasons.On the other hand, methods such as process modelingallow detailed description at an individual level, but theinteractions and adaptive consequences of interactions arebeyond the assumptions identiÞed in the process models. Insuch simulations agents do not adjust, adapt, or upgradebehaviors based on either previous information or speciÞcsituations presented in their environment.

It is important to acknowledge any advantage in ßexi-bility ABMs hold over other modeling methods disappearif links between the data and agent development cannotbe audited. A critical beneÞt using ethnographic data toprogram ABMs relates to the depth and nuance the dataprovides in this endeavor. ABMs essentially rely on pro-gramming social processes and then using such processesto generate behavior, and output.

Complex models require numbers. But it is not somuch a question of how to measure phenomena;instead, it is a question of how to express qualitieslearned through anthropological research, usingfunctions instead of words as the language for thatexpression. (Agar2001, p. 364)

Here the beneÞt and descriptive power of ethnographycan be fully exploited. For example, in many drug usestudies, including those of illicit drug markets, a common

Fig. 1 The generic structure of an illicit drug economy noting thatbelow the organized framework of a cartel such economies functionlike a complex adaptive system


123

and important measure is ÔÔfrequency of drug use.ÕÕ Tra-ditional survey research methods collect frequency of usedata in preset formats, (e.g., daily, weekly, monthly, life-time). Heroin use by agents in our model was determinedvery differently and by incorporating a more extensive setof elements, including: (1) the drugÕs pharmacokinetics andpharmacodynamics (i.e., creating a dynamic level ofÔÔheroinÕÕ in an agents system); (2) an agentÕs drug ÔÔhabitÕÕ(i.e., specifying a targeted amount of heroin the agentwanted to use); (3) deÞning the availability of the drug and(4) including the means to purchase it. If an agent obtainedexcess funds, it could purchase and use increased amountsof heroin, thereby increasing its drug habit and potentiallyusing more frequently. This, in turn, could lead the agent toexhaust its funds and enter a withdrawal state. A desperateneed for money could also potentially lead the agent tocrime.

This, and other chains of consequences, are not obviousand are difÞcult to identify, represent, or model from surveyresearch (Byrne1998; Goldspink2002). Converting deci-sions and behaviors identiÞed ethnographically intodynamic probabilities allowed the authors to intuitivelyreproduce and explore causal paths. Unlike static models,ABMs allow agents to modify decisions and behavior basedon past experience and/or new information. For instance,each customer agent had an individualized list of drugdealers and/or market locations to purchase heroin. Theselists were continuously updated by information collected bythe agent. If transactions with particular dealers end upbeing unsuccessful, that dealer was moved down an agentÕslist and the probability of using them diminished.

Procedures

Inherently transdisciplinary in nature, computationalmodeling requires close collaboration. The research team,which included a social scientist, computer programmer/statistician, and mathematician, jointly evaluated all agentschematics, model parameters, and logic to insure Þdelityand de-bug computer code. However, as with all compu-tational modeling projects the IDMS project bears thesignature of the methodology used to inform the model.How the ethnographic data was extracted and utilized wascritical.

In some instances, narrative and descriptive accountsdirectly informed programming the application. Whenpossible, behaviors, how activities were organized, deci-sion-making processes, interactive activity, and relationsbetween agents were programmed as described by partic-ipants. For example, private dealers described the processof using brokers to gain introductions to customers.Behaviors were also programmed based on participantobservation. For example, witnessing street sellers being

relatively immobile when selling drugs and other obser-vations of the actors within their environment made criticalcontributions to the validity of the data (see Hoffer2006)and model.

Ethnographic data was also extracted after it had beenprocessed analytically. Such Þndings are not accessibleusing other methods. For example, the explanatoryframeworks detailed inJunkie Business required consid-erable Þeldwork, bringing together multiple instances andexamples of behaviors. This process involved interpretingrelationships embedded in the market such as the ÔÔstreet-brokersÕÕ role. Brokers were deÞned as a social group byinterviewing, observing, and carefully investigating abroad range of issues, i.e.,thick description (Geertz1973).These Þndings produced a composite representation thatwas accessible to programming. In other words, the eth-nography allowed the behaviors of brokers to be mappedby their interactions and the environment.

Agent probabilities were translated into model parame-ters. This translation was a two-step process: First, datawere structured to identify the parameters necessary for themodel. Second, calibration of parameters was conducted tomatch known marginal values. We monitored globalcharacteristics such as means, standard deviations andshapes of histograms for several variables over the lengthsof the simulations to make sure that they were within thereasonable range. In this way, we balanced simplicity (i.e.,refraining from introducing unnecessary noise) and realism(i.e., ensuring that the collective behavior of all agentsreproduces believable results). Some parameters varied byagent while others were Þxed for entire agent groups. Forsome parameters, values were unknown and could only beestimated from the data (see ÔÔLimitationsÕÕ). For example,the probability that a customer knew a private dealer wasestimated from the data, while the threshold for entering aÔÔdesperateÕÕ state was calibrated to produce the observablerates of drug arrests and treatment admission. Using thishybrid combination of ethnography and ABM prospec-tively, researchers could accurately guide and targetresearch questions to better calculate these estimates.

The Market Model

The IDMS project incorporates six different agent types.Constructed to reproduce the operation of the market, thebehaviors focus on heroin sales and consumption.Cus-

tomers, brokers, sellers, and private dealers are the mostbehaviorally complex agents. These agents can learn aboutmarket, change their level of addiction based on heroin use,choose transaction partners, and modify their activitiesbased on past experiences.Police andhomeless agents areless complex. These agents are used more as proxies for


123

market conditions and are ÔÔreßexiveÕÕ in design. Agentbehaviors are outlined below:

1. Customers: Customers are the modelÕs most complexagents. All customer agents purchase, use and haveaddictions to heroin. Each agent incorporated pharma-codynamic characteristics that allowed their addictionto grow or diminish based on their use. All customershave an inventory of heroin that diminishes when theyuse. To use heroin, customers must purchase speciÞcamounts of the drug based on its full retail price (e.g., agram for $130). Retail drug prices were extracted fromthe ethnographic data. Customers attempt to use theamount of heroin that most closely matches theiraddiction level, i.e., their targeted use amount. Inpurchasing drugs, customer agents canpurchase theirtargeted amount or as much as possible. Customersreceive an income replenished weekly, bi-weekly, ormonthly. They can also receive money randomly.Customers keep track of the agents (sellers andbrokers) they purchase drugs from and update lists ofwhere to go to purchase based on previous successfulor unsuccessful sales.

2. Brokers: Like customer agents, broker agents also useheroin. They interact with customers by transportingmoney to sellers and returning to customer agents withheroin. Brokers do not have incomes, nor do theypurchase heroin with their own money. They onlyreceive heroin through a tax on the drug they purchasefor customers. Once they receive heroin from custom-ers they use it.

3. Sellers: Seller agents do not use heroin. Each has agiven inventory of heroin to sell. For all transactions,sellers charge and receive the full retail price for thedrug. Sellers stop selling when they are out of drug orwhen their shift ends, whichever occurs Þrst. If sellersare arrested they are sent to jail and removed from themarket. The length of time a seller is removed islinearly correlated to the amount of heroin they arearrested possessing.

4. Private Dealers: Private dealer agents also sell heroinin standard units and for the full retail price like sellers.However, a private dealer operates outside the open-air(public) market environment. Private dealer agentsonly sell heroin to customers, or brokers, they know. Abroker is the only agent that can introduce a customeragent and a private dealer.

5. The Police: Police agents patrol the market, randomlyinspecting and subsequently arresting agents whoposses heroin. Police ÔÔbustsÕÕ are represent by short-term increases in the numbers of police agentspatrolling the area, and by altering their inspectionand arrest rates.

6. The Homeless: Homeless agents take no active role inthe market or its transactions. Rather, homeless agentsprovide ÔÔnoiseÕÕ in the market by populating it. Policetreat homeless agents as if they were sellers, brokers,or customers.

All agents interact on a grid in space designated publicor private. In public space, representing the Larimer open-air drug market, all agents freely moved and interactedwithin this space. Agents were only able to ÔÔseeÕÕ otheragents in public space. Customer and private dealer agentsresided in private space. When customers wanted to pur-chase heroin they either traveled into the public space (i.e.,open-air market) or directly to a private dealer known tothem. Private dealers only resided in the simulationÕs pri-vate space.

Initial Conditions

Unlike theoretical ABM, the IDMS had basic estimates toset initial conditions. For example, the area of the simu-lation represented a square geographic area that was a half-mile by half-mile in size. To determine the number ofcustomers to populate the model, the authors worked fromestimates of brokers and street-sellers. These groups weresmaller and easier to estimate from the ethnography. Themodel was then calibrated with different numbers of cus-tomers and ratios of failed and successful deals wereevaluated. The Þnal version of the model included 345agents:n = 200 customers,n = 25 private dealers,n = 20street dealers, andn = 100 homeless agents. The numberof street brokers was varied from zero to 50 (see below).The public area of the model was also ÔÔpatrolledÕÕ by onepolice ofÞcer. This agent moved around the market,inspected agents randomly within a 100 ft. radius, andremoved agents who were found with heroin in their pos-session. The number of police agents was varied forexperiments (see below). Police agents removed otheragents for a time according to the amount of heroin foundin the agentÕs possession, i.e., the more heroin, the moretime. After serving the ÔÔjail term,ÕÕ agents returned to themarket.

Results

Outcomes in ABM research represent the aggregated out-put of agent behaviors. Because the researcher controls theenvironment, ABMs allow a researcher to set up virtualexperiments, such as comparing two scenarios with dif-ferent numbers of agents or different behavioral parame-ters. To extend the ethnography, model experimentsconcentrated on two research questions: (1) quantifying the


123

role of the broker relative to the marketÕs output, measuredby drug-related revenue; and (2) determining the inßuencepolice ÔÔbustsÕÕ had on the marketÕs long-term function,market dynamics and agent behaviors. Experiments weredescriptive and not intended to formally test speciÞchypotheses.

Because the behavior of agents is independent, eachsimulation run generates unique outcomes. To identifytrends, each 12-month scenario was replicated 50 times.Data shown at daily time-steps represents average valuesfor agents. All graphs start at day 100 because the Þrst3 months of each simulation represented a transient stateoccurring prior to the quasi-equilibrium that emerged.

Figure2 shows the number of daily customer transac-tions by seller in the absence of street brokers. In additionto noting the number of transactions the number of heroinunits sold is also indicated. Differences between thenumber of transaction and number of units sold slightlyincreases over time, indicating a change in customersÕpurchase amounts while their drug habits are maintained.Sales involving private dealers are low because the primaryway in which customers learned of private dealer locationswas through the activity of brokers.

The second Þgure shows daily customer transactions byseller when brokers are present in the market, Fig.3includes 25 street brokers and Fig.4 includes 50. Streetbrokers change market dynamics by increasing the totalamount of drugs sold and enhancing the role of the privatedealers. After approximately 90 days, private dealer salesovertake market sales from street dealers, ultimatelyreducing the number of sales by street dealersÕ by almosthalf. These results are dramatized when 50 street brokersare in the market. This illustrates the convenience ofknowing a private dealer in making the drug more acces-sible. Interestingly, the brokers inclusion in the market

results in more transactions; however, doubling the numberof brokers does not markedly increase transaction totals.Overall, when the number of street brokers is very high weobserve the brokers make more transactions than streetdealers. This does not mean that the role of the streetdealers is drastically decreased. The brokers still usestreet dealers as a source of drug supply.

Figures5 and6 involve evaluating the effect of a drugbust on market dynamics. On day 120 of the simulation weadded 29 police agents into the market (n=30) for a 24-hperiod. For this experiment we included 50 street brokers.

In the short-term, the drug bust scenario had an effect onall market agents. Figure5 notes a sharp reduction in streetdealer sales at the time of the bust, i.e., as seller agentswere arrested. However, simultaneous to this change, saleswith private dealers increase. Although the bust results in asharp short-term drop in total transactions in the market,brokers play a key role in mitigating effects. In Figure6 wesee during the bust, brokers transfer customer sales fromstreet sellers to private dealers. In fact, a drug bust can

Number of customer transactions by sales source with no brokers

100 150 200 250 300 350

010

020

030

040

050

0

Time, days

Nu

mb

er o

f tr

ansa

ctio

ns

and

un

its

All TransactionsAll Units/50Street DealerPrivate Dealer

Fig. 2 Number of daily customer transactions by seller when nostreet brokers are present

Number of customer transactions by sales source with 25 brokers

100 150 200 250 300 350

010

020

030

040

050

0

Time, days

Nu

mb

er o

f tr

ansa

ctio

ns

All TransactionsStreet DealerPrivate DealerStreet Broker

Fig. 3 Number of daily customer transactions by seller when 25street brokers are present

Number of customer transactions by sales source with 50 brokers

100 150 200 250 300 350

010

020

030

040

050

0

Time, days

Nu

mb

er o

f tr

ansa

ctio

ns


Fig. 4 Number of daily customer transactions by seller when 50street brokers are present


123

remove nearly all the street dealers with little long-termeffect. During these timeframes the market relies on bro-kers transitioning customer sales to private dealers. Afterthe incarceration of street dealers ends and they return tothe market, they are unable to generate the same revenue asa large portion of their ÔÔbusinessÕÕ has been taken over byprivate dealers.

Limitations

Like other modeling efforts, in ABM research the inves-tigator is ultimately responsible for developing the modeland model limitations often reßect what the modeler wants,or can, include. Limitations vary in concert with whatfeatures of social behavior the researcher decides to model(Gilbert and Toritzsch1999). The objective of the IDMSproject focused on reproducing the operation and history ofa local heroin market informed by ethnographic data.Capturing the nuances associated with agent/environmentinteractions, the simulation involves a considerable amount

of complex behavior. However, it is important toacknowledge how this ÔÔmirror to realityÕÕ is imperfect.

First, agents in this model only interact in limited ways.Customers do not communicate or cooperate with oneanother, nor do street sellers operate in coordinated ways,e.g. when sellers sell all the heroin they have, they do notgo to fellow sellers and get more heroin. Although addingthese interactions would have increased the simulationÕsrealism, decisions and computational limitations constrainthe behavioral repertoire of agents in ABM research.

In our simulation customers could not purchase heroinfor less than full retail price, nor could they use credit topurchase heroin. The primary authorÕs ethnography detailsthat both these activities were important in reducing acustomerÕs cost for drugs; both also involve dealers closelymonitoring sales relationships and customer behavior. Inother words, heuristics were available to program. How-ever, incorporating algorithms to reßect elastic pricing andissuing credit would have exponentially complicatedagentÕs simulated interactions and it was decided to includethese behaviors in subsequent model extensions, or in adifferent model.

This simulation focuses on demand and not on supply.Mechanisms for supplying the market with heroin are notincorporated, i.e., when selling agents run out of inventorythey are automatically re-supplied. This decision was datadriven. Despite high levels of rapport with the heroindealers inJunkie Business, the ethnographer was unable tointerview members of the Mexican DTO supplying thesedealers with heroin.

Unique features of the Larimer market are integrated in themodel and potentially limit its generalizablity. For instance,variations in heroin quality (i.e., purity) are not included. Atthe time, most of the heroin in Denver was Mexican black tar,of low purity, and supplied from homogeneous sources.Customers had a general sense that heroin sold by privatedealers was better in quality than heroin purchased on thestreet, but substantive reports ofpurity variationswere rare. Asa result, agentsÕ assessments of the product are minimal andonly indirectly affected decision-making.

Finally, the available data often determines what ismodeled. The IDMS project attempted to employ a data-driven protocol, grounding model components in the data.Nonetheless, it must be acknowledged that the originalresearch wasnot originally collected for the purposes ofconstructing an ABM. Gaps in data were informed by acombination of literature review and consultation withsubject matter experts. Such substitute techniques areclearly not ideal. An important secondary objective of thisexperimental project was to identify prospective researchmethods combining ethnography and ABM to minimizethese limitations in future studies implementing thisapproach.

Number of customer transactions by sales source

100 150 200 250 300 350

010

020

030

040

050

0

Time, days

Nu

mb

er o

f tr

ansa

ctio

ns


Fig. 5 Number of daily customer transactions by seller when 50street brokers are present and a police bust occurs at day 120

Number of broker transactions by sales source

100 150 200 250 300 350

050

100

150

200

Time, days

Nu

mb

er o

f tr

ansa

ctio

ns

All TransactionsStreet DealerPrivate Dealer

Fig. 6 Number of daily broker transactions by sales source


123

Discussion

Experimental results presented are the Þrst in a series usingthis computational methodology to explore the dynamics ofthe Larimer area heroin market. Thus far, experimentsconÞrm the importance of the role of the broker in themarket, as well as the limited effectiveness of police bustson the marketplace. Combined, these Þndings suggest thatbrokers stabilize heroin sales when the police attempt todisrupt market activities by arresting street-based dealers.

One novel policy implication suggested by these resultswould be to speciÞcally target brokers with interventions,removing them from the market. As brokers do not actuallybuy or sell signiÞcant amounts of heroin, arrest wouldlikely not yield success. Our model already includes bro-kers being arrested. However, because these users tend tohave highly variable heroin habits, limited economicresources, and are homeless and/or live transient lifestyles,maybe other interventions (and/or incentives) might work.Future experiments with our market simulation couldidentify a functional threshold of brokering activity. Inother words, how much brokering (in the aggregate and asan average per broker) is required for the market to suc-cessfully adapt to police intervention? Sample size, speciÞcoutcome objectives, as well as inclusion/exclusion criteriacould then be developed for an intervention. Next, usingABM the potential broker-targeted intervention could beprototyped and evaluated before being rolled-out into theÞeld. After such an intervention is in operation, statisticaland ethnographic data could feedback into the ABM toupdate behaviors, extrapolate potential unintended conse-quences, and project results. To date, this sort of real-timeintegrated relationship between data collection, interven-tion and ABM is yet to be realized. However, the capacityis in place to make it happen.

As a feasibility study, additional contributions the IDMSproject makes relate to future research projects integratingABM. In our project, programming agent speciÞcationsthat incorporate environmental conditions providesresearchers with a novel framework to identify and eval-uate parameters essential to market performance. Forexample, in brokered sales the threshold of transactionsoccurring before private dealers sold directly to customerswas critical. If this threshold were too high customerswould be unable to connect with private dealers. Identi-fying parameters and how they contribute to the function ofthe system greatly enhances opportunities for futureresearch. Because data collection and analyses are iterative,this type of feedback has particular utility in ethnographicresearch. In other words, an ethnographer could conductpreliminary analyses using ABM to guide follow-upinterviewing and observation. An important next step incomplex systems research will be to prospectively collect

both the descriptive data required to identify parameters, aswell as the systematic data necessary to set them.

Finally, because this project was speciÞcally orientedto an ethnographic case study of a heroin market thatoperated in Denver, Colorado, 1992Ð1996, its generaliz-ability needs to be considered. As a feasibility study of ahybrid methodology, this discussion is bifurcated by theproduct described, i.e., the market model, and the tech-niques used in its development, i.e., combining ethnog-raphy and ABM.

Although this was an ethnographic study of a localheroin market operating in a very speciÞc place and time,the methodology this manuscript describes is entirelytransportable. Independent of topic, in ethnographic studiesthat: (1) describe and situate social behaviors of partici-pants within an environment, and (2) delineate change overtime, opportunities exist to integrate ABM. This is becausesocial processes, i.e., what ABMs model, are mandated inconnecting these themes. Importantly, programming ABMsusing data extracted from ethnography is not forsakingcultural complexity or the relevance of culture. Rather, asin the case presented here, the ethnographer is distillinginteractions and/or belief systems that generated theobserved outcome(s) of interest for quantiÞcation and moredetailed analysis. As a potential procedure to observequantitative output from qualitative inquiry, this method isparticularly promising for ecologically oriented anthropol-ogists.

For cultural anthropologists and other researchers usingqualitative methods, the process of constructing ABMsfrom ethnographic data opens exciting new horizons.SpeciÞcally, it provides opportunities to quantify out-comes. During the process of building an ABM researcherscan identify the data required to produce a veriÞablemodel. Currently, quantitative aggregated data on behavioris only available via epidemiological surveys. Yet thesemethods alone cannot demonstrate how or why patterns indata occur. As in the IDMS project, ABMs offer a linkbetween ethnographic details, i.e., an explanatory frame-work, and aggregated outcomes. While this new ethno-graphic tool needs further reÞnement to make suchquantiÞcation commonplace, it offers a unique procedurefor detailed investigation and experimentation.

The value of our market model to the Þeld of illegal drugmarket research ultimately involvescreating a framework

in which drug markets can be compared. Reßecting one ofthe monumental challenges to this Þeld, the generalizabilityof the Larimer market model to other illegal drug marketcontexts is impossible to fully determine. Currently,researchers who ethnographically characterize differentdrug markets, for different drugs, in different times, andusing different theoretical perspectives, are highly chal-lenged in comparative analysis (Curtis and Wendel2000).


123

Clearly, drug markets in other geographic settings areunique from Denver. One immediate difference in theLarimer area was the absence of gang activity associatedwith heroin dealing. But customer, dealers, brokers, and thepolice exist in almost all local drug markets; in the cases ofmarijuana and methamphetamines, producers may alsoexist. In our model, most activities of and relationshipsbetween agents are generic. Customer agents representingheroin users have a drug addiction, inventory of drugs,supply of money to buy drug, and an understanding ofwhere to go to connect with dealers or brokers. Brokeragents simply mediate heroin sales between buyers andsellers for a tax. With rudimentary behaviors of customersand dealers programmed, organizational and interactiverelationships among agents and between agents and theenvironment (i.e., what makes markets unique) can becompared and contrasted. Future iterations of this hybridapproach can also be designed to link simulation outcomesto community epidemiological indicator data such as drugarrests, treatment entry, and hospital admissions in an effortto track trends. Using ABM to provide a framework forcomparing drug markets, dealerÕs behaviors, and outcomescould greatly enhance the application of data from ethno-graphic studies of illegal drug markets.

This paper presented an experimental research projectthat used detailed ethnographic data to simulate thedynamics of a local heroin market and enhance ourunderstanding of this market. While research objectivespresented in this manuscript clearly introduce new meth-odological and collaborative challenges, they exploit thebeneÞts of using this ecological approach to model socialsystems, evaluate interventions, and project outcomes. Associal and behavioral scientists continue to develop ways toexpress the importance of environmental factors on thebehaviors of individuals, as well as how individuals modifytheir environment, our research supports these objectivesthrough synthesizing ABM and ethnographic research.

Acknowledgments The authors would like to acknowledge supportfor this research from the National Institutes of Health, NationalInstitute of Drug Abuse, R21 DA019476.

References

Adler, P. A. (1985).Wheeling and dealing: An ethnography of anupper-level drug dealing and smuggling community. New York:Columbia University Press.

Agar, M. (1973).Ripping and running. New York: Academic Press.Agar, M. (1999). Complexity theory: An exploration and overview

based on John HollandÕs work.Field Methods, 11(2), 99Ð120.Agar, M. (2001). Another complex step: A model of heroin

experimentation.Field Methods, 3(13), 353Ð369.Agar, M. (2005). Agents in living color: Towards emic agent-based

models.Journal of Artificial Societies and Social Simulation,8(1). http://jasss.soc.surrey.ac.uk/8/1/4.html.

Agar, M., Bourgois, P., French, J., & Murdoch, O. (1998). Heroinaddict habit size in three cities: Context and variation.Journal ofDrug Issues, 28(4), 921Ð940.

Bankes, S. C. (2002). Agent-based modeling: A revolution?Pro-ceedings of the National Academy of Sciences of the UnitedStates of America, 99, 7199Ð7200.

Bonabeau, E. (2002). Agent-based modeling: Methods and techniquesfor simulating human systems.Proceedings of the NationalAcademy of Sciences of the United States of America, 99, 7280Ð7287.

Bourgois, P. (1997).In search of respect: Selling crack in El Barrio.New York: Cambridge University Press.

Byrne, D. (1998).Complexity theory and the social sciences. London:Rutledge.

Carley, K. M. (2002). Computational organization science: A newfrontier. Proceedings of the National Academy of Sciences of theUnited States of America, 99, 7257Ð7262.

Caulkins, J. (1997). Modeling the domestic network for illicit drugs.Management Science, 43(10), 1364Ð1371.

Caulkins, J., & Reuter, P. (2006). Illicit drug markets and economicirregularities.Socio-Economic Planning Sciences, 40, 1Ð14.

Conte, R. (2002). Agent-based modeling for understanding socialintelligence.Proceedings of the National Academy of Sciences ofthe United States of America, 99, 7189Ð7190.

Conte, C., & Gilbert, N. (1995). Introduction. In N. Gilbert &C. Conte (Eds.),Artificial societies: The computer simulation ofsocial life (pp. 1Ð15). London: UCL Press.

Conte, C., Hegselmann, R., & Terna, P. (1997).Simulating socialphenomena. Berlin: SpringerÐVerlag.

Curtis, R., & Wendel, T. (2000). Toward the development of atypology of illegal drug markets. In M. Natarajan & M. Hough(Eds.),Illegal drug markets: From research to prevention policy(pp. 121Ð152). Monsey, NY: Criminal Justice Press.

Doran, J. E., & Gilbert, N. (1994).Simulating society: The computersimulation of social phenomena. London: UCLA Press.

Eck, J. E., & Gersh, J. (2000). Drug trafÞcking as a cottage industry. InM. Natarajan & M. Hough (Eds.),Illegal drug markets: Fromresearch to policy (pp. 241Ð271). Monsey, NY: Criminal JusticePress.

Epstein, J. M. (2006).Generative social science: Studies in agent-basedcomputational modeling. Princeton: Princeton University Press.

Epstein, J. M., & Axtell, R. (1996).Growing artificial societies.Cambridge: MIT Press.

Forrester, J. (1994). Policies, decisions, and information sources formodeling. In J. Morecroft & J. Sterman (Eds.),Modeling forlearning organizations (pp. 51Ð84). Portland, OR: ProductivityPress.

Geertz, C. (1973).The interpretation of cultures. New York: BasicBooks.

Gilbert, N. (2007).Agent-based models. London: Sage Publications.Gilbert, G. N., & Conte, R. (1995).Artificial societies: The computer

simulation of social life. London: UCL Press.Gilbert, N., & Toritzsch, K. G. (1999).Simulation for the social

scientist. Philadelphia: Open University Press.Goldspink, C. (2002). Methodological implications of complex systems

Approaches to sociality: Simulation as a foundation for knowl-edge.Journal of Artificial Societies and Social Simulation, 5(1).http://jasss.soc.surrey.ac.uk/5/1/3.html.

Green, L. W. (2006). Public health asks of systems science: Toadvance our evidence-based practice, can you help us get morepractice-based evidence?American Journal of Public Health, 96,406Ð409.

Hamid, A. (1991a). From ganja to crack: Caribbean participation inthe underground economy in Brooklyn, 1976Ð1986. Part 1.Establishment of the marijuana economy.The Internationaljournal of the addictions, 26(6), 615Ð628.


123

http://jasss.soc.surrey.ac.uk/8/1/4.html

http://jasss.soc.surrey.ac.uk/5/1/3.html

Hamid, A. (1991b). From ganja to crack: Caribbean participation inthe underground economy in Brooklyn, 1976Ð1986. Part 2.Establishment of the cocaine (and crack) economy.The Inter-national journal of the addictions, 26(7), 729Ð738.

Hoffer, L. (2006).Junkie business: The evolution and operation of aheroin dealing network. Belmont, CA: Thompson Wadsworth.

Holland, J. H. (1995).Hidden order: How adaptation buildscomplexity. New York: Addison-Wesley.

Holland, J. H. (1998).Emergence: From chaos to order. RedwoodCity, CA: Addison-Wesley.

Homer, J. B., & Hirsch, G. B. (2006). System dynamics modeling forpublic health: Background and opportunities.American Journalof Community Psychology, 96(3), 452Ð458.

Hough, M., & Natarajan, M. (2000). Introduction: Illegal drugmarkets, research, and policy. In M. Hough & M. Natarajan(Eds.),Illegal drug markets: From research to prevention policy(pp. 1Ð17). Monsey, NY: Criminal Justice Press.

Johnson, B. D. (2003). Patterns of drug distribution: Implications andissues.Substance Use and Misuse, 38(11Ð13), 1789Ð1806.

Johnson, B. D., Dunlap, E., & Tourigny, S. (2000). Crack distributionand abuse in New York. In M. Natarajan & M. Hough (Eds.),Illegal drug markets: From research to prevention policy(pp. 19Ð57). Monsey, NY: Criminal Justice Press.

Johnson, B. D., Goldstein, P. J., Preble, E., Schmeidler, J., Lipton, D. S.,Spunt, B., et al. (1985).Taking care of business: The economics ofcrime by heroin abusers. Lexington, MA: D.C. Heath andCompany.

Koester, S. K. (1994). Copping, running, and paraphernalia laws:Contexual variables and needle risk behavior among injectiondrug users in Denver.Human Organization, 53(3), 287Ð295.

Koester, S. K., & Hoffer, L. (1994). Indirect sharing: Additional risksassociated with drug injection.Aids and Public Policy, 9(6),100Ð105.

Leischow, S. J., & Milstein, B. (2006). Systems thinking andmodeling for public health practice.American Journal of PublicHealth, 96, 403Ð405.

Lempert, R. (2002). Agent-based modeling as organizational andpublic policy simulators.Proceedings of the National Academyof Sciences of the United States of America, 99, 7195Ð7196.

Levin, G., Gilbert, E. B., & Hirsch, G. B. (1975).The persistentpoppy: A computer-aided search for heroin policy. Cambridge,MA: Ballinger Publishing Company.

Levy, D. T., Bauer, J. E., & Lee, H. (2006). Simulation modeling andtobacco control: Creating more robust public health policies.American Journal of Public Health, 96, 494Ð498.

Mabry, P., Olster, D., Morgan, G., & Abrams, D. (2008). Interdis-ciplinary and systems science to improve population health.American Journal of Preventive Medicine, 35(S2), S211ÐS224.

Marney, J. P., & Tarbert, H. F. E. (2000). Why do simulation?Towards a working epistemology for practitioners of the darkarts.Journal of Artificial Societies and Social Simulation, 3(4).http://www.soc.surrey.ac.uk/JASSS/3/4/4.html.

McLeroy, K. (2006). Thinking of systems.American Journal ofPublic Health, 96, 402.

McManus, R. (2002) A new kind of science: Wolfram goes fromsimple rules to complex forms.NIH Record, 1Ð4.

Preble, E., & Casey, J. (1969). Taking care of business: The heroinuserÕs life on the street.The International journal of theaddictions, 4(1), 1Ð24.

Reuter, P., MacCoun, R., & Murphy, P. (1990).Money from crime: Astudy of the economics of drug dealing in Washington. D.C.Santa Monica, CA: RAND-Drug Policy Research Center.

Sawyer, K. R. (2003). ArtiÞcial societies: Multiagent systems and themicroÐmacro link in sociological theory.Sociological Methodsand Research, 31(3), 325Ð363.

Sawyer, K. R. (2006).Social emergence: Societies as complexsystems. New York: Cambridge University Press.

Schelling, T. C. (1978).Micromotives and macrobehavior. NewYork: Norton.

Singer, M. (2006).Something dangerous: Emergent and changingillicit drug use and community health. Long Grove, IL: WavelandPress.

Sterman, J. D. (2006). Learning from evidence in a complex world.American Journal of Public Health, 96, 505Ð514.

Waldrop, M. M. (1992).Complexity: The emerging science at theedge of order and chaos. New York: Simon & Schuster.

Wolfram, S. (1986).Theory and applications of cellular automata:Including selected papers, 1983–1986. Singapore:World ScientiÞc.


123

http://www.soc.surrey.ac.uk/JASSS/3/4/4.html

Researching a Local Heroin Market as a Complex Adaptive System

Documents

Transcript of Researching a Local Heroin Market as a Complex Adaptive System