Computational Ethnography and Agent-based Modeling · number, just use the pager” ... Analyze the ABM, Understand the relationships . Conduct a focused survey Update model parameters

1

RTI International is a trade name of Research Triangle Institute

3040 Cornwallis Road ■ P.O. Box 12194 ■ Research Triangle Park, North Carolina, USA 27709 Phone 919-541-6167 e-mail [email protected] 919-541-6722

Computational Ethnography and Agent-based Modeling

Georgiy V. BobashevJoey Morris (RTI), Daniel Heard (Duke), and Lee Hoffer (Case Western)

Presented at SAMSI 2013

The Problem

2

Causality and Surveys

A typical solution to a problem:Increase law enforcement

Restrict access to prescription drugs

Why policies don’t work? Adaptive behavior?

Not a straightforward solution

Lack of understanding of what is going on in the community

Responses are often based on static surveys

How to understand causality and possible solutions?3

Complex Systems

In engineering scientists stopped being linear and one-dimensional long time ago. Rocket science ☺

Social sciences could be more complex than most of mechanical systems, but the approach is often way too simplistic

4

What is Ethnography?

Ethnography (from greek ἔθνος ethnos "folk, people" and γράφω grapho "to write") is a qualitative research method designed to represent graphically and in writing, the culture of a people. The results (e.g. a report) reflects the knowledge and the system of meanings in the lives of a cultural group.

Ethnography, as the empirical data on human societies and cultures, was pioneered in the biological, social, and cultural branches of anthropology

5

An Ethnographic Study of Heroin Use

Lee Hoffer “Junkie Business”

How does heroin market recover and why police busts don’t work?

"The difference here is subtle but clear: dealing heroin is a social behavior with economic outcomes rather than an economic behavior with social outcomes"

6

Turning Narratives into Rules

“That guy buys a gram and a half every #@$&% day, you know. Every day. I mean he’s doing $130 worth of @#$& business every day, seven days a week. He’s got some $35,000 or $40,000 inheritance”

What does it mean for a modeler?Probability P(Windfall), Amount X~Distribution(P), P buy the max amount (1.5 gram).

Where do we get the values? Ask the ethnographer (Lee)

7

Turning Narratives into Rules

“There is Neighborhood Watch and #$@% like that here. The neighbors pay attention to what’s happening. At 8p. we stop the business. We don’t want no traffic over here, So don’t ever come here with anybody else. Don’t call the home number, just use the pager”

What does it mean for a modeler?No deals with private dealers happen after 8pm. The probability that a drug user is getting a drug after 8pm is small.

8

9

Actors in the Agent-Based Model

Customer (Drug User)Dose size depends on the habit

Street Dealer

Private Dealer

Street Broker

Homeless

Police officer

10

Schematic ABM Market Model

Street Dealer

Broker

Police Officer

Private Dealer

Homeless

Customer

11

Drug User (Customer)

Individual Addiction LevelsSimple Neurobiological Process

12

Exogenous factors: Drug availability

Effect(concentration)

A(t,τ1)HabitY/N

use

Endogenous factors: Dose size, Time since the last dose

Exogenous factors: Competition with other reinforcers, External Cues, Stress, etc.

Step jump in dose concentration

Exponential concentration decay

Habit level depends on the average concentration

“Desperation” as a measure of negative reinforcement

0

100

200

300

0 1 2 3 4 5 6 7 8 9 10 11 12

Addi

ctio

n Le

vel (

in u

nits

)

Simulation Months

Agent #59

addiction Avg Addiction

Example of a Habit Dynamics

14

Street Dealer

15

Street Broker

The Role of Street Brokers and Private Dealers

16

17

The Role of Street Brokers and Private Dealers

18

The Role of Street Brokers and Private Dealers

19

The Role of Street Brokers and Private Dealers

20

Results

Street Brokers do not necessarily compete with Street dealers but rather compliment their activities and thus increase market performance

Street Brokers buffer from the police busts

Street Brokers increase market volume especially after the market busts

After the busts the Street Brokers are taking away some business from the Street dealers

Complex Systems and Protocols

On 30 October 1935, Army Air Corps test-pilot Major Ploer Peter Hill and Boeing employee Les Tower took the aircraft on a second evaluation flight; however, the crew forgot to disengage the airplane's "gust lock," a device that held the bomber's movable control surfaces in place while the aircraft was parked on the ground. Having taken off, the aircraft entered a steep climb, stalled, nosed over and crashed, killing both, Hill and Tower.

“Flying Fortress” B-17 the largest and the most sophisticated aircraft of the time based on the Boeing Model 299.

ODD “Standard” Protocol (Grimm et al. 2006, Railsback and Grimm, 2011)

PurposeOverview Entities, state variables and scales

Process overview and schedulingEmergenceObjectivesAdaptation

Design Learningconcepts Prediction

SensingInteractionStochasticityCollectivesObservationInitialization

Detail Input dataSubmodels

Can We Simplify the Model?(timestep)

23

Can We Simplify the Model?(complexity)

24

Not a Trivial TaskNot a Simple Gaussian Emulator

Need to estimateDistributions (e.g. Beta, Gamma)

Dependencies (regressions)

Reduced model doesn’t fit?

How to account for learning?

How to account

for police busts?

25

Nonlinear Relationship Between the Variables

26

How to Test Whether Models Are Equivalent?

27

Log p-values Characterizing Equivalence

Compare the Full and the Reduced models pointwise

28

Can We Simplify the Reduced Model?(timestep)

29

30

Lessons Learned

• Heroin market is adaptive• Need to consider multiple players and relationships

between them• These relationships could be explored through simulation

modeling• If we want to apply social network interventions need to

understand concepts of trust • Don’t need to collect a lot of useless survey data but rather

collect small amount of correct data that ethnography can teach us

• Collaborate!

31

Impact on Survey Methodology and Social Science

• Current approach Hypothesize a relationship Conduct a survey Analyze using regression model Develop Policy

• New proposed approachConduct an ethnographic study Build an ABM Analyze the ABM, Understand the relationships Conduct a focused survey Update model parameters

Develop Policy

32

Publications and AcknowledgementsHoffer, L. (2006) Junkie Business: The Evolution and Operation of a Heroin Dealing Network, Belmont, CA: Thompson Wadsworth.

Bobashev, G. V., Costenbader, E. M., & Gutkin, B. S. (2006). Comprehensive mathematical modeling in drug addiction sciences. Drug and Alcohol Dependence, 89(1), 102–106.

Hoffer L. & Bobashev G. V. (2009). Researching a local heroin market as a complex adaptive system. American Journal of Community Psychology.

Hoffer L., Bobashev G.V., Morris R.J., (2011) Simulating patterns of heroin addiction within the social context of a local heroin market. In Gutkin B. & Ahmed S. (Ed.), The computational neuroscience of drug addiction. Springer Verlag. pp. 313-331

Newlin D.B., Regalia P.A., Seidman T.I., and Bobashev G.V. (2011) Control Theory and Drug Addiction, In Gutkin B. & Ahmed S. (Ed.), The computational neuroscience of drug addiction. Springer Verlag. pp.: 58-107

• NIH/NIDA R01 (PI Hoffer and Bobashev)“Researching social dynamics of local methamphetamine market”

• NIH/NIDA R21 (PI Bobashev) “Systems Approach to Modeling of Drug Use Recovery ”

Contact: [email protected]

Additional Slides

33

Entities, State variables, and Scales

What are the model components?

Agents (Who?)

Environments (Where?)

Major variables (global and state variables) (What?)

Major rules (How?)

Temporal and spatial scales (discrete and continuous)

Populations and boundaries (time, space, population) 34

Process Overview and Scheduling

How are the components functioning?

What are agents doing? (e.g., searching, mating, fighting, competing)

How is environment changing? (e.g., providing resource, restricting activities)

What is the observer doing? (e.g., collects the data)

35