Dynamic Simulation of Community Crime and Crime-Reporting Behavior Michael Yonas, Jeffrey Borrebach, Jessica Burke, Shawn Brown, Richard Garland, Katherine.

Dynamic Simulation of Community Crime and Crime-Reporting Behavior

Michael Yonas, Jeffrey Borrebach, Jessica Burke,Shawn Brown, Richard Garland, Katherine Philp, Donald Burke,

and John GrefenstetteGraduate School of Public Health

University of Pittsburgh

ISSH 2011May 26 2011

Community Crime and Public Health

• Violence is 2nd leading cause of death for youth ages 15-24• Leading cause of death for African American youth• Community interventions include encouraging citizens and

victims to report crime• However, approximately half of violent crimes go unreported• We are developing models to examine and evaluate

potential community crime intervention strategies

3

Background: Collective Efficacy

• Collective efficacy refers to the effectiveness of informal mechanisms by which residents achieve public order (Sampson et at, Science 1997) A key mechanism influencing interpersonal violence in a

neighborhood Contrasts with formally or externally induced actions

e.g. police intervention, administrative programs• Examples

monitoring children's play groups willingness to intervene against truancy, loitering Local intervention to control social disorder, such as

graffiti

4

Research Questions

Q1: How do these factors operate in a dynamic environment?

Q2: Can we capture other theoretical constructs in a spatially explicit model (i.e., dynamics of neighborhood contagion)?

Q3: Can we develop metrics to permit comparison and evaluation of possible intervention programs prior to implementation?

Q4: Can modeling approach help us develop novel community intervention programs to increase collective efficacy?

In addition, can a modeling approach be used as a tool to cultivate community-partnered research (community-based participatory research (CBPR))?

5

Why Use an Agent-Based Model?

• Collective efficacy depends on individual decisions and actions

• Capture effects of individual experience change in attitude of witnesses or victims change in attitude of offenders due to punishment (or not)

• Reflect influence of interpersonal relationships family, friends, community Engage respected members of the neighborhood victims and friends of victims

6

Graffiti serves as a proxy for community crime

7

Agents in the Model

• Adults do not commit offenses in this model may witness a nearby incident may report a witnessed incident

• Juveniles may commit an offense if reported, may receive punishment

assumed to reduce likelihood to offend in future

8

Community Model

• Blocks are different colors to aid visualization• Agents are colored to show their current activity

Two-dimensional space (100 x 100 toroid – wraps around)

1000 Agents• 90% adults• 10% juveniles• similar to Pittsburgh, PA

Community is subdivided into 25 blocks to measure geographic statistics

9

Crime Reporting Behavior for Adults

• Have a fixed home location

• May witness incidents near their home each adult i has a probability of witnessing pwitness,i

• May report incidents they witness each adult i has a probability of reporting preport,i

• Both probabilities are drawn from uniform random distributions, such that 50% of incidents are witnessed 50% of witnessed incidents are reported

10

Offending Behaviors for Juveniles

• Based on Theory of Reasoned Action (Ajzen, 1980): if perceived reward > perceived risk, then take action

• Perceived reward assigned randomly to individuals and declines with age

• Perceived risk depends on experience: Initial values assigned randomly to individuals If an offense is committed and is punished, perceived risk

increases If an offense is committed and is not punished, perceived risk

decreases

• Juveniles may move through the community attracted toward areas associated with unpunished offenses

11

Daily Time Step of Simulation

1. Each juvenile moves a small distance attracted toward areas of high opportunity

2. Each juvenile decides whether to commit an offense based on the individual's current preceived risk and reward

3. Each adult probabilistically witnesses any nearby offense based on the individual's current witnessing probability

4. Witnesses decide whether to report each offense based on the individual's current reporting probability

5. If an offender is reported, he/she may received punishment with a global probability of punishment

6. If an offender is punished, increase perceived risk

7. If an offender is not punished, decrease perceived risk

12

Movement of Juvenile Agents

• Assumptions: They share information about detection risks at locations Offenders prefer locations with low risk of consequences

• Mechanisms: Each location has an associated "opportunity index"

increases if there are unpunished offenses in the nearby surrounding area

decreases otherwise Each offender moves each time step:

50% chance of random direction 50% chance of moving to the highest opportunity

location within a few steps of current location

13

Visualization of current agent activity

• Blue = adult, not witnessing• Yellow = witnessing, not

reporting• Green = witness and reporting

• White = offender not being reported• Red = offender being reported• Orange = offender being reported

punished

14

Calibrating the Offenders

Pathways to Desistance Study (Shubert and Mulvey, 2011)

• Longitudinal survey of N = 1,354 active juvenile offenders over a three-year period

• Collected self-reported frequency of offenses, perceived rewards, perceived risk, at 6 month intervals

• Analysis explored differences in risk perception based on prior offending experience Did perceptions shift as the result of arrest?

15

Risk, Reward Perception by Offender Frequency

Observations (Shubert and Mulvey, 2011)

• The most frequent offenders perceived significantly less risk and more reward from crime than those with medium frequency of offenses

• Less frequent offenders perceived significantly more risk and less reward • Individuals decrease the level of perceived risk when offending is undetected or avoids

punishment• Individuals tend to increase the level of perceived risk when they are arrested• As individuals age, perceived reward appears to decrease for all levels of offender

frequency

16

Risk, Reward in the Model: Step 1 – Fixed individual levels

• Assume each offender has random level of perceived risk and reward drawn from U(0.25, 0.75)

• Assume no changes due to detection or punishment• Result: More frequent offenders perceive less risk and more reward than less

frequent offenders Plots show mean of 5 runs of simulation

17

Risk, Reward in the Model: Step 2 – Getting away with it

• Assume individuals decrease their perceived risk when offense is undetected or avoids punishment:

riski = riski + a, where a > 0

For example: a = 0.0002 * riski • Result: More frequent offenders reduce their perceived risk over time more than less

frequent offenders Plots show mean of 5 runs of simulation

18

Risk, Reward in the Model: Step 3 – Effects of punishment

• Assume individuals increase perceived risk when punished:riski = riski – b, where b > 0

for example, b = 0.005 * riski • Result: More frequent offenders increase their perceived risk over time more

than less frequent offenders Plots show mean of 5 runs of simulation

19

Risk, Reward in the Model: Step 4 – Rewards decline with age

• Assume individuals decrease perceived reward over time, for all levels of offendersrewardi = rewardi – c, where c > 0

for example, c = 0.00025 * rewardi • Result: More frequent offenders increase their perceived risk over time more than

less frequent offenders Plots show mean of 5 runs of simulation

20

Calibrated Model

Shubert and Mulvey, 2011 Simulation Data

21

Community Interventions to Enhance Collective Efficacy

• Enhancing Witness Ability Improved street lighting Encouraging neighborhood outdoor activity

• Enhancing Reporting Police hotlines Anonymous reporting systems Phone trees

• Enhancing Offender Intervention Intervention groups Parent education

22

Geographic Heterogeneity

• Like real neighborhoods, the blocks within the model community vary in several factors number of adults density of houses individual differences in probability of witnessing and of

reporting

• These factors lead to block-level differences in incidence rates and punishment rates

• As a result, some area are expected to attract offenders who seek low risk areas

23

Visualization of Patch Incidence Levels

Colored irregular regions are associated with high numbers of recent incidents.

Offenders follow local gradients of "high opportunity", leading to clustering offenses.

24

Spatial distributions of incidents

(a) (b) (c)

(a) Incident Level by Block, showing heterogeneity(b) Histogram of perceived opportunity, splits into high- and low-

opportunity areas(c) Histogram of Offense Level per location, results from perceived

opportunity distribution

25

Comparing Community Intervention Programs

• Control: no intervention

• Global Intervention Select T% of adults from the entire community, and

influence them to always witness and report nearby offenses

• Targeted Intervention Select T% of adults from the area surrounding the blocks

with the highest number of offenses, and influence them to always witness and report nearby offenses

26

Activated Adults – Global Intervention

27

Activated Adults – Targeted Intervention

28

Which intervention works best?

• What is "best"? ....

30

Which intervention works best?

• Targeted intervention provides more saturated reporting opportunities, and may be more effective at reduce offenses than global intervention

• Global intervention may affect more offenders• Targeted intervention may encourage offenders to

move to other neighborhoods• Which methods has the largest effect?

31

Experimental Design

• Three interventions: Global Targeted None (control)

• Each intervention activated the same number of adults Always witness nearby offenses Always report witness offsenses Offenders reported by activated witness are always punished

• Measurements: Overall number of offenses post intervention The distance moved by offenders affected by the intervention

those offenders who are reported by activated adults• 20 runs of 270 days, with intervention starting on day 90

32

Intervention Effects

Conclusions:• Global and targeted intervention reduce overall offenses• Targeted intervention may have a larger effect on the

targeted offenders• Targeted intervention leads to offenders moving to

surrounding area

Intervention Activated Adults

Offenses After Intervention

Offenses by Targeted

Offenders

Final Distance by Targeted Offenders

None 4% 5902.4 (526.7) NA 13.26 (2.24)

Global 4% 5352.4 (538.2) 59.6 (7.3) 15.17 (2.47)

Targeted 4% 5542.5 (566.4) 53.8 (12.4) 23.12 (6.99)

33

Effects of Target Radius and Percent Activated

• For target radius = 20 or 30, no significant difference in offenses after global or targeted interventions, over activated range of 1% - 5%

• For activated range > 2%, targeted intervention with radius 20 result in significant movement of offenders to neighboring blocks

• With larger target radius, targeted intervention approaches global intervention effects on offender movement

Discussion: Future Extensions

• More detailed behavior for both witnesses and offenders• Interpersonal offenses and retaliatory behavior• Detailed social networks

e.g., friends and families• Explicit elements of the law enforcement system

e.g., increasing police presence in neighborhoods experiencing high crime rates

• Turnover in the population New offenders become active as others retire Inter-neighborhood movement of people

Discussion: Conclusions

• Models can serve as a useful tool for confirming and generating(?) behavioral theory

• Models can provide useful insights into which interventions at the neighborhood level might be most effective in reducing the presence of crime in a community

• Models provide a tool for engaging diverse expertise in understanding and preventing crime and violence

36

Discussion: Lessons Learned

• Importance of developing a common language

• Involvement of community partner

• Finding data against which to validate agent-based model is a challenge!

37

References

1. Center for Disease Control and Prevention, National Center for Injury Prevention and Control. Web-based Injury Statistics Query and Reporting System (WISQARS) [accessed 2010 June 14]: www.cdc.gov/injury/wisqars/

2. Center for Disease Control and Prevention, National Center for Injury Prevention and Control. Web-based Injury Statistics Query and Reporting System (WISQARS), Youth Violence Data Sheet [accessed 2010 June 14]:www.cdc.gov/violenceprevention/pdf/YV-DataSheet-a.pdf

3. Kellerman, A.L., Fuqua-Whitley, D.S., Rivara, F.P., Mercy, J.: Preventing Youth Violence: What Works? Annual Review of Public Health 19, 271-292 (1998)

4. National Youth Violence Prevention Resource Center (NYVPRC): www.safeyouth.gov/Resources/Prevention/Pages/PreventionStrategies.aspx

5. Anderson, E.: Code of the Street: Decency, Violence and the Moral Life of the Inner City. Norton, New York (1999)6. Earls, F. J.: Violence and Todays Youth: Critical Health Issues for Children and Youth. Future of Children 4(3), 4-23 (1994)7. Sampson, R. J., Raudenbush, S. W., Earls, F.: Neighborhoods and Violent Crime: A Multilevel Study of Collective Efficacy.

Science 277, 918-924 (1997)8. Taylor, R. B., Gottfredson, S. D., Brower, S.: Understanding Block Crime and Fear. Journal of Research in Crime and

Delinquency 21, 303-331 (1984)9. Wilson, J.M., Chermak, S., McGarrell, E.F.: Community-Based Violence Prevention: An Assessment of Pittsburgh's One Vision

One Life Program. RAND Corp.,Santa Monica, CA (2010)10. Bibb, M.: Gang Related Services of Mobilization for Youth. In: Klein, M.W. (ed) Juvenile Gangs in Context: Theory, Research,

and Action. Prentice-Hall, Englewood Clis, NJ (1967)11. Bennett, T.H., Holloway, K.R., Farrington, D.P.: Effectiveness of Neighborhood Watch in Reducing Crime. National Council on

Crime Prevention, Stockholm (2008)12. Reiss, A.J., Roth, J.A.: Measuring Violent Crime and Their Consequences. In: Understanding and Preventing Violence, pp.

404-429. National Research Council, National Academy Press (1993)

38

Acknowledgments

• Supported by the National Institute of General Medical Sciences MIDAS grant 1U54GM088491-01

• Richard Garland, Executive Director for One Vision One Life (Pittsburgh, PA)

39

Extra Slides

40

Preliminary Model Dynamics

• Effects of Witness Rate on Incidence• Effects of Reporting Rate on Incidence• Effects of Punishment Rate on Incidence

• Purpose: check the internal consistency and face validity of model

41

Discussion: Boundaries of the Model

• Exogenous Parameters Proportion of adults, juveniles Probability of offenses being reporting Initial decision probabilities for adults and juveniles

unknown, need to test sensitivity

• Endogenous Parameters Change is incident rates over time Changes in spatial distributions of incidents Effects of community interventions

• Important Excluded Parameters Social capital Socio-economic status Individual prior history Social networks Law enforcement policies Realistic risk-reward assessment by potential offenders

42

Testing Model Dynamics: Witness Rate

• Assumes all witnessed offenses are reported

43

Targeted Intervention: Offenders Move to Adjacent Areas

Click to play

44

Testing Model Dynamics: Report Rate

• Assumes are offenses are witnessed

45

Testing Model Dynamics: Punishment Rate

• Assumes all offenses are witnessed and reported