Desistance and Legitimacy: Effect Heterogeneity in a Field Experiment on High Risk Groups

1
Desistance and Legitimacy Effect Heterogeneity
in a Field Experiment on High Risk Groups

• Jeffrey Fagan
• Andrew Papachristos
• Danielle Wallace
• Tracey Meares
• American Society of Criminology
• St. Louis
• November 2008

2
Background

• PSN as demand-side intervention designed to
  reduce gun carrying and gun use
• Recent policy preferences lean toward specific
  deterrence and forced compliance intensive
  cost-based strategies of close surveillance
  coupled with low threshold interventions to
  incapacitate offenders.
• Risk increase by changing police surveillance
  strategies to focus on weapons and intensifying
  parole and probation supervision
• Cost increase by threat of federalizing gun
  crimes
• Longer sentences, higher conviction rates with
  tougher plea negotiations

3

• PSN combines applies alternate strategy to engage
  offenders in more complex rational choice
  calculus where elevated risks and costs of
  offending are offset by simultaneous increases in
  the benefits of compliance and crime avoidance.
• Federal prosecution for gun crimes
• Longer sentences
• Gun-focused policing and seizures
• Serious offers of social, economic and other
  human services in group dynamic (forums)
• Providing capital to high risk groups in
  resource-starved neighborhoods is designed to
  legitimate legal sanctions in context-neutral
  forums that clarify and promote more balanced
  deterrence signals
• Designed for and implemented in two Chicago
  neighborhoods with very high homicide and gun
  crime rates.

4
(No Transcript)
5
Assignment Groups

• Guns and Homicides are not randomly distributed
• Treatment
• 24 beats on West Side
• Control
• 30 beats on South side

6
Research Design

• Quasi-Experimental Panel Design
• Near-equivalent Control Group
• Panel data of every neighborhood in Chicago
  from 1999 to 2006
• Multi-level modeling strategy to assess within
  and between group variation in neighborhoods
• Neighborhood and individual data
• Data
• Multiple sources CPD, ATF, IDOC, and the Census
• Crime reports geocoded to police beat
• Quarterly from January 1999 to December 2006
• Individual data reported by zip code (!_at_)

7
Neighborhood Results

• Strong declines in homicides in target areas
  compared to control areas and citywide rates
• No evidence of displacement to adjacent police
  districts
• We use propensity scores to adjust for
  differences in target and control areas as well
  as secular trends in the city
• Declines evident in both homicide and other gun
  crimes
• Compared to other PSN components, the number and
  saturation of PSN forums exerts the strongest
  influence on recidivism rates
• Aggregate effects on social networks of gun
  offenders? Aggregation produces generalized
  effects (general deterrence) via social influence
  model?

8
Figure 4. Fitted Model Summary with 95 CI
9
Results 1. Summary of PSN Neighborhood Effects
PSN Predictor Homicides (logged) Homicides (logged) Gun Homicides (logged) Gun Homicides (logged) Gang Homicides (logged) Gang Homicides (logged) Aggravated Battery (logged) Aggravated Battery (logged)
PSN (Dummy) Coeff Coeff -0.052 -0.052 -0.053 -0.053 -0.011 -0.011 -0.012 -0.012
PSN (Dummy) Exp(B) Exp(B) 0.949 0.949 0.948 0.948 0.989 0.989 0.988 0.988
PSN (Dummy) SE SE 0.013 0.013 0.013 0.013 0.008 0.008 0.008 0.008
PSN (Dummy) p-value p-value 0.000 0.000 0.000 0.000 0.235 0.235 0.159 0.159

Percent Offenders Attend Forum Coeff Coeff -1.17 -1.17 -0.782 -0.782 -0.951 -0.951 -0.063 -0.063
Percent Offenders Attend Forum Exp(B) Exp(B) 0.310 0.310 0.457 0.457 0.386 0.386 0.939 0.939
Percent Offenders Attend Forum SE SE 0.528 0.528 0.431 0.431 0.285 0.285 0.193 0.193
Percent Offenders Attend Forum p-value p-value 0.026 0.026 0.072 0.072 0.001 0.001 0.744 0.744

10
Results 2. Summary of PSN Neighborhood Effects
PSN Predictor Homicides (logged) Homicides (logged) Gun Homicides (logged) Gun Homicides (logged) Gang Homicides (logged) Gang Homicides (logged) Aggravated Battery (logged) Aggravated Battery (logged) Aggravated Battery (logged)
ATF Seizures Coeff -0.002 -0.002 -0.002 -0.002 -0.00009 -0.00009 -0.00009 0.003 0.003
ATF Seizures Exp(B) 0.998 0.998 0.998 0.998 1.000 1.000 1.000 1.003 1.003
ATF Seizures SE 0.001 0.001 0.001 0.001 0.0007 0.0007 0.0007 0.003 0.003
ATF Seizures p-value 0.075 0.075 0.042 0.042 0.894 0.894 0.894 0.325 0.325

Prosecutions (logged) Coeff -0.031 -0.031 -0.024 -0.024 -0.017 -0.017 -0.017 0.007 0.007
Prosecutions (logged) Exp(B) 0.969 0.969 0.976 0.976 0.983 0.983 0.983 1.007 1.007
Prosecutions (logged) SE 0.017 0.017 0.017 0.017 0.011 0.011 0.011 0.007 0.007
Prosecutions (logged) p-value 0.075 0.075 0.150 0.150 0.128 0.128 0.128 0.368 0.368

Person-Month Sentences (logged) Coeff -0.003 -0.003 -0.003 -0.003 0.002 0.002 0.002 -0.02 -0.02
Person-Month Sentences (logged) Exp(B) 0.997 0.997 0.997 0.997 1.002 1.002 1.002 0.980 0.980
Person-Month Sentences (logged) SE 0.005 0.005 0.005 0.005 0.003 0.003 0.003 0.002 0.002
Person-Month Sentences (logged) p-value 0.658 0.658 0.943 0.943 0.490 0.490 0.490 0.298 0.298

11
Recidivism Effects

• Methods and Data
• Criminal history data on 3,092 ex-offenders from
  experimental and control Chicago police districts
  
• Analyzed over four years to assess differences in
  timing of recidivism
• Cox regressions
• Disaggregated to examine separate effects on gang
  and non-gang members and first offenders versus
  repeat offenders
• Disaggregate by type of crime

12

• Propensity Scores ?
• Cox Models
• S (t) Pr (T gt t)
• Overlapping risks, neither competing nor
  exclusive
• Covariates
• Race African American
• Age years
• Education High school graduate
• Family ties Married or cohabitating
• Prior record first offense
• Gang member

13
Sample
Comparison PSN Total
N 2,677 (100) 415 (100) 3,092 (100)
Male 2,635 (98.4) 407 (98.1) 3,042 (98.4)
African American 2,531 (94.5) 383 (92.3) 2,914 (94.2)
Age (mean) 35.1 35.0 35.1
Married or Cohabitating 208 (7.8) 18 (4.4) 226 (7.3)
Prior Arrest 1,185 (44.3) 277 (66.7) 1,462 (47.3)
Gang Member 1,297 (48.4) 91 (21.9) 1,388 (44.9)
Any Recidivism 950 (35.5) 67 (16.1) 1,107 (35.8)
14
(No Transcript)
15
(No Transcript)
16
PSN Effects Within-District Comparisons
Offense Category Non-Gang Gang
All Crimes -1.547(.38) -.839(.20)
Murder -43.74(.005) -.287(1.01)
Total Violent Crime -.350(.58) -.006(.36)
Property-Violent -.424(.80) -1.467(.66)
Drug Sales -.193(.71) -.145(.23)
Drug Possession -.355(.37) -.145(.23)
Weapon -46.07(.005) .113(.39)
Drug Conspiracy .000(1.52) 1.643(.97)
17
PSN Effects Between-District Comparisons
Offense Category Non-Gang Gang
All Crimes -1.649(.40) -.795(.19)
Murder 54.81(1.00) -.276(1.00)
Total Violent Crime .290(.61) .404(.37)
Property-Violent .649(.086) -.563(.54)
Drug Sales -.225(.68) .110(.35)
Drug Possession -.936(-.586) -.586(.22)
Weapon -36.36(.37) .123(.37)
Drug Conspiracy -34.56(1.52) 2.001(.91)
18
PSN Effects Between-District Comparisons
First Offenders versus Priors
Offense Category Repeaters First Offenders
All Crimes -.945(.18) -37.29(.19)
Murder -.559(1.03) -36.94(1.01)
Total Violent Crime -.070(.31) -34.27(.44)
Property-Violent -.987(.50)
Drug Sales -.392(.33)
Drug Possession -.106(.20) -41.88(.30)
Weapon -.357(.38)
Drug Conspiracy -1.643(.097) .000(.00)
19
Next Steps

• Competing risk hazards model
• More detail on prior records
• Temporal analysis to estimates effects of
  degrading of intervention
• Contemporaneous neighborhood change