The author(s) shown below used Federal funds provided by the U.S.
Department of Justice and prepared the following final report:
Document Title: Monitoring High-Risk Gang Offenders with GPS
Technology: An Evaluation of the California
Supervision Program Final Report
Author(s): Stephen V. Gies, Randy Gainey, Marcia I. Cohen,
Eoin Healy, Martha Yeide, Alan Bekelman,
Amanda Bobnis
Document No.: 244164
Date Received: November 2013
Award Number: 2009-SQ-B9-K018
This report has not been published by the U.S. Department of Justice.
To provide better customer service, NCJRS has made this Federally-
funded grant report available electronically.
Opinions or points of view expressed are those
of the author(s) and do not necessarily reflect
the official position or policies of the U.S.
Department of Justice.
Monitoring High-Risk Sex Offenders With GPS Technology: An Evaluation of the California Supervision Program
ii
MONITORING HIGH-RISK GANG OFFENDERS
WITH GPS TECHNOLOGY: AN EVALUATION OF
THE CALIFORNIA SUPERVISION PROGRAM
FINAL REPORT
September 30, 2013
Prepared for
National Institute of Justice
810 Seventh Street NW
Washington, DC 20531
Prepared by
Stephen V. Gies
Randy Gainey
Marcia I. Cohen
Eoin Healy
Martha Yeide
Alan Bekelman
Amanda Bobnis
Development Services Group, Inc.
7315 Wisconsin Avenue, Suite 800E
Bethesda, MD 20814
www.dsgonline.com
This project was supported by Grant No. 2009-SQ-B9-K018 awarded by the National
Institute of Justice, Office of Justice Programs, U.S. Department of Justice. Points of view
in this document are those of the authors and do not necessarily represent the official
position or policies of the U.S. Department of Justice.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
iii
Monitoring High-Risk Gang Offenders
with GPS Technology: An Evaluation of
the California Supervision Program
Final Report
September 30, 2013
Authors:
Stephen V. Gies
Randy Gainey
Marcia I. Cohen
Eoin Healy
Martha Yeide
Alan Bekelman
Amanda Bobnis
Development Services Group, Inc.
7315 Wisconsin Avenue, Suite 800E
Bethesda, MD 20814
http://www.dsgonline.com/
This project was supported by Grant No. 2009-SQ-B9-K018 awarded by the National Institute of
Justice, Office of Justice Programs, U.S. Department of Justice. The opinions, findings, and
conclusions or recommendations expressed in this publication are those of the author(s) and do not
necessarily reflect those of the Department of Justice.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
v
Authors Contact Information
Stephen V. Gies
Development Services Group, Inc. (DSG)
7315 Wisconsin Avenue, Suite 800E
P: 301.951.6600; f: 301.951.3324
SGies@dsgonline.com
Randy Gainey
Old Dominion University
5115 Hampton Boulevard
Norfolk, VA 23529
P: 757.683.4794; f: 757.683.5634
RGainey@odu.edu
Marcia I. Cohen
DSG
P: 301.951.0056
MCohen@dsgonline.com
Eoin Healy
DSG
P: 301.951.5373
EHealy@dsgonline.com
Martha Yeide
DSG
P: 301.951.6619
MYeide@dsgonline.com
Alan Bekelman
DSG
P: 301.951.0056
ABekelman@dsgonline.com
Amanda Bobnis
DSG
P: 301.951.5382
ABobnis@dsgonline.com
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
vii
Abstract
Despite the overall decline in violent crime nationally, gang violence rates throughout the country
have continued at exceptional levels over the past decade. Therefore, it is vital for parole
departments to have effective tools for maintaining public safety. The purpose of this evaluation is to
determine the effectiveness of global positioning system (GPS) monitoring of high-risk gang
offenders (HRGOs) who are released onto parole.
This study integrates outcome, cost, and process evaluation components. The outcome component
assesses the impact of the California Department of Corrections and Rehabilitation’s Division of
Adult Parole Operations (DAPO) GPS supervision program by employing a nonequivalent-group quasi-
experimental design, with a multilevel discrete-time survival model. A propensity score matching
procedure is used to account for differences between the treatment and comparison groups. The
study population is drawn from all HRGOs released from prison between March 2006 and October
2009 in six specialized gang parole units in the State of California. The final sample includes 784
subjects equally divided between the treatment and control groups. The treatment group consists of
HRGOs who were placed on GPS monitoring, and the control group consists of matched gang
offenders with a similar background. The resulting sample shows no significant differences between
the groups in any of the propensity score matching variables.
The effectiveness of the program is assessed using an intent-to-treat (known as ITT) approach, with
two main outcomes of interest: compliance and recidivism. Compliance is measured through parole
violations; recidivism is assessed using rearrests and rearrests for violent offenses. Each outcome is
assessed with a survival analysis of discrete-time recidivism data, using a random intercept
complementary loglog model. In addition, frailty modeling is used to account for the clustering of
parolees within parole districts.
The findings indicate that during the two-year study period, subjects in the GPS group, while less
likely than their control counterparts to be arrested in general or for a violent offense, were much
more likely to violate their parole with technical and nontechnical violations. Descriptive statistics
and summary analysis revealed more GPS parolees were returned to custody during the study
period. These results will be studied further in a forthcoming follow-up report.
The cost analysis indicates the GPS program costs approximately $21.20 per day per parolee, while
the cost of traditional supervision is $7.20 per day per paroleea difference of $14. However, while
the results favor the GPS group in terms of recidivism, GPS monitoring also significantly increased
parole violations. In other words, the GPS monitoring program is more expensive, but may be more
effective in detecting parole violations.
Finally, the process evaluation reveals the GPS program was implemented with a high degree of
fidelity across the four dimensions examined: adherence, exposure, quality of program delivery, and
program differentiation.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
ix
Contents
Acknowledgments ........................................................................................................................... xi
Executive Summary ......................................................................................................................... xii
1. Background ............................................................................................................................... 11
A. Introduction ............................................................................................................................ 11
B. Literature Review ................................................................................................................... 14
C. The California Department of Corrections and Rehabilitation’s Global Positioning
System Supervision Program .............................................................................................. 112
D. The Study Goals ..................................................................................................................... 117
2. Methodology ............................................................................................................................. 21
A. Overview .................................................................................................................................. 21
B. Participants ............................................................................................................................. 21
C. Data Sources .......................................................................................................................... 22
D. Measures ................................................................................................................................ 25
E. Statistical Overview ................................................................................................................ 27
3. Results ...................................................................................................................................... 31
A. Baseline Characteristics .......................................................................................................... 31
B. Record of Supervision (ROS) ................................................................................................... 33
C. Outcome Analysis ..................................................................................................................... 35
D. Cost Analysis ........................................................................................................................... 39
4. Process Evaluation ................................................................................................................... 41
A. Overview ................................................................................................................................... 41
B. Data Sources ............................................................................................................................ 41
C. Program Fidelity ....................................................................................................................... 43
D. GPS Monitoring ....................................................................................................................... 413
E. Summary ................................................................................................................................. 414
5. Discussion and Recommendations .......................................................................................... 51
A. Summary .................................................................................................................................. 51
B. Policy Implications ................................................................................................................... 53
C. Limitations ................................................................................................................................ 57
D. Next Steps ............................................................................................................................... 58
References ...................................................................................................................................... R1
Attachments
A. California Department of Corrections and Rehabilitation Map
B. Parole Agent Survey
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
xi
Acknowledgments
evelopment Services Group, Inc. (DSG), has many people to thank for helping us complete
this study. First and foremost, Dan Stone, former Parole Administrator of the Electronic
Monitoring Unit (EMU) and current Division of Adult Parole Operations (DAPO) Director, was
our original contact with the California Department of Corrections and Rehabilitation (CDCR),
and was instrumental in securing the funding and getting this project off the ground. In addition, we
thank Denise Milano and Steve Marshall who each served as the Parole Administrator EMU DAPO
during the course of this project, for their cooperation and support. Both Denise and Steve made all
CDCR resources available to us; without them, this project would not have been possible.
We also owe a debt of gratitude for the tremendous assistance and cooperation given by numerous
other CDCR staff. These individuals include Adela Gonzales, Parole Agent III EMU DAPO; Denise
LeBard, Parole Agent III EMU; Titus Quinn, Parole Agent II EMU; Daniel Ramirez, Parole Agent II EMU;
Jill Rivera, Parole Agent II EMU; Gabriel Rogers, Parole Agent II EMU; Rugina Garcia, Parole Service
Associate; and Carrie Daves, Data Processing Manager III, Enterprise Information. Adela and Denise
helped arrange each of our site visits and were gracious enough to accompany our modest
entourage from unit to unit throughout the state to track down parolee case files. Rugina personally
printed each and every RAP sheet that was used in this study and diligently sent them to us on a
regular basis. Carrie obtained and transmitted the vast majority of electronic data used in this report
and patiently humored and effectively responded to all of our inquiries. Overall, CDCR was a true
partner in this research, taking part in everything from the design of the study to the provision of a
vast array of data to the facilitation of effective communication through regular conference calls and
electronic communication. We cannot thank them enough. We would also like to thank both GPS
monitoring vendors: Satellite Tracking of People (STOP) LLC and Pro Tech for both providing the data
for the study and entertaining our staff during the GPS parole agent training activity.
One of us (Gainey) utilized his graduate students for the coding and data entry of the RAP sheets. We
thank these young scholars for their valuable contributions and wish them well in their careers.
These Old Dominion University graduate students include Maryann Stone, and Jeff Toussaint.
We appreciate the support of the National Institute of Justice (NIJ) leadership in allowing us to
complete this important study. We thank NIJ for its backing and supportespecially current Program
Manager Dr. Marie Garcia.
Last but not in any way least we thank our own staff and colleagues at DSGparticularly Research
Associate Eoin Healy, writereditors Martha Yeide, Carrie Nathans, and Research Assistant Amanda
Bobnis. In addition, Research Assistants Micah White and Chase Montagnet assisted with data
coding and quality control. Finally, Marcia Cohen, Vice President for Research and Evaluation and
Alan Bekelman, President, kept us all on task throughout the process and contributed to the thinking
and writing of this report. All of these people helped in the analysis, writing, rewriting, and reviewing
of multiple iterations of the chapters herein and stayed with us as we put sections into a coherent
narrative.
It was a privilege to work with CDCR, attend the GPS training, and watch the program in action, and
to be able to apply the science and tools of evaluation on such a strong program.
Stephen V. Gies, Principal Investigator
Randy Gainey, CoInvestigator
D
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
xii
Executive Summary
PURPOSE
Los Angeles has been dubbed by some as the “the gang capital
of the world” (The Advancement Project 2007, p. 1). However,
gangs in California are not limited to the City of Los Angeles.
There are roughly 250,000 members statewide in 336 different
gangs (NGTA 2011). Although these street gangs are typically
not highly organized (Howell 2012), the individual members are
involved in a host of violent criminal activities, including assault,
drug trafficking, extortion, firearms offenses, home invasion
robberies, homicide, intimidation, shootings, and weapons
trafficking. In fact, a recent analysis conducted by the National
Gang Intelligence Center indicates gang members are
responsible for an average of 48 percent of violent crime in
most jurisdictions, and for a much greater percentage of violent
crime in jurisdictions like California with a large concentration of
gang members, where it is estimated that gangs are
responsible for at least 90 percent of crime (NGTA 2011).
Consequently, street gang activity and the criminal justice
response in California and other jurisdictions throughout the
United States remain important and significant inquiries. One
response in California has been to use Global Positioning
System (GPS) monitoring of high-risk gang offenders
(HRGOs)
who are placed on parole. The purpose of this evaluation is to
determine the effectiveness of this strategy.
STUDY GOALS AND OBJECTIVES
The overall purpose of this study is to conduct a quasi-
experimental evaluation of the California Department of
Corrections and Rehabilitation (CDCR) GPS monitoring program
of HRGOs. Specifically, the goals of this study are to
Assess the fidelity of the program.
Assess the cost of the GPS program.
Assess the effectiveness of the GPS program for gang
offenders.
Objectives
To meet these goals, this project has set several highly specific objectives to measure the success of
each goal. The specific objectives of the project organized by goal are as follows:
In this report, the term gang offender refers to an individual identified as either a prison or criminal street gang member.
See page 1-13 for more details.
Highlights
Purpose: The purpose of this evaluation is
to determine the effectiveness of the global
positioning system (GPS) monitoring of
high-risk gang offenders (HRGOs) who are
placed on parole.
Design: This study integrates both outcome
and process evaluation components. The
outcome component assesses the impact
of the California Department of Corrections
and Rehabilitation (CDCR) GPS supervision
program by employing a nonequivalent-
group quasi-experimental design with a
multilevel survival model. In addition, a
propensity score matching procedure was
used to account for the differences
between the treatment and comparison
groups.
Outcomes: This study provides evidence
that GPS is an effective suppression tool to
remove individual gang members from the
community. The odds of a technical
violation are 36 percent greater among the
GPS group, while the odds of a
nontechnical violation are 20 percent
greater. Conversely, the GPS group is less
likely to be rearrested overall (the chance
of being rearrested is 26 percent lower)
and for violent crimes (32 percent lower).
Cost: The cost of the GPS program is
roughly $14.00 per day per parolee more
expensive than traditional supervision.
However, the outcome results favored the
GPS group. In other words, the GPS
monitoring program is more expensive but
more effective.
Fidelity: The GPS program was
implemented with a high degree of fidelity.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
xiii
1. Assess the Fidelity of the GPS Program.
Determine the program adherence to all core components (i.e., program staffing qualifications,
caseload restrictions, parolee orientation specifications, and parole supervision specifications).
Determine the degree to which the prescribed level of program exposure was obtained.
Determine the quality of program delivery (e.g., skill of the staff in using techniques or methods
prescribed by the program and preparedness or attitude of staff toward the program).
Determine the degree to which program components were reliably differentiated from one
another.
2. Assess the Cost of the Program.
Determine the cost of monitoring HRGOs with the GPS system.
Determine the cost of monitoring HRGOs without the GPS system.
3. Assess the Effectiveness of the GPS Program for Gang Offenders.
Determine the effect of GPS monitoring on offenders’ subsequent occurrence of noncompliance
with parole conditions (i.e., technical violation and nontechnical violation).
Determine the effect of GPS monitoring on offenders’ subsequent occurrence of criminal
behavior (i.e., rearrest for any offense and rearrest for a violent offense).
DATA AND METHODOLOGY
To accomplish our goals and objectives, this study integrates both outcome and process evaluation
components. The outcome component assesses the impact of the CDCR GPS supervision program by
employing a nonequivalent-group quasi-experimental design with a multilevel survival model. We also use
a propensity score matching procedure to account for the differences between the treatment and
comparison groups. The study population is drawn from HRGOs (as determined by the GPS Monitoring
Gang Eligibility Assessment Criteria Form) who are released from prison and residing in the State of
California. The effectiveness of the program is assessed using an intent-to-treat (known as ITT) approach,
with two main outcomes of interest: noncompliance and recidivism. Noncompliance is operationalized as
a violation of parole. Recidivism, on the other hand, is operationalized as an arrest for a new crime. Each
outcome is assessed with a survival analysis of discrete-time data, using a random intercept
complementary loglog model. In addition, frailty modeling is used to account for the clustering of
parolees within parole districts. The outcome component also includes a cost-effectiveness analysis of
each outcome. The process component (see chapter 4) uses both quantitative and qualitative methods to
provide a rich context to the program treatment and structure and to assess program fidelity (i.e., whether
the program was designed well and implemented as intended).
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
xiv
PARTICIPANTS
This study focuses on HRGOs who are released from prison and placed on parole supervision with GPS
monitoring in six California jurisdictions. This group (N=407) includes all HRGOs placed on GPS
monitoring technology from March 2006 through October 2009 in each of the six specialized gang units
located in the City of Los Angeles and the following California counties: Fresno, Los Angeles, Riverside,
Sacramento, and San Bernardino. To identify comparison individuals likely to have pretreatment risk
characteristics similar to those in the treatment group, a propensity score procedure was performed using
a sample of offenders drawn from each of the same six communities that maintained specialized gang
units, but who were not placed on GPS at the time of data collection. The initial sample included more
than 145,000 subjects. The matching procedure resulted in a final sample of 784 subjects (392
treatment
and 392 control subjects). The two groups did not significantly differ on any variable.
DATA SOURCES
We used six primary sources to collect data: 1) the CDCR data management system, 2) official arrest
records, 3) parole supervision records, 4) GPS monitoring data, 5) a CDCR parole agent (PA) survey,
and 6) CDCR cost information.
California operates a data management system that houses numerous databases relevant to the
supervision of HRGO parolees. The majority of data used for this study were derived from three
databases: CalParole, the Revocation Scheduling and Tracking System (RSTS), and the Offender-Based
Information System (OBIS). A central feature of the California system is that offenders are linked across
all of these systems through a unique identifier that permits users to identify the same individual in
different contexts or data systems.
Another principal data source for this study was the official record of arrests, convictions, and custody
(commonly known as a RAP sheet) of each study subject. These data were provided in a hardcopy format
and coded by hand into a database developed specifically for the study.
A third data source included the record of supervision for each parolee. Specifically, the parole agent
notes the date and the specific type of contact. These data were collected to measure the level of
supervision received by each offender and to assess the California GPS program model.
The fourth data source was the GPS monitoring data from the two vendors: Satellite Tracking of People
(or STOP) LLC and Pro Tech. These data were used for descriptive purposes and to assess the California
GPS program model. Each vendor provided the following data: a profile of the offender; a record of each
event (inclusion/exclusion zone, strap tamper, low battery, cell communication gap, and no GPS
communication) that includes the event start and stop times and duration during a specified period; and
the assignment history of the device.
A survey instrument was also developed to collect process data from CDCR parole agents. The final
version contained questions in seven areas: 1) program staffing, 2) agent information, 3) equipment
issues, 4) caseload specifications, 5) enrollment and orientation, 6) collaborative engagement, and 7)
general summary.
The treatment group was slightly reduced (15 subjects) because there was no admit status in the data.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
xv
The final category of data was cost information. The primary sources for such data were written reports,
observations, and interviews. To facilitate the identification and specification of each cost, all expenditure
items were divided into four broad categories that have common properties: 1) personnel (all fulltime and
parttime staff and consultants), 2) facilities (i.e., the physical space required for the program), 3)
equipment and materials (furnishings, instructional equipment, etc.), and 4) other inputs (all other costs
that do not fit the other categories).
RESULTS
1. Assess the Fidelity of the GPS Program.
This study provides evidence regarding the degree to which the program services were delivered as
designed. Overall the process evaluation reveals the GPS program was implemented with a high degree
of fidelity across the four dimensions examined: adherence, exposure, quality of program delivery, and
program differentiation. A summary of each dimension is provided below:
Adherence refers to whether the program service or intervention is being delivered as it was designed. In
this case, the program was composed of five core components: program staffing requirements, caseload
restrictions, HRGO parolee screening, parolee enrollment and orientation specifications, and parole
supervision specifications. The findings demonstrate that while there was some variation across districts,
the overall program fidelity was high in terms of adherence to program staffing requirements, caseload
specifications, parolee orientation, GPS supervision, and field supervision.
Exposure refers to the measured quantity of a program. However, unlike the California GPS program for
sex offenders, where each subject is required to be continuously monitored by GPS (i.e., 365 days), there
is no prescribed GPS dosage level for the offenders in the gang program, resulting in a wide variation in
the number of days offenders are placed under GPS supervision (GPS supervision days ranged from 0 to
727). Consequently, dosage, while relevant for understanding the operation of the program, is not
applicable as a measure of fidelity in this study.
Quality of program delivery is the manner in which a teacher, volunteer, or staff member delivers a
program (e.g., skill in using the techniques or methods prescribed by the program, enthusiasm,
preparedness, or attitude). Overall, these findings suggest that in terms of quality of delivery, the GPS
program was delivered with proficient skill and a positive attitude.
Program differentiation identifies the unique features of different components or programs that are
reliably differentiated from one another. The single difference between traditional parole supervision and
GPS supervision is the use of GPS technology as a monitoring tool. The findings indicate that the
significant difference between the groups in terms of GPS monitoring shows that the GPS program is
visibly differentiated from traditional parole supervision.
2. Assess the Effectiveness of the GPS Program.
The GPS and control groups were well matched in this study after the use of propensity score adjustments
for numerous pretreatment characteristics. At baseline, mean scores on a wide range of demographic
and pretreatment characteristics are remarkably similar between the groups. Despite these baseline
similarities, a curious pattern of divergence in outcomes emerges during the two-year study period. The
odds of a technical violation are 36 percent greater among the GPS group, while the odds of a
nontechnical violation are 20 percent greater. Conversely, the GPS group is less likely to be rearrested
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
xvi
overall (the chance of being rearrested is 26 percent lower) and for violent crimes (32 percent lower).
At first glance, these findings appear contradictory. However, an interpretation of the findings through the
lens of a suppression program framework offers lucidity to the paradox. Suppression programs often use
a combination of policing, prosecution, and incarceration to remove individual gang members from the
community (Howell 2000). The goal of these programs is to influence the behavior of gang members by
dramatically increasing the certainty, severity, and swiftness of criminal justice sanctions (Braga and
Kennedy 2002). The use of GPS technology to monitor HRGOs falls within this context. In fact, one of the
most common gang suppression programs involves the operation of special gang probation and parole
caseloads with high levels of surveillance and more stringent revocation rules for gang members (Klein
2004). The program described in this study offers many of the same features but overlays GPS
monitoring as an added level of surveillance.
With this in mind, the data suggests that CDCR utilizes GPS as a suppression program where the
technology is used to monitor offenders with the goal of placing them back into custody for any
injudiciousness. Specifically, CDCR utilizes parole violations (in lieu of an arrest and the associated court
proceedings) as a means of returning GPS-monitored gang members back into custody.
3. Assess the Cost of the Program.
This study also provides details on the cost of the GPS monitoring program in comparison with the cost of
traditional supervision. The analysis found that the cost of the GPS program is $21.20 per day per
parolee, while the cost of traditional supervision is $7.20 per day per paroleea difference of about $14.
However, the results favored the GPS group in terms of the goal of the programremoving dangerous
gang members from the community. In other words, the GPS monitoring program is more expensive but
more effective. Specifically, when compared with traditional parole supervision, GPS monitoring costs
$1.49 per day per offender more than traditional parole to obtain a 1 percent decrease in arrests.
Conversely, due to the positive effect of GPS monitoring on technical and nontechnical violations, the GPS
program costs $10.77 per day per offender to obtain a 1 percent increase in technical violations and
$12.73 per day per offender to obtain a 1 percent increase in nontechnical violations.
POLICY IMPLICATIONS
Given the extreme nature of the gang problem, the response of criminal justice agencies to gang activity
in California and other jurisdictions throughout the United States is a vital public safety concern. As
indicated earlier, these responses can generally be grouped into three broad categories: prevention,
intervention, and suppression. Suppression programs are generally considered the least effective gang
program type (Decker 2002), but relatively few gang programs, regardless of strategy type, have been
found to reduce the criminal behavior of gang members (Klein and Maxson 2006, Howell 1998, Spergel
1995), and little serious evaluation research has concentrated specifically on gang suppression
strategies (Klein 1995). This research helps address fill this gap. Moreover it provides evidence that
suppression programs designed to keep high-risk offenders off the street may offer benefits by
decreasing community violence and increasing public safety. However, the cost analysis suggests that the
GPS monitoring program is more expensive. Specifically it costs roughly $4 per offender per day more
than traditional supervision. Is the increase in public safety worth the cost? While policymakers will
ultimately be faced with the harsh decision of how much they are willing to pay for a safer community,
there are a number of policy recommendations borne from the observations and findings of this study
that could improve the effectiveness and/or reduce the costs of the program to make it more cost
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
xvii
effective and thus more attractive to policymakers. These recommendations are summarized below.
Not All Gang Offenders Are Created Equal
Unlike the GPS program for sex offenders in California, the GPS program for gang offenders does not
utilize a standardized risk instrument to identify potential subjects for inclusion. Given that the goal of the
program is to place dangerously violent gang offenders back into custody, it can be assumed that the
most appropriate offender is an identified gang member with a high propensity toward violence. As a
result, we recommend the adoption of a more formalized decision process that ensures that the targeted
population is being served by the program. Specifically, we recommend incorporating a risk instrument
designed to predict violent offending into the existing decision making process. For this purpose, some
of the most common instruments include the Historical, Clinical, Risk Management-20 (HCR-20), the
Violence Risk Appraisal Guide (VRAG), and the Structured Assessment of Violence Risk in Youth (SAVRY).
Another possibility would be to employ the California Static Risk Assessment (CSRA) tool, an actuarial
instrument specifically developed for and already in use in California.
Going Beyond Crime Mapping
While CDCR currently has the capacity to use their GPS monitoring program to run crime correlations, the
use of GPS monitoring holds the potential for an unprecedented insight into gang-related activity. By its
very nature, GPS technology allows for an exceptional gain in the geographic intelligence of gang member-
activity by specifically tracking the mobility and engagements of a parolee. We recommend moving
beyond traditional crime scene correlations to conduct social network analysis in order to identify the
contacts, ties, and attachments that one gang has to another.
Conduct a Cost Analysis on Outsourcing the Monitoring Center Function
Creating a monitoring center function is critical to the smooth operations of GPS programs, since the GPS
supervision of paroles can generate an overwhelming amount of information. However, it should be noted
that there are numerous ways to configure a monitoring center, some of which may be more or less costly
to CDCR. Considering the volume of offenders on GPS monitoring in California and the cost associated
with outsourcing the operation of the monitoring center, we recommend that CDCR conduct a study to
determine the marginal cost of internalizing the monitoring center.
Push Criminal Prosecution
While back-end sentencing is not without some merit (e.g., swiftly removing potentially violent criminals
from the community), the practice used in California permits some dangerous offenders to dodge the
more severe penalties that would have been imposed had the cases been prosecuted in the criminal
court system as opposed to being handled by the parole board. We recommend that whenever possible
parolees who commit new crimes, particularly crimes of a serious nature, be prosecuted to the fullest
extent of the law in criminal courts.
Continue to Emphasize the Use of GPS Monitoring as a Tool
The final recommendation has been offered elsewhere (Gies et al. 2012), but it bears repeating here.
Public officials should bear in mind that GPS monitoring is merely a tool useful in the larger context of
parole practice. It is not a panacea for all things criminal. This recommendation is borne from the
inflated expectations of GPS monitoring attributable to the misconceptions about what GPS monitoring
can actually accomplish (Payne and DeMichele 2011). While California recognizes this concept and
integrates this principle into its training, its importance cannot be overstated.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
xviii
Thoughtfully Grow the Program
Recent evidence suggests that GPS monitoring is a useful supervision tool. However, little research has
investigated the use of GPS technology as a tool to deter criminal behavior by removing serious and
violent offenders from the streets. While not conclusive, this study provides promising evidence that GPS
technology offers increased public safety by potentially removing dangerous criminals from the streets
before they commit more violent crimes. It is recommended that CDCR carefully weigh the benefits and
detriments of the program, but consider expanding the GPS monitoring of HRGO to additional units. The
main benefit appears to be the potential for increased public safety. The key detriment rests on the
increased costs: not only the costs of operating the GPS program, but also the costs associated with
returning these offenders to custody.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
11
1. Background
A. INTRODUCTION
Purpose
The street gang culture in Los Angeles is legendary (Howell et al. 2011) and has been famously depicted
in numerous Hollywood films such as Boyz n the Hood, Training Day, and Colors, to name a few. In fact,
Los Angeles has been dubbed by some as the “the gang capital of the world” (The Advancement Project
2007, p. 1). However, gangs in California are not limited to the City of Los Angeles. There are roughly
250,000 members
statewide in 336 different gangs (NGTA 2011). Although these street gangs are
typically not highly organized (Howell 2012), the individual members are involved in a host of violent
criminal activities, including assault, drug trafficking, extortion, firearms offenses, home invasion
robberies, homicide, intimidation, shootings, and weapons trafficking. In fact, a recent analysis conducted
by the National Gang Intelligence Center indicates gang members are responsible for an average of 48
percent of violent crime in most jurisdictions, and for a much greater percentage of violent crime in states
like California with a large concentration of gang members, where it is estimated that gangs are
responsible for at least 90 percent of crime (NGTA 2011).
Moreover, despite the overall dramatic declines in violent crime nationally, Howell and colleagues (2011)
found overwhelming evidence that gang violence rates have continued in California and throughout the
country at exceptional levels over the past decade. In fact, they suggest gang violence is rather
commonplace in very large cities and seems largely unaffected by, if not independent from, other crime
trends (Howell et al. 2011).
Consequently, street gang activity and the criminal justice response in California and other states and
localities throughout the United States remain important and significant inquiries. One response in
California has been to use Global Positioning System (GPS) monitoring of high-risk gang offenders
(HRGOs)
who are placed on parole. It is hypothesized that the GPS monitoring technology deters
offenders from engaging in criminal behavior and encourages parolees to be more compliant because it
increases probability of detection by law enforcement. The purpose of this evaluation is to determine the
effectiveness of this strategy.
Background
The impetus for this project began in July 2005 when the California Department of Corrections and
Rehabilitation (CDCR) began a pilot program in San Diego testing the use of GPS technology to monitor
high-risk sex offenders on parole. The success of the pilot project prompted CDCR to expand the program
across the state. Implementation of the full statewide program was completed in December 2008 after
phasing in 4,800 GPS monitoring units (Gies et al. 2012). This figure nearly triples the 1,800 GPS units
used by Florida, the second-leading state to use the devices. As of August 2011, there were 9,912 sex
offenders on parole in California (9 percent of all parolees under the jurisdiction of the CDCR). Roughly
7,022 of these sex offenders were living in the community and 6,968 (99.2 percent) were monitored by
GPS technology.
The NGIC report estimates there are six gang members per 1,000 people in the state. The population of California is
roughly 38 million. Thus, we estimated the gang population in the following manner (38,000,000/1,000)*6=228,000).
In this report, the term gang offender refers to an individual identified as either a prison or criminal street gang member.
See page 1-13 for more details.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
12
The generally positive experiences among parole agents (PAs) with the sex offender monitoring program
spiked interest in applying the same technology to monitor other types of offenders. After thoughtful
consideration and contemplation, CDCR took steps to use this technology to address the severe gang
problem that plagued the state. In March 2006, CDCR’s Division of Adult Parole Operations entered into a
partnership with the city of San Bernardino to implement a pilot project similar to the San Diego program
to track the movements of known gang members. The San Bernardino pilot program established a 20-
unit specialized gang parole caseload that uses GPS technology as a supervision tool for active gang
member parolees who have a history of violence and weapons possession and who are identified as a
public safety risk to the city.
In May 2007, Gov. Arnold Schwarzenegger proposed an antigang initiative known as the California Gang
Reduction, Intervention, and Prevention (CalGRIP) program to provide more than $48 million in state and
federal funding for local antigang efforts, including job training, education, and intervention programs.
CalGRIP also expanded the pilot program in San Bernardino to an 80-unit program by adding 20 units
each in the City of Los Angeles and the following California counties: Fresno, Los Angeles, Riverside,
Sacramento, and San Bernardino.
This study focuses on HRGOs who were released from prison and
placed on parole supervision with GPS monitoring in these six original jurisdictions from March 2006
through October 2009.
How GPS Works
GPS is a space-based global navigation satellite system that provides location and time information in all
weather, anywhere on or near the earth. The initial GPS project was developed in 1973 as a military
application to overcome limitations of previous navigation systems, integrating ideas from several
predecessors, including numerous classified engineering design studies from the 1960s. However, in the
1980s, the government made the system available for civilian use, and GPS became fully operational in
1994. The system is freely accessible by anyone with a GPS receiver (although some of the more
sophisticated technologies are reserved for military users).
The GPS system consists of three major segments. These are 1) the space segment (SS), 2) the control
segment (CS), and 3) the user segment (US). The U.S. Air Force developed, maintains, and operates the
SS and CS. The SS segment comprises 24 to 32 satellites orbiting the earth at an altitude of
approximately 20,000 kilometers. The CS comprises a master control station, an alternate master control
station, and six monitoring stations around the globe. Finally, the US comprises hundreds of thousands of
U.S. and allied military users of the secure GPS Precise Positioning Service and tens of millions of civil,
commercial, and scientific users of the Standard Positioning Service.
These three segments work in concert to produce accurate time and position information. The GPS
satellites (SS) circle the earth twice a day in a precise orbit and continuously transmit signal information
(i.e., the time the message was transmitted, precise orbital information, and general system health).
Notably, all GPS satellites synchronize operations so these repeating signals are transmitted at the same
instant. The synchronized signals, moving at the speed of light, arrive at the GPS receiver (US) at slightly
different times because some satellites are farther away than others. The distance to the GPS satellites
can be determined by estimating the amount of time it takes for their signals to reach the receiver. When
The GPS supervision of HRGOs subsequently expanded to a number of other jurisdictions and then contracted due to
budget considerations during the course of this study. This research focuses on the original six jurisdictions.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
13
the receiver estimates the distance to at least four GPS satellites, it can calculate its position in three
dimensions (latitude, longitude, and altitude). However, a receiver can determine a two-dimensional
position (latitude and longitude) from only three satellites. Regardless of method, this position is then
displayed on a map for the user. Many GPS receivers also show derived information such as direction and
speed, which are calculated from position changes. Finally, the monitoring stations (CS) are used to
precisely track each satellites orbit and synchronize the signals. The flight paths of the satellites are
tracked by dedicated U.S. Air Force monitoring stations in Hawaii; Kwajalein in the West Pacific; Diego
Garcia in the Indian Ocean; Ascension Island in the South Atlantic; Cape Canaveral, Fla.; and Colorado
Springs, Colo. The tracking information is sent to the Air Force Space Command in Colorado Springs,
which contacts each satellite regularly with a navigational update. These updates synchronize the atomic
clocks on board the satellites to within a few nanoseconds of one another and adjust the orbital
information of each satellite.
ACCURACY. The accuracy of a position determined with GPS depends on the type of GPS receiver. Most
handheld GPS units are accurate to within 15 meters on average. Other types of receivers use
enhancement methods such as Differential GPS (DGPS) to obtain much higher accuracy. DGPS requires a
network of fixed, ground-based reference stations to broadcast the difference between the positions
indicated by the satellite systems and the known fixed positions. Observations made by the stationary
receiver are used to correct positions recorded by the roving units, producing an accuracy greater than 1
meter. Other methods such as Real Time Kinematic and Post Processing can enhance accuracy even
further but at a significantly increased cost. Consequently, these enhancement methods are typically
used only in more advanced applications such as land surveying. When used properly under ideal
conditions, the accuracy of each method is approximated as follows:
Autonomous: <10m
Differential GPS: 0.32.0m
Real Time Kinematic: 0.050.5m
Post Processing: 0.020.25m
LIMITATIONS. GPS receivers require an unobstructed view of the sky and often do not perform well because
of interference from buildings, terrain, electronics, or sometimes even dense foliage. These obstructions
can cause position errors or possibly no position reading at all. Consequently, GPS units typically do not
work well indoors, underwater, or underground. Other factors that can degrade the GPS signal and thus
affect accuracy include the following:
Atmospheric disturbances. This error occurs when the satellite signal slows as it passes through
the atmosphere. The GPS system uses a built-in model that calculates an average amount of
delay to partially correct for this type of error.
Signal multipath. This error occurs when the GPS signal is reflected off objects such as tall
buildings or large rock surfaces before it reaches the receiver. This increases the travel time of
the signal, thereby causing errors.
Receiver clock errors. This error occurs when the receivers built-in clock is not as accurate as
the atomic clocks onboard the GPS satellites, resulting in very slight timing errors.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
14
Orbital errors. This error is due to inaccuracies of the satellites reported location.
Satellite geometry/shading. This error refers to the relative position of the satellites at any given
time. Ideal satellite geometry exists when the satellites are located at wide angles relative to each
other. Poor geometry results when the satellites are located in a line or tight grouping.
B. LITERATURE REVIEW
Research on gangs has grown tremendously since the 1980s, providing a wealth of information on who
joins gangs and why, what types of criminal activities gangs encourage, recidivism rates for ex-offenders
who are gang affiliated, and what works in prevention and intervention/reentry. Still, continuing rates of
gang activity and violence make it clear that we are still seeking effective ways to interrupt gang activity
and manage ex-offenders as they reenter their home communities to reduce the adverse impacts.
Gang Prevalence, Membership and Activity
According to the National Gang Center, the number of gangs has grown from a low in 2003 of around
20,000 to an estimated 30,000 in 2011; this represents a 12 percent increase from 2006 and is the
highest annual estimate since 1997. As of 2011, there were approximately 782,500 gang members.
These gangs and their activities constitute a pervasive problem throughout the country, as demonstrated
by a recent trend analysis of U.S. gang problems from 2002 to 2009 (Howell et al. 2011). Although the
nation has experienced an overall decline in rates of violent crime, this trend has not affected gang
violence. Rather, rates of gang violence have continued relatively unchanged during this period for most
cities with populations of 50,000 or more. In some of the largest cities, the percentage of homicides that
are gang related is very highin 2009, one third of homicides in Chicago and one half of homicides in Los
Angles were gang related.
The peak age range for gang membership is roughly 14 to 15 (Huff 1998). This finding is remarkably
consistent across self-report studies, regardless of the risk level of the sample, the restrictiveness of the
gang definition, and the study location (Klein and Maxson 2006). However, the peak age range may be
older in cities where gangs have existed longer (Curry and Decker 1998). For instance, in 2011, law
enforcement reported that more than three out of every five gang members were adults (National Gang
Center). The proportion of adult member to juveniles was larger for larger cities and suburban counties
than for smaller cities and rural counties. The typical range for gang members is ages 12 to 24.
The gender and racial/ethnic composition of gangs has remained relatively stable over the past decade.
Although female gang membership may be increasing (Klein 1995), virtually all studies agree that males
join gangs at higher rates. In fact, the prevalence rates for males are to 2 times as high as those for
females in most studiesa pattern that transcends different study approaches (Klein and Maxson 2006).
Data from the National Youth Gang Survey indicate females continue to make up less than 10 percent of
gang membership. Data also indicate that the ethnic composition of gang members remained relatively
stable during the 19962011 survey period, although there is also a wide ethnic/race differential in gang
membership. According to the National Youth Gang Survey, in 2011 the ethnicity of gang members was
roughly 46 percent Hispanic, 35 percent African American, 11 percent white, and 7 percent other
race/ethnicity. This pattern is consistent regardless of the definition of gang and the nature of the sample
approaches (Klein and Maxson 2006). The disproportionate representation of minority groups in gangs is
not a result of a predisposition toward gang membership; rather, minorities tend to be overrepresented in
areas overwhelmed with gang activity (Bursik and Grasmick 1993).
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
15
While most cities and jurisdictions generally do not record criminal offenses other than homicides and
graffiti as gang related, research has provided insight into the nature of gangs’ criminal activity. The
research demonstrates that although gang members commit a fair share of violent crime, gang members
do not necessarily specialize in violence. Instead, they tend to be “generalist in nature, spanning the
range of the cafeteria of delinquency choices” (Klein and Maxson 2006, see also Thornberry et al. 2003).
Gang members do, however, commit a disproportionate number of offenses compared with non-gang
members (Klein and Maxson 2006, Thornberry et al. 2003, Miller 2001). For instance, in a recent
comparison of patterns of offending among gang and non-gang youth in Dutch and U.S. youth samples,
Esbensen and Weerman (2005) found gang members are four to six times as likely as non-gang youth to
engage in minor and serious delinquency. Data from the Rochester Youth Development Study indicate
that gang members are seven times as likely as non-gang youth to commit delinquent offenses
(Bjerregaard and Smith 1993). This relationship is robust across a wide variety of definitions of gang and
across different measurements of offending (Klein and Maxson 2006); it also holds up when gang
members are compared with other highly delinquent non-gang youth (Thornberry 1998, Huizinga 1997).
Recidivism of Gang Members
Many of these gang offenders go to prison, but what happens once they are released? During 2010, state
and federal prisons released more than 700,000 prisoners (Reentry Facts). While no national estimates
indicate what percentage of ex-offenders are former and current gang members, the reentry of these
former and current gang members helps drive gang activity and violence in their home communities.
More than one third of law enforcement agencies identified the return of gang members from secure
confinement as a factor in local crime activity. The percentage of agencies identifying this return from
secure confinement as a factor rose from 42 percent in 2006 to almost 53 percent in 2011 (National
Gang Center). As Olson, Dooley, and Kane (2004) note, the practical impact of these ex-offendersreentry
can be substantial: the return of gang members released in Illinois during 2000 translated to nearly
11,000 adult gang-involved ex-offenders reentering home communities in that one year.
A growing number of quantitative and qualitative studies are assessing the recidivism rates of ex-
offenders who are current or former gang members. Most studies have found ex-offenders are associated
with a higher risk of recidivism than non-gang ex-offenders, whether the ex-offenders are juveniles, young
adults, or adults.
The relationship between gang affiliation and juvenile offender recidivism has been assessed in multiple
studies. For instance, using event history analysis for a sample of 2,435 state incarcerated delinquents,
Caudill (2010) found gang affiliates have a significantly higher risk of recidivating within 6 months of
release compared with non-gang ex-offenders. However, the risk ratios of the two groups converge shortly
after that. In a study of 1,804 serious and violent delinquents released from a large southern correctional
facility, Trulson and colleagues (2012) found gang murderers had a higher risk of rearrest and any felony
re-arrest than non-gang murderers, after controlling for youth characteristics, delinquent background, and
social history measures.
A similar relationship between gang affiliation and increased recidivism is found for young adult and adult
samples. Huebner, Varano, and Bynum (2007) assessed recidivism among 322 young men aged 17 to
24. Proportional hazard models indicated gang membership is one of the critical factors in predicting the
timing of reconviction (other factors included race, drug dependence, and institutional behavior). Fifty-six
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
16
percent of gang members and drug-dependent offenders recidivated, with an average time of 29 months
to reconviction, compared with the 28 percent of non-gang and non-drug-involved offenders who
recidivated, with an average time of 37 months to reconviction. Analysis revealed that, compared with
drug-dependent individuals, gang members recidivated at higher rates (45 percent compared with 29
percent) and more quickly (40.34 months compared with 32.59 months). In a study of California
parolees, McShane, Williams III, and Dolny (2003) found gang members have higher recidivism rates
across all commitment offense categories compared with the general parole population, even after
controlling for age. It was also found that gang members were more likely to engage in certain types of
crimes that result in greater recidivism. The researchers concluded that although gang membership was
not one of the strongest predictors of recidivism, it had an independent, negative effect on recidivism.
A study by Olson, Dooley, and Kane (2004) also looked at the relationship between recidivism and gang
membership among 2,534 adult inmates released from prison in Illinois in November 2000. Similar to
the studies cited above, this study found that, compared with non-gang ex-offenders, gang members were
more likely to get rearrested, be rearrested more quickly after release, and be arrested for violent and
drug offenses. This relationship held even after controlling for variables such as inmate and neighborhood
characteristics. Given the large sample, Olson, Dooley, and Kane were able to examine rearrest rates for
four age groups (1724, 2531, 3239, and 39+). Contradictory to the findings by Huebner, Varano, and
Bynum (2007), the relationship between recidivism and gang membership was weak for the young adult
group and the 3239 aged group: gang and non-gang members in these age groups were rearrested at
similar rates (however, Olson, Dooley, and Kane did not treat drug-dependent offenders as a separate
group). The relationship was strongest for the 2531 and 39+ groups (both almost twice as likely to be
rearrested compared with non-gang members). They conclude that the evidence from their study suggests
the agecrime curve better characterizes non-gang members, where older age was associated with lower
recidivism.
Although such quantitative studies have established the link between gang affiliation and increased
recidivism, a number of qualitative studies have explored why ex-offenders recidivate. Olson, Dooley, and
Kane (2004) found gang members in their sample were more likely to be characterized by risk factors
that predict recidivism. Thus, in the Illinois sample, gang members being released were more likely to be
younger, male, minority, single/never married, and have lower levels of education. The community to
which offenders return also affects recidivism. Huebner, Varano, and Bynum (2007) found men who
returned to disadvantaged communities were reconvicted more quickly. Olson, Dooley, and Kane also
found a higher proportion of gang members in their sample returned to neighborhoods in Cook
County/Chicago compared with a higher proportion of non-gang members who returned to other urban
areas of the state.
Social networks and social identity play a role in continued criminal activity. Scott (2004) found through a
series of ethnographic interviews with 12 ex-inmates that the draw for individuals to street-level drug work
stemmed less from the money involved than from the social network and activities involved in that work.
According to interviews with 39 Chicago-area ex-convicts, many ex-offenders essentially “cocoon”
themselves in close family and friend networks (Scott, Dewey, and Leverentz 2005). While these close
ties afford ex-offenders material and non-material benefits in the short term, they limit the extent to which
ex-offenders develop weak ties” with the extended community. It is these “weak ties” that enhance
social capital (e.g., by connecting them to individuals who can support attainment of personal goals, such
as employment). Moreover, this link to former gang networks is reinforced in returning ex-offenders by
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
17
community mistrust and the abrasive tactics used by police toward ex-offenders (Scott 2004). Decker and
Pyrooz’s (2011) study of 177 ex-offenders indicate the difficulty of disrupting ties with a gang: Even when
former gang members have shifted on measures of embeddednessthe degree to which gang members
are immersed in activities of and feel commitment to the gangand no longer consider themselves gang
members, the social networks around themas represented by rival gang members and policemay still
consider them gang members. Seventy-four percent of former gang members reported that police
continued to treat them as gang members when they had left the gang (including being stopped and
questioned about gang activity, their names being retained in gang databases, and being arrested;
Decker and Pyrooz 2011).
Scott, Dewey, and Leverentz (2005) note the tendency of much reintegration research to embrace
binaries (e.g., either successes or failures at reintegration; either working or not working), which rarely
captures the reality of the lives of ex-offenders. Fleisher and Decker (2001) emphasize the gradual nature
of disengagement from gang membership, as self-identification to a gang may persist for years and gang
identity provides important social ties. Decker and Pyrooz (2011) also found evidence for a long-term
disengagement process: although concerns of former gang members being harassed by rival gang
members and police decreased over time, concerns remained high over time. In short, being recognized
as a former member is complex, gradual, and perceived differentially by different groups” (15).
Programs
Gang programs can generally be grouped into three broad categories: prevention, intervention, and
suppression. In general, prevention strategies keep youth from joining gangs, while intervention strategies
seek to reduce the criminal activities of gangs by pulling youth away from gangs. These strategies typically
include community organization, early childhood programs, school-based interventions, and afterschool
programs. Interestingly, relatively few intervention programs target reentering populations of ex-offenders
with gang affiliations (for instance, in the What Works in Reentry Clearinghouse, there is no topic area
devoted to gang offenders, and a search on “gang” identifies only two programs), although materials have
been developed to help jurisdictions create gang desistance plans (e.g., see Young and Gonzalez 2013).
Suppression programs use the full force of the lawgenerally through a combination of policing,
prosecution, and incarcerationto deter criminal activities of entire gangs, dissolve gangs, and remove
individual gang members from gangs (Howell 2000). Typical suppression programs include street sweeps,
school-based law enforcement programs that use surveillance and buybust operations, civil procedures
that use gang membership to define arrest for conspiracy, prosecution programs, and special gang
probation and parole caseloads with high levels of surveillance and more stringent revocation rules for
gang members (Klein 2004).
The use of GPS technology to monitor HRGOs falls within the suppression category, given that the goal is
to influence behavior of gang members by dramatically increasing the certainty, severity, and swiftness of
criminal justice sanctions (Braga and Kennedy 2002). Although suppression is universally considered to
be the most fashionable response to gangs, it is also perceived to be the least effective (Decker 2002).
However, relatively few gang programs, regardless of strategy type, have been found to reduce the
criminal behavior of gang members (Klein and Maxson 2006, Howell 1998, Spergel 1995), and little
serious evaluation research has concentrated specifically on gang suppression strategies (Klein 1995).
Moreover, one of the most successful gang programs noted in the literature is primarily a suppression
strategy. The Tri-Agency Resource Gang Enforcement Team (TARGET) is a gang crimeintervention
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
18
program in Orange County, Calif., intended to provide a strong criminal justice response to offenses
committed by gang members. Similar to the GPS program, the goal of TARGET is to reduce gang crime by
selectively incarcerating the most violent gang offenders. It accomplishes this goal by identifying repeat
gang offenders based on their criminal record and monitoring them closely for new offenses. When a
gang member is arrested, the offender is prosecuted by the district attorney assigned to the TARGET unit
to obtain the lengthiest period of incarceration possible to deter future criminal offending. An evaluation
of the program found the placement of repeat gang offenders in custody appears to have had an effect
on reducing gang crime (Kent, Donaldson, Wyrick, and Smith 2000). During the first year of the program
(1992), gang crime decreased by 11 percent. The cumulative reduction in gang crime was 64 percent
through 1993, 59 percent through 1994, and 47 percent through 1997.
Despite these encouraging findings, suppression programs are still perceived to be less effective than
some other strategies designed to reduce criminal behavior of gang members. Given this discrepancy and
many other unanswered questions regarding the effectiveness of gang programs, there is still a critical
need for high-quality evaluation research on gang programs (Decker 2002). This research helps address
this need.
Electronic Monitoring
Electronic monitoring (EM) devices have increasingly been used in prison diversion and release programs
over the past decades. Such devices include polygraphs, random calling and voice verification, remote
alcohol monitoring, sleep pattern analysis, motion detection analysis, check-in kiosks, and GPS systems
(IACP 2008). EMparticularly GPS deviceshas become a popular tool for monitoring paroled offenders.
BACKGROUND. The first electronic monitoring devices were developed in the 1960s by a group of
researchers at Harvard University, with the main purpose of providing feedback to offenders fitted with
the units. The feedback was meant to provide social support and facilitate rehabilitation (Burrell and
Gable 2008). However, this device failed to gain acceptance, and it was not until the 1980s that EM
reemerged. The climate had changed considerably, with the emergence of a more punitive model of
offender treatment. Technology made possible increased surveillance and enforcement in the community
setting.
The decision of New Mexico State District Judge Jack Love in 1983 to sentence three offenders to home
detention with EM has taken on an almost mythic status. To fulfill his vision, he first had to convince
someone to manufacture the transmitter devices. Since those early days, the pool of manufacturers and
service providers has been in flux (Burrell and Gable 2008), but part of the dramatic growth in the use of
EM is due to the aggressive marketing of these private companies (Black and Smith 2003; Lilly, 2006).
From those first three offenders in 1983, it has been estimated that approximately 100,000 offenders
were on EM in 2006 (Conway 2006, as cited in Burrell and Gable 2008). The usability of these units was
enhanced considerably when the military discontinued the policy in 2000 of “selective availability,” which
had made civilian receivers significantly less accurate than military receivers (Florida Senate Committee
on Criminal Justice 2004).
RADIO FREQUENCY AND GPS MONITORING. Two types of EM are used most frequently for monitoring
offenders. The first, radio frequency monitoring (RF), is used to determine whether an offender on house
arrest is at home. The offender wears a tamper-resistant small transmitter that communicates with a
small receiving unit connected to the phone line. If the signal is lost, the receiving unit communicates with
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
19
the monitoring station, which in turn can notify the probation officer. These systems can accommodate
work or religious schedules, so offenders can be off site at scheduled times. Officers can also use a “drive
by” monitoring device to check if the offender is at home or in treatment as scheduled. A radio frequency
(RF) unit is the least expensive form of monitoring and costs about $2.75 per day (Florida Senate
Committee on Criminal Justice 2004).
The second system, GPS monitoring, uses a network of satellites to calculate the physical position of the
offender. The offender wears a tamper-resistant bracelet that receives transmissions from the satellites
and calculates his or her location. With a passive GPS system, this information is stored and transmitted
at appointed times to the monitoring station. With the active GPS system, information on the individual’s
location is transmitted to the monitoring station in near “real time.” This allows the station to alert the
probation officer immediately when a violation occurs. Both active and passive GPS systems allow certain
zones to be excluded (such as crime hot spots or rival gang territories) or included (such as a work zone)
and provide information on where and when an individual has been throughout the course of the day. The
passive GPS system costs about $4 per day, and the active system costs about $9 per day (Florida
Senate Committee on Criminal Justice 2004).
GPS has garnered an increasing amount of attention. But though there are multiple benefits to its use,
officials in the justice and corrections systems, as well as the general public, need to be aware of
potential shortcomings. The International Association of Chiefs of Police (IACP) has identified four main
benefits of GPS:
1. Flexibility. GPS offers an alternative to incarceration, which is expensive. It also can be tailored
for individual offenders so specific geographic areas can be selected for inclusion (the offender
can visit that area) or exclusion (the offender must avoid that area to avoid a violation alert being
sent).
2. Reintegration. GPS may promote compliance with the conditions of supervision and treatment,
since locations can be tracked.
3. Control. The criminal justice system retains the ability to track and respond to the movement
patterns of offenders, and the equipment provides a tangible and continuing reminder to the
offender that monitoring is ongoing.
4. Investigation. It is possible to use location information to confirm an individual is or is not a
suspect for a particular crime.
While these benefits make GPS attractive, there still are concerns about GPS. IACP has identified the
following four issues:
1. Limited empirical support. Findings from research studies on EM’s impact on recidivism are
mixed (see below).
2. Increased officer workload. Though early advocates of EM believed this tool could increase the
manageable caseload under supervision, experience with the technology has suggested the
opposite. This workload increase stems from multiple factors, such as the need for officers to
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
110
monitor GPS equipment, to respond to alerts (many of which can be “false” alerts [Elzinga and
Nijboer 2006]), to teach offenders how the equipment works, and to ensure the equipment is
maintained and replaced when it fails. Sachwald (2007) noted that in Maryland’s experience of
implementing GPS, hardware failures occurred for about half of the offenders placed on GPS, and
the equipment occasionally had to be replaced two or even three times before it worked. In light
of such realities, the Florida Department of Corrections recommended the total caseload burden
diminish with the introduction of EM, so supervising officers have a caseload of 25:1 with no EM,
22:1 for radio frequency monitoring, 17:1 for active GPS monitoring, and 8:1 for passive GPS
monitoring (Florida Senate Committee on Criminal Justice 2004).
3. False sense of security. The public may not understand the limitations of this technology and
assume it is a panacea. In truth, GPS is a tooland one that can fail. For instance, in a pilot study,
some intentional violations on the part of volunteers were not detected by the system (Elzinga
and Nijboer 2006). There are also documented instances of “false” readings, such as when
offenders were recorded in one place when officers knew them to have been elsewhere
(Sachwald 2007). Also, although GPS units may be able to track where offenders are, they cannot
provide information on what offenders are doing.
4. Legal concerns. Courts have not yet decided how to resolve challenges to the use of GPS. If
equipment malfunctions and a crime is committed, will departments be held responsible? What
happens if the department fails to respond to an alert? Lawsuits over such matters could cost
departments millions of dollars in court costs and damages.
USES OF EM. EM can be used at different points in the judicial systemfor example, for pretrial
supervision as an alternative to jail, as an alternative to incarceration for selected offenders, or as part of
a mandated supervision program after release from prison.
It also can be used for different purposes, including
Public safety
Safety of individual victims
Accountability of offenders
Behavior change and recidivism reduction
Reduction of jail or prison populations
Reducing costs
Notably, not all of the purposes are mutually compatible (Florida Senate Committee on Criminal Justice
2004). Thus, departments using EM should clarify their goals at the start of the program.
Some critics (e.g., Lilly 2006, Nellis 2006) have noted the absence of rehabilitation as an overall goal for
the use of EM, contrary to the intent of its earliest developers. In response to this absence, Burrell and
Gable (2008) have proposed the development of an incentive-based model of EM that could integrate a
rehabilitation component through the use of positive reinforcement. They note this type of model fits into
the framework of evidence-based practices and point to the success of drug courts in using positive
reinforcement to shape offender behavior and facilitate rehabilitation.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
111
EFFECTIVENESS OF EM. The research on the effectiveness in reducing recidivism is still somewhat mixed,
although the base confirming GPS effectiveness is growing. This result stems in large part from the
limitations of many extant studies. For instance, the 1997 report to the U.S. Congress (Sherman et al.
1997) categorized home detention with EM as an approach that “doesn’t work.” This conclusion,
however, was based on the only two studies deemed to have adequately rigorous designs (Burrell and
Gable 2008). A 2005 meta-analysis of 119 studies on the use of EM with moderate- to high-risk
offenders, conducted by Marc Renzema and Evan MayoWilson, faced a similar problem with study
limitations. They concluded “all studies [of EM] in moderate- to high-risk populations have serious
limitations and matched studies of EM in moderate- to high-risk populations are of very low quality.” Only
3 of the 119 studies considered by Renzema and MayoWilson incorporated a control or comparison
group in their research design, and all three produced inconclusive results on the value of EM. (For
example, Finn and Muirhead–Steves’ 2002 study of the EM program in Georgia found sex offenders on
EM were less likely to reoffend than their counterparts in the comparison group, but Renzema and Mayo
Wilson also found “evidence that EM may not have produced the observed differences.”)
However, some recent studies, with rigorous research designs, suggest the optimism about the potential
of EM may not be groundless. For instance, a 2006 study conducted by researchers at the University of
Florida makes a slightly stronger case for EM. Padgett, Bales, and Blomberg analyzed data from 75,661
serious offenders in Florida who had been placed on home confinement between 1998 and 2002 and
found that “Both radio frequency and global positioning system monitoring significantly reduce the
likelihood of technical violations, reoffending, and absconding for this population of offenders.” The
positive effect was particularly noteworthy since the population placed on EM was a significantly higher
risk population. However, Padgett, Bales, and Blomberg also found EM had a lesser impact on sex
offenders than on other offender groups. Although violent offenders on GPS monitoring were 91.5 percent
less likely to commit a new offense than violent offenders who were not electronically monitored, sex
offenders were only 44.8 percent less likely to commit a new offense. This small treatment effect is
probably the result of most sex offenders’ relatively low tendency to be rearrested. They also noted that,
given the efficacy of both RF and GPS, the price differential for their use is substantialan important
consideration for policymakers.
A 2010 study conducted by researchers at Florida State University also offers evidence for the
effectiveness of EM. For their quantitative analysis, Bales and colleagues analyzed data on 5,034
medium- and high-risk offenders on EM and 266,991 offenders not placed on EM over a 6-year period;
they used propensity score matching to minimize selection bias. The researchers found EM reduced
offenders’ risk of failure by 31 percent; within the EM group, GPS monitoring resulted in 6 percent fewer
supervision failures, compared with RF. They noted that all categories of offenders, regardless of offense
type, experienced fewer supervision violations as a result of EM; however, the effect was reduced for
violent offenders. For their qualitative analysis, the researchers conducted interviews with 105 offenders,
36 supervising officers, and 20 administrators from throughout Florida. They found that offenders and
their families suffered negative consequences, including poorer relationships with significant others and
children and offenders’ more frequent inability to obtain and retain employment. They also concluded that
EM appeared to be a cost-effective method for dealing with offenders.
Two studies released in 2012 provide growing evidence for the effectiveness of using GPS for different
populations. A 2012 study of 516 high-risk sex offenders in California assessed the outcomes of those
receiving traditional parole supervision compared with those receiving GPS supervision (Gies et al. 2012).
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
112
The two main outcomes of interest were noncompliance (measured through violations of parole) and
recidivism (measured through rearrest, reconviction, and return to prison). Gies and colleagues used a
survival analysis of time-to-event recidivism data, using a Cox proportional hazards model. The study
found the hazard ratio of a sex-related violation was nearly three times as great for subjects who received
traditional parole supervision than for subjects who received the GPS supervision. In terms of recidivism,
compared with subjects who received the GPS monitoring supervision, the hazard ratio for any arrest was
more than twice as high among subjects who received traditional parole supervision.
A 2012 study by Erez and colleagues examined the use of GPS for enforcing court-mandated “no contact
orders in domestic violence cases. The study examined the outcomes for more than 3,600 defendants
referred to the GPS program across three sites. Outcomes of interest included short-term outcomes
(defendants’ program violations and rearrests during the pretrial period) and long-term outcomes
(rearrests during a 1-year follow-up period after the case). The results indicated GPS was associated with
practically no contact attempts. Furthermore, defendants enrolled in GPS monitoring had fewer program
violations compared with those placed in traditional electronic monitoring (EM) that used RF technology.
Erez and colleagues also found defendants on GPS had similar conviction rates across the three sites to
those who remained in jail during the pretrial period.
These findings provide promising evidence that EM can reduce recidivism. Still, none of the existing
studies on EM has shown EM does more than postpone recidivism. Parolees appear to be compliant
while subject to monitoring, but, in the words of Peckenpaugh and Petersilia (2006), when the bracelets
come off, other studies have found that monitored offenders perform no better than offenders [who] were
never subject to monitoring.” Gainey, Payne, and O’Toole (2000) have raised the related issue of whether
time spent on EM affects recidivism. Their review of the limited research on the relationship between
recidivism and time served and the relationship between time on EM and program completion led to
mixed findings. Their own study found the more time offenders spent on EM, the lower the likelihood of
recidivism. However, this result varied by type of offender. These findings are provocative but provide only
a starting point for answering questions about the impact of EM on recidivism after the removal of EM.
They also suggest the need for further research about the impact of EM when it is part of a
comprehensive program and is not studiedas do Padgett, Bales, and Blomberg (2006) and Bales and
colleagues (2010)in isolation from other program components.
C. THE CALIFORNIA DEPARTMENT OF CORRECTIONS AND REHABILITATIONS
GLOBAL POSITIONING SYSTEM SUPERVISION PROGRAM
CDCR is charged with the responsibility of administering the program to monitor with GPS technology all
offenders released from prison, living in the community, and placed on a specialized gang caseload.
According to the original CDCR policy protocols, the sanctioned goals of this program are identical to
those of the sex offender program. They include the following:
1. Use the technology to gather information that can enhance supervision.
2. Provide PAs and local law enforcement with the ability to monitor the location and movement of
targeted parolees.
3. Aid in the investigation of parole violations and criminal investigations.
4. Strengthen partnerships with local law enforcement agencies.
Informally, however, the goal of the gang program is very different from that of the sex offender program.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
113
Whereas GPS is used to legitimately manage sex offenders in the community, anecdotal evidence
indicates GPS is used to monitor gang offenders with the intent of returning them to prison as soon as
possible as a public safety measure. Another difference between the programs is a differentiation in
supervision requirements by risk level. For example, in contrast to the sex offender program, where the
tracks of low-risk offenders are reviewed less frequently, the tracks of all offenders placed on a
specialized gang offender caseload are reviewed daily. In addition, specialized gang caseloads
CANNOT exceed 20 cases.
Can include ONLY active GPS cases.
Can include ONLY GPSmonitored parolees.
Can include ONLY gang offender parolees.
Eligibility for and Designation of High-Risk Gang Offender
All GPS cases are assigned to a specialized caseload that has specific distinct requirements
differentiating it from a traditional parole caseload. To be eligible to be placed on the specialized gang
offender caseload, a parolee must meet at least one of the mandatory criteria listed in CDCR Form 2203
(Rev. 0411 0), GPS Monitoring Gang Eligibility Assessment Criteria, prior to assignment. If the parolee
meets any of these criteria, the Agent of Record (AOR) shall hold a case conference with the Unit
Supervisor (US) to determine if placement on the caseload is appropriate
. The criteria include the
following:
The parolee has been verified as a currently active member/associate of a prison disruptive
group pursuant to the California Code of Regulations, Title 15, Crime Prevention and Corrections,
Division 3, Section 3378(c)(1).
The parolee has been validated as a prison gang member/associate pursuant to the California
Code of Regulations, Title 15, Crime Prevention and Corrections, Division 3, Section 3378(c)(1).
The parolee has a special condition of parole to not associate with any prison gang, disruptive
group, or street gang member pursuant to the California Code of Regulations, Title 15, Division 2,
Section 2513(e).
The parolee was convicted pursuant to Penal Code (PC) Sections 182.5 and 186.22 (i.e., active
participation in any criminal street gang) and currently has a special condition of parole to not
associate with any prison gang, disruptive group, or street gang member.
The parolee is subject to gang registration requirements pursuant to PC Sections 186.30
186.32.
The parolee is a party to an active court civil gang injunction.
Unfortunately, this process is not standardized and no data exists on the details of the decision-making process.
However, anecdotal evidence from formal and informal interviews suggests that each time an eligible subject is released
from prison, the AOR and US discuss the merits and detriments of supervising the offender with GPS. Among the factors
considered are the current size of the specialized gang caseload, the risk of violence in comparison to those already under
GPS supervision, the offender’s gang affiliation and status within the gang, the California Static Risk Assessment score,
and collateral information received from other stakeholders (local law enforcement, special task forces, institutional/prison
gang units, etc.).
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
114
The parolee has been identified by CDCR staff or local law enforcement to be or have previously
been involved in gang activity.
The PA and US may also use the following additional criteria to make a final determination:
The AOR has reason to believe the parolee is not in compliance with current parole conditions,
and enhanced parole supervision is required in the interest of public safety.
The parolee has been identified by local law enforcement as being a suspect in a felony crime
involving violence and/or weapons.
The parolee has a controlling or non-controlling commitment offense(s) included in PC Section
667.5(c) (i.e., a violent felony).
The parolee has a risk number value of 5 (i.e., high risk to commit violent offenses as determined
by the California Static Risk Assessment).
The Program Components
The CDCR HRGO Monitoring Program is composed of two distinct elements: GPS monitoring and intensive
supervision.
These components are described in detail below.
GPS MONITORING. The GPS monitoring component employs the tracking system of two different vendors:
Satellite Tracking of People (STOP) LLC and Pro Tech. STOP is used in the southern portion of California
(Regions 3 and 4),
and Pro Tech
is responsible for the northern areas (Regions 1 and 2). Although the
terminology of the vendors differs, the capabilities of hardware and software are virtually identical. Each
vendor employs an active monitoring system that combines cellular and GPS technology to automatically
track the location of a parolee. The unit takes a data point every minute and transmits the location data
every 10 minutes. The tracking device is a single-piece GPS unit that weighs about six ounces and is
roughly the size of a computer mouse. The device is worn flush around the left ankle, secured by a
tamper-resistant, fiber-optic technology strap and specialized security screws
to secure the strap to the
device. The battery can operate longer than 48 hours on a single charge, and recharging takes roughly 1
hour from any standard 110-volt electrical outlet. The batterys lifespan is typically 1 to 2 years.
The software system of each vendor employs a combination of data integration, geomapping, and GPS
technology to monitor parolees. Each vendor tracks information about parolee activities supplied by the
GPS technology and transmits it to the supervising PA through the monitoring center. The monitoring
center provides the PA with information in two basic forms: daily summary reports (DSRs) and immediate
alert (IA) notification. For each parolee, a DSR is emailed to the PA each morning. The notification details
It should be noted that unlike the GPS monitoring program for sex offenders, there is no mandatory treatment
component.
CDCR is organizationally and operationally divided into four distinct regions, with numerous districts within each region and
numerous parole units within each district. Region 1 consists of the Central Valley, ranging from Bakersfield to the Oregon
border, while Region 2 encompasses the coastal counties from Ventura to the Oregon border. Region 3 includes only Los
Angeles County, and Region 4 consists of the southern counties of Imperial, Orange, Riverside, San Diego, and San Bernardino.
(See attachment A for a map of CDCR regions.)
Strap clips and bridge clips have replaced the screws to secure the strap since the time period of this study.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
115
all activity recorded by the GPS unit, including charging activity, zone violations, strap tampers, and other
violations. The PA must review all recorded activity and note any actions that stem from the notification.
The notification also includes a direct link to a Web-based data system for a review of the “tracks” or
movement patterns of any offender on any GPS caseload. The software plots the location and movement
on an interactive Web mapping service application, allowing the PA to see the movements of a parolee
and investigate any unusual or suspicious movement patterns. PAs are provided with laptops enabled
with wireless Internet cards to allow access to software from the field.
An IA notification is automatically generated by the monitoring center and transmitted to the supervising
PA
through a text message when the GPS unit records specific types of violations
. Upon receipt of an IA
notification, the supervising PA must analyze and appropriately respond to the information contained
within the notification. This investigation typically begins with the transmission of a signal that forces the
unit to beep or vibrate, indicating the offender must either telephone or physically appear before the PA
immediately. If these methods fail to resolve the problem and the event is regarded as a serious threat to
public safety, the PA follows response protocols including but not limited to responding to inspect the
device, attempting to locate the offender, etc. The PA may also contact local law enforcement for
assistance. Each situation is different and while guidelines are given, agents and supervisors must use all
available information to decide on the appropriate level of response.
The GPS monitoring technology includes numerous other features that aid the PA in monitoring the
offender, including the following:
Inclusion zone: A geographic location that an offender is required to occupy during certain times
of day. The application of an inclusion zone enables the PA to be alerted to a parolees movement
out of the specific location. Inclusion zones may include but are not limited to the parolees
residence, employment, or treatment location.
Exclusion zone: A geographic location that an offender is prohibited from entering at all or during
certain times of day. Contrary to the inclusion zone, the application of an exclusion zone enables
the PA to be alerted to a parolees movement into a specific location. Exclusion zones may
include but are not limited to the victims residence, areas of known narcotic activity, prior arrest
locations, or areas of restricted travel.
Track mapping: Tools and procedures for analyzing an offenders movements on a map
Status call button: A feature that initiates an audible tone and/or vibration from the receiver
During the study period, the alert notification protocol operated as described. However, CDCR has subsequently altered
this operational model to reduce the burden placed on PAs to respond to a multitude of minor alerts. Effective October
2012, alert notifications are now triaged through a Vendor Monitoring Center (VMC). The VMC follows preestablished
protocols to triage GPS alert information. For less urgent alerts, the VMC will attempt to resolve issues directly with the
parolee prior to PA involvement. In the event the VMC cannot resolve the alert with the parolee, it is escalated to the PA.
For more urgent GPS alerts, the VMC will provide immediate notification to the PA in accordance with established IA
notification protocols.
The specific types of violations are known to the researchers. However, at the request of CDCR and to preserve the
integrity of the parole program, this detail is omitted from the final report.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
116
Crime scene correlation: The intersection of crime incident data with GPS tracks to determine
whether any offender was in the vicinity of a crime.
The specific requirements for the GPS supervision component calls for a PA assigned a specialized gang
caseload to
Review the DSR for each GPS-monitored parolee at regular intervals.
Conduct a track review for each GPS-monitored parolee at regular intervals using specific
methods.
Immediately respond to all GPS alert notifications (exclusion, inclusion, tamper, gap, cell, battery).
Resolve all GPS alert notifications and note actions taken to clear the event (exclusion, inclusion,
tamper, gap, cell, battery).
Assign a residence inclusion zone (or transient inclusion zone for homeless parolees), a travel
restriction zone, and a victim exclusion zone (if applicable) (residence, travel, victim).
All PAs involved in the GPS program (whether or not directly supervising parolees) must be trained by the
Division of Adult Parole Operations’ EM unit in the use of GPS technology as a parole supervision
monitoring tool. The training program covers a variety of information, including policies, procedures, and
protocols when using GPS as a supervision tool. PAs must attend GPS training before supervising
parolees using GPS.
INTENSIVE SUPERVISION. The intensive supervision component involves recurrent contact with HRGOs by
PAs. The PA meets face-to-face with the parolee on the first working day after release and informs the
parolee that GPS monitoring technology is being added as a special condition of parole and that
participation in the program is mandatory (refusal will result in immediate revocation of parole and return
to prison). Specifically, the traditional intensive supervision component requires a PA assigned a
specialized gang caseload to
Establish first contact with the parolee within a specific number
of days after release.
Conduct the initial interview within a specific number
*
of days after release.
Meet at the parolee’s residence within a specific number
*
of days after release.
Conduct a minimum number
*
of face-to-face contacts with the parolee each month.
Conduct a minimum number
*
of collateral contacts per month.
Meet with law enforcement to update parole information a minimum number of times
*
per year.
The exact number is known to the researchers. However, at the request of CDCR and to preserve the integrity of the
parole program, this figure is omitted from the final report.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
117
Conduct a minimum number
*
of random drug tests each month.
Conduct a case conference review a minimum number
*
of times each year.
D. THE STUDY GOALS
Goals
The overall purpose of this study is to conduct a quasi-experimental evaluation of the CDCR GPS
monitoring program of HRGOs. Specifically, the goals are to
Assess the fidelity of the program.
Assess the cost of the GPS program.
Assess the effectiveness of the GPS program for gang offender.
Objectives
This project has set several highly specific objectives to measure the success of its goals. It should be
noted that the objectives for the effectiveness goal do not predict a direction. This lack of direction is
intentional due to questions (discussed above) regarding the gang program’s intent. A deterrence-based
program like the sex offender program would view a decline in criminal behavior as desirable because it
results in fewer crimes. Conversely, a suppression-based program would likely see an initial increase in
recidivism among the treatment subjects as a positive outcome because returning these offenders to
custody may ultimately increase public safety. Nevertheless, the specific objectives of the project,
organized by goal, are as follows:
1. Assess the Fidelity of the GPS Program.
Determine the program adherence to all core components (i.e., program staffing qualifications,
caseload restrictions, parolee orientation specifications, and parole supervision specifications).
Determine the degree to which the prescribed level of program exposure was obtained.
Determine the quality of program delivery (e.g., skill of the staff in using techniques or methods
prescribed by the program and preparedness or attitude of staff toward the program).
Determine the degree to which program components were reliably differentiated from one
another.
2. Assess the Cost of the Program.
Determine the cost of monitoring HRGOs with the GPS system.
Determine the cost of monitoring HRGOs without the GPS system.
3. Assess the Effectiveness of the GPS Program for Gang Offenders
Determine the effect of GPS monitoring on offenders’ subsequent occurrence of noncompliance
with parole conditions (i.e., technical violation and nontechnical violation).
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
118
Determine the effect of GPS monitoring on offenderssubsequent occurrence of criminal behavior
(i.e., rearrest for any offense and rearrest for a violent offense).
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
21
2. Methodology
A. OVERVIEW
This study integrates outcome and process evaluation components. The outcome component assesses
the impact of the California Department of Corrections and Rehabilitation (CDCR) GPS supervision
program by employing a nonequivalent-group quasi-experimental design with a multilevel survival model.
We also use a propensity score matching procedure to account for differences between the treatment and
comparison groups. The study population is drawn from high-risk gang offenders (HRGOs) (as determined
by the GPS Monitoring Gang Eligibility Assessment Criteria Form) who have been released from prison
and are residing in the State of California. The effectiveness of the program is assessed using an intent-
to-treat (known as ITT) approach, with two main outcomes of interest: noncompliance and recidivism.
Noncompliance is operationalized as a violation of parole. Recidivism, on the other hand, is
operationalized as an arrest for a new crime. Each outcome is assessed with a survival analysis of
discrete-time data, using a random intercept complementary loglog model. In addition, frailty modeling
is used to account for the clustering of parolees within parole districts. The outcome component also
includes a cost-effectiveness analysis of each outcome. The process component (see chapter 4) uses
quantitative and qualitative methods to provide a rich context to the program treatment and structure and
to assess program fidelity (i.e., whether the program was well-designed and implemented as intended).
B. PARTICIPANTS
California is notorious for having a substantial population of gang members. In fact, the 2010 Organized
Crime in California Annual Report to the California Legislature suggests gangs operate in cities of all sizes
throughout the state (Office of the Attorney General [OAG] 2010). In concordance with the OAG report, the
2011 National Gang Threat Assessment (NGTA) found California, Arizona, and Illinois are the states with
the highest number of gang members in the country. With data collected by the National Drug Intelligence
Center through the National Drug Threat Survey and National Gang Intelligence Center, the NGTA
estimates there are more than six gang members per 1,000 people in the State of California (NGTA
2011).
In response to this problem, CDCR’s Division of Adult Parole Operations in March 2006 entered into a
partnership with the City of San Bernardino to implement a pilot project to track the movements of known
gang members. In May 2007, Governor Schwarzenegger implemented the California Gang Reduction,
Intervention and Prevention initiative (CalGRIP) to provide more than $48 million in state and federal
funding for local antigang efforts, including prevention, intervention, enforcement, job training, and
education strategies (Governor’s Office of Gang and Youth Violence Policy 2010). CalGRIP also expanded
the 20-unit pilot program in San Bernardino to an 80-unit program by adding 20 units each in five
additional jurisdictions. This study focuses on HRGOs who are released from prison and placed on parole
supervision with GPS monitoring in six
California jurisdictions. This group (n=407) includes all HRGOs
placed on GPS monitoring technology from March 2006 through October 2009 in each of the six
specialized gang units located in the City of Los Angeles and the following California counties: Fresno, Los
Angeles, Riverside, Sacramento, and San Bernardino.
The GPS supervision of HRGOs expanded to a number of other jurisdictions and then contracted due to budget
considerations during the course of this study. This research focuses on the original six jurisdictions.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
22
Propensity Score Adjustment
To identify comparison individuals likely to have pretreatment risk characteristics similar to those in the
treatment group, a propensity score procedure was performed using a sample of offenders drawn from
each of the same six communities that maintained specialized gang units, but who were not placed on
GPS at the time of data collection. The initial sample included more than 145,000 records. This sample
was narrowed down to nearly 11,000 subjects by eliminating duplicate records (15,324 records) as well
as subjects who were a) paroled outside the 200609 time frame (10,576 records); b) paroled out of
state (554 records); c) deported (14,237 records); d) paroled to a “nonrevocable” parole status (5,157
records); e) paroled to a unit outside of the study strictures (52,277 records); f) placed on parole with no
conditions relating to gang membership or association (34,791 records); g) parolees with discharge dates
before Jan. 1, 2009, because it was unlikely these Record of Supervision (ROS) files would be available
(1,444 records); and h) parolees with a unit code designated as MNRP, an administrative code (383
records). This working sample included 10,963 subjects (407 treatment and 10,556 control subjects).
The working sample was used to match the treatment group with a similar group of control subjects using
the STATA propensity score procedure PSMATCH2 (Leuven and Sianesi 2003). The PSMATCH2
implements full Mahalanobis matching and a variety of propensity score matching methods to adjust for
pretreatment observable differences between a group of treated and a group of untreated subjects.
Matching methods to choose from include one to one (nearest neighbor or within caliper; with or without
replacement), knearest neighbors, radius, kernel, local linear regression, and Mahalanobis matching.
This study used the one-to-one nearest-neighbor method. The treatment group was matched on race, age,
gender, admit status, controlling offense type, controlling offense severity, registration as a violent
offender, narcotics offender or sex offender, drug and alcohol testing requirements, date of parole, and
parole district. A propensity score was generated for each parolee. The PSMATCH2 program for STATA
matched control and treatment group parolees to unique nearest neighbors whose propensity score was
within a certain caliper.
Because parolees’ were assigned a single match, the data were sorted randomly
before the procedure was run. Parolees who could not be matched were dropped. The matching
procedure resulted in a final sample of 784 subjects (392 treatment
and 392 control subjects). The two
groups did not differ significantly on any variable.
Independent samples ttests and chi-squared tests were run to investigate differences between the
matched sample (n=784) and parolees who were not selected in the matching process (n=10,179). The
matched sample had significantly fewer African American parolees, as well as fewer sex offender
registrants. It was also significantly more male, Hispanic, and had more violent offender registrants. In
terms of controlling offenses, the matched sample had significantly more violent and other offenses,
and more charges, as well as significantly fewer drug and property offenses.
C. DATA SOURCES
Once the treatment and control groups were established, we used six primary sources to collect data: 1)
the CDCR data management system, 2) official arrest records, 3) parole supervision records, 4) GPS
monitoring data, 5) a CDCR parole agent (PA) survey, and 6) CDCR cost information.
We experimented with various calipers. We chose the caliper that resulted in the largest number of possible cases for
analysis while eliminating selection biases in the variables included in the matching (i.e., there were no significant
differences between the groups).
The treatment group was slightly reduced (15 subjects) because there was no admit status in the data.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
23
CDCR Data Management System
California operates a data management system that houses numerous databases relevant to the
supervision of HRGO parolees. These databases include but are not limited to the Automated Release
Date Tracking System, the Correctional Offender Management Profiling for Alternative Sanctions (known
as COMPAS), CalParole, the Parole Law Enforcement Automated Data System, the Revocation
Scheduling Tracking System (RSTS), the Offender-Based Information System (OBIS), the Distributed Data
Processing Systems, and the California Law Enforcement Telecommunications System. Most data used
for this study were derived from three databases: CalParole, RSTS, and OBIS. The CalParole tracking
system stores a variety of information on offenders released from prison and placed on parole, including
birth date, gender, race, residency information, the date the parolee was released from prison, the date
the parolee is scheduled to be discharged from parole, any special conditions linked to parole, and the
unit and agent to which the parolee is assigned. RSTS stores a vast array of data regarding parole
revocations, including information on the date and type of parole violation and the result of the parole
revocation hearing. OBIS maintains a rich database of information concerning prior criminal history (date
of arrest, arrest charges, disposition date, disposition charges, disposition, and length of sentence) for all
adult offenders in California. A central feature of the California system is that offenders are linked across
all three of these systems through a unique identifier that permits users to find the same individual in
different contexts or data systems.
Official Arrest Records
Another principal data source for this study was the official record of arrests, convictions, and custody
(commonly known as a RAP sheet) for each study subject. Official records are frequently used in research
on recidivism. However, there are many methodological issues involved in assembling and interpreting
data from RAP sheets. These sources of error include but are not limited to linking dispositions to arrests,
false negative errors in arrest records, definitional problems in interpreting RAP sheets, handling events
with multiple charges, and dealing with technical violations. To minimize such errors, researchers in this
study worked closely with CDCR staff to correctly interpret RAP sheets. All records were manually entered
into a database specifically developed for this study.
Parole Supervision Records
PAs maintain a record of supervision for each parolee under their supervision. Specifically, the PA notes
the date and specific type of contact. A contact may be categorized as follows: a) initial interview, b)
office, c) residence, or d) collateral contact. The ROS is stored only in hard copy format in the parolee’s
case file, which is typically located in the parole unit of record. Consequently, a set of site visits was
conducted to obtain the record of supervision data from the PA case files. Again, all data were keyed
directly into a database specifically developed for this study. These data were collected to measure the
level of supervision received by each offender and to assess the California GPS program model.
GPS Monitoring Data
The GPS monitoring data were used to categorize subjects in groups as well as for descriptive purposes
and to assess the California GPS program model. The GPS monitoring system into which HRGO parolees
are enrolled was operated by two vendors (Satellite Tracking of People [STOP] LLC and 3M) during the
course of this study.
STOP is used in the southern portion of California (Regions 3 and 4
), and 3M is
It should be noted that CDCR recently discontinued its relationship with 3M and placed all parolees under the STOP LLC
system.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
24
responsible for the northern areas (Regions 1 and 2). While the terminology of the vendors differs, the
capabilities of hardware and software are virtually identical. As described in chapter 1, each vendor
employs an active monitoring system that combines cellular and GPS technology to automatically track
the location of a parolee. Each vendor provided an assignment history for each parolee to indicate the
date and time an offender was monitored with GPS technology. In addition, each vendor provided a
record of each GPS event
(Inclusion Zone, Exclusion Zone, Battery, Strap/Device, Cell Communication
Gap, and No GPS Communication) that included the event start and stop times and duration during a
specified period.
CDCR Parole Agent Survey
The survey instrument was developed to collect process data from CDCR PAs. To facilitate comparisons
between the two studies, it was adapted from the survey instrument used in the study of the CDCR GPS
monitoring program for sex offenders (Gies et al. 2012). The final version of this survey contained
questions in seven areas:
1. Program staffing
2. Agent information
3. Equipment issues
4. Caseload specifications
5. Enrollment and orientation
6. Collaborative engagement
7. General summary
The instrument was emailed to all PAs in August 2012. It was used to question PAs about core program
components and administered in a Web-based format, in which an email was sent to agents by CDCR
encouraging their participation. The email also contained a note introducing the anonymous Web-based
survey, instructions for taking it, a link to the survey embedded within the text, and a password to
securely access it. PAs were sent numerous requests to complete the survey throughout the month; the
survey was closed at the end of September 2012. The request received 24 unique and eligible responses,
a figure that roughly corresponds to the number of agents carrying gang-related GPS caseloads at the
time of the survey. At that time, there were roughly 30 level 1 GPS PAs
with existing gang offender
caseloads, yielding a good response rate (83.3 percent) for GPS PAs.
CDCR originally implemented six specialized gang units in the following California communities: Fresno,
the City of Los Angeles, Los Angeles County, Riverside, Sacramento, and San Bernardino. Subsequently,
during the course of this study, CDCR first added (due to the high number of gang offenders under
supervision) then withdrew (due to budget concerns) several specialized gang units.
Overall, the survey
provided a good representation of the GPS PAs (see table 4.1 in chapter 4). An analysis of the survey data
CDCR is organizationally and operationally divided into four distinct regions, with numerous districts within each region and
numerous parole units within each district. Region 1 consists of the Central Valley, ranging from Bakersfield to the Oregon
border, while Region 2 encompasses the coastal counties from Ventura to the Oregon border. Region 3 includes only Los
Angeles County, and Region 4 consists of the southern counties of Imperial, Orange, Riverside, San Diego, and San Bernardino.
GPS event data for the study period were available from a single vendor (i.e., STOP). The second vendor (i.e., 3M) was
unable to provide the data at the date of publication (n=281). Although 3M provided a file designated to include event
data, the majority of this data was missing. Efforts are ongoing to retrieve this data from 3M.
Level 1 PAs directly supervise parolees.
The sample in this study is restricted to the original six specialized gang units.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
25
by district suggests the volume of responses favored the regions
encompassing the original six
specialized units, with Region 1 representing 29 percent of respondents, Region 2 representing 16
percent of respondents, Region 3 representing 29 percent of respondents, and Region 4 representing 25
percent of respondents. Notably, Region 3 (Los Angeles County) is the smallest geographic unit. The
distribution across the original six districts was comprehensive, with at least two responses (most have
many more) from each of the parole districts.
Cost Information
The cost information elements used in the analysis are grouped into four broad categories: 1) personnel
(all fulltime and parttime staff and consultants), 2) facilities (i.e., the physical space required for the
program), 3) equipment and materials (furnishings, instructional equipment, etc.), and 4) other inputs (all
remaining costs that do not fit into the other categories). This information was obtained through
communications with CDCR staff and a review of budget documents. A cost-effectiveness analysis
worksheet was developed that divided all cost elements into one of the four broad categories. This
worksheet was transmitted to CDCR by electronic communication with a request to add the monetary
values to each category along with explicit instructions to add any cost element that was missing from the
initial draft. Follow-up discussions by electronic communication were used to refine cost elements and
associated monetary values. For verification and to correct the cost elements, a final version of the
worksheet was transmitted to a CDCR budget analyst.
D. MEASURES
Outcomes
The two main outcomes of interest were noncompliance and recidivism. Noncompliance was
operationalized as violations of parole. CDCR tracks numerous different types of prohibited parolee
behavior that can be divided into technical and nontechnical violations. Technical violations refer to
behavior by an offender under supervision that is not by itself a criminal offense and generally does not
result in arrest. This type of non-criminal behavior includes absconding, access to weapons, association
with known gang members, and various other violations of the parole process. Nontechnical violations
refer to behavior that constitutes a new criminal offense. Nontechnical violations can range from less
serious types of violations such as drug possession to very serious violations such as assault, rape, and
homicide. This study examines both types of violations. The data were obtained from the CDCR Data
Management System. Recidivism, on the other hand, was operationalized as an arrest for a new crime.
The distinction between a nontechnical violation and a new arrest is the method of processing the event.
Specifically, parole violations are processed through a parole board hearing while a new arrest is
processed through traditional court proceedings. Arrest data were obtained from official records (RAP
sheets). Each outcome was measured in terms of the month the event occurred. Each subject was
tracked for 2 years after the initial parole date (month 1 through 24).
Independent and Control Variables
The main variable of interest was the use of GPS monitoring (i.e., GPS status). Group differences between
GPS and control condition subjects were minimized on a range of pretreatment characteristics, including
sociodemographic and criminal history measures through the use of the aforementioned propensity score
CDCR is organizationally and operationally divided into four distinct regions, with numerous districts within each region and
numerous parole units within each district. Region 1 consists of the Central Valley, ranging from Bakersfield to the Oregon
border, while Region 2 encompasses the coastal counties from Ventura to the Oregon border. Region 3 includes only Los
Angeles County, and Region 4 consists of the southern counties of Imperial, Orange, Riverside, San Bernardino, and San Diego.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
26
adjustment procedure. However, subsequent analyses of the arrest and custody data (only collected and
coded for treatment and control subjects subsequent to the matching procedure) revealed statistically
significant differences between the groups in terms of the number of drug and weapons arrests. Thus,
these two control variables were added to the model.
GPS STATUS. The main variable of interest is the use of GPS monitoring (i.e., GPS status). Unlike with sex
offenders in California, there is no legislative mandate to supervise gang offenders with GPS monitoring.
In other words, CDCR has the discretion to use GPS monitoring with any offender released onto parole
who is eligible under the Gang Eligibility Assessment Criteria. Parolees eligible for the specialized gang
caseload can go on and off the caseload (and thus GPS monitoring) at any time while on active parole,
depending on the caseload demand of the parole unit and discretion of the unit supervisor.
As a result, GPS status is measured in two ways. First, it is measured dichotomously, by noting group
membership (0=control group; 1=GPS group) as a continuous or static variable. The GPS group includes
HRGOs who received traditional parole supervision plus placement on GPS monitoring technology, while
the control group includes HRGOs who received only traditional parole supervision during the study
period.
This measurement specification takes an ITT approach, where all offenders who were assigned to
GPS supervision are considered to be part of the GPS group, regardless of whether the parolee received
the treatment. In general, an ITT approach offers a more conservative estimate of the treatment effect,
for a subject may be arrested while removed from GPS, but still assigned to the GPS group.
Discrete-time event history (survival) models offer a convenient and intuitive way of incorporating both
repeated events and time-varying covariates into models. The approach begins by creating a person
(parolee) X time (i.e., month) data set. Such an approach naturally lends itself to adding repeated events
(e.g., arrests) as well as explanatory variables that are constant but may vary in their effect over time (i.e.,
may have a greater or lesser effect over time) or that actually vary over time (i.e., change values over
time). To begin the analyses, we first calculated an at-risk variable that indicated whether within any given
month, the offender was at risk of an event occurring (i.e., arrest or parole violation). If the offender was
“on the street” at any point during the month, the at-risk variable was coded 1; if the offender was in
CDCR custody for the entire month, the at-risk variable was coded zero. This variable was used to
“censor” (code as missing) offender months in which the offender was not at risk of experiencing the
event. It allowed us the ability to create a person X time data set that allowed the predication of the
likelihood of an event occurring when an individual was at risk of having that event occur.
OTHER MEASURES. A good deal of literature suggests unstable environments are associated with criminal
recidivism (Walters 2003, Davies and Dedel 2006, Tille and Rose 2007). The CDCR maintains data on
the point at which offenders change their location of residence. We calculated the number of moves
offenders made in any given month and used this as a time-varying independent variable to predict arrest
and parole violations.
Although this approach makes group comparisons straightforward, the real-world management of high-risk offenders in
the community did not result in a clean delineation between the groups. The GPS tracking data revealed that 42 offenders
(10.7 percent) assigned to the treatment group, while exposed to GPS monitoring during the parole period, did not receive
GPS monitoring during the two-year tracking period following the current release from custody event. In addition, CDCR
placed 21 (5.4 percent) of control group offenders on GPS monitoring subsequent to the group assignment for the study.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
27
E. STATISTICAL OVERVIEW
Missing Data
No baseline item included in the propensity score matching procedure contained missing data. The
demographic information described in the matching procedure was collected for all subjects. In addition,
while the official records did not contain out-of-state and juvenile criminal histories, these events were
corroborated in other CDCR data sources. Thus, all subjects were confirmed to have been previously
arrested, convicted, and placed in prison at least once.
Treatment Outcome Analyses
A series of analyses was performed in sequential phases to assess the impact of the CDCR GPS
supervision program. The first phase of analyses explored the differences (or lack thereof) between
groups in numerous pretreatment characteristics, as well as outcomes at baseline. Independent samples
ttests were used to test for significance between the groups.
The second phase assessed the impact of the GPS program on each measure of recidivism. Here we used
a survival analysis model to predict time until each event. In a discrete-time model, time is treated not as
a continuous variable but as divided into discrete units or chunks (e.g., weeks, months, years). The model
is characterized by few possible survival (or censoring) times, with many people sharing the same survival
time (RabeHesketh and Skrondal 2008).
Discrete-time analysis is common in social and behavioral science applications because it can easily
accommodate both time-constant and time-varying covariates (i.e., covariates that change between the
time a person becomes at risk and experiences the event) (Muthén and Masyn 2005). For instance,
continuous-time models are predicated on the often unrealistic assumption that the effect of a covariate
on event occurrence is constant over time (Singer and Willet 1993). Yet in criminological research, the
effects of covariates such as marital status and employment may vary over time, with the risk of
reoffending lower during periods of marriage and employment compared with periods of separation and
unemployment. Finally, discrete-time models do not require a hazard-related proportionality assumption
that is commonly used in continuous survival analysis. Instead, they become models for dichotomous
responses when the data are expanded to personperiod data. Logit and probit models can then be used,
as well as complementary loglog models (RabeHesketh and Skrondal 2008).
REPEATED EVENTS. Up to this point, only nonrepeatable events have been discussed. However, in
longitudinal research, an event may occur more than once throughout an individual’s lifetime. For
example, a subject may be arrested, go to prison for a specified duration, and then return again to the
communityat which point the subject is again at risk for arrest. In such a case, the durations between
events may be correlated because of the presence of unobserved individual-level factors. Repeated
events are usually handled by including individual-specific random effects in an event history model,
pointing to the requisite for a multilevel modeling approach. The discrete modeling approach we chose is
also amenable to modeling repeated events.
CLUSTERING. Parolees are monitored by agents who operate within an explicit parole district, creating
clusters of subjects. In other words, each parolee is clustered or nested within a parole district. In
clustered data, it is usually important to allow for dependence or correlations among the responses
observed for units that belong to the same cluster (RabeHesketh and Skrondal 2008). For example, in
the present application, it is possible that recidivism outcomes for parolees from the same parole district
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
28
are correlated because parolees have been supervised within the political and regulatory environment of
the same district. To account for the data clustering, random-effects models (also called multilevel,
hierarchical linear, or mixed models) provide a useful approach for simultaneously estimating the
parameters of the regression model and the variance components that account for the data clustering.
MODEL. In this study, we use multilevel discrete-time survival models, where random effects, often called
frailties in this context, are included to handle the unobserved heterogeneity between clusters and within-
cluster dependence (RabeHesketh and Skrondal 2008). That is, to accommodate dependence among
survival times of parolees within the same district, after controlling for observed covariates, a random
intercept is included for each district. The frailty approach provides a means to examine heterogeneity
among subjects and to estimate the distribution of subsequent failure time with the use of failure times
and covariate information from other members in the cluster. For these reasons, frailty models have been
widely used for the analysis of clustered survival data (Hougaard 1995, Duchateau and Janssen 2008).
Discussions on the use of frailties models can be found in Hougaard (2000), Therneau and Grambsch
(2000), and Wienke (2010).
Specifically, the observations of the district measure with equal value are assumed to have shared (the
same) frailty. Across groups, the frailties are assumed to have a Gumbel distribution commonly found in
survival and event history analyses. According to RabeHesketh and Skrondal (2012: 782) the standard
Gumbel distribution has a mean of about .577 (called Euler’s constant) and a variance of π
2
/6, and is
asymmetric.In the present application, we specify a random-intercept complementary loglog model
using the xtcloglog command in STATA with a shared frailty option. This model is appropriate due to our
use of interval-censored survival times (and as a proportional hazards model would hold in continuous
time). These models include dummy variables for each period and do not include a constant. As
discussed above, a shared frailty is specified by parole district. The resulting exponentiated coefficients
produced by these models can be interpreted as hazard ratios in continuous time” (Rabe–Hesketh and
Skrondal 2008, p. 356).
The next chapter examines the results of these analyses.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
31
3. Results
A. BASELINE CHARACTERISTICS
Several demographic and baseline characteristics of the sample are displayed in table 3.1. In addition,
the groups were compared on parole district to account for the geographic diversity of the State of
California.
Table 3.1. Comparison of Baseline Characteristics: GPS and Control Groups
MEASURE
CONTROL GROUP
GPS GROUP
TVALUE
Sex
Male
99%
99%
.379
Race
African American
28%
29%
-.238
Hispanic
58%
59%
-.362
White
9%
8%
1.33
American Indian
1%
1%
.000
Other
4%
3%
.769
Age at Parole
Age
29.03 yrs
28.70 yrs
.648
Controlling Offense
Violent
36%
36%
.000
Drug
23%
22%
.511
Property
14%
15%
-.411
Other
27%
27%
-.161
Registrations
Narcotics Register
38%
.35%
.891
Drug Testing
85%
84%
.391
Alcohol Testing
23%
21%
.687
Violent Offender Register
19%
18%
.367
Arrest
a
Any Arrest
11.88
11.43
.906
Violent Arrest
2.92
2.78
.798
Drug Arrest
3.54
2.95
2.10*
Weapons Arrest
1.64
1.91
-2.28*
Gang Arrest
.265
.337
-1.67
Property Arrest
3.25
3.15
.454
Prior Custody
b
Days in Prison
1,522
1,681
1.51
Custody Events
3.69
3.90
-.801
Offender Status
New Admit
55%
55%
.072
Other
45%
45%
-.072
Note: Sample size: GPS group=392; control group=392.
a
Juvenile arrests are not included in the analysis. One subject had a single
juvenile arrest and no adult arrests. The subject was coded as having 0 prior arrests.
b
Custody includes only prison events (i.e., jail events
are excluded). Eight subjects were sentenced to a custody term but awarded time served and subsequently spent 0 days in CDCR custody.
*
p < .05;
**
p < .01;
***
p < .001.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
32
While there were no significant differences in the baseline characteristics used in the propensity score
matching procedure, some data were unavailable electronically and could not be collected for each
offender until the sample was tapered to a manageable scope. For instance, after the introduction of
arrest history, it was noted that there were small but significant differences between the treatment and
control groups in terms of prior drug arrests and prior weapons arrests. Specifically, the GPS group
tended to have more prior drug arrests in their criminal history, while the control group tended to have
more prior weapons-related arrests. Considering the population being researched in this study includes
high-risk gang offenders (HRGOs), it is very likely that these differences are due to chance. However, in
order to account for all group differences, all multivariate models include time constant covariates for
prior drug arrests and prior weapons arrests.
Gender, Race, and Age
Overall, the sample was 99 percent
male and consisted of more Hispanic
offenders (59 percent) than any other
race, but also included substantial
proportions of African American (28
percent) and white (8 percent)
offenders. The vast majority of
offenders (60 percent) were between
21 and 30 years old (see table 3.2).
The mean age of the full sample was
29 years at the time of parole. There were no statistically significant differences between the groups in
any of these characteristics.
Prior Arrests
Table 3.3 demonstrates an overall
long history of criminal behavior
among the subjects in the study. The
data indicate that only 18 percent of
the full sample was arrested less than
six times before the start of the study
period, with 34 percent arrested 6 to
10 times previously. In fact, nearly half
(48 percent) was arrested 11 or more
times, with nearly one fourth (24.1
percent) arrested 15 or more times.
Overall, the sample was, on average,
arrested about 12 times previously. There were no statistically significant differences between the groups
in any of these characteristics.
Table 3.2. Comparison of Age at Parole: GPS and Control Groups
MEASURE
CONTROL GROUP
GPS GROUP
Age
<20
6.4%
8.2.%
2125
32.4%
26.0%
2630
27.3%
33.4%
35+
33.9%
32.4%
Note: Sample size: GPS group=392; control group=392. No significant differences.
Table 3.3. Comparison of Prior Arrests:
GPS and Control Groups
MEASURE
CONTROL GROUP
GPS GROUP
Arrest Events
15
16.8%
18.1%
610
33.4%
34.2%
1115
25.0%
24.2%
15+
24.7%
23.5%
Note: Sample size: GPS group=392; control group=392. Juvenile arrests are not
included in the analysis. One subject had a single juvenile arrest and no adult arrests.
The subject was coded as having 0 prior arrests. No significant differences.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
33
Prior Custody
An examination of prior custody events
further confirms the frequency with
which the sample subjects participate
in criminal activity. Table 3.4 shows
that nearly two thirds (62 percent) of
the sample had been in prison more
than once prior to the study period,
with nearly 40 percent incarcerated
more than four times previously.
Overall, the sample was, on average,
incarcerated 2.5 times previously for
1,602 days before being paroled into our study. There were no statistically significant differences
between the groups in any of these characteristics.
Controlling Offense and Registrations
Not only did offenders in the sample demonstrate an elevated number of arrests and custody events, but
an analysis of the controlling offenses and registrations indicate a proclivity for serious and violent
behavior. The controlling or primary offense is designated by the court as the base termusually the
offense that keeps the offender in custody for the longest period of time. Overall, the data indicate that
the largest proportion (36 percent) of the sample was placed in custody for violent offenses. Offenders
were also placed in custody for drug (23 percent), property (14 percent), and a range of other offenses
(27 percent). The most frequent condition of parole was drug testing, which was a condition for 84
percent of the sample, with 22 percent having to submit to alcohol testing. Moreover, 36 percent and 19
percent of the sample were required to sign on to the narcotics and violent offender registry, respectively.
There were no statistically significant differences between the groups in any of these characteristics.
Summary
The previous tables provided information on several pretreatment characteristics of the sample. The
group comparison of these characteristics indicates that the two groups are very similar. In fact, the only
significant differences are that the control group had a greater number of subjects with prior drug arrests.
Conversely, the GPS group had a greater number of subjects with weapons-related arrests.
B. RECORD OF SUPERVISION (ROS)
Parolees are released into the community under very specific conditions, which often include
requirements such as obeying the law, refraining from drug and alcohol use, avoiding contact with the
parolees victims or other gang members, obtaining employment, and maintaining required contacts with
a parole agent (PA). To optimize the level of supervision for a population of HRGOs, CDCR standardized
the minimum number of specific contact types to which PAs are required to adhere (see chapter 5 for
more details on adherence). Specifically,
A PA assigned a specialized gang caseload MUST conduct a minimum number
of face-to-face
contacts with the parolee each month.
The exact number is known to the researchers. However, at the request of CDCR and to preserve the integrity of the
parole program, this figure is omitted from the final report.
Table 3.4. Comparison of Prior Prison Events:
GPS and Control Groups
MEASURE
CONTROL GROUP
GPS GROUP
Prison Events
1
38.8%
37.8%
2
13.5%
13.5%
3
9.7%
8.7%
4+
38.0%
40.1%
Note: Sample size: GPS group=392; control group=392. Custody includes only prison events
(i.e., jail events are excluded). Eight subjects were sentenced to a custody term but awarded
time served, and subsequently spent 0 days in CDCR custody. No significant differences.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
34
A PA assigned a specialized gang caseload MUST conduct a minimum number
*
of collateral
contacts per month.
A PA assigned a specialized gang caseload MUST conduct a minimum number
*
of random drug
tests each month.
A supervisor MUST conduct a case conference review a minimum number
*
of times each year.
Table 3.5 provides a comparison of the control and GPS groups in terms of supervision level. For each
contact type, the table displays the rate of contact per month (adjusted for the number of days under
supervision) for both the offenders who received GPS supervision and those who received traditional
parole supervision. A face-to-face contact is any visit or contact in which the PA meets directly with the
parolee. As such, the initial interview, office visits, and residence visits are all considered face-to-face
contacts. Collateral contacts are any contacts in which the PA checks up on the parolee indirectly through
family, friends, associates, and neighbors. In addition, drug tests are performed during office visits and
case reviews are conducted by parole unit supervisors. Both supervision activities are designed as
additional safeguards to ensure parolees are adhering to established guidelines.
While every effort was made to locate and code the ROS file for each subject in the study to control for
supervision, approximately 42 percent of the sample (326 subjects) were missing ROS records. The
reasons for these missing data varied, but the majority were no longer available because parole
administrators either purged the data from the file after a return-to-custody event or completely destroyed
the entire file shortly after discharge. Nevertheless, for those subjects whose ROS information was
obtained, the data demonstrate a significantly greater number of face-to-face and collateral contacts and
drug tests for the GPS group compared with control group subjects. For instance, agents meet face to
face with GPS parolees on average almost two times (1.84) per month compared with 1.54 times per
month for non-GPS parolees. These findings suggest the GPS group received an overall higher level of
supervision compared with subjects in the control group.
Summary
Table 3.5 provides information on the supervision of HRGOs in California. The group comparison of these
supervision elements indicates that (perhaps not surprisingly) subjects in the GPS group are supervised
more closely than the control group in terms of face-to-face and collateral contacts. In addition, agents
appear to more diligently review the case file of offenders placed on GPS monitoring compared with
agents who have traditional offenders. In short, while CDCR PAs attempt to strictly monitor known gang
offenders on parole, the activities embedded within the structure of the GPS monitoring program
necessitate more supervision.
In theory, the level of supervision is an important variable as it relates to recidivism, because offenders
who are monitored more closely in the community are less likely to engage in illegal behavior. Thus, one
may want to include a measure of supervision in a model predicting the effect of GPS supervision on
recidivism. In practice, however, this may not always be the case. A recent study by the Urban Institute
found parole supervision has little effect on rearrest rates of released prisoners. Specifically, the study
found mandatory parolees fared no better on supervision than similar prisoners released without
supervision. In fact, in some cases they fared worse (Solomon, Kachnowski, and Bhati 2005).
More problematic for this study, the level of supervision is likely directly related to the effect of GPS
supervision, as simply being placed on GPS monitoring necessitates more contact between agent and
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
35
offender. The increased contact comes in many forms, ranging from innocuous maintenance issues and
equipment failure to the investigation of parole violations directly related to GPS monitoring (e.g., zone or
curfew violations). These interactions are not a form of more intensive supervision, but rather intricately
intertwined into the operation of the GPS program, where controlling for the level of supervision may
actually remove a portion of the GPS program effect.
Given the inconclusive nature regarding supervision and the inability to disentangle the effect of
traditional supervision and GPS supervision, we do not attempt to control for it here.
C. OUTCOME ANALYSIS
Studies of criminal behavior typically use one or more of the following three measures to assess
reoffending:
Violation of parole
Rearrest
Return to prison custody
These measures are indicators of the occurrence of offending behavior. Each has strengths and
weaknesses. Violations of parole typically used to measure parolee noncompliance may or may not
constitute a new crime because offenders may commit acts that violate only the technical aspects of
parole (i.e., missing an appointment with a PA). It should be noted that as part of their parole, gang
members often receive special conditions not to associate with certain persons, typically known
associates, and other gang members. Violations of this rule, associating with persons prohibited” are
deemed to be technical parole violations. This specificity makes technical parole violations a more
interesting outcome for the population of interest to this study. Arrests are the most popular and
convenient measure of crime available, but an arrest does not prove a new offense actually occurred, as
occasionally the charges against an offender are dropped and the offender is released without further
incident. In addition, arrests account only for crimes that have been detected by law enforcement. Finally,
a return to custody is the narrowest measure of recidivism, as it accounts for only the most serious crimes
and violations that result in a prison term. This report uses measures of technical parole violations,
nontechnical parole violations, arrests, and arrests for violent offenses to assess the offending behavior
Table 3.5. Comparison of Supervision Record: GPS and Control Groups
CONTACTS PER MONTH
a
CONTACT TYPE
CONTROL
GPS
TVALUE
Supervision
Face to Face
1.57
1.84
-3.51
***
Collateral
1.05
1.25
-2.90
***
Total (Face and Collateral)
2.62
3.09
-3.48
***
Drug Test
.69
.076
-1.89
Case Review
.23
.28
-2.27
*
Note: n=458: 211 control, 247 GPS; 326 subjects (42 percent) were excluded from the supervision analysis because of the unavailability of ROS
data. ROS data are often purged from the file after a return-to-custody event and completely destroyed shortly after discharge from parole.
a
The number of contacts divided by the number of days in the community.
*
p < .05; ** p < .01; *** p < .001
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
36
of HRGO parolees. It also provides descriptive statistics for the sample’s return to custody during the
study period, the analysis of which will be featured in a forthcoming follow-up report.
Descriptive Statistics
Before running the discrete-time analysis models, the raw outcome data is described and summarily
assessed using chi-squared tests to look for group differences in categorical data. These are shown in
Table 3.6.
When considering parole violations, this study assesses technical and nontechnical parole violations
separately. As discussed above, an outcome of interest to this particular population is the technical
parole violations, which would document (among other violations) infringements to restrictions on a
parolees association with other known gang members. As shown in table 3.6, there were similar
proportions of parolees in both conditions experiencing such a violation in their 2 years from release to
parole, with 42.6 percent for the control group and slightly more (43.9 percent) for the GPS group. While
these differences were not significant, we can hypothesize that the GPS surveillance may help increase
detection of technical infringements, in particular when attempting to establish a person’s whereabouts
(e.g., determining they spent some time in the home of a known gang member). In terms of nontechnical
parole violations, more than half of both conditions experienced such an event in their 2 years post-
release to parole, with 53.9 percent for the GPS group and slightly more, 54.9 percent, for the control
group. These differences were also not significant.
When looking at arrests during the study period, 46.7 percent of the GPS group was rearrested at least
once, compared with 56.1 percent of the control group. The chi-squared test shows the treatment group
was significantly less likely to be rearrested than the control group [X
2
(1, n=784) = 6.99, p<.01]. These
effects can also be seen in the arrests for violent offenses, with the GPS group experiencing fewer arrests
for violent behavior (12.5 percent) compared with the control group (19.6 percent) during the 2 years
following their release from prison. Similarly, these group differences were also statistically significant
[X
2
(1, n=784) = 7.41, p<.01].
Finally, in terms of a return to custody during the 2 years after their release from prison, we note 79.3
Table 3.6. Comparison of Outcomes: GPS and Control Groups
EXPERIENCE OF THE EVENT DURING THE STUDY PERIOD
EVENT TYPE
CONTROL
GPS
CHI-SQUARE
Technical Parole Violation
167 (42.60%)
172 (43.88%)
.1299
Nontechnical Parole Violation
215 (54.85%)
211 (53.83%)
.0823
Arrest
220 (56.12%)
183 (46.68%)
6.9902
**
Arrest for a Violent Offense
77 (19.64%)
49 (12.50%)
7.4137
**
Return to Custody
272 (69.39%)
311 (79.34%)
10.1761
***
Note: n=784: 392 control and 392 GPS parolees.
*
p < .05;
**
p < .01;
***
p < .001
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
37
percent of the GPS group was reincarcerated compared with 69.4 percent of the control group. These
differences were shown to be statistically significant [X
2
(1, n=784) = 10.18, p=.001]. This is somewhat
surprising in view of fewer treatment parolees rearrested in general and rearrested for a violent offense;
however, we can hypothesize that this difference may be related to the increased ability to detect and
investigate crimes and parole violations using GPS tracking technology. This particular aspect of the
program, as well as a discrete-time survival analysis of the return to custody finding, will be the subject of
a forthcoming follow-up report.
These findings, however, only measure if an event occurred at least once during the study period for the
parolees. They do not measure the number of times the events occurred, nor do they account for a
number of observable independent variables that may moderate or contribute to these rates. In the
following section, discrete-time survival analyses, in the form of random-intercept complementary loglog
models, are used to assess the hazard of recidivism for the GPS intervention for arrest and parole
violation outcomes.
RESULTS OF DISCRETE-TIME SURVIVAL MODELS
Each outcome is assessed using multilevel discrete-time survival models, controlling for the geographic
mobility during the study period, prior weapons offenses, prior drug offenses, and the district into which
gang offenders are paroled using a shared frailty approach (to accommodate dependence among the
survival times of parolees within the same district). (See chapter 2 for a detailed description of each
measure.) Before running multivariate models, the dataset is declared to be panel data using the unique
parolee identifiers, and months, ranging from 1 to 24, as the time variable. As per discrete-time survival
analysis design, the model includes dummy variables for each of the 24 months from the subjects
release from prison.
The outcomes are modeled in a random-intercept complementary loglog model and produce output in
terms of regression coefficients, which, when exponentiated as reported here, can be interpreted as
hazard ratios (RabeHesketh and Skrondal 2008). The hazard ratio is an estimate of the differential rate
of a recidivism event for the GPS group compared with the control group. It should be noted that four
subjects died during the study period. Their outcome data is censored after their death as they are no
longer at risk of a recidivism event.
Technical and Nontechnical Parole Violations
We focus first on technical and nontechnical violations of parole, shown in table 3.7. In contrast to the
bivariate analyses, more robust multivariate models indicate that in both cases GPS is associated with a
greater likelihood of experiencing parole violations. The odds of a technical violation are 36 percent
greater among the treatment group, while the odds of a nontechnical violation are 20 percent greater. In
terms of control variables, geographic mobility (a time-varying independent variable), is as expected,
positively associated with both technical and nontechnical violations. In addition, prior drug arrests are
not associated with technical violations, but are significantly and positively associated with nontechnical
violations. This is unsurprising, as drug possession and drug use (as measured by mandatory narcotics
testing) are considered nontechnical parole violations. Prior weapons arrests are not associated with
technical or nontechnical parole violations.
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
38
Not presented in the table are 24 dummy variables (with no constant) representing the hazard of a
technical parole violation for each of the 24 months of the study. While interesting and potentially
important, these coefficients are not presented, as the tables would become unwieldy. For technical
violations, the coefficients are all statistically significant but relatively small and consistent (i.e., ranging
from .04 to .06). The hazards ratios for nontechnical violations are much larger and much more varied
ranging from .03 to .32. There are various interpretations of this finding, but it appears nontechnical
violations start relatively small (exp β = .03) and grow more or less linearly to about month 9 (exp β .32)
and taper off to the mid .20s thereafter. It is unlikely that this effect is due to procedural changes unless
there is a concerted effort to focus attention on offenders as they progress on parole. That is, since
parolees are released at different dates, this systematic change is likely not related to programmatic or
policy effects put in place at a certain time. Rather, the effect is likely due to variations in parolee
behavior that changes over time.
We next focus on overall arrests and arrests for violent crimes, presented in table 3.8. In contrast to the
positive effect of GPS on parole violations, the treatment group (GPS) is less likely to be rearrested overall
(the chance of being rearrested is 26 percent lower) and for violent crimes (32 percent lower). Among the
control variables, the only substantive effect was for prior arrests for drug offenses predicting overall
arrests. For overall arrests, the month variables, not presented in the table, are all statistically significant
but small, ranging from .02 to .07, showing no systematic variation over the two-year period. As true by
definition, the chances for violent arrests are even smaller, ranging from .004 to 0.4.
Table 3.7. Multivariate Random-Intercept Complimentary LogLog Models for Technical and
Nontechnical Parole Violations
TECHNICAL VIOLATION
NONTECHNICAL VIOLATION
VARIABLES
EXP β
SE
EXP β
SE
GPS
1.364
***
.091
1.203
**
.079
Moves
1.291
**
.117
1.341
**
.117
Prior Drug Arrest
1.013
.009
1.038
***
.008
Prior Weapons Arrest
.994
.021
1.038
.020
Log-likelihood: -3,512.91
Log-likelihood: -10,534.811
N = 784 over 24 months
Wald chi-square: 714.48(28)
***
Wald chi-square: 780.51(28)
***
Note: These models include 24-month dummy variables (no constant), and a shared frailty by parole district.
*
p<.05;
**
p<.01;
***
p<.001
This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Monitoring High-Risk Gang Offenders with GPS Technology: An Evaluation of the California Supervision Program
39
Summary
The results are mixed but reasonable. While speculative, the multivariate models suggest the GPS group
is significantly more likely to be violated for both technical and nontechnical conditions of their parole.
Presumably, this is the result of the greater restrictions offenders placed on GPS receive, and the
increased ability of PAs to detect violations. The higher levels of supervision discussed in section B above
may explain these disparities. Alternatively, offenders placed on GPS are significantly less likely to be
arrested overall, especially for violent offenses. This is an important finding confirming the bivariate
results and controlling for periods at risk of being arrested, other individual predictors of recidivism, and
the district into which offenders were paroled. Moreover, these are very conservative tests based on an
intent-to-treat model, where the experimental group was not always on GPS when they were at risk of
violating conditions of their parole or being rearrested.
D. COST ANALYSIS
This section performs a cost effectiveness (CE) analysis based on the findings above to ascertain which
program alternative (GPS monitoring supervision or traditional supervision) can achieve the most efficient
result (i.e., the most effective outcome at the lowest cost). The underlying assumption is that different
program alternatives are associated with different costs and different results. By choosing those with the
lowest cost for a given outcome, policymakers can use their resources more effectively (Levin and
McEwan 2001).
The basic technique of CE is to derive results for the effectiveness of each alternative by using standard
evaluation procedures (Rossi and Freeman 1985) and to combine such information with cost data
derived from the ingredients approach to provide a systematic way for evaluators to estimate costs of
social interventions (Levin 1983). The strength of this approach lies in its simplicity. Most important is
that it merely requires combining cost data with effectiveness data that are ordinarily available to create a
CE comparison. Further, it lends itself well to an evaluation of alternatives being considered. The major
disadvantage is that one can compare costs only among alternatives with similar goals. Fortunately, this
drawback does not have any bearing on the current study, as both alternatives focus on noncompliance
and recidivism.
The costs of an intervention are defined as the value of the resources dedicated to an intervention. These
are referred to as the ingredients of the intervention, and it is the social value of these ingredients that
Table 3.8. Multivariate Random-Intercept Complimentary Log-Log Models for Overall Arrests and
Arrests for Violent Offenses
ARRESTS
VIOLENT ARRESTS
VARIABLES
EXP β
SE
EXP β
SE
GPS
.842
*
.063
.675
*
.108
Moves
.960
.119
1.006
.256
Prior Drug Arrest
1.033
**
.008
1.005
.019
Prior Weapons Arrest
1.027
.022
1.060
.049
Log-likelihood: -2,939.22
Log-likelihood: -886.41
N = 784 over 24 months