Special Report
C alif o r n i a D e pa rtm ent of Corr ect ion s an d
Re ha b i l i tat i o n ’s I m pl em entat ion of t he
N o n- R e vo c a b l e Pa rol e Prog ram

Office of the
Inspector General

S tat e o f C a l i f o r n i a
M ay 2 0 11

Bruce A. Manfross, Inspector General (A)

Office ofthe Inspector General

May 25,2011

Matthew 1. Cate, Secretary
California Department of Corrections and Rehabilitation
1515 S Street, Room 502 South
Sacramento, CA 95814
Dear Mr. Cate:
Enclosed is the Office of the Inspector General's special report of the California Department of
Corrections and Rehabilitation's (CDCR) implementation of the non-revocable parole program.
We conducted this review under the authority of California Penal Code section 6126, which
assigns the Office of the Inspector General responsibility for oversight of CDCR.
The special report concludes the following with respect to CDCR's implementation of the nonrevocable parole program: (1) the automated California Risk Assessment (CSRA) instrument
inaccurately assesses a number of offenders; (2) the automated CSRA instrument uses
incomplete conviction data; (3) the automated CSRA instrument inconsistently applies juvenile
data when calculating risk assessment scores; and (4) CDCR's initial policy regarding juveniles
convicted of serious or violent felonies was incorrect.
We would like to thank you and your staff for the cooperation extended my staff in completing
this special report. If you have any questions concerning this report, please contact me at (916)
830-3600.
Sincerely,

. YJY:x.1:~ 1/-;;./ c
o,.(2' )Bruce A. Monfross

Inspector General (A)

Enclosure

Edm/lnd G. Brown. Jr., Governor
P.O. Box 348780. SACRAMENTO. CALIFORNIA 95834-8780 PHONE (916) 830-3600 FAX (916) 928-4684

Contents
Executive Summary .................................................................................................................... 1
Introduction ................................................................................................................................. 5
Parameters of Review ............................................................................................................... 10
Results of Special Report ........................................................................................................... 11
Finding 1 .............................................................................................................................. 11
The Automated Assessment Tool Inaccurately Assesses a Number of Offenders
Finding 2............................................................................................................................... 17
The Automated Assessment Tool Uses Incomplete Conviction Data
Finding 3 .............................................................................................................................. 21
The Automated Assessment Tool Inconsistently Applies Juvenile Data when
Calculating Risk Assessment Scores
Finding 4............................................................................................................................... 23
CDCR’s Initial Policy Regarding Juveniles Convicted of Serious or Violent
Felonies was Incorrect
Recommendations ...................................................................................................................... 24
California Department of Corrections and Rehabilitation’s Response
to the Special Report .................................................................................................................. 25
The Office of the Inspector General’s Comments on the Response from the California
Department of Corrections and Rehabilitation............................................................................28

Executive Summary
On October 11, 2009, a major change occurred in California statutory parole requirements with
the passage of Senate Bill X3 18. This landmark legislation was intended to help alleviate the
endemic overcrowding within California prisons – and the numerous constitutional violations
and budgetary demands occasioned by such overcrowding – by providing a system whereby
non-violent parole offenders would not be returned to prison unless they were convicted of
another felony offense.
On April 4, 2011, additional significant legislation
was enacted with the passage of Assembly Bill
109. When funded, this legislation will ultimately
shift non-violent parolees from oversight by
the California Department of Corrections and
Rehabilitation (CDCR) to oversight by local
governmental agencies. Consequently, the nonrevocable parole program established by Senate
Bill X3 18 can now be viewed as an interim
measure that will be in place only until such
time as local governmental agencies assume
supervision over non-violent parolees.

Findings in Brief
The Office of the Inspector General
finds that:
•	The Automated CSRA Instrument
Inaccurately Assesses a Number of
Offenders
•	The Automated CSRA Instrument
Uses Incomplete Conviction Data
•	The Automated CSRA Instrument
Inconsistently Applies Juvenile Data
when Calculating Risk Assessment
Scores

Effective January 25, 2010, CDCR began placing
eligible convicted felons on non-revocable parole,
commonly referred to as NRP, in compliance
•	CDCR’s Initial Policy Regarding
with Penal Code section 3000.03. Before the
Juveniles Convicted of Serious or
law’s enactment, when these types of inmates
Violent Felonies was Incorrect
were released from California prisons, they were
typically subject to parole terms of one to three
years and were under some level of supervision
by CDCR. Since the enactment of Penal Code section 3000.03, however, paroled inmates who
meet certain criteria must be placed on non-revocable parole. Parolees on non-revocable parole
are not supervised. Moreover, unlike supervised parolees, they are not subject to arrest or reincarceration in prison for parole violations.
The California Department of Corrections and Rehabilitation has very little authority over
offenders once they are placed on non-revocable parole. In the interest of public safety, then,
and to comport with the legislative intent that only qualified, non-violent offenders be placed
on non-revocable parole, the screening process used to determine an inmate’s eligibility
for non-revocable parole must be accurate. The screening process for non-revocable parole
excludes the following inmates and parolees: registered sex offenders; offenders with current
or prior serious, violent or sexually violent felony convictions; offenders who are known
prison gang members; and other offenders determined to have a high risk to reoffend. To
determine an inmate’s risk of reoffending, CDCR has developed a validated risk assessment

State of California • May 2011	

Page 1

instrument referred to as the California Static Risk Assessment (CSRA).1 However, flaws in the
CSRA’s implementation have resulted in flawed assessments. The CSRA has understated some
offenders’ risk of reoffending; some of these high-risk offenders have been placed on nonrevocable parole.
The Office of the Inspector General (OIG) estimates that approximately 23.5 percent of the
offenders assessed for possible placement on non-revocable parole between January and July
2010 were scored inaccurately, and that approximately 15 percent of the more than 10,000
offenders placed on non-revocable parole were inappropriately placed on non-revocable parole
during that same time period.2 Over 450 of these ineligible offenders carry a high risk for
violence, and some of these ineligible offenders may have already been discharged from nonrevocable parole after completing 12 months of parole, thereby precluding CDCR from taking
action to correct the parolee’s inappropriate placement on non-revocable parole. It should be
noted, however, that CDCR reports it has improved scoring tables used in the automated scoring
process and that these corrections would have reduced the error rate from approximately 23.5
percent to approximately eight percent if these corrections had been in place before July 2010.
It appears that CDCR’s error rate in assessing offenders for non-revocable parole results from
several factors. First, the computer-generated automated scoring system does not take into
account information from CDCR’s Offender Based Information System (OBIS) database,
which contains information related to prior supervision violations. Although CDCR’s manual
scoring team uses OBIS information for scoring purposes, and although CDCR has recognized
the need for the automated scoring system to eventually use data from the OBIS system, the
automated scoring system currently does not do so. Second, as discussed below, the automated
scoring system, unlike the manual scoring team, is not designed to take into account certain
data concerning juvenile offenses. Further, some offenders whose CSRA score should have
been completed manually because multiple sources had reported criminal history information
on the offender were for unknown reasons scored instead by the automated scoring system.
During our review, we also discovered that the primary database that CDCR relies upon when
calculating inmate scores for possible placement on non-revocable parole provides incomplete
data. This database, the Automated Criminal History System (ACHS), is maintained by the
California Department of Justice and is the most complete relevant data source currently
available. Almost half of the 16.4 million arrests reported in the Department of Justice’s database
between 2000 and 2009, however, lack documented outcomes, such as convictions or other
dispositions. Incomplete conviction history data reduces the assessment tool’s accuracy in
predicting an offender’s likelihood of reoffending. In addition, untimely entered conviction data
reduces the CSRA tool’s accuracy in assessing the offender’s eligibility for placement on non1 Although it is anticipated that the enactment of Assembly Bill 109 will eventually result in the non-revocable
parole program being phased out as local governmental agencies assume oversight responsibility for non-violent
offenders, CSRA scores will still be used for Correctional Offender Management Profiling for Alternative
Sanctions (COMPAS) and in the Parole Violation Decision Making Instrument (PVDMI) purposes. Consequently,
it will remain important for CSRA scoring to be as accurate as possible.
2 These figures reflect only those offenders with computer-generated automated scores. There were approximately
2,300 offenders also approved for non-revocable parole based on a manually calculated CSRA score.
Bureau of Investigations, Office of the Inspector General	

Page 2

revocable parole.3 Despite these significant shortcomings in the ACHS database, CDCR had no
choice but to use that database, however inadequate, as the best available source of data.
We also determined that the CSRA inconsistently uses data concerning juvenile offenses.
Because CSRA developers were concerned about using incomplete or unreliable data for
juvenile offenders, they intended to exclude juvenile offender criminal history data in their
construction of the CSRA. However, the automated scoring process was implemented in such a
way as to include some – but not all – juvenile offenses in the risk assessment scoring process.
In addition to inconsistently including juvenile convictions in the automated scoring process,
CDCR initially incorrectly issued a policy that ignored the juvenile records of adult offenders
who, when they were minors, were tried as adults and convicted of serious and violent felonies.
The unintended result of the incorrect policy is that some adult offenders with histories of serious
and violent felonies could be placed on non-revocable parole. Although CDCR has revised the
policy, our interviews with field staff indicate that at least some CDCR staff members continue to
follow the incorrect policy.

3 Individual agencies, not the Department of Justice, are ultimately responsible for entering data into the
ACHS, and the agencies are not uniform in their manner of entering data into the system, or in their timeframe for
entering that data. As a result, it is possible that an offender’s conviction data may not be entered into the system
until years after the fact, and after the offender was scored for non-revocable parole purposes. Such untimely
entered conviction data, which might render the offender ineligible for non-revocable parole, would not be
considered for scoring purposes.
State of California • May 2011	

Page 3

Recommendations
In this special report, the Office of the Inspector General identifies deficiencies in CDCR’s
implementation of non-revocable parole. To address the issues identified in this special report,
we recommend that CDCR take the following actions:
•	 Develop a quality control program for its CSRA scoring to ensure consistency,
accuracy and timely reviews.
•	 Collaborate with the California Department of Justice and the California Judicial
Council to obtain more complete criminal history information.
•	 Review the CSRA implementation to determine whether juvenile offender criminal
history is considered appropriately.
•	 Ensure that CDCR staff is trained according to the revised policy regarding the
inclusion of juvenile data for serious and violent felonies.
•	 Review CSRA scores of offenders already placed on non-revocable parole to verify
that they were accurately assessed.
•	 Retain criminal history records used to determine eligibility for non-revocable parole.

Bureau of Investigations, Office of the Inspector General	

Page 4

Introduction
Historically, the California Department of Corrections and Rehabilitation (CDCR) has
supervised parolees who are subject to particular rules, known as conditions of parole, upon
their release from prison. Conditions of parole are written requirements that parolees must
follow, such as meeting with their parole agents on a regular basis, contacting their parole
agents regarding changes in employment or residence, obeying all laws, and not possessing
weapons. Although no longer incarcerated, parolees remain under CDCR’s supervision until
they have completed their term of parole. Parolees who violate their parole conditions can have
their parole status revoked and be returned to prison even though they have not been convicted
of any new criminal charges. On October 11, 2009, a major change in parolee statutory
requirements occurred with the passage of Senate Bill X3 18, which resulted in the enactment
of California Penal Code section 3000.03. The new law mandates CDCR to place parolees on
non-revocable parole, effective January 25, 2010.
Under this Penal Code section, offenders who are already entitled to parole and who meet
the criteria for non-revocable parole are released into the community. They are not subject to
parole holds, supervision by a parole agent, or any reporting of parole violations to the Board
of Parole Hearings. For the duration of their parole period, parolees on non-revocable parole
remain subject to search by any law enforcement officer. However, they cannot be returned to
prison for parole violations. Like any other citizen, they must be convicted of a felony criminal
offense and receive a new sentence in order to be sent back to prison.
According to CDCR, implementation of the nonrevocable parole program reduces prison overcrowding
and focuses parole supervision on paroled sex offenders
and high-risk parolees with serious and violent
commitment histories. Moreover, CDCR asserts that the
new program will give parole staff additional time to
work more closely with local law enforcement agencies,
an improvement which should enhance overall public
safety. As such, it is clear that non-revocable parole is
designed to serve a compelling public purpose.4

(Offenders on
non-revocable parole) will
not be returned to prison for
parole violations. They must
be found guilty of a crime
and receive a new sentence
to return to prison.

During CDCR’s implementation of non-revocable
parole, however, problems arose. In April 2010, CDCR
reported that it had incorrectly placed over 600 offenders
on non-revocable parole. Public officials expressed concern as they became aware of offenders
on non-revocable parole who had either previously committed significant crimes or committed
additional crimes after their placement on non-revocable parole. Requests from the Legislature
4 It is also important to note that Assembly Bill 109 was enacted on April 4, 2011. When funded, this legislation
will, among other things, ultimately remove non-violent parolees from oversight by CDCR to oversight by local
governmental agencies. Consequently, the non-revocable parole program established by Senate Bill X3 18 can
now be viewed as an interim measure that will be in place only until such time as local governmental agencies
assume supervision over non-violent offenders.
State of California • May 2011	

Page 5

to investigate the matter, as well as our concern for public safety, prompted the Office of
the Inspector General to inquire about CDCR’s application of the eligibility criteria for nonrevocable parole and inquire specifically about the development of a validated assessment
instrument as required by law. We not only reviewed the assessment instrument development
process but also tested the accuracy of CDCR’s automated assessment system.
Eligibility criteria for non-revocable parole
In order to qualify for non-revocable parole, an offender must meet the eligibility criteria as
described in Figure 1. To determine an offender’s likelihood to reoffend, CDCR developed,
with assistance from the Center for Evidence-Based Corrections at the University of California,
Irvine, as well as from the Washington State Institute for Public Policy, a risk assessment
instrument that is patterned after a risk assessment instrument developed for Washington. The
Washington model was particularly appealing to CDCR because it uses static risk indicators,
such as gender, date of birth, and criminal history.
To replicate the Washington predictive instrument, CDCR applied Washington’s methods to
data related to California offenders. Because the California Department of Justice advised
CDCR that the criminal history records for California juveniles were unreliable and only
sporadically recorded, CDCR excluded these factors from consideration as it developed its own
predictive instrument. Consequently, CDCR performs its assessment of inmates by using 22
Figure 1: Eligibility criteria for non-revocable parole.
Ineligible for NRP
Place offender on
supervised parole.

Is the offender eligible for non-revocable parole (NRP)?
Required to register as a sex offender?

NO



Committed to prison for or have a prior conviction for a serious or
violent felony?

NO



Committed to prison for or have a prior conviction for a sexually
violent offense?

NO



Guilty of a serious disciplinary offense during the current prison term?

NO



A validated prison gang member or associate?

NO



Refused to sign any written notification of parole requirements or
conditions?

NO



Determined to pose a high risk to reoffend (based on CRSA score)?

NO



YES 
YES 
YES 
YES 
YES 
YES 
YES 

Eligible for NRP
Place offender on unsupervised parole.

Bureau of Investigations, Office of the Inspector General	

Page 6

factors, or static indicators, rather than using the 26 factors employed in the Washington model,
which is more fully discussed later in this report.
The resulting assessment instrument, called the California Static Risk Assessment (CSRA),
uses the 22 static indicators to measure the risk or probability that an offender will reoffend
within three years of his or her release from prison. The CSRA was developed to measure the
risk of reoffending in two ways: it measures the likelihood of a felony arrest within three years
of the inmate’s release into the community and the likelihood of a felony conviction within
three years of the inmate’s release into the community.
These static indicators are items of data taken from the offender’s demographic descriptors,
commitments to prison, adult felony and misdemeanor records, and sentence and supervision
violations. Based on numerical values assigned to these static indicators, a weighted scoring
algorithm calculates the offender’s risk of reoffending and assigns the offender to the
corresponding risk group. The risk of reoffending is assigned a score, as follows:
	
	
	
	

1 – Low risk to reoffend
2 – Moderate risk to reoffend
3 – High risk to reoffend, and in particular, to commit drug offenses
4 – High risk to reoffend, and in particular, to commit property offenses
5 – High risk to reoffend, and in particular, to commit violent offenses

Indicating a low risk to reoffend, a score of “1” predicts that 48 percent of offenders in this
group will be arrested for a felony, or that 18 percent of offenders in this group will be
convicted of a felony within three years of being released into the community.
Indicating a moderate risk to reoffend, a score of “2” predicts that 69 percent of offenders in
this group will be arrested for a felony, or that 31 percent of offenders in this group will be
convicted of a felony within three years of being released into the community. Arrest rates
are higher than conviction rates because not all offenders who are arrested are convicted, for
various reasons.
Only offenders assessed at low and moderate risk levels are eligible for non-revocable parole.
Offenders assessed at high risk levels–scores “3” (High Drug), “4” (High Property) and “5”
(High Violent)—are not eligible for non-revocable parole because they are considered highly
likely to reoffend.
The development of the CSRA instrument included a validation process to ensure that it would
produce the same results when applied to offenders beyond the group of offenders on which it
was tested. Based on the results of this validation process, researchers determined the CSRA’s
predictive accuracy to be “moderately predictive” for future arrest and “weak” for future
conviction, and concluded the CSRA was a valid risk assessment instrument.

State of California • May 2011	

Page 7

To implement the non-revocable parole program, in January 2010 CDCR formed two teams of
experienced parole agents to determine the non-revocable parole eligibility for approximately
24,000 pre-screened parolees among the more than 110,000 supervised parolees. In addition, the
two parole teams were tasked with screening inmates scheduled for parole up to April 1, 2010,
when eligibility screening would be incorporated into CDCR’s release program.
The two parole teams’ screening process included reviewing inmate central files, CSRA scores,
current criminal identification and information reports, Federal Bureau of Investigation Record
of Arrest and Prosecution sheets (for multi-state records), Penal Code section 290 registration
requirements or eligible sexually violent offenses, prison disciplinary histories, and prison gang
identification forms that document possible validated prison gang membership or association.
Based on the results of this review, eligible inmates and parolees were placed on non-revocable
parole.
After installing upgrades to the CSRA computer program on March 29, 2010, CDCR reported
that over 600 offenders were actually at a higher risk to reoffend than originally projected.
According to CDCR, several factors contributed to the updated assessment tool’s increased
accuracy at calculating scores, including the following:
•	 The code-mapping table was updated to list additional charges that can be extracted
from criminal histories.
•	 The algorithm that extracts convictions from criminal histories was revised to also
extract any conviction data contained in the comments sections of criminal histories.
•	 The California Department of Justice started recording additional data on crimes that
can be classified as either misdemeanors or felonies.
However, CDCR did not uniformly implement the instrument upgrade at the several locations
where it maintained CSRA scores. Some non-revocable parole assessments, therefore,
continued to be based on old scores rather than on scores calculated after the March 29 update.
This problem has since been remedied by eliminating the locations that contained outdated
information.
Of the 600 offenders identified as erroneously classified after the March 29 update,
approximately 400 had been released to non-revocable parole and were subsequently reclassified
as requiring supervised parole. The remaining offenders were scheduled for non-revocable parole
but had their status changed to supervised parole prior to their release.
In May 2010, CDCR determined that it had erroneously placed an additional 77 offenders on
non-revocable parole and sought to return them to supervised parole. This error was attributed
to a modification in the CSRA instrument on May 12, 2010, as well as to a failure to promptly
update scores when manual calculations were necessary. These errors have been remedied by
CDCR.
The initial screening teams of experienced parole staff have all returned to their normal duties.
Screening for non-revocable parole is now conducted by each prison as part of the process of
Bureau of Investigations, Office of the Inspector General	

Page 8

placing offenders on parole. A small parole staff is
available to handle non-revocable parole screening for
offenders placed directly on parole from the courts.
Using the CSRA instrument as the mandated risk
assessment tool, CDCR had 14,859 offenders on nonrevocable parole as of December 9, 2010.

“If offenders have not
been convicted of any
offenses listed under these
three sections, and if they
meet the other eligibility
requirements listed in
the non-revocable parole
statute, they are eligible for
non-revocable parole.”

A further concern with the implementation of nonrevocable parole arose when a law enforcement official
from a major metropolitan area and a member of
the California Legislature expressed concern about
CDCR’s inclusion of offenders whose prior crimes
appeared to be serious or violent felonies. However,
certain crimes that may seem serious or violent are
not considered so under state law. Eligibility for
non-revocable parole includes the requirement that
offenders cannot have been convicted of a serious or
violent felony. Serious or violent felonies are specified
in the non-revocable parole statute as those defined in Penal Code sections 1192.7, 1192.8 and
667.5. If offenders have not committed any offenses listed under these three sections, and if
they meet the other eligibility requirements listed in the non-revocable parole statute, they are
eligible for non-revocable parole.

State of California • May 2011	

Page 9

Parameters of Review
To develop the information contained in this special report, the Office of the Inspector General
completed the following activities:
•	 Interviewed various CDCR staff members and reviewed reports from CDCR’s Office
of Research, Division of Adult Parole Operations, Division of Juvenile Justice and
Office of Legal Policy.
•	 Interviewed staff and reviewed reports from the Center for Evidence-Based
Corrections, University of California, Irvine.
•	 Contracted with a statistics expert from California State University, Sacramento, to
review CDCR’s development of the CSRA instrument and validate the OIG sampling
methods.
•	 Interviewed staff and reviewed data from the California Department of Justice.
•	 Conducted an assessment of the automated and manual CSRA processes.
•	 Consulted with an expert affiliated with the Washington study.

Bureau of Investigations, Office of the Inspector General	

Page 10

Results of Special Report
Finding 1
The Automated Assessment Tool Inaccurately Assesses a Number of
Offenders
Each week, after receiving updated electronic criminal history data from the California
Department of Justice, CDCR calculates the California Static Risk Assessment (CSRA) scores
for all offenders. Since there are approximately 165,000 offenders in CDCR custody and
approximately 108,000 parolees, CDCR has over 273,000 risk assessment scores to evaluate
weekly, with about four percent of offenders flagged for manual scoring. These CSRA scores,
which project an offender’s likelihood of reoffending within three years after being released
from prison, are not only used to place offenders on non-revocable parole but are also a
component in CDCR’s decisions about offenders’ programs, placements and supervised parole.
CSRA scores of 1 to 5 measure the risk or probability of reoffending as follows:
	
1 – Low risk to reoffend
	
2 – Moderate risk to reoffend
	
3 – High risk to reoffend, and in particular, to commit drug offenses
	
4 – High risk to reoffend, and in particular, to commit property offenses
	
5 – High risk to reoffend, and in particular, to commit violent offenses
As described in Figure 2, CDCR uses an algorithm developed by the University of California,
Irvine, to produce a risk assessment score. The algorithm searches the electronic criminal
history data of each offender for convictions, which are identified as “score-able events.” This
data is then fed into a CSRA calculator. The resulting score for each offender is made available
to staff through a CSRA link on CDCR’s computer network.
About 96 percent of offenders’ risk assessment scores can be calculated through the automated
system. Data from the remaining four percent (approximately 10,000 offenders) must be
manually reviewed for reasons that include evaluating criminal histories that come from
Figure 2: Overview of CDCR’s workflow for generating CSRA scores.

DOJ data for…
165,000
in custody

Quality
control
process
for manual
scoring

UC Irvine
algorithm

108,000
on parole

CSRA
calculator

Quality
control
process for
automatic
scoring

Individual
CSRA
scores

CDCR
decision
makers
statewide

Missing element
State of California • May 2011	

Page 11

sources outside California, such as other states or the Federal Bureau of Investigation. The task
of manually reviewing data is assigned to a specially trained and experienced team, which is
separate from the team performing the automated scoring. Among its other duties, CDCR’s
manual scoring unit is tasked with performing quality control oversight of automated scoring.
The automated CSRA scoring process produces a number of inaccurate risk assessments
According to CDCR, it operates an ongoing program that verifies and authenticates the automated
CSRA calculations by comparing the automated scoring with manual scoring. However, there is
no formal policy specifying the frequency of the comparisons, so comparisons are performed at
the discretion of the manager. Even so, CDCR was unable to produce any documentation that this
process had ever been done. Noting that the automated CSRA system was a work in progress,
CDCR explained that it continually receives questions and comments from the manual scoring
team, and that many of these inquiries have resulted in instrument changes to the automated
CSRA algorithm. As described in detail in this section, we note that the manual CSRA scoring
team has access to more information than the automated CSRA process.
To evaluate the effectiveness and error rate of the CSRA scoring process, we conducted a test
of the automated CSRA scores. We were provided with access to the manual CSRA scoring
input and assessment calculator, and our staff was trained by CDCR’s manual CSRA scoring
unit. We compiled a list of the 10,134 auto-scored offenders on non-revocable parole in July
2010. From this list of auto-scored offenders, we selected a statistically valid random sample
of 200 offenders and retrieved their criminal histories from the Department of Justice. Our
sample, verified by a statistics expert at California State University, Sacramento, permits us
to make statistically valid projections regarding the statewide population of offenders on nonrevocable parole. The confidence level for our samples is 95 percent with a margin of error of
plus or minus 5.83 percent. We then manually calculated the CSRA score of each offender’s
file in our sample, using CDCR’s manual scoring process.
We determined that CDCR’s automated scoring system produced inaccurate risk assessment
scores for 47 of the 200 offenders, or 23.5 percent of our sample. To confirm our assessments,
we gave CDCR’s manual scoring team, which is tasked with quality control for automated
CSRA scores, approximately one-third of the files from our sample––including all 47
erroneously scored files––and asked them to score the files independently. Our manual scoring
team and CDCR’s manual scoring team agreed on 67 of the 68 files manually reviewed by each
team. As noted in Figure 3 and corroborated by CDCR’s manual scoring team, 15.5 percent of
the offenders sampled are ineligible for non-revocable parole because they are highly likely to
reoffend. An additional eight percent of the offenders’ files had a score error, but the error did not
affect the offenders’ non-revocable parole eligibility.5
When we made our sample selection in July 2010, there were 10,134 auto-scored offenders
5 There were an additional seven offenders in the auto-scored sample that we manually scored as “2” (moderate
risk) and were therefore appropriately placed on non-revocable parole by CDCR. However, these seven offenders
were discharged from non-revocable parole before our review, and CDCR was unable to produce their original
CSRA scores. Therefore, these seven offenders were excluded from the score error calculation because we did not
have an automated CSRA score with which to compare.
Bureau of Investigations, Office of the Inspector General	

Page 12

Figure 3: Projected error rates based on OIG auto-scored sample of 200 files.
OIG random sample
of NRP files
auto-scored July 2010
200

Total NRP files
auto-scored July 2010
10,134

Number with score
errors

47 (23.5%)

2,381†

Should not be NRP

31 (15.5%)

1,571†

CSRA Score
High Drug (3)

13 (6.5%)

659

CSRA Score:
High Property (4)

9 (4.5%)

456

CSRA Score:
High Violent (5)

9 (4.5%)

456

These are our best estimates of the errantly-scored parolees on non-revocable parole and those who should
not be on non-revocable parole, based on the results of our evaluation of a statistically valid sample. Applying the
margin of error of 5.83 percent and 4.98 percent accordingly, the number of errantly-scored parolees could be as
high as 2,972 or as low as 1,791, and the number of parolees inappropriately placed on non-revocable parole could
be as high as 2,075 or as low as 1,066.

†

on non-revocable parole. Projecting the error rates from our statistically valid sample to the
total population, we found that there were an estimated 1,571 auto-scored offenders on nonrevocable parole in July 2010 who were not eligible for non-revocable parole because their
actual risk assessment scores identify them as highly likely to reoffend.
The cause of the error rate is central to the CSRA tool implementation. Unlike errors that may
occur during the manual scoring process, which can be attributed to staff errors in interpreting
or entering data into the CSRA, errors that occur during the automated scoring process should
be traceable to a specific automated procedure or algorithm in the CSRA auto scoring process.
The information we developed does not indicate that errors in the automated scoring process
are uncorrectable. In fact, of the 47 scoring errors found in the OIG sample of 200, CDCR has
acknowledged that 31 of the errors were directly related to the CSRA tool’s mapping table.
According to CDCR staff, CDCR updated the tool’s mapping table to resolve this problem on
July 1, 2010. Reportedly, this update would have reduced the error rate in our sample from 23.5
percent to approximately eight percent.6
6 CDCR did not reassess the offenders it had previously placed on non-revocable parole––as it did after
improvements it made in March and May 2010. As discussed in the Introduction of this report, CDCR sought
to recall 400 and 77 offenders, respectively, from non-revocable parole after it assessed the impact of its system
changes on previously released offenders. Based on the results of our statistically valid sample, we estimate that
had CDCR reassessed the offenders it had released to non-revocable parole prior to its July 2010 system change,
CDCR would have discovered that it had inappropriately released approximately 811 additional offenders to nonrevocable parole.
State of California • May 2011	

Page 13

We are not in agreement with CDCR on the cause of the remaining 16 errors; the Office of the
Inspector General attributes the remaining 16 errors to the following causes:
•	 Eleven errors relate to supervision violations recorded in CDCR’s Offender Based
Information System (OBIS) database, to which the CSRA automated scoring process
does not have access.
•	 Two errors relate to scoring events found in multi-source out-of-state criminal
histories.
•	 Two errors relate to juvenile convictions.
•	 One error relates to counting a CDCR custody event as a felony when it appears to
actually be a supervision violation.
Identification of the causes of these 47 errors further reinforces the need for CDCR to develop an
ongoing quality control process for timely assurance that the CSRA tool is working as intended.
Regarding the 11 errors relating to supervision violations, according to CDCR staff, the
automated CSRA tool was designed to only consider offender information contained in the
California Department of Justice’s Automated Criminal History System (ACHS). This design
excludes from consideration important risk prediction information contained in other systems,
including CDCR’s own systems. For example, CDCR maintains information in two of its
computer systems on probation and parole violations––which the CSRA tool refers to as
supervision violations. Supervision violations are one of the 22 factors the CSRA considers
to determine offenders’ risk of reoffending. In fact, after age and gender, it is the most heavily
weighted remaining factor in the CSRA’s felony scale. Acknowledging this deficit in its
evaluation of the CSRA completed in November 2009, CDCR described a next step in the
development of the CSRA:
Investigating the use of parole violations that do not appear on automated “rap sheets,”
but are recorded on CDCR automated systems. This would include violations that do
not result in a criminal arrest or conviction, but may be important markers for risk
prediction.
According to CDCR, it has not yet completed this step, despite the fact that the manual scoring
team has access to and considers such information when calculating an offender’s score.
Our review indicates that the inclusion of this important risk prediction information would
significantly reduce the error rate we discovered in the automated scoring process.
Conviction data is dynamic and subject to delays in reporting
During our review, we became aware of the dynamic nature of criminal history reporting in
California “rap sheets” and multi-source criminal history data from other states. An offender’s
rap sheet is a report of the information contained in the ACHS. The Department of Justice
depends on the local jurisdictions to report convictions or probation violations for all offenders.
Local jurisdictions do not always report accurately, completely, or promptly. It is therefore
possible for an offender to have one or more events in his or her criminal history that have not
Bureau of Investigations, Office of the Inspector General	

Page 14

yet been reported to the Department of Justice in the ACHS. This means CDCR would not
have access to the unreported conviction or probation violation when calculating an offender’s
CSRA score. It is therefore possible for an offender to be deemed eligible for non-revocable
parole, but to soon thereafter have previously unrecorded convictions and supervision
violations reported to the Department of Justice, rendering the offender ineligible for placement
on non-revocable parole.
By design, CDCR does not re-check the CSRA scores of offenders after placement on nonrevocable parole. The CDCR Office of Research explained that once offenders have been
placed on non-revocable parole, their CSRA scores are locked and removed from the weekly
automated CSRA calculations done on all other inmates and supervised parolees. Again, if an
offender’s eligibility changes after release to non-revocable parole, CDCR has no mechanism
to be alerted or to recalculate the CSRA score.
During our preparation for this special report, we reviewed inmate central files and parolee
field files from our sample. We learned that CDCR does not always retain the criminal history
record used during their review to determine non-revocable parole eligibility. We consider
these criminal history records a valuable resource if eligibility for non-revocable parole is ever
questioned; accordingly, we advise that these criminal records be retained.
Some offenders erroneously placed on non-revocable parole may have already been
discharged from parole
From January 25, 2010, through October 27, 2010, over 8,000 offenders were discharged from
non-revocable parole. Penal Code section 3001 requires that certain paroled offenders who
meet the conditions of their parole supervision be discharged from parole within 30 days
following their twelfth month on parole. Since offenders on non-revocable parole have no
conditions of supervision to violate, this statute guarantees that offenders placed on nonrevocable parole will be discharged from parole within 13 months of their original parole date
unless they are sentenced to prison for committing a new felony offense.
Offenders on supervised parole before January 25, 2010, and then approved for non-revocable
parole became eligible for discharge one year from their original parole date, due to the
provisions of Penal Code section 3001. This means it is possible that some offenders may have
spent just a few weeks on non-revocable parole before they were eligible for discharge and
released from all parole conditions.
Based on the error rate found in our statistical review, we found it likely that a portion of these
over 8,000 offenders were ineligible for non-revocable parole and should have been placed
on supervised parole. In October 2010, CDCR reported that almost half of all supervised
parolees—47 percent—are returned to prison for parole violations. It is therefore probable that
some of the discharged parolees inappropriately placed on non-revocable parole would have
violated their parole conditions and returned to prison, had they been on supervised parole.

State of California • May 2011	

Page 15

CDCR’s use of the CSRA instrument for other purposes could also affect the validity of
inmate placement in other programs
Because CDCR uses an offender’s CSRA score for purposes beyond determining eligibility for
non-revocable parole, the deficiencies identified in the CSRA automated scoring process could
have unforeseen negative impacts in other CDCR operations. Two prominent applications
of CSRA scores occur in the Correctional Offender Management Profiling for Alternative
Sanctions (COMPAS) risk and needs assessment tool and in the Parole Violation Decision
Making Instrument (PVDMI). Among other purposes, CDCR uses COMPAS to decide which
programs to provide to incarcerated offenders. The CSRA instrument has been integrated into
the COMPAS risk and needs assessment tool. The PVDMI assesses a supervised parolee’s risk
for recidivism using the CSRA score in conjunction with the severity of the parole violation
(based on a severity index) to determine an appropriate and proportionate response to the
parole violation, such as returning the offender to prison or imposing alternative sanctions. The
extent of the negative impact of the CSRA automated scoring deficiencies on the COMPAS and
PVDMI outcomes is as yet unknown.

Bureau of Investigations, Office of the Inspector General	

Page 16

Finding 2
The Automated Assessment Tool Uses Incomplete Conviction Data
The automated CSRA instrument uses an offender’s felony and misdemeanor conviction
record, among other factors, to determine the offender’s risk to reoffend. That risk is described
as a score, which CDCR uses to determine whether the offender is eligible to be considered for
non-revocable parole. The risk assessment score is also a factor in determining the offender’s
placement on supervised parole and in various CDCR programs.
The conviction data that forms the basis of the risk assessment score comes from the California
Department of Justice’s Automated Criminal History System (ACHS) database. This database
is a compilation of arrest, conviction, and custody data for California offenders and is the
only source of such data currently available to CDCR to complete a risk assessment. Arrest
information is entered into the ACHS database by over 600 law enforcement agencies
statewide and by courts in the 58 counties. Conviction and dismissal information is entered into
the ACHS database by the respective California courts that adjudicated an offender’s charges.
Each week, CDCR sends the Department of Justice a list of all offenders incarcerated in prison
and on parole. The Department of Justice then transfers the corresponding ACHS arrest and
conviction data to CDCR, as shown in Figure 4.
Figure 4: Overview of agencies reporting to DOJ’s ACHS.
Charges are reported at arrest:
County law
enforcement
agencies
Federal law
enforcement
agencies
Other charging
law enforcement
agencies

Dispositions are reported as cases progress:
Arresting law
enforcement agency
drops charges

City law enforcement
agencies

Consistent
arrest
reporting
process

Inconsistent
disposition
reporting
process

ACHS
DOJ
data …

State of California • May 2011	

Prosecuting agency
decides against
prosecuting
Prosecuting agency
prosecutes charges
Court adjudicates
charges

Database compilation
of arrest, conviction
and custody data for
California offenders

… forwarded
to CDCR

Page 17

Conviction data is incomplete
Arrest information is normally entered into the ACHS database when an offender is fingerprinted
using the Department of Justice’s LiveScan system. Because fingerprints are used to identify
arrested offenders, the Department of Justice is confident that arrest information in ACHS is
relatively complete, meaning that the identity of everyone who is arrested is likely entered into
the system. The accuracy of the data, however, such as the specific charges in an arrest, cannot
be confirmed. The Department of Justice relies on local agencies, which retain the actual arrest
documents, to enter the data into the database; accordingly, the Department of Justice cannot
confirm the data’s accuracy.
The Department of Justice is less confident in the ACHS data’s accuracy and completeness
regarding arrest dispositions. An arrest is generally disposed of in one of three ways: (1)
the arresting law enforcement agency decides to drop the charges; (2) the arresting law
enforcement agency forwards charges to the prosecuting agency, which decides against
prosecuting the charges; or (3) the prosecuting agency prosecutes the charges and the court
adjudicates the charges. Unlike arrest information, however, disposition information is not
collected by LiveScan. Instead, the Department of Justice relies on each of the criminal
justice agencies included in each of the above scenarios to report the disposition of arrests.
Unfortunately, this reporting oftentimes does not occur.
According to the Department of Justice, California law enforcement agencies reported more
than 16.4 million arrests between 2000 and 2009. However, criminal justice agencies reported
dispositions for only 8.7 million of the arrests––and failed to report dispositions on 7.7 million
arrests (47 percent). Included in these 16.4 million
arrests are 7.3 million felony arrests, for which 3.3
The Department of Justice
million (45 percent) have no reported disposition. The
Department of Justice ACHS data, therefore, does not
ACHS data … does not
contain the outcome of almost half of all arrests in
California. The implications for non-revocable parole
contain the resulting
are plain: more recorded convictions would result
outcome of almost half of
in increased accuracy of risk assessment scores and
fewer high-risk offenders erroneously placed on nonall arrests in California.
revocable parole.
In addition to having incomplete records of the arrest
dispositions, the Department of Justice is not always able to match the disposition records it has
with the corresponding arrest records. The Department of Justice advised us that it currently has
approximately 1.9 million dispositions for which it cannot identify the corresponding offender
arrest records. As a result, these dispositions are not included in the ACHS database. This
omission further weakens the incomplete ACHS conviction data reported to CDCR for its use in
the automated CSRA instrument.

Bureau of Investigations, Office of the Inspector General	

Page 18

Current alternatives are limited
Although the ACHS data in the automated CSRA instrument is incomplete, CDCR has no viable
alternative sources to provide more complete information. Currently, the Department of Justice’s
ACHS database is the only central repository of California conviction information. Opportunities
may be on the horizon, however, for CDCR to obtain more complete conviction information.
The California Judicial Council, which functions as the administrative arm of the California
judiciary, has begun an effort to establish a statewide information system that, among other
activities, would capture information on cases adjudicated by all 58 of the California superior
courts. This new information system currently operates in only a few California counties. The
Judicial Council reports that in the future, all of its trial courts will be brought into the new
information system. However, the council added that due to budgetary constraints, further
implementation of the information system is currently on hold.
The Judicial Council’s new information system may present an opportunity for CDCR to obtain
more complete conviction information for its automated CSRA instrument. This information
would likely be more complete because it would come directly from the adjudicating court rather
than through an intermediary (DOJ), as is the case with the ACHS database.
More complete conviction information would improve CDCR’s ability to predict recidivism
Obtaining more complete offender conviction data would increase the effectiveness of the
CSRA in predicting the likelihood that a parolee will reoffend after his or her release from
prison. Both our statistical expert and the researcher who oversaw the development of
Washington’s predictive instrument, which CDCR used as a model for the CSRA, agreed that
better data would increase the predictive accuracy of the CSRA.
The measure of the predictive accuracy used by the Washington researchers is called the area
under the receiver operating characteristic (AUC). The Washington researchers provided the
following description of the AUC:
The best measure for determining how accurately a score predicts an event like
recidivism is a statistic called the area under the receiver operating characteristic
(AUC). The AUC ranges from .500 to 1.000. This statistic is .500 when there is no
association and 1.000 when there is perfect association. AUCs in the .500s indicate
little or no predictive accuracy, .600s weak, .700s moderate, and above .800 strong
predictive accuracy.

State of California • May 2011	

Page 19

As shown in Figure 5, CDCR’s version of the Washington instrument is based on subsequent
felony convictions and produced AUC scores ranging from .650 to .681. These scores indicate
that the model is weak in predicting whether an offender will be convicted of a felony within
the three years after being released. The AUC scores for future arrests ranged from .673
to .704, which is weak to moderate in predicting future arrests. Both of these measures are
weaker than the results Washington obtained in its predictive tool. That Washington’s accuracy
rate is higher than California’s accuracy rate may be due in part to Washington’s smaller
and less diverse population and cleaner data. Other states’ instruments are more typical of
the CSRA, like the Wisconsin Risk Assessment Instrument, which has an AUC of .660. Our
statistics expert from California State University, Sacramento, concluded that the CSRA is a
useful and helpful instrument to predict future recidivism. However, it could be improved. One
reason the California AUC scores are lower than Washington’s is that the criminal history data
that California uses is incomplete. Both our statistical expert and the researcher who oversaw
the development of Washington’s predictive instrument believe that CDCR could increase
the predictive accuracy of the CSRA if it had access to and was able to use more complete
conviction data.
Figure 5: The predictive accuracy of the CSRA instrument is lower than the accuracy of the Washington
instrument.
.800

	

.750

.700 - .800: Moderately Predictive Range

.757

.756

.745
Less than .700:
Weak Predictive Range

.700

.704
.675

.650

.670

.673

.681

.650

.600

	

Any Felony	

Washington Initial Sample

Property/Violent Felony	

Violent Felony

CSRA Initial Sample (Arrest)
CSRA Initial Sample (Conviction)

Bureau of Investigations, Office of the Inspector General	

Page 20

Finding 3
The Automated Assessment Tool Inconsistently Applies Juvenile Data
when Calculating Risk Assessment Scores
When it developed the CSRA instrument, CDCR patterned it after the Washington instrument,
which was based on offender demographics and criminal history. However, unlike the
developers of the Washington instrument, CDCR did not include separate juvenile data
categories in the CSRA. Of the 26 items in the Washington instrument, as seen in Figure 6,
four relate to juvenile offender criminal history. The CSRA developers chose not to include
juvenile offender criminal history because, as they explained in their CSRA working paper, the
California Department of Justice felt the data would be “unreliable and sporadically recorded.”
As a result, the CSRA instrument contains 22 rather than 26 items in the assessment.
According to CDCR, the CSRA algorithm ignores juvenile adjudications by filtering out the
Welfare and Institutions Code section 602 prefix, which assigns jurisdiction for most juvenile
offenses to the juvenile courts. However, we note that some juvenile offender convictions
do not have a Welfare and Institutions Code section 602 prefix and are, therefore, used
during the scoring process to calculate offenders’ risk assessment scores. Even though the
CSRA instrument was developed to ignore juvenile data, we have determined that the CSRA
instrument does, in fact, capture some juvenile data while ignoring other juvenile data when
calculating risk assessment scores.
We do not know how this juvenile data affects the risk assessment scores for the entire
CDCR offender population, but we believe that CDCR should investigate its impact. In the
200 randomly sampled auto-scored CSRA files, we identified six offenders who had juvenile
adjudication data that counted towards their risk assessment scores. For example, we found in
an adult offender’s criminal history a juvenile adjudication for a felony Health and Safety Code
violation relating to the possession of a controlled substance. The CSRA instrument handles
this juvenile adjudication as an adult event, but the Washington model treats this separately as
a juvenile occurrence. This difference could be significant because juvenile adjudications in the
Washington study impact recidivism predictions differently from adult convictions.

State of California • May 2011	

Page 21

Figure 6: Comparison of CSRA instrument items and Washington instrument items.
CSRA Instrument Items

Washington Instrument Items
Demographics

Age at time of release

Age at time of current sentence

Gender

Gender
Juvenile Record

Ignored

Felony convictions

Ignored

Non-sex violent felony convictions

Ignored

Felony sex convictions

Ignored

Commitments to state juvenile institution
Total Felony Convictions

Total felony convictions

Commitments to Department of Corrections
Adult Felony Record

Felony homicide

Felony homicide

Felony sex

Felony sex

Felony violent property

Felony violent property

Felony assault offense––not domestic violence

Felony assault offense––not domestic violence

Felony domestic violence assault or protection
order violation

Felony domestic violence assault or protection
order violation

Felony weapon

Felony weapon

Felony property

Felony property

Felony drug

Felony drug

Felony escape

Felony escape
Adult Misdemeanor Record

Misdemeanor assault––not domestic violence

Misdemeanor assault––not domestic violence

Misdemeanor domestic violence assault or
violation of a protection order

Misdemeanor domestic violence assault or
violation of a protection order

Misdemeanor sex

Misdemeanor sex

Misdemeanor other domestic violence

Misdemeanor other domestic violence

Misdemeanor weapon

Misdemeanor weapon

Misdemeanor property

Misdemeanor property

Misdemeanor drug

Misdemeanor drug

Misdemeanor escapes

Misdemeanor escapes

Misdemeanor alcohol

Misdemeanor alcohol
Adult Sentence Violations

Total sentence/supervision violations

Total sentence/supervision violations

Bureau of Investigations, Office of the Inspector General	

Page 22

Finding 4
CDCR’s Initial Policy Regarding Juveniles Convicted of Serious or Violent
Felonies was Incorrect
Penal Code section 3000.03, subsection (b) prohibits any offender convicted of a serious or
violent felony as defined in the Penal Code from being eligible for non-revocable parole.
Figure 7 presents a partial listing of serious or violent felonies listed in the code. On January
21, 2010, CDCR issued policy memorandum 10-01, Implementation of Penal Code Section
3000.03 – Non-Revocable Parole. This memorandum explains the criteria that make an
offender ineligible for non-revocable parole. Under the sections for both violent and serious
felony convictions, the policy specifies that juvenile sustained convictions are not disqualifying
factors. Accordingly, CDCR staff members who conduct eligibility screening for non-revocable
parole calculate the age of the offender and then exclude any serious or violent felonies that the
offender committed as a juvenile. However, according to the legal staff in CDCR’s policy unit,
the policy was worded incorrectly: it should have read, “juvenile sustained petitions,” rather
than “juvenile sustained convictions.”
The difference between a “sustained petition” and
a “sustained conviction” is significant. A sustained
petition is the outcome of a guilty adjudication
in juvenile court; a “sustained conviction” is the
outcome of a guilty verdict in adult court. According
to Penal Code section 3000.03 (b), any person
committed to prison, or having a prior conviction,
for a serious or violent felony is not eligible for nonrevocable parole. Therefore, juveniles tried as adults
and convicted of serious or violent felonies should
not be eligible for non-revocable parole.

Figure 7: Partial listing of serious or violent
felonies.
Violent felonies include:
•	 Murder
•	 Rape
•	 Lewd and Lascivious act
•	 Robbery
•	 Arson
•	 Car Jacking
Serious felonies include:
•	 Providing certain narcotics to minors
•	 Grand theft with a firearm
•	 Intimidation of a witness/victim
•	 Gross vehicular manslaughter

On November 5, 2010, following the Office of the
Inspector General’s identification of the inaccurate
policy, CDCR took corrective action and changed the
language of the policy to correctly read, “juvenile sustained petitions.” When asked how it would
correct past implementation of the faulty policy, CDCR replied that it had already trained staff to
properly apply the policy, since the staff’s training material had been correct from the beginning.
However, during our discussions with field staff, we discovered this was not always the case.
Specifically, we found that the teams responsible for screening the initial group of eligible
offenders had been following the incorrect policy and had consequently ignored all convictions
relating to serious and violent felonies that adult offenders had committed as juveniles.

State of California • May 2011	

Page 23

Recommendations
In this special report, the Office of the Inspector General identifies deficiencies in CDCR’s
implementation of non-revocable parole. To address the issues identified in this special report,
we recommend that CDCR take the following actions:
•	 Develop a quality control program for its CSRA scoring to ensure consistency,
accuracy and timely reviews.
•	 Collaborate with the California Department of Justice and the California Judicial
Council to obtain more complete criminal history information.
•	 Review the CSRA implementation to determine whether juvenile offender criminal
history is used appropriately.
•	 Ensure that CDCR staff is trained according to the revised policy regarding the
inclusion of juvenile data for serious and violent felonies.
•	 Review CSRA scores of offenders already placed on non-revocable parole to verify
that they were accurately assessed.
•	 Retain criminal history records used to determine eligibility for non-revocable parole.

Bureau of Investigations, Office of the Inspector General	

Page 24

California Department of Corrections and
Rehabilitation’s response to the special report

(page 1 of 3)

†

† Circled numbers correspond to OIG’s comments (on page 28) to CDCR’s response text.
State of California • May 2011	

Page 25

California Department of Corrections and
Rehabilitation’s response to the special report

(page 2 of 3)













Bureau of Investigations, Office of the Inspector General	

Page 26

California Department of Corrections and
Rehabilitation’s response to the special report

(page 3 of 3)









State of California • May 2011	

Page 27

The Office of the Inspector General’s Comments on
the Department’s Response (page 1 of 3)
Although we are not responding to all of CDCR’s statements contained in its response, we are
commenting on the following specific issues to provide clarity and perspective.
‡  As

our report states, our concerns are not with the CSRA instrument, but rather with
CDCR’s automation of the CSRA instrument.

 We appreciate CDCR’s desire to save money by automating the CSRA instrument,
and encourage it to continue to seek opportunities to be more efficient throughout its
organization. However, CDCR should not compromise public safety in doing so, as it
does by understating offenders’ risk of reoffending and releasing high-risk offenders to
unsupervised parole.
 The OIG understands that a system such as CDCR’s automated CSRA will be refined and
modified over time. However, as discussed in Finding 1 of our report, when improvements
allow CDCR to identify high-risk offenders that it previously erroneously released to
unsupervised parole, we expect CDCR to recall such offenders to supervised parole in the
interest of public safety.
 Contrary to CDCR’s assertion, the OIG has specifically noted corrections or improvements
that CDCR has made to the automated scoring process since July 2010, such as noting
that changes made by CDCR to its mapping table would reduce the error rate found in our
sample from 23.5 percent to eight percent. In this report, the OIG simply notes that during
the specific time period that the OIG sampled––January through July 2010––a specific error
rate was detected in the automated scoring process. The report also, however, gives
credit to CDCR for subsequent improvements that would serve to reduce the error
rate. In short, this is not a case of “critics…[seizing] upon the change as an opportunity to
focus in hindsight upon the ‘deficiencies’ of the prior version.” Instead, this is a case of the
OIG reporting upon those prior deficiencies, while also acknowledging those improvements
that CDCR has since made to the automated scoring process, subsequently reducing the
error rate.
 Although CDCR advised us that a July 2010 modification addressed many of the errors we
identify in our report, we have not verified this assertion.
 Statistical samples are always taken at a given point in time. To review CDCR’s
implementation of non-revocable parole, we selected a statistically valid sample of offenders
on non-revocable parole in July 2010. At the time of our review, this was the latest data
available.
 CDCR is misstating the facts. As stated in Finding 1 of our report, the experts at UC Irvine
with whom CDCR contracted to develop the CSRA state that information not contained in
automated rap sheets but recorded on CDCR automated systems––including supervision
‡ Circled numbers correspond to CDCR’s response text beginning on page 25.

Bureau of Investigations, Office of the Inspector General	

Page 28

The Office of the Inspector General’s Comments on
the Department’s Response (page 2 of 3)
violations––may be important markers for risk prediction and should be addressed in the
next steps of development of the CSRA. Furthermore, CDCR’s manual CSRA scoring team
includes supervision violations reported in CDCR’s automated systems in its evaluations of
offenders.
 The CSRA identifies 22 factors that are significant to consider in estimating the risk of an
offender’s propensity to reoffend. As discussed in our report, these factors include age,
gender, felony convictions, and supervision violations. Nevertheless, when assessing an
offender with its automated CSRA system, CDCR does not utilize information in its own
data systems that establish the presence of these factors.
 When the Office of the Inspector General determines it requires expertise beyond that of its
staff, it consults with experts in the appropriate fields. In this report, we retained a statistics
expert at California State University, Sacramento and consulted with the researcher who
oversaw the development of Washington’s predictive instrument, upon which the CSRA is
based. We do not state that the CSRA is flawed, rather, we identify the CSRA as a validated
risk assessment tool. What we point out is that the automated scoring process employed
by CDCR does not take into account information that CDCR’s own experts believe to
be important risk indicators, and that CDCR’s manual scoring team routinely takes into
consideration during the manual scoring process.
 While the Office of the Inspector General did mistakenly code one offender as a male
when in fact she was female, we corrected this error and determined that the change had no
impact on our conclusion that CDCR had inappropriately released the high-risk offender to
unsupervised parole. We attempted to meet with CDCR to discuss its concerns with this and
other offenders, but CDCR did not avail itself to meet as of the time of this reporting.
 In the example cited, CDCR’s automated score and manual score did not match. The OIG
score matched CDCR’s manual score, which concluded that the offender should not have
been released to non-revocable parole. As discussed in number 10 above, we attempted to
discuss this concern with CDCR.
 CDCR has missed the point. Although these offenders’ criminal histories should have
triggered a manual CSRA, CDCR failed to do so, instead relying on the incomplete
automated CSRA score to release these high risk offenders to non-revocable parole.
 The Office of the Inspector General acknowledges that, at present, the ACHS is the
most viable rap sheet database available to CDCR. We disagree, however, that
this finding is of “scant value,” since all experts consulted agree that the CSRA’s predictive
capacity will be enhanced if and when a more accurate database is made available. This
finding simply points out the need for such a database, as almost half of the 16.4 million
arrests between 2000 and 2009 lacked dispositions, such as convictions.

State of California • May 2011	

Page 29

The Office of the Inspector General’s Comments on
the Department’s Response (page 3 of 3)
 As discussed in our report, CDCR appropriately excludes from its assessments juvenile
sustained petitions, which are adjudicated in juvenile court. However, CDCR is inconsistent
in its treatment of juvenile sustained convictions, which are adjudicated in adult court.
Our analysis found that CDCR treats some juvenile convictions as adult convictions, and
ignores other juvenile convictions. The designers of the CSRA acknowledge that juvenile
convictions impact an offender’s risk of reoffending differently than an adult conviction, and
therefore, should be considered separately. Nevertheless, CDCR in some cases combines
juvenile convictions with adult––thereby diluting their impact––and in other cases excludes
juvenile convictions––ignoring their impact altogether.

Bureau of Investigations, Office of the Inspector General	

Page 30

SPECIALREPORT
REPORT
SPECIAL
INMATE CELL PHONE USE ENDANGERS PRISON
SECURITY AND PUBLIC SAFETY

California Department of Corrections and Rehabilitation’s
Parole Program
Implementation of the Non-Revocable
OFFICE OF THE
INSPECTOR GENERAL

OFFICE OF THE INSPECTOR GENERAL
DAVID R. SHAW

INSPECTOR GENERAL

STATE OF CALIFORNIA

Bruce A. Monfross

MAY 2009

INSPECTOR GENERAL (A)



David Chriss

DEPUTY INSPECTOR GENERAL, SENIOR

Larry Sewell
DEPUTY INSPECTOR GENERAL

Dave Biggs
DEPUTY INSPECTOR GENERAL

STATE OF CALIFORNIA
May 2011
WWW.OIG.CA.GOV