Skip navigation
Disciplinary Self-Help Litigation Manual - Header

Washington's Offender Accountability Act Outcomes, WSIPP, 2005

Download original document:
Brief thumbnail
This text is machine-read, and may contain errors. Check the original document to verify accuracy.
Washington State
Institute for
Public Policy
110 Fifth Avenue Southeast, Suite 214

•

PO Box 40999

•

Olympia, WA 98504-0999 •

(360) 586-2677

•

www.wsipp.wa.gov

July 2005

WASHINGTON’S OFFENDER ACCOUNTABILITY ACT:
A FIRST LOOK AT OUTCOMES‡
On an average working day in the Superior Courts
of Washington, about 125 adults are convicted for
a felony crime. Over the course of a year, this
means that close to 30,000 adult felony sentences
are handed down statewide.1

Key Finding: Preliminary results indicate
that recidivism rates have declined
slightly since passage of Washington’s
Offender Accountability Act (OAA).
Enacted in 1999, the OAA requires the
Department of Corrections to classify
adult felony offenders and re-allocate
community-based resources by putting
more effort on higher-risk offenders and
less effort on lower-risk offenders.

Except for those who serve a life sentence in
prison, all of these felony offenders will re-enter
the community, either immediately after
sentencing or eventually after serving time in a
county jail or a state prison.2
In 1999, the Washington State Legislature passed
the Offender Accountability Act (OAA) to affect how
the state provides community supervision to these
adult felony offenders. In broad terms, the OAA
directs the Washington State Department of
Corrections (DOC) to:

We estimate that the two-year felony
recidivism rates of higher-risk offenders
have dropped by 3.5 percentage points,
while the rates for lower-risk offenders
have fallen a more modest 1.2 points.
These findings are preliminary; definitive
results will require four more years of
observation. While it is too early to
conclude that the OAA “caused” the drop
in recidivism, these initial outcomes can
be interpreted as promising.

9 Classify felony offenders according to their risk
for future offending as well as the amount of
harm they have caused society in the past; and
9 Deploy more staff and rehabilitative resources
to higher-classified offenders and, because
budgets are limited, spend correspondingly
fewer dollars on lower-classified offenders.
When the Legislature enacted the OAA, it defined a
straight-forward goal for the Act: to “reduce the risk of
reoffending by offenders in the community.”3

‡

For information on this report, contact the authors: Steve Aos
and Robert Barnoski at, respectively, <saos@wsipp.wa.gov> or
<barney@wsipp.wa.gov>.
1

Washington Administrative Office of the Courts, data for 2004.
Depending on the nature of an offender’s crime and criminal
history, roughly 30 percent of felony sentences result in a
commitment to state prison while the remaining 70 percent
involve a local non-prison sanction, which most often includes
serving time in a county jail. Also, during fiscal year 2004, there
were 32 life sentences issued. Data sources: Washington
Administrative Office of the Courts; Washington Sentencing
Guidelines Commission, Statistical Summary of Adult Felony
Sentencing, Fiscal Year 2004.
3
RCW 9.94A.010.
2

To determine whether the OAA results in lower
recidivism rates,4 the Legislature directed the
Washington State Institute for Public Policy (Institute)
to evaluate the impact of the Act. This is the
Institute’s fifth annual report on the OAA. The final
evaluation on long-term outcomes is due in 2010.
This year’s report provides the initial opportunity to
examine the effect of the OAA’s first year of
operation on short-run recidivism rates. We
emphasize that the information provided here is
preliminary; we will only have definitive results after
four more years of observation. This year’s report is
a bit like the earliest returns on election night—the
reader should be aware that initial outcomes can
change significantly as time unfolds.

4

“Recidivism” refers to the re-commission of a new criminal
offense.

Exhibit 1

Summary of Findings
After we summarize our preliminary findings in
this short section, the balance of the report
describes in more detail how the OAA operates as
well as the multivariate statistical methods we
used to carry out the analysis.
Under the OAA, DOC classifies offenders into four
groups and allocates more resources to the higherrisk groups (and corresponding fewer resources to
the lower-risk groups). To test whether this strategy
lowers recidivism, we analyze the reconviction rates
of all offenders released to the community during
the first full year of implementation of the OAA—
between July 1, 2001, and June 30, 2002. We
compare this initial OAA group with similar offenders
released prior to the OAA.
In this preliminary analysis, we made two types of
comparisons: 1) we combined the two higher-risk
OAA groups and compared their recidivism with
their higher-risk pre-OAA counterparts; and 2) we
combined the two lower-risk OAA groups and
compared them with similar lower-risk offenders
from the pre-OAA period.
Exhibits 1 and 2 show the key findings from our
preliminary analyses. Twenty-four months after
re-entering the community, we estimate that both
higher-risk and lower-risk offenders supervised
under the OAA have slightly lower recidivism
rates than their non-OAA comparison groups.
9 For the higher-risk offenders under the
OAA, we find that the OAA group had a 39.9
percent overall (felony and misdemeanor)
recidivism rate, while the comparison group
had a 44.9 percent rate—a 5.0 percentage
point drop favoring the OAA group. Focusing
only on felony reconvictions, the OAA and
comparison groups had 23.1 and 26.6 percent
recidivism rates, respectively—a 3.5
percentage point drop favoring the OAA
group. Narrowing the focus further to just
violent felony recidivism, we find that the OAA
and comparison groups had 5.5 and 5.9
percent recidivism rates—a 0.4 percentage
point drop favoring the OAA group.
9 For the lower-risk offenders under the OAA,
we also find lower recidivism rates than for the
comparison group, although the difference in
rates is smaller. For example, we find that the
OAA group had a 35.1 percent overall (felony
and misdemeanor) recidivism rate, while the
comparison group had a 37.5 percent rate—a
2.4 percentage point drop favoring the OAA
group.
2

Higher-Risk Offenders (RMA and RMB):
2-Year Reconviction Rates for the
OAA and Comparison Groups
Comparison (N = 4,389)
OAA (N = 4,389)

44.9%
39.9%

26.6%
23.1%

5.9%
Any Recidivism*

Felony Recidivism*

5.5%

Violent Felony
Recidivism**

* = Statistically significant difference ** = Not statistically different

Exhibit 2

Lower-Risk Offenders (RMC and RMD):
2-Year Reconviction Rates for the
OAA and Comparison Groups
Comparison (N = 10,413)
OAA (N = 10,413)

37.5%

35.1%

21.7% 20.5%

3.9%
Any Recidivism*

Felony Recidivism*

3.4%

Violent Felony
Recidivism*

* = Statistically significant difference

All differences are statistically significant with one
exception: the violent felony recidivism rates for
the higher-risk OAA group.
For this preliminary recidivism analysis of the
OAA, we did not attempt to conduct a benefit-cost
analysis; that economic study will be completed in
a future report on the OAA. If, however, the
incremental cost of the OAA is zero—that is, if the
OAA simply reallocates existing DOC resources—
then any statistically significant reduction in
recidivism attributed to the OAA will be costbeneficial.
It is, however, too early in our evaluation of the
OAA to conclude that the preliminary recidivism
reductions shown on Exhibits 1 and 2 are
“caused” by the OAA. That is, there may be other
factors that are unobserved to us that caused the

estimated reductions in recidivism. Additionally,
in this preliminary analysis we were only able to
use a two-year follow-up period to measure
recidivism, and we have found that more reliable
results can be measured with at least a three-year
follow-up period. In subsequent reports on the
OAA, we will be able to observe recidivism over
longer time frames and follow additional cohorts
of OAA offenders. This will increase our ability to
determine whether the OAA is a causal factor in
the drop in recidivism.

Elements of the 1999 Offender
Accountability Act
The OAA requires DOC to take two broad steps:
1) DOC must classify all offenders using a
research-based assessment tool, and 2) the
agency must use this information to allocate
supervision and treatment resources.5 These
basic elements are described in this section.
OAA Element 1: DOC’s Offender
Classification System
The OAA instructs DOC to classify felony
offenders according to the risk they pose to
reoffending in the future and the amount of harm
they have caused society in the past. To give
operational direction to this classification, the OAA
defines the assessment with this language:
“Risk assessment” means the application of
an objective instrument supported by
research and adopted by the department for
the purpose of assessing an offender's risk
of reoffense, taking into consideration the
nature of the harm done by the offender,
place and circumstances of the offender
related to risk, the offender's relationship to
any victim, and any information provided to
the department by victims.6
With this language, the legislature directed DOC
to classify offenders by taking into account two
broad concepts: the “risk of reoffense” and the
“nature of the harm done.” These two concepts
do not necessarily address the same factors.

5

The OAA also gave DOC new authority to hold timely hearings
and to sanction offenders with crimes committed after July 1,
2000, who violate conditions of community custody.
6
RCW 9.94A.030(32), emphasis added.

The “risk of reoffense” concept is forward looking.
A classification system that measures the risk of
reoffense is designed to predict whether an
offender is likely to commit another crime in the
future. The “harm done” concept, on the other
hand, is backward looking. A classification system
that measures harm done measures how much
damage an offender has already caused victims
and society, regardless of what he or she is likely
to do in the future.
The DOC designed its “Risk Management
Identification” (RMI) system—the formal name of
DOC’s classification system—to include two sets of
assessments and decision rules that together
attempt to measure and balance both of these
OAA concepts. First, DOC adopted a formal risk
assessment tool to measure the likelihood of future
reoffending. Second, DOC adopted additional
criteria to gauge how much harm the offender’s
prior criminal activity caused victims and society.
Each of these two classification tools is
summarized here.
Element 1a: DOC’s “Risk of Reoffense”
Assessment Tool.7 Prior to the OAA, DOC began
using a formal risk assessment tool called the
“Level of Service Inventory-Revised (LSI-R).”
Canadian researchers developed this 54-question,
copyrighted instrument in the 1980s. Previous
research (outside of Washington State) indicated
the LSI-R is a valid instrument for predicting
whether an offender is likely to reoffend.8 DOC
adopted the LSI-R as one of the key parts of its
Risk Management Identification system.
The questions on the LSI-R cover ten areas of an
offender’s life.9 After DOC staff administers the
LSI-R, an offender’s combined LSI-R score is
tabulated. An offender’s LSI-R score can range
from 1 to 54, where higher numbers indicate a
higher probability of reoffending.
Exhibit 3 provides a snapshot on how the LSI-R
relates to recidivism in Washington. The chart
shows that the higher the LSI-R score, the higher
7

This section draws on our formal study of the LSI-R as
implemented by DOC. See, R. Barnoski, S. Aos, Washington’s
Offender Accountability Act: An Analysis of the Department of
Corrections’ Risk Assessment. Olympia: Washington State
Institute for Public Policy, 2003.
8
Prior research associated with the LSI-R is discussed in: D. A.
Andrews & J. L. Bonta (1995) The Level of Service InventoryRevised, Manual. North Tonawanda, New York: Multi-Health
Systems, Inc.
9
These 54 questions include: ten questions on prior criminal
history; ten on an offender’s education and employment; two on
finances; four on an offender’s family situation; three on an
offender’s living situation; two on leisure and recreation activities;
five on peers; nine of alcohol and drug problems; five on
emotional or personal problems; and four on an offender’s
attitude.

3

the chance that an offender will recidivate. An
analysis of these data indicates that a one-point
increase in the LSI-R score results in a 1.5
percentage point increase in the misdemeanor and
felony recidivism rate, a 1.1 percentage point
increase for felony recidivism, and a 0.3 percentage
point increase for violent felony recidivism.

Exhibt 3

Recidivism Rates by LSI-R Score
80%
70%
M isdemeano r o r
F elo ny R ecidivism

60%
50%
40%

F elo ny
R ecidivism

30%
Vio lent F elo ny
R ecidivism

20%
10%
0%
0

5

10

15

20
25
LSI-R Score

30

35

40

45

Unfortunately, Exhibit 3 also reveals no distinct
changes in recidivism rates from one score to the
next—that is, the lines on Exhibit 3 increase at fairly
steady rates. This is a significant finding, because
it means that the LSI-R provides no naturally
occurring “cut-off scores” to create low- and highrisk categories. Nonetheless, DOC uses LSI-R
specific cutoff scores, in conjunction with its harmdone rules, to classify offenders. The specific
scoring rules used by DOC are summarized on
page 12 of this report.
In our previous study testing the predictive validity
of the LSI-R as used by DOC, we presented four
findings:10
9 The LSI-R predicts recidivism moderately well.
9 The predictive power of the LSI-R can be
improved significantly by adding several readily
available measures.
9 An enhanced prediction instrument would
improve the classification of DOC offenders by
specifically measuring the likelihood of the most
serious form of recidivism—violent felonies.
9 Finally, as reported above, there are no distinct
changes in recidivism rates at specific risk
scores, thus there are no obvious “cut-off
scores” to create risk categories.

Following up on these conclusions, the Institute
and DOC are currently discussing alternative
approaches to risk-for-reoffense classifications for
the RMI system.
Element 1b: DOC’s “Harm Done” Criteria. As
mentioned, the LSI-R was designed to predict
whether an offender will commit another crime. It
was not constructed to measure the level of prior
harm caused by an offender—a key requirement
in the OAA legislation. To implement this aspect
of the OAA, DOC adopted an additional set of
rules to gauge how much damage an offender
has caused in his or her prior criminality. DOC
developed these “harm-done” rules from
recommendations by DOC, the Victims Council,
and criteria established by the Washington
Association of Sheriffs and Police Chiefs.11
Examples of the RMI questions used to determine
prior harm done include the following: Is the
offender classified as a Level I, II, III sex
offender? Is the offender designated as a
Dangerous Mentally Ill Offender? Did the
offender commit a violent offense involving a
stranger? If an offender scores a “yes” on any of
these conditions, then, regardless of the
offender’s LSI-R score, the offender is regarded
as needing higher levels of community custody.
The Product of the Classification System: RMA,
RMB, RMC, and RMD Offender Classifications.
Together, the LSI-R and the harm-done criteria
make up DOC’s RMI classification system; DOC
uses the system to classify each offender. As
mentioned, DOC’s specific scoring rules are listed
at the end of this report.
With the RMI scoring system, each felon under
DOC supervision is classified into one of four
categories: RMA, RMB, RMC, or RMD. The RMA
category is the highest risk and highest harmdone classification, while the RMD category is the
lowest risk and lowest harm-done group.
Exhibit 4 shows the distribution of all RMIclassified offenders for the sample of offenders
used in this study. The sample includes felons
released to the community during Fiscal Year
2002—those released from state prison as well as
those sentenced to local sanctions (usually county
jail). The chart shows that 30 percent of offenders
are classified as either RMA or RMB (the highest
risk and harm-done categories) with the majority
(70 percent) of offenders classified in the less
risky RMC and RMD categories.
11

10

4

Barnoski and Aos, 2003.

Washington State Department of Corrections, “Risk
Assessment and the Offender Accountability Act, November 5,
2001,” presentation to the House Criminal Justice and Corrections
Committee, November 30, 2001.

Exhibit 4

All DOC Offenders by RMI Level

RMA
15%

RMB
15%

RMD
29%
RMC
41%
Source: See Exhibit 7

OAA Element 2: Resource Allocation Pursuant
to the OAA
As discussed, the OAA requires that DOC not
only classify offenders, but also re-allocate
resources according to this classification. The
OAA directs DOC to deploy more of its
community-based resources to the higher-risk
RMA and RMB offenders, with correspondingly
fewer resources devoted to the relatively lowerrisk RMC and RMD offenders. Thus, the RMI
designation is central to the OAA and its ability to
reduce overall recidivism rates.

DOC budget data indicate that the higher-risk
(and highest harm-done) offenders receive more
resources than lower-risk offenders. Exhibit 5
indicates that $5,500 per year is budgeted for
community supervision for the average RMA and
RMB offender. This contrasts with $1,249 and
$505 for each lower-risk RMC and RMD offender,
respectively. This differential in the level of effort
is also seen in the monthly amounts of community
supervision staff time that DOC budgets for
different risk-classified offenders. Exhibit 6 shows
that 9.2 hours of community supervision per
month are budgeted for RMA offenders. The
slightly lower-risk RMB offenders are budgeted to
receive a similar level of supervision: 7.6 hours of
staff attention per month. RMC offenders are
budgeted to receive less supervision at 5.4
budgeted monthly hours while RMD offenders
receive considerably less supervision at 1.6
hours.
These DOC data provide a clear indication that
more community-based resources are being
budgeted for the higher-risk offenders and
correspondingly fewer resources are being spent
on lower-risk offenders. Whether the OAA is able
to reduce overall recidivism rates and provide a
net benefit to Washington will depend, in part, on
the effectiveness of this resource re-allocation
that DOC has made pursuant to the legislative
direction of the OAA.

Exhibit 5

Exhibit 6

DOC Budgeted Annual Cost per Offender
on Community Supervision, by RMI Level

DOC Supervision Hours Budgeted per
Offender per Month, by RMI Level
9.2
7.6

Supervison Hours

Fiscal Year 2005 Dollars

$5,500

$1,249

5.4

1.6

$505

A&B

C

RMI Classification Level
Source: Personal communication with DOC staff.

D

A

B

C

D

RMI Classification Level
Source: Personal communication with DOC staff.

5

The Institute’s Evaluation of the OAA:
Technical Description
The Basic Research Question. In passing the
OAA, the intention of the Legislature was to lower
recidivism rates by reallocating a portion of DOC’s
budgeted resources, especially those used to
supervise offenders in the community. The OAA
directs that more DOC resources be spent on higherrisk offenders and, because of budget restrictions,
fewer resources be spent on lower-risk offenders. If
supervision and treatment resources are efficacious,
then, under the OAA, recidivism rates might be
expected to decrease for higher-risk offenders and
increase for lower-risk offenders. If, however,
community supervision and treatment resources do
not have a statistically significant effect on recidivism
rates, then, of course, the OAA will not change the
recidivism rates of either the higher-risk or lower-risk
offenders. It is also possible that resources affect
higher-risk offenders differently than lower-risk
offenders. In particular, it is possible that supervising
lower-risk offenders less may reduce their recidivism
rates. Thus, the basic research tasks for the
evaluation of the OAA are to estimate whether the
Act produces significant effects on the recidivism
rates of high-risk and low-risk offenders.

day before OAA Legislation went into effect). The
total pre-OAA sample includes 19,995 offenders.
2) The OAA group includes all offenders released
to the community between July 1, 2001, and June
30, 2002—the first full year of effective OAA
implementation. We selected the July 1, 2001, start
date for the OAA group because that was when,
according to DOC, the RMI classification system
was fully operational.12 The June 30, 2002, cutoff
date for the OAA group was set so that offenders in
the sample would have a fixed two-year follow-up
period in the community, with an additional sixmonth period to allow any cases to be adjudicated
by the courts. The total OAA sample includes
14,879 offenders.
Data. For all offenders in this study, we collected
administrative data on the following factors:
9 Age at release to the community
9 Gender
9 Ethnicity
9 Adult and juvenile criminal history
9 Crime for the current offense
9 LSI-R scores
9 Sentence type (prison or community)

Research Design. An ideal way to estimate
whether the OAA achieves a reduction in
recidivism would be to assign randomly some
offenders to the new OAA regime and some to the
old pre-OAA approach to community supervision.
Any observed difference in recidivism rates could
then be confidently attributed to the effect of the
OAA. Unfortunately, a random assignment
approach was not possible for this evaluation
because the implementation of the OAA was
statewide and, as we explain, it became fully
operational at one time (mid-2001).
Lacking the ability to employ a random assignment
evaluation design, the Institute uses several “nextbest” non-experimental approaches to address the
basic question of interest for the evaluation.
Two Groups of Offenders for the Evaluation:
the Pre-OAA Group and the OAA Group. Our
approach to evaluating the OAA derives from
administrative data on two groups of offenders
who were released to the community during two
different time periods:
1) The pre-OAA group includes all DOC
offenders released to the community (from either
prison or jail, or directly to community placement)
between October 1, 1998 (the day the first LSI-R
scores were available) and June 30, 2000 (the
6

9 Recidivism data
The administrative data were obtained from two
principal sources: electronic files from the
Department of Corrections and electronic courtbased data from the Administrative Office of the
Courts. The Institute obtains quarterly data
updates from these sources and combines them
for research purposes. For the evaluation
presented in this report, the data are through
December 2004.
For the OAA group, we also have each offender’s
RMI classification. We do not, of course, have
information on the RMI classification for the preOAA group since the OAA system was not in
place during that time period.
In the multivariate models we use for this
evaluation, we include three statisticallydeveloped risk scores we have computed for each
offender in the two groups. These scores are
alternatives to the LSI-R and were constructed
specifically to measure “static” risk factors in a
more comprehensive and consistent way than the
12

The OAA legally became effect on July 1, 2000. According to
DOC, however, the RMI classification system was not fully
operational until the late spring of 2001. Therefore, we selected
July 1, 2001, as the beginning release-to-the-community date for
our evaluation.

Exhibit 7

Descriptive Statistics
OAA and Pre-OAA Groups

Demographic Variables
Age at Release
Male %
White %
Black %
Asian %
Native Amer %
Risk for Reoffense Scales
LSI-R Score
Prison Sentence %
Static Felony Risk Score
Static NonDrug Fel Score
Static Violent Fel Score
Unadjusted Recidivism
Felony & Misdemeanor %
Felony %
Violent Felony %
Sample Size

OAA Group by RMI Classification

Statistically
Significant
Difference?
(p-value)

RMA

RMB

RMC

RMD

RMA
and
RMB

RMC
and
RMD

32.09
78.2%
77.2%
16.4%
2.3%
2.9%

No (0.27)
Yes (0.03)
Yes (0.02)
Yes (0.00)
No (0.37)
No (0.79)

32.70
92.2%
67.3%
24.5%
2.8%
3.8%

32.98
81.4%
75.8%
17.7%
1.4%
4.0%

32.11
74.4%
80.9%
13.2%
1.8%
3.1%

31.71
71.7%
81.4%
11.5%
3.6%
1.8%

32.84
86.7%
71.6%
21.1%
2.1%
3.9%

31.94
73.7%
81.1%
12.5%
2.5%
2.5%

26.57
17.1%
59.21
35.29
35.34

22.03
21.8%
58.03
34.12
34.53

Yes (0.00)
Yes (0.00)
Yes (0.00)
Yes (0.00)
Yes (0.00)

31.55
33.8%
66.21
40.33
46.75

35.19
27.8%
69.35
40.82
42.35

29.72
16.2%
62.24
35.97
34.47

15.32
4.6%
46.15
28.70
26.94

33.39
30.8%
67.79
40.58
44.53

23.69
11.3%
55.50
32.92
31.32

41.7%
26.1%
6.8%
14,879

41.8%
26.3%
7.0%
19,995

No (0.90)
No (0.78)
No (0.52)

48.8%
29.9%
12.4%
2,192

52.7%
34.0%
9.9%
2,236

47.8%
30.9%
6.6%
6,072

24.2%
13.6%
2.7%
4,379

50.8%
32.0%
11.1%
4,428

37.9%
23.6%
4.9%
10,451

OAA
Group

PreOAA
Group

32.21
77.3%
78.3%
15.0%
2.4%
2.9%

Data Sources: Washington State Department of Corrections and Administrative Office of the Courts. The OAA group includes those offenders released to the
community during Fiscal Year 2002. The pre-OAA group includes those offenders released to the community between October 1, 1998, and June 30, 2000.

LSI-R allows. Static risk factors are those
observed characteristics of an offender that do not
change over time, such as an offender’s gender
and prior criminal history. Each of the three
scores—for felony risk, violence risk, and nondrug felony risk—were calculated with multivariate
logistic regression models predicting the three
types of recidivism outcomes as a function of a
number of variables including information on prior
and current adult and juvenile criminal history.13
Recidivism Outcome Measures. All references
to “recidivism” in this study refer to an offender
who is reconvicted for a new offense in the courts
in Washington State. That is, recidivism is a
conviction for an offense committed after
placement in the community. For offenders
sentenced to prison or jail, placement in the
community begins at the time of release from
confinement. For offenders sentenced directly to
community supervision, placement begins at the
time of sentencing. Adequately measuring
recidivism for adult offenders requires a sufficient
follow-up period for reoffending as well as another
period to allow for reoffenses to be formally
adjudicated.14 In this study, we calculate a 24month follow-up period and allow six months for
adjudications to be decided in the court system.
Both these time periods are shorter than
13

More information on the construction of these scores can be
obtained from one of this report’s authors, Robert Barnoski.
14
This definition of recidivism is consistent with that developed for
the legislature in R. Barnoski, Standards for Improving Research
Effectiveness in Adult and Juvenile Justice. Olympia:
Washington State Institute for Public Policy, 1997.

appropriate for adult offenders. In future reports
on the OAA we will be able to calculate 36-month
recidivism rates and allow for a 12-month
adjudication period. In this preliminary OAA
report, however, we were constrained to the
shorter time intervals.
We report three types of recidivism rates in this
study: 1) any recidivism which records a new
felony or misdemeanor conviction; 2) felony
recidivism which measures only new felony
convictions; and 3) violent felony recidivism which
includes only new felony convictions for homicide,
sex offenses, robbery, or aggravated assault.
In upcoming annual reports on the OAA, we will
also examine two other commonly-measured
recidivism outcomes: returns to prison and
technical violations. The principal focus of the
evaluation, however, will continue to be whether
the OAA affects the rate at which offenders
commit new crimes.
Descriptive Statistics. Basic descriptive
information on the OAA and pre-OAA groups of
offenders is presented in Exhibit 7. The Exhibit
also provides descriptive information for the OAA
group broken out by RMI classification.
Exhibit 7 shows that there are some statistically
significant differences between offenders released
prior to the OAA and the offenders released after
the OAA, although many of the differences are
quite small. For example, 78.2 percent of the preOAA group are male while 77.3 percent of the
7

OAA group are male. More significant for our
evaluation of the OAA, several risk-for-reoffense
scales are statistically different between the two
groups. The three static risk scales indicate that
the pre-OAA is about two percent less risky than
the OAA group.
There is a large difference, however, in the LSI-R
scores of the two groups: the pre-OAA group is
about 17 percent less risky than the OAA group
as measured by the LSI-R. This large
discrepancy raises concerns about the
comparability of the LSI-R during the two time
intervals. In the pre-OAA period, the LSI-R was a
new tool for DOC and it may not have been
completed as thoroughly as it was during the OAA
period. We confirmed this in personal
communication with DOC staff. Because of this, it
is inappropriate to use the LSI-R score, in
aggregate, to control for differences in the preOAA and OAA groups in our analyses. Thus, in
the statistical models we developed for this
evaluation, we use the three static risk scales that
we developed to measure risk for reoffense.
Three Statistical Models. Lacking the
opportunity to employ a random assignment
research design, we developed three statistical
modeling approaches to test whether the OAA
achieves reductions in recidivism.15 These
modeling approaches provide a range of
estimates of the effect of the OAA on recidivism;
each model offers advantages and
disadvantages. From these estimates, we
attempt to draw preliminary inferences about
whether the OAA has affected the recidivism rates
of adult felony offenders in Washington.
1) Basic Multivariate Model. First, we
estimated a standard multivariate logistic
regression model where 24-month recidivism
outcomes are a function of OAA group
membership along with a variety of control
variables. The model takes this form:
Recidivism = f(OAA, X, error)
In this basic model, we test simply for any
estimated overall effect of the OAA on recidivism,
after controlling for information we have on each
offender (the “X” term in the equation). As shown
on Exhibit 7, there are pre-existing differences
between the pre-OAA and the OAA groups with
the OAA appearing to be slightly riskier for
reoffense. The controlling variables, which
include the observed variables described earlier,
15

A standard text describing some of our modeling approaches is:
J. M. Wooldridge, Econometric Analysis of Cross Section and
Panel Data. Cambridge, MA: MIT Press, 2002.

8

are used to adjust statistically for these preexisting differences. This basic model tests for an
overall OAA effect, but cannot estimate separate
effects for the high-risk and low-risk offenders as
classified by the OAA. This is a limitation to this
first simple modeling approach since the purpose
of the OAA is to separate offenders into higherand lower-risk classifications. Nonetheless, we
estimated this basic multivariate model to provide
an initial examination of the effect of the OAA.
2) Propensity Score Matching Models. Next,
we implement two propensity score matching
models. This modeling approach involves two
steps. First, we develop a model to predict which
OAA offenders are classified by DOC as either a
higher-risk RMA or RMB offender or as a lowerrisk RMC or RMD offender. That is, for the OAA
sample, we estimate:
RMAB-hat = f(X, error)
RMCD-hat = f(X, error)
In these models, an OAA offender’s actual group
membership is a function of a variety of X
variables that DOC uses to classify offenders (see
page 12). We use this information to predict
which offenders in the pre-OAA period would
have been classified as an RMA or RMB offender
(RMAB-hat) or as an RMC or RMD offender
(RMCD-hat) had the RMI system been in place
prior to the OAA. Based on these propensity
scores, we then used a matching algorithm to
select matched OAA and pre-OAA groups
(matched on each offender’s propensity score).
After selecting the OAA and pre-OAA matched
groups based on propensity score matching, we
then used multivariate logistic regression to
estimate effects.
Recidivism-AB Group = f(OAA, X, error)
Recidivism-CD Group = f(OAA, X, error)
The advantage of the propensity score models
over the basic multivariate model is that they
allow an estimate of the separate effects of the
OAA on the higher-risk (AB) and lower-risk (CD)
offenders. That is, it models explicitly the
selection process DOC uses to assign an offender
to the RMA and RMB classifications as well as the
RMC and RMD classifications.
3) Risk Factor Matching Models. Finally, we
estimate another form of matching models where
we create matched groups not based on a single
propensity score but, rather, on the three separate
risk scales that we developed to predict
recidivism. This approach allows the creation of
comparison groups based on the specific risk

factors shown to predict recidivism. In this
approach, we use the matching algorithm to
select, for each OAA AB or CD offender, a unique
matched offender from the pre-OAA group with
nearly identical scores on the three risk factor
scales. Once these risk-scale matched groups
are created, we use, as before, multivariate
logistic regression to estimate OAA effects.
Recidivism-AB Group = f(OAA, X, error)
Recidivism-CD Group = f(OAA, X, error)
The advantage of the risk scale matching
approach is it uses specific information about
factors shown to predict recidivism as the basis
for selecting the comparison groups for the
analysis. The disadvantage is it does not model
explicitly the classification process DOC uses to
assign offenders to the RMA, RMB, RMC, and
RMD categories.

A Note on the Matching Algorithm
The matching algorithm used by the Institute for
this analysis assigns absolute-difference values
for each matching factor chosen. For example, if
age were a factor, then a 35-year-old case record
would have an age distance value of 0 when
compared with a 35-year-old control, and an age
distance value of 1 when compared with controls
aged 34 or 36. Each matching factor is assigned
a maximum allowable distance. Using the age
example again, we could assign a maximum
allowable distance of 5, thus allowing a match of
the 35-year-old case record with controls between
30 and 40. Using multiple matching factors, we
match each case to the control with the lowest
sum of difference values.
For this study we used the following distances for
the static risk factor matching models and the
propensity score matching models:
Matching
Factor
------------Static Felony Risk Score
Static NonDrug Felony Score
Static Violence Score
Propensity Score

Maximum
Distance
----------------7
5
5
.01

9

The lower panel of Exhibit 8 shows the regression
results for a model without the covariates in the
first model. Without the other control variables, the
OAA has no significant difference on any measure
of recidivism. This finding is consistent with the
descriptive information shown in Exhibit 7. That is,
the OAA sample appears to be a slightly riskier
group than the pre-OAA group.

Preliminary Evaluation Results
Exhibit 8 displays the findings from the first
modeling approach—the basic multivariate model.
As mentioned, this model is presented as the first
step in the analysis; its simple structure does not
allow a refined look at the question of whether the

The regression results in Exhibit 8
confirm that a significant OAA
effect on recidivism emerges only
after adjusting for the higher-risk
nature of offenders in the OAA
group.

Exhibit 8

Basic Multivariate Model:
Mean-Adjusted 2-Year Recidivism Rates

Method and Outcome

OAA
Group

All OAA and Pre-OAA Offenders
Comparison
DifferEffect
pGroup
ence
Size
value

N

The results of the second and third
modeling approaches are shown
34.4%
38.1%
-3.6%
-0.08
.00
34,830
in Exhibit 9. The top panel shows
19.9%
21.9%
-2.0%
-0.05
.00
34,830
the results for the propensity score
3.4%
4.0%
-0.5%
-0.03
.00
34,830
matching methods, while the
bottom panel shows the results for
41.7%
41.8%
-0.1%
-0.00
.90
34,874
the risk factor matching approach.
26.1%
26.3%
-0.1%
-0.00
.78
34,874
In each case, the risk factor
6.8%
7.0%
-0.2%
-0.01
.52
34,874
matching method produces
Note: Results are from separate logistic regressions for each recidivism outcome, with effects calculated
smaller OAA effects than the
at the means of the samples. Effect sizes are calculated using the arcsine transformation described in:
M.W. Lipsey and D. Wilson. (2001) Practical meta-analysis. Thousand Oaks: Sage Publications, Table
propensity score matching
B10, formula (22).
method. For example, for the
higher-risk group (RMAB
OAA affects recidivism rates for higher- and
offenders),
the
propensity
score method estimates
lower-risk offenders. Nonetheless, the results
a
statistically
significant
reduction
of 7.6 percentage
shown in the upper panel of Exhibit 8 (the model
points
in
overall
recidivism
while
the
risk factor
with all covariates) indicate that overall recidivism
method
indicates
a
5.0
percentage
point
reduction.
rates are 3.6 percentage points lower for the OAA
Except
for
the
violent
felony
recidivism
estimates
group than for the pre-OAA comparison group.
for the AB group, all reductions are statistically
For felony-only recidivism, the rates are 2.0
significant.
percentage points lower. For violent felony
Basic Multivariate Model
With Covariates
Any Recidivism
Felony Recidivism
Violent Felony Recidivism
Basic Multivariate Model
Without Covariates
Any Recidivism
Felony Recidivism
Violent Felony Recidivism

recidivism, the rates are lower by about half a
percentage point. All these results are small, but
they are statistically significant given the large
size of the sample in the analysis (n = 34,830).

Exhibit 9

Propensity Score and Risk Factor Matching Models:
Mean-Adjusted 2-Year Recidivism Rates
Higher-Risk Group (RMA and RMB Offenders)
ComOAA
parison
DifEffect
pGroup
Group
ference
Size
value
N

Lower-Risk Group (RMC and RMD Offenders)
ComOAA
parison
DifferEffect
pGroup
Group
ence
Size
value
N

Propensity Score Matching
Any Recidivism
Felony Recidivism
Violent Felony Recidivism

42.8%
24.2%
7.1%

50.5%
29.0%
8.1%

-7.6%
-4.8%
-1.0%

-0.15
-0.11
-0.04

.00
.00
.09

7,404
7,404
7,404

33.4%
19.5%
3.0%

35.9%
20.9%
3.5%

-2.5%
-1.4%
-0.5%

-0.05
-0.03
-0.03

.00
.02
.03

20,860
20,860
20,860

Risk Factor Matching
Any Recidivism
Felony Recidivism
Violent Felony Recidivism

39.9%
23.1%
5.5%

44.9%
26.6%
5.9%

-5.0%
-3.5%
-0.4%

-0.10
-0.08
-0.02

.00
.00
.41

8,778
8,778
8,778

35.1%
20.5%
3.4%

37.5%
21.7%
3.9%

-2.4%
-1.2%
-0.5%

-0.05
-0.03
-0.02

.00
.05
.05

20,826
20,826
20,826

Method and Outcome

Note: Results are from separate logistic regressions for each recidivism outcome, with effects calculated at the means of the samples. Effect sizes are calculated using the
arcsine transformation described in: M.W. Lipsey and D. Wilson. (2001) Practical meta-analysis. Thousand Oaks: Sage Publications, Table B10, formula (22).

10

Exhibit 10

Group Comparability for the Risk Factor Matched Groups
(Descriptive Statistics for the Matched Groups)
Higher-Risk Offenders (RMA&B)

Demographic Variables
Age at Release
Male %
Black %
Asian %
Native Amer %
Risk for Reoffense Scales
Prison Sentence %
Static Felony Risk Score
Static NonDrug Fel Score
Static Violent Fel Score
Unadjusted Recidivism
Felony & Misdemeanor %
Felony %
Violent Felony %
Sample Size

OAA
Group

Pre-OAA
Group

32.8
86.7%
21.0%
2.1%
3.9%

Statistically
Significant
Difference?

Lower-Risk Offenders (RMC&D)

Statistically
Significant
Difference?

(p-value)

OAA
Group

Pre-OAA
Group

31.4
88.8%
19.1%
2.3%
3.7%

No (.97)
Yes (.00)
Yes (.03)
No (.61)
No (.58)

31.9
73.2%
12.4%
2.6%
2.5%

31.8
74.7%
15.2%
2.5%
2.7%

No (.68)
Yes (.02)
Yes (.00)
No (.93)
No (.37)

30.6%
67.7
40.6
44.5

28.5%
67.6
40.5
44.5

Yes (.04)
No (.99)
No (.94)
No (.97)

11.3%
55.43
40.33
46.75

19.8%
55.42
40.82
42.35

Yes (.00)
No (.96)
No (.95)
No (.92)

50.7%
31.9%
11.2%
4,389

50.0%
32.5%
9.6%
4,389

No (.52)
No (.54)
Yes (.02)

37.9%
23.6%
4.9%
10,413

39.3%
24.6%
5.9%
10,413

Yes (.04)
No (.12)
Yes (.00)

(p-value)

Data Sources: Washington State Department of Corrections and Administrative Office of the Courts. The OAA group includes those offenders released to
the community during Fiscal Year 2002. The pre-OAA group includes those offenders released to the community from October 1, 1998, to June 30, 2000.

Exhibit 10 examines descriptive statistics for the
matched groups created using the risk factor
approach to matching. We report this group
because it is our preferred modeling approach at
this juncture in the evaluation of the OAA. Exhibit
10 shows that after matching RMA and RMB
offenders with offenders in the pre-OAA period—
matching on the three risk scores for felony, nondrug felony, and violence—the two groups are
quite comparable, although several important and
statistically significant differences remain. The
offenders in the OAA group have slightly fewer
males, slightly more blacks, and a slightly higher
percentage of previous prison sentences. As
expected, there were no significant differences on
the static risk scores, since those are the variables
on which we created the matched samples. The
matched groups for the lower-risk RMC and RMD
offenders were similarly comparable, although
there were also significant differences on the same
variables, but in the opposite direction. For both
these matched groups, we rely on the multivariate
controls to adjust for the differences in estimating
the average treatment effect for the OAA.
The logistic regression results for felony
recidivism for the risk factor matching model is
shown in Exhibit 11 to demonstrate the general
approach taken in this analysis. The regression
results for each of the other models shown in
Exhibits 8 and 9 use the same set of covariates
as shown in Exhibit 11. Detailed regression
results for each model are available from the
authors.

Exhibit 11

Regression Output for the
Risk Factor Matched Sample: High-Risk Group
Dependent Variable: Felony Recidivism
Method: ML - Binary Logit (Quadratic hill climbing)
Included observations: 8778 after adjustments
QML (Huber/White) standard errors & covariance
Variable
CONSTANT
OAA (AB)
MALE
BLACK
ASIAN
NATAMER
ETHOTHER
AGE
FELONYSCORE
FELONYSCORE^2
FELONYSCORE^3
NONDRUGFELONYSCORE
NONDRUGFELONYSCORE^2
VIOLENCESCORE
PRISON
LSI41
LSI3240
ADULTVIOLENCE
LSI10_VIOLENCE
LSI47
LSI48
LSI49
LSI50
HOMICIDE
Mean dependent var
S.E. of regression
Sum squared resid
Log likelihood
Restr. log likelihood
LR statistic (23 df)
Probability(LR stat)

Coefficient

Std. Err

Z-Statistic

Prob.

-4.405
-0.186
0.035
0.484
-0.200
0.024
-0.555
-0.012
0.090
-0.001
0.000
0.054
0.000
0.003
0.278
0.766
0.547
-0.206
-0.189
-0.053
-0.005
-0.236
-0.133
-0.301

0.467
0.061
0.085
0.061
0.192
0.127
0.249
0.003
0.019
0.000
0.000
0.014
0.000
0.002
0.056
0.084
0.065
0.059
0.074
0.084
0.059
0.067
0.062
0.220

-9.436
-3.061
0.413
7.900
-1.042
0.187
-2.225
-4.111
4.755
-4.189
3.997
3.907
-2.810
1.490
4.996
9.076
8.432
-3.500
-2.559
-0.634
-0.078
-3.513
-2.134
-1.367

0.000
0.002
0.680
0.000
0.297
0.852
0.026
0.000
0.000
0.000
0.000
0.000
0.005
0.136
0.000
0.000
0.000
0.001
0.011
0.526
0.937
0.000
0.033
0.172

0.322055 S.D. dependent var
0.435748 Akaike info criterion
1662.175
Schwarz criterion
-4880.065 Hannan-Quinn criter.
-5516.173
Avg. log likelihood
1272.216 McFadden R-squared
0.000000

0.467291
1.117354
1.136711
1.123949
-0.555943
0.115317

11

Exhibit 12

DOC’s Criteria for Risk Management Levels A – D
RISK MANAGEMENT A (RMA)
Offenders will be assigned Risk Management Level A if they
meet one or more of the following criteria:
1)

An LSI-R score of 41 or over and have been convicted of a
violent crime

2)

Level III sex offenders

3)

Designated as Dangerous Mentally Ill Offender (DMIO)

4)

Do not meet the above criteria but through documented
history meet any of the following:
a) Committed a violent act involving a victim who was
unknown to the offender.
b) Committed a predatory act of violence directed toward
strangers or individuals with whom a relationship has
been established or promoted for the primary purpose of
victimization.
c) Committed a violent act where the victim was vulnerable
due to age (5 years or younger), physical condition,
mental disability, or ill health where the victim was
incapable of resisting the offense, or with significantly
impaired ability to protect him/herself.
d) Committed violent acts or made threats of violence
directed toward institutions or groups in the community,
including, but not limited to, religious, ethnic, or racial
groups.
e) History of violent acts and continue to exhibit behavior
demonstrating a current threat to the victim(s) including,
but not limited to, domestic violence or sexual offenses.

RISK MANAGEMENT B (RMB)
Offenders who do not meet the criteria to be assigned to RMA,
will be assigned Risk Management Level B if they meet one or
more of the following criteria:
1)

An LSI-R score of 41 or over;

2)

An LSI-R score of 32 to 40 and have been convicted of a
violent crime;

3)

Level II sex offenders; and/or

4)

Offenders with identified high level of needs including, but not
limited to, those who are developmentally disabled or
seriously mentally ill as determined by a qualified service
provider.
RISK MANAGEMENT C (RMC)

1)

Offenders who do not meet the criteria to be assigned to
RMA or RMB, with a LSI-R score of 24 to 40, will be assigned
to Risk Management Level C.

2)

Level I sex offenders will be assigned to RMC.

RISK MANAGEMENT D (RMD)
Offenders who do not meet the criteria to be assigned to RMA,
RMB, or RMC with an LSI-R score of 0 to 23 will be assigned to
RMD.

The Institute’s
firstState
four OAA reports are available on our website: http://www.wsipp.wa.gov/
Washington
Institute for
R. Barnoski, S. Aos, Washington’s Offender Accountability Act: An Analysis of the Department of
Public Policy
Corrections’ Risk Assessment. Olympia: Washington State Institute for Public Policy, 2003.

The Washington Legislature created the Washington State Institute for Public Policy in 1983. A Board of Directors—representing the
legislature,
theWashington's
governor, and public
universities—governs
the Institute
and guidesand
the development
of all activities.
The
Institute’s
mission
S. Aos,
Offender
Accountability
Act: Update
Progress Report
on the
Act's
Evaluation.
is to carry out practical research, at legislative direction, on issues of importance to Washington State.

Olympia: Washington State Institute for Public Policy, 2003.

S. Aos, Washington's Offender Accountability Act: An Evaluation of the Department of Corrections' Risk
Management Identification System. Olympia: Washington State Institute for Public Policy, 2002.
S. Aos, P. Phipps, R. Barnoski, R. Lieb, Evaluation Plan for the Offender Accountability Act. Olympia:
Washington State Institute for Public Policy, 2000.

Document No. 05-07-1202
Washington State
Institute for
Public Policy
The Washington Legislature created the Washington State Institute for Public Policy in 1983. A Board of Directors—representing the
legislature, the governor, and public universities—governs the Institute and guides the development of all activities. The Institute’s mission
is to carry out practical research, at legislative direction, on issues of importance to Washington State.

12

 

 

CLN Subscribe Now Ad 450x600
Advertise here
PLN Subscribe Now Ad