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

November 2007

DOES PARTICIPATION IN WASHINGTON’S
WORK RELEASE FACILITIES REDUCE RECIDIVISM?

In 2007, the Washington State Legislature
passed an adult offender reentry initiative with
the goal of reducing recidivism.1 As part of that
effort, the Legislature directed the Washington
State Institute for Public Policy (Institute) to
evaluate the Department of Corrections’ (DOC)
work release program to determine its impacts
on key outcomes, such as recidivism.

Summary
Work release facilities enable certain offenders
under the jurisdiction of the Washington State
Department of Corrections (DOC) to serve up to
six months of their prison sentence in a
residential facility while employed in the
community. Today, there are 15 work release
facilities that house about 700 offenders
statewide.

This report includes findings from our
recidivism analysis. Recidivism is defined as
any offense committed after release to the
community that results in a Washington State
conviction. We analyzed three types of
recidivism: felony, violent felony, and total
recidivism, which includes misdemeanors.

The Institute was directed by the 2007
Legislature to evaluate whether participation in
Washington’s work release facilities impacts
recidivism. Our time period of study includes
offenders who released from DOC between
January 1998 to July 2003.

A future report will assess the impact of
offender’s participation in work release on
employment.

Findings from the study indicate participation in
Washington’s work release facilities:
• lowers total recidivism, by 2.8 percent
• has a marginal effect on felony recidivism;
by 1.8 percent; and
• has no effect on violent felony recidivism.

Legislative Direction
The Institute was directed to:
•

Evaluate DOCs’ work release program on
key outcomes.

•

Identify the programs that show the
greatest effectiveness on key outcomes
and which services should be provided
for effective reintegration.

•

Of the 15 facilities operating in 1998 to 2003, we
found that participation in some contributes to
greater reductions in recidivism than others.
We ran our economic model to determine if the
marginal benefits of work release outweigh the
cost. Based upon the felony recidivism findings,
participation in work release generates $3.82 of
benefits per dollar of cost. The benefits (about
$2,300 per work release participant) stem from
the future benefits to taxpayers and crime
victims from the reduced recidivism.

Examine work release practices inside
and outside of Washington State.

This report is divided into three sections based
upon these legislative directives.

For more information, please contact Elizabeth Drake at
(360) 586-2767 or ekdrake@wsipp.wa.gov.
1

ESSB 6157, Chapter 483, Laws of 2007.

1

and clerical staff, depending upon each
facility’s contract. DOC staff include the work
release supervisor, case management staff,
and administrative support.

Section I: Evaluation of DOC’s Work
Release Program
Our first legislative directive was to evaluate
DOC’s work release program on key
outcomes. For this study, we analyzed the
impact of work release on recidivism.

Offenders are responsible for finding a job
within about ten days of arrival at the facility
and are typically required to work 40 hours a
week. Some work release facilities have
established informal partnerships with local
employers. Many work release facilities have
a job specialist who helps offenders with
interviewing techniques, resume writing, and
job preparation. These specialists often come
from the local Employment Security
Department or are provided through the
contracted staff. Offenders are responsible for
their own transportation—typically public
transportation—to and from work.

DOC’s Work Release Program
Washington’s work release program was
created by the Legislature in 1967.2 Work
release facilities enable certain offenders
under the jurisdiction of DOC to serve up to six
months of their prison sentence in a residential
facility while employed in the community.3
Today, 15 work release facilities house about
700 offenders statewide.4
DOC eligibility criteria restrict who participates
in work release. Current statewide policy5
prohibits offenders convicted of First Degree
Murder and offenders convicted of First
Degree Rape6 from participating in work
release unless approved by DOC’s screening
committee.

By law, wages earned by the offender can be
deducted for the following reasons: vocational
training expenses, room and board, financial
support for dependents, legal financial
obligations, payments to creditors, and
personal savings to be used upon release.8

In addition, each work release facility has its
own local eligibility criteria. For example, some
facilities house both male and female
offenders, while others are gender specific.
Some facilities serve as a therapeutic
community for chemically dependent
offenders, while others do not. In addition,
some work release facilities may accept some
sex offenders, only treated sex offenders, or no
sex offenders. Appendix A contains more
detailed information on facility characteristics.

Evaluation Design
The best way to determine a program’s
effectiveness is to compare the outcomes of
offenders who participate in work release with
similar offenders who do not participate. In an
ideal research setting, offenders would be
randomly assigned to a work release or
comparison group. We did not have that
option for this evaluation; thus, we constructed
an appropriate comparison group by
minimizing differences between the groups and
adjusting statistically when differences
remained.

DOC hires contractors to provide custodial
staff for security.7 In addition, contractors
typically provide food service, maintenance,

Evaluations that measure recidivism are
“retrospective” by design, which means that we
did not evaluate the effectiveness of work
release as it operates today. The study groups
selected were as recent as possible, while
allowing sufficient time for a 36-month followup period. Our time period of study includes
offenders who were released from prison

2

RCW 72.65
RCW 9.94A.728 (6)
4
DOC considers Lincoln Park and Rap House one work release
facility. In this report, these facilities are analyzed separately,
because they serve two different populations. Thus, when
counted separately, there are 16 work release facilities.
5
DOC Policy Directive 300.500.
6
Policy prohibits participation by offenders convicted of First
Degree Rape who are within their first three years of
confinement.
7
The Tri-Cities Work Release is the only work release facility in
Washington that is entirely state-operated.
3

8

2

RCW 72.65.050

between January 1, 1998 and July 31, 2003
and we measured recidivism through
September 2007.

Technical Note:
Measures of Association Strength
The AUC
A statistic called the area under the receiver
operating characteristic (AUC) is the best
measure for determining how accurately a
characteristic predicts an event, such as
participation in work release or recidivism.†

Selecting the Work Release Group
There were 35,475 offenders released from a
DOC facility from January 1, 1998, through
July 31, 2003. Of these, 32 percent, or 11,413
offenders, participated in work release.9 This
group of offenders became our study group for
the evaluation. The remaining 24,062
offenders did not participate in work release.

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 to no predictive accuracy,
.600s weak, .700s moderate, and above .800,
strong predictive accuracy.

To understand how work release participants
differ from the general prison population, we
compared the two groups on key
characteristics such as criminal history, offense
seriousness, sentence length, and
demographics.

Standardized Estimate
A statistic called the standardized estimate is
used to compare the relative strength between
the dependent variable, such as participation in
work release or recidivism, and certain
characteristics. Negative numbers indicate a
negative association and positive numbers
indicate a positive association with the
dependent variable. The larger the estimate is,
the greater the association.

Exhibit 1 shows that offenders who participate
in work release tend to have more criminal
history, but less serious and less violent
offenses than the general prison population.
Thus, offenders who participate in work
release have shorter sentences and spend
less time in prison than the general prison
population. There are more African American
and female offenders in the work release
group, and they are also slightly older than the
general prison population.

†

V. Quinsey, G. Harris, M. Rice, & C. Cormier. (1998).
Violent offenders: Appraising and managing risk. Washington
D.C.: American Psychological Association; P. Jones. (1996).
Risk prediction in criminal justice. In A. Harland (Ed.),
Choosing correctional options that work. Thousand Oaks, CA:
Sage, pp. 33–68.

We also conducted a logistic regression
analysis which included all characteristics in
the model to see if, collectively, these
characteristics are associated with participation
in work release. Results of the full model,
however, indicate that the characteristics have
a weak association with participation in work
release.

To determine which characteristics are
predictive of participation in work release, we
ran a logistic regression analysis using the
characteristics shown in Exhibit 1. A statistic
produced from the logistic regression, called
the area under the receiver operating
characteristic (AUC), helps determine how
strongly these characteristics are associated
with participation in work release (for a further
explanation of the AUC, see the sidebar on this
page). The AUCs in Exhibit 1 indicate that,
individually, none of the characteristics is
predictive of participation in work release.

In summary, based upon the characteristics in
Exhibit 1, it is difficult to predict who
participates in work release.

9

This figure includes offenders who spent at least one full day in
work release; it does not include offenders who transferred
through a work release facility en route to another facility on the
same day. This figure also excludes offenders who entered
DOC for sanctioning purposes for a violation of community
supervision.

3

Exhibit 1
Work Release Group Versus the General Prison Population:
Key Characteristics

Work
Release
Group

Offenders
Released
Between 1998 to
2003 (excluding
work release
Group)

11,413

24,062

Means
Felony risk scorea
Non-drug risk scorea
Violent risk scorea
Prior adult felony adjudications
Minimum sentence yearsb
Maximum sentence yearsb
Mandatory sentence days
Actual prison days
SRA offender scorec
SRA severity levelc
Age at release

74
52
32
4.0
3.0
3.2
20
514
4.4
4.6
34

Percentages
Male
Caucasian American
African American

82%
69%
26%

Logistic regression AUC

0.678

Number of Releases

Significance
Level

AUC

Std.
Estimate

71
52
34
3.3
3.1
3.2
29
701
3.8
5.0
33

0.00
0.38
0.00
0.00
0.00
0.09
0.01
0.00
0.00
0.00
0.00

0.547
0.511
0.521
0.566
0.500
0.500
0.504
0.575
0.551
0.536
0.560

0.082
0.006
-0.066
0.118
-0.023
-0.011
-0.019
-0.161
0.103
-0.063
0.090

89%
73%
20%

0.00
0.00
0.00

0.533
0.524
0.532

-0.104
-0.058
0.083

a

The risk scores shown are calculated based upon the scoring methods of DOC’s static risk instrument. For more
information, see: R. Barnoski & E. Drake (2007). Washington's Offender Accountability Act: Department of Corrections'
static risk instrument. Olympia: Washington State Institute for Public Policy, Document No. 07-03-1201.
b
110 observations were omitted from the sample for this statistic because they were outliers—offenders sentenced to
life in prison or death.
c
The Sentencing Reform Act (SRA) of 1984 established a “sentencing grid,” which is based upon the offender score
and offense severity level. The offender score is calculated primarily on prior convictions (0 to 9 plus) and the severity
level is reflective of the current offense of conviction and ranges from a low of 1 to a high of 16.

Based on these DOC screening criteria, the
following offenders were excluded from
participation in work release:

Selecting the Comparison Group
The next step of the evaluation was selecting
an appropriate comparison group. To do this,
we first reviewed historical DOC work release
screening policies used during our study
period—1998 through 2003—to determine who
was eligible to participate.

•

10

Those convicted of First Degree Murder,
First Degree Rape, First Degree Assault,
First Degree Assault of a Child, First
Degree Kidnapping, Homicide by Abuse,
Second Degree Murder, First Degree
Manslaughter.10

Unless approved by DOC’s Headquarter’s Community
Screening Committee.

4

•

Those with an out-of-state release plan;
those wanted by law enforcement for
another felony; and offenders who were
released to the custody of federal
authorities, such as the Immigration and
Customs Enforcement.

•

Offenders with violations or infractions
that were violent, such as assault or sex
offenses.

criminal history but less serious and less
violent offenses than the comparison group.
Offenders who participated in work release had
longer sentences and spend more time in
prison than the comparison group. There were
fewer Caucasians and more African Americans
in the work release group, and they were also
slightly older than the comparison group.
These differences may reflect the local
screening policies of work release facilities.
Unfortunately, however, some of these local
criteria are unobserved to the researchers.

In addition to the exclusion criteria above,
offenders also had to meet the following
inclusion criteria to participate in work release:
•

Must have had six months or less until
their early release date.

•

Must have had the lowest custody
classification level—Minimum Custody
Level 1.

To further test the differences between the
study groups, we ran logistic regression
analyses to determine if we can predict who
participates in work release based upon the
characteristics in Exhibit 2. Shown in the
exhibit are the individual AUCs—the strength
of association between the characteristic and
participation in work release. The AUCs
indicate that, individually, none of the variables
have an association with participation in work
release.

The comparison group selected for this study
included offenders who did not participate in
work release, but met all of the aforementioned
eligibility criteria according to DOC statewide
policy. Offenders in the comparison group
were released from prison during the same
time period as the work release group—
January 1, 1998, through July 31, 2003. A
total of 3,913 offenders were included in our
comparison group.11

We also conducted a logistic regression
analysis, which included all of the
characteristics in the model to see if,
collectively, they are associated with
participation in work release. Results of the full
model indicate that the characteristics have a
weak association with participation in work
release (AUC=.668).

Offenders who refuse to participate in work
release, by DOC policy, have their custody
level overridden to a higher level. Due to this
criterion, offenders in both the work release
and the comparison groups have volunteered
to participate in the program, therefore
reducing the possibility of self-selection bias
threatening the validity of the study design.

If any statistical bias remains in our multivariate analysis, it would be in the direction of
showing work release to be more effective at
reducing recidivism.

We compared the work release group with the
comparison group to estimate how different the
two groups are on key characteristics such as
criminal history, offense seriousness, sentence
length, and demographics. Exhibit 2 shows
that there are some statistically significant
differences between characteristics of the work
release and comparison groups. Offenders
who participated in work release had more
11

Appendix B details the selection process for the study groups.
It also shows how many offenders would have been eligible for
work release under DOC’s current screening policy.

5

Exhibit 2
Work Release Group versus Comparison Group: Key Characteristics

Number of Releases

Work
Release
Group
11,413

Comparison
Group

Significance
Level

AUC

Std.
Estimate

3,913

Means
Felony risk scorea
Non-drug risk scorea
Violent risk scorea
Prior adult felony adjudications
Minimum sentence yearsb
Maximum sentence yearsb
Mandatory sentence days
Actual prison days
SRA offender scorec
SRA severity levelc
Age at release

74
52
32
4.0
3.0
3.2
20
514
4.4
4.6
34

74
54
34
3.8
2.5
2.7
12
498
4.0
4.3
33

0.91
0.00
0.00
0.00
0.00
0.00
0.08
0.16
0.00
0.00
0.00

0.547
0.511
0.521
0.566
0.500
0.500
0.504
0.575
0.551
0.536
0.560

0.082
0.006
-0.066
0.118
-0.023
-0.011
-0.019
-0.161
0.103
-0.063
0.090

Percentages
Male
Caucasian American
African American

82%
69%
26%

82%
72%
22%

0.31
0.00
0.00

0.533
0.524
0.532

-0.104
-0.058
0.083

Logistic regression AUC

0.668

a

The risk scores shown are calculated based upon the scoring methods of DOC’s static risk instrument. For
more information see: R. Barnoski & E. Drake (2007). Washington's Offender Accountability Act: Department
of Corrections' static risk instrument. Olympia: Washington State Institute for Public Policy, Document No.
07-03-1201.
b
110 observations were omitted from the sample for this statistic because they were outliers—offenders
sentenced to life in prison or death.
c
The Sentencing Reform Act (SRA) of 1984 established a “sentencing grid,” which is based upon the
offender score and offense severity level. The offender score is calculated primarily on prior convictions (0 to
9 plus) and the severity level is reflective of the current offense of conviction and ranges from a low of 1 to a
high of 16.

At-Risk Date and Follow-up Period for
Recidivism

Defining Recidivism
Recidivism is defined as any offense
committed after release to the community that
results in a Washington State conviction.
Three types of recidivism are reported:
•

Violent felony convictions;

•

Felony convictions, including violent
felonies;

•

Total recidivism, including
misdemeanors, felonies, and violent
felony convictions.

Offenders who participate in work release are
partially confined, meaning they are free in the
community during working hours but are
confined in a facility at night.12 Since offenders
are not fully at-risk to reoffend, two
complexities are created in conducting
recidivism analysis for the work release group.
First, we needed to determine if the at-risk date
should be the date offenders started work
release or the date they were released from
DOC confinement into the community. We
12

RCW 9.94A.731 defines partial confinement as confined in a
facility for at least eight hours per day.

6

Exhibit 4
Timing of New Felony Conviction for People
Who Recidivated by Month

examined whether work release participants
committed offenses while on work release.
Exhibit 3 shows that 2.6 percent, or 296
offenders, committed a felony offense while on
work release. Thus, the decision was made to
use the work release start date as the
recidivism at-risk date.13 The at-risk date for
the comparison group is the date of release
from prison.

Cumulative Percent Recidivism

100%

Exhibit 3
Percent Actual Recidivism
While On Work Release

Work Release
Comparison

80%
60%
40%
20%
0%
0

Type of
Recidivism
Total
Felony
Violent Felony

Total
Number
11,413
11,413
11,413

Number
Recidivated
368
296
19

Percentage
Recidivism
3.2%
2.6%
0.2%

6

12
18
24
Months Follow-up

30

36

Recidivism Rates14
We used multivariate regression analysis to
adjust for observed differences that exist
between the study groups (see Exhibit 2). This
enables us to calculate adjusted recidivism
rates, which gives a clearer picture of whether
work release affects recidivism.15

Second, since the at-risk date is the date
offenders started work release, our study
group was still partially confined during the
follow-up period. Thus, we needed to
determine an appropriate follow-up period for
the work release group. In order to address
this issue, we analyzed the timing of those
people who recidivated.

Exhibit 5 displays multivariate-adjusted
recidivism rates for felony, violent felony, and
total recidivism at the 36-month follow-up (The
results of the logistic regression analyses for each
type of recidivism are shown in Appendix C).

Exhibit 4 shows there was a suppressed
recidivism rate for work release offenders
within the first 6 months of their at-risk date. It
should be noted that offenders can serve up to
6 months on work release; however actual
length of stay in work release is an average of
104 days.

•

To account for this suppression, we extended
the recidivism follow-up period for each
individual offender in the work release group
by the number of days they spent in work
release. For example, if an offender spent 3
months on work release, that offender’s
recidivism follow-up was extended to 39
months.

14

Total recidivism—for offenders who
participated in work release, we found
that 58 percent had a new conviction for
any offense. Without work release, we
calculated that 61 percent were
reconvicted for any new offense within
three years—a statistically significant
difference.

It may also be of interest to note that 22 percent of the work
release group released from prison rather than work release,
meaning that these offenders were unsuccessful in work release.
15
Specifically, we used logistic regression and included the
independent variables listed in Exhibit 2. The recidivism rate for
the comparison group was adjusted using the odds ratio from the
logistic regression. For example, using the actual recidivism rate
of the work release group (45 percent) and the effect size
(-0.0736), we do the following calculation to get a recidivism rate
of 47 percent for the comparison group: (45/(1-45))/exp(-0.0736)/
(1+(45/(1-45))/exp(-0.0736)).

13

Ninety-three percent of the offenders who committed a felony
while on work release were classified by DOC’s risk assessment
tool as high-risk (drug, property, or violent) offenders.

7

•

•

similar to other adult corrections’ programs we
have reviewed.16 Further, these effects are
very stable over time.

Felony recidivism—for offenders who
participated in work release, we found
that 45 percent had a new felony
conviction. Without work release, 47
percent of offenders were convicted for a
new felony within three years.
Statistically, this is a marginally
significant difference (p=.1187).

After a 24-month follow-up, there was no effect
on violent felony recidivism. Exhibit 6 also
shows that, while there were large effects of
work release participation on violent felony
recidivism at the 4-month follow-up period,
those differences decreased by 12-months and
there was no difference by the 36-month
follow-up period.

Violent felony recidivism—there was
no difference between the two groups.

Exhibit 5
36-Month Adjusted Recidivism Rates for
Work Release and Comparison Groups
70%

Work Release
Comparison

-0.3

Total Recidivism
Felony
Violent Felony

58% 61%

40%

45%

47%

-0.2

30%

Effect Size*

Percent Reconvicted

60%
50%

Exhibit 6
Size of Recidivism Reductions for Each Followup Period by Type of Recidivism

-0.1

20%
10%

10% 10%

0%

0.0
Total*

Felony**
Type of Recidivism

Violent

Months Follow-up

0.1

*

Statistically significant difference at the p<.05 level.
**
Marginally significant at p=.1187.

4

12

24

36

*

A negative number indicates work release lowers recidivism
relative to the comparison group.

We also conducted multivariate-adjusted
recidivism rates measured at intervals from 4to 36-months follow-up to see how participation
in work release impacted recidivism over time.
Significant differences were found between the
work release and comparison groups for total
recidivism at every follow-up, but were found
only at the 12-month follow-up for felony
recidivism. For violent felony recidivism, a
significant difference was found only at the 4month follow-up period.

In summary, work release lowers recidivism
rates for total reconvictions, has a marginal
effect on felonies, and has no effect on
violent felonies.

Another way of portraying the effect of
participation in work release on recidivism is to
look at the “effect size.” Effect sizes measure
the degree to which a program has been
shown to change an outcome for program
participants relative to the comparison group.
Exhibit 6 shows that work release appears to
decrease non-violent recidivism. These
effects, for felony and total recidivism, are

16

S. Aos, M. Miller, & E. Drake (2006). Evidence-based public
policy options to reduce future prison construction, criminal
justice costs, and crime rates. Olympia: Washington State
Institute for Public Policy, Document No. 06-10-1201.

8

The first step in conducting a benefit-cost
analysis is to determine the cost of the
program versus the cost of not participating in
the program. This is calculated by multiplying
the total length of stay by the cost per person,
per day. Exhibit 7 shows that the average
total cost for an offender to participate in
Washington’s work release is $43,071
compared with $42,456 to not participate.18

Benefit-Cost Analysis
In addition to estimating whether work release
reduces recidivism, it is important to estimate
whether the benefits of participation in work
release outweigh the costs. We do this with
the economic model used in our other benefitcost analyses of corrections programs.17

Exhibit 7
Work Release versus Non-Participation: Costs Comparison
Work
Release

Comparison

Average Length of Stay in Days
Prison days—before start of work release

514

641

Prison days—after start of work release, but before DOC release datea
Days on work release
Total length of stay

38
104
656

0
0
641

Average prison cost in a minimum facility/per person, per dayb
Total prison cost

$ 66
$ 36,538

$ 66
$ 42,456

Average work release cost/per person, per dayb
Total work release cost
Total cost per person

$ 63
$ 6,533
$ 43,071

—
—

Cost

$ 42,456

a

We estimated the number of days in prison for the comparison group had they received the same average
minimum term as the work release group. Thus, 614 days is the product of 903 days (average minimum days for
work release group) multiplied by 71 percent (the percentage of the minimum term that the comparison group
served). The length of stay for the work release group is the actual number of days served in prison before the start
of work release.
b

Estimate from DOC, November 2007, in 2007 dollars.

18

17

If the number of days spent in prison after the start of work
release were zero, the total cost for work release group would be
$40,558.

See Aos, et al., 2006.

9

Section (1) of Exhibit 8 shows the effect size
for participants relative to the comparison
group.19 The effect size translates into a 1.4
percent reduction in crime, shown in section
(2). The economic benefits of participating in
work release are shown in section (3) of the
exhibit.

The second step in conducting a benefit-cost
analysis is to determine the monetary benefits
of participation in the program. Crime
reductions result in an economic benefit to
both taxpayers and to crime victims. It is
important to note that the benefits in this
section are based upon our findings of the
impact of participation in work release on
felony recidivism: it had a marginal effect.

The final step in conducting a benefit-cost
analysis is to compare the benefits to the costs
in order to determine the bottom-line estimate.
We find that participation in work release
generates $3.82 of benefits per dollar of cost.

To estimate the benefits of participation in the
program, we first estimated how the effect size
is related to future crimes avoided and how
much taxpayers and crime victims save when
crime is reduced.

Exhibit 8
Work Release Program Benefit-Cost Analysisa
(1)

(2)

(3)

(4)

(5)

Effect Size
Unadjusted effect size
Adjusted effect size

-0.040
-0.020

Effect on Crime Outcomes
Percentage change in crime outcomes

-1.4%

Benefits
Crime victim costs avoided
Taxpayer costs avoided
Total crime-related costs avoided
Costs
Total work release cost per program participant
Benefit-Cost
Benefit-Cost Ratio
Total benefits minus costs per participant
Internal Rate of Return on Investment

$1,161
$1,140
$2,301

$ 603

$ 3.82
$1,698
33%

a

For methods on adjusted effect size and benefit-cost analysis: S. Aos,
et al., 2006.

19

The adjusted effect size is also displayed. It reflects
assumptions we make concerning research design quality and
whether the program operated in the "real world.” See Aos, et
al., 2006.

10

Multivariate regression analysis was used to
adjust for observed differences between the
study groups. This enables us to adjust the
recidivism rate of the comparison group based
upon the control variables for each work
release group, which are shown in column
(4).22 If statistical significance was not
obtained, there was no difference in recidivism
rates between the work release and
comparison groups; thus, there was no need to
adjust the comparison group rate.

Section II: Identification of Facilities
With the Greatest Effectiveness on
Recidivism
Our second directive was to identify programs
that show the greatest effectiveness on key
outcomes. In this study, we identified the work
release facilities with the greatest reductions in
participant recidivism.
For this part of the analysis, ideally we would
select a comparison group for each individual
facility in order to determine whether
participation in that particular facility had a
significant impact on recidivism. This
approach could not be done because there
were not enough offenders in the comparison
group to select a comparison group for each
individual facility. Thus, we compared each
facility with the entire comparison group using
a logistic regression.20 A logistic regression
analysis was run for felony recidivism at the
36-month follow-up to determine how individual
work release facilities impact felony recidivism
relative to the comparison group.

It is difficult, quantitatively, to determine why
participation in these four facilities contributes
to a reduction in recidivism. Discussions with
DOC indicate that facilities with negative
parameter estimates are urban facilities
compared with those with positive estimates,
which tend to be rural facilities. Individual
facility findings may become clearer after the
employment analysis is conducted.

Exhibit 9 summarizes the results of the
analysis. The number of offenders who
participated in each work release facility is
displayed in column (1).21
A negative parameter estimate, shown in
column (2), indicates a decrease in felony
recidivism relative to the comparison group.
A positive estimate indicates an increase in
recidivism. It is also important to note the
significance level of the individual facilities.
From January 1998 through July 2003,
participants in Rap House, Brownstone,
Bishop Lewis, and Progress House had
significantly reduced felony recidivism rates;
participants in Peninsula had a significantly
increased felony recidivism rate.
Column (3) shows the actual recidivism rates
for each of the work release facilities.
20

When using logistic regression, one variable of mutually
exclusive independent variables, must be omitted to serve as the
“reference category.” For this analysis, the comparison group
serves as the reference category.
21
The total number of offenders in column (1) is more than the
total number of offenders in the work release group (11,413)
because offenders participated in multiple work release facilities.

22

Control variables included in the logistic regression analysis
are shown in Technical Appendix C.

11

Exhibit 9
Logistic Regression for 36-Month Adjusted Felony Recidivism
By Work Release Facility

Facility

Number
Participating
(1)

Parameter
Estimate
(2)

Actual
Facility
Recidivism
(3)

Adjusted
Comparison
Recidivism
(4)

Statistically Significant (p<=.05)
Rap House
Brownstone
Bishop Lewis
Progress House
Peninsula

235
1,044
1,159
1,540
694

-0.446
-0.289
-0.178
-0.126
0.174

33%
43%
49%
49%
45%

43%
50%
53%
52%
41%

Not Statistically Significant
Ahtanum View
Bellingham
Clark County
Eleanor Chase House
Helen B. Ratcliff
Lincoln Park
Longview
Madison Inn
Olympia
Pioneer
Reynolds
Tri-Cities

766
411
503
411
601
466
811
502
466
5
1,594
383

-0.110
0.042
0.075
0.109
-0.168
-0.175
0.153
-0.075
0.183
0.781
-0.096
0.064

43%
45%
33%
35%
34%
45%
46%
52%
49%
40%
49%
46%

43%
45%
33%
35%
34%
45%
46%
52%
49%
40%
49%
46%

Logistic regression AUC
Number observations

0.837
15,309

12

Section III: Examination of Work
Release Practices

and a positive effect size indicates an increase
in recidivism.

Our final directive was to examine work
release practices inside and outside of
Washington State. In order to do this, we
conducted a systematic review of all the
literature that examines the impact of
participation in work release on recidivism.23 In
this section, we report the findings of studies
that have utilized a rigorous methodology.

Three of the four studies have found that
participation in work release reduces
recidivism.25 One study, which utilized the
strongest level of research design, random
assignment, found no difference between the
work release and the comparison groups.
The Institute has stated in previous reports that
more research needs to be conducted on work
release before it can be determined if
participation in work release reduces
recidivism because the findings are mixed and
there have been too few recent evaluations. 26

Exhibit 10 shows the adjusted effect size for
each study.24 The effect size measures the
degree to which a program has been shown to
change an outcome for program participants
relative to the comparison group. A negative
effect size indicates a decrease in recidivism

Exhibit 10
Rigorous Studies Evaluating the Impact of Participation of Work Release on Recidivism
Jeffrey &
Woolpert

LeClair &
Guarino-Ghezzi

Turner &
Petersilia

Waldo &
Chiricos

Study Information
Year published
Research design levela

1974
3

1991
3

1996
3

1977
5

Program Information
State
Number in work release
Number in comparison

California
109
92

Massachusetts
212
211

Washington
112
106

Florida
188
93

-0.172

-0.049

-0.049

0.021

Adjusted effect size

Citations:
1) Jeffrey, R., & Woolpert, S. (1974). Work furlough as an alternative to incarceration. The Journal of' Criminology, 65(3),
405-415.
2) LeClair, D. P., & Guarino-Ghezzi, S. (1991). Does incapacitation guarantee public safety? Lessons from the
Massachusetts furlough and prerelease programs. Justice Quarterly, 8(1), 9-36.
3) Turner, S. M., & Petersilia, J. (1996). Work release in Washington: Effects on recidivism and corrections costs. Prison
Journal, 76(2), 138-164.
4) Waldo, G. P., & Chiricos, T. G. (1977). Work release and recidivism: An empirical evaluation of a social policy. Evaluation
Quarterly, 1(1), 87-108.
a

Studies are rated based upon the Maryland scale of rigor—1 is the lowest quality and 5 is the highest quality, random
assignment. In our analysis of the literature, we only report findings of studies rated a 3 or higher.

23

For more details on our methodology of systematic reviews,
see: Aos et al. 2006.
24
The adjusted effect size reflects assumptions we make
concerning research design quality and whether the program
operated in the "real world". See Aos et al. 2006.

25

The study conducted by Jeffery & Woolpert was a jail work
release program. The remaining studies were prison work
release.
26
Aos et al. 2006.

13

Appendix A
Characteristics of Work Release Facilities

Work Release
Ahtanum View
Bellingham
Bishop Lewis

City
Yakima
Bellingham
Seattle

County Catchment Area
Yakima, Grant, Douglas, Chelan, Klickitat, Kittitas
Snohomish, Skagit, Whatcom, Island
King, Snohomish

Brownstone

Spokane

Adams, Asotin, Ferry, Garfield, Lincoln, Okanogan,
Pend Oreille, Spokane, Stevens, Whitman

Clark County

Vancouver

Clark, Skamania

Eleanor Chase House

Spokane

Adams, Asotin, Ferry, Garfield, Lincoln, Okanogan,
Pend Oreille, Spokane, Stevens, Whitman

Helen B. Ratcliff
Lincoln Park

Seattle
Tacoma

King, Snohomish
Pierce

Longview

Longview

Cowlitz, Whakiakum, Lewis, Grays Harbor, Pacific,
Clark overflow

Madison Inn

Seattle

King, Snohomish

Olympia
Peninsula
Pioneer
Progress House

Olympia
Port Orchard
Seattle
Tacoma

Thurston, Lewis, Mason, Grays Harbor, Pacific
Kitsap, Jefferson, Clallam
King, Snohomish
Pierce

Rap House

Tacoma

Pierce

Reynolds
Tri-Cities

Seattle
Kennewick

King, Snohomish, Island
Benton, Franklin, Walla Walla, Columbia, Garfield

14

Year
Operation
Began
1972
1975
1970

Current
Capacity
60
25
61

Special Characteristics of the Population
Served
Coed
Coed
Males

1988

72

Males

27

Coed

1993

23

Females

early 80’s

22
30

Females
Mentally ill offenders

1993

54

Coed

early 80’s

25

Males
Therapeutic community for chemically dependent

1978

25
60
N/A
75

Coed
Coed
Coed
Coed

early 80’s

20

Coed
Developmentally and physically disabled

1972

100
30

Coed
Coed

Appendix B
Study Group Selection Process from the Release Cohort
And DOC Eligibility Criteria

DOC Releases:
January 1998 through
July 2003
N=35,475

Participated in
Work Release
N=11,413

Did not participate
in Work Release
N=24,062

After excluding
offenders not
released in
Washington
N=19,393

After excluding
offenders who do not
meet current
eligibility criteria
N=19,316

After excluding
offenders who do not
meet historic eligibility
criteria:
N=18,901

After excluding
offenders who do not
meet custody eligibility
criteria
N=3,957

After excluding
offenders who do not
meet custody eligibility
criteria
N=3,913

15

Appendix C
Exhibit C1
Logistic Regression Results for 36-Month Follow-up:
Felony Recidivism
Standardized
Estimate

Odds
Ratio

Sig.
Level

Work release

-0.018

0.93

0.12

Felony risk score

0.124

1.01

0.00

Non-drug risk score

-0.148

0.99

Violent risk score

0.113

1.02

Age at release

-0.207

0.96

Prior adult felony adjudications

0.997

1.71

Male

0.070

1.39

Variable

Exhibit C2
Logistic Regression Results for 36-Month Follow-up:
Violent Felony Recidivism
Standardized
Estimate

Odds
Ratio

Sig.
Level

Work release

0.001

1.00

0.95

Felony risk score

-0.540

0.95

0.00

0.00

Non-drug risk score

-0.035

1.00

0.31

0.00

Violent risk score

0.733

1.13

0.00

0.00

Age at release

-0.270

0.95

0.00

0.00

Prior adult felony adjudications

0.476

1.29

0.00

0.00

Male

0.076

1.43

0.00

Variable

Caucasian

-0.021

0.92

0.07

Caucasian

-0.020

0.93

0.23

Actual prison days

-0.108

1.00

0.00

Actual prison days

-0.016

1.00

0.63

SRA severity level

0.065

1.04

0.00

SRA severity level

0.026

1.02

0.33

SRA offender score

-0.215

0.90

0.00

SRA offender score

-0.049

0.98

0.07

Minimum sentence years

0.209

1.00

0.00

Minimum sentence years

0.009

1.00

0.89

Maximum sentence years

-0.177

0.86

0.00

Maximum sentence years

0.030

1.03

0.61

Mandatory sentence days

-0.016

1.00

0.37

Mandatory sentence days

-0.023

1.00

0.44

Logistic regression AUC

0.845

Logistic regression AUC

0.843

Number observations

15,309

Number observations

15,309

Exhibit C3
Logistic Regression Results for 36-Month Follow-up:
Total Recidivism

Exhibit C4
Control Variables for Facility Logistic Regression
Results in Exhibit 9

Standardized
Estimate

Odds
Ratio

Sig.
Level

Variable

Work release

-0.028

0.89

0.01

Felony risk score

0.356

1.03

0.00

Non-drug risk score

-0.049

1.00

Violent risk score

0.192

Age at release

-0.072

Prior adult felony adjudications
Male

Variable

Standardized
Estimate

Odds
Ratio

Sig.
Level

Felony risk score

0.126

1.01

0.00

Non-drug risk score

-0.146

0.99

0.00

0.05

Violent risk score

0.112

1.02

0.00

1.03

0.00

Age at release

-0.203

0.96

0.00

0.99

0.00

Prior adult felony adjudications

1.000

1.71

0.00

0.614

1.39

0.00

Male

0.074

1.42

0.00

0.014

1.07

0.25

Caucasian

-0.033

0.88

0.01

-0.107

1.00

0.00
0.00

Caucasian

-0.058

0.80

0.00

Actual prison days

Actual prison days

-0.136

1.00

0.00

SRA severity level

0.071

1.04

SRA severity level

0.043

1.03

0.01

SRA offender score

-0.213

0.91

0.00

SRA offender score

-0.169

0.92

0.00

Minimum sentence years

0.218

1.00

0.00

Minimum sentence years

0.192

1.00

0.00

Maximum sentence years

-0.186

0.85

0.00

Maximum sentence years

-0.182

0.86

0.00

Mandatory sentence days

-0.019

1.00

0.31

Mandatory sentence days

0.002

1.00

0.90

Logistic regression AUC

0.838

Number observations

15,309

For further information, contact Elizabeth Drake at
(360) 586-2767 or ekdrake@wsipp.wa.gov

Logistic regression AUC

0.837

Number observations

15,309

Document No. 07-11-1201

Washington State
Institute for
Public Policy
The Washington State 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.