Predicting the effect of substance abuse treatment on probationer recidivism
Transcript of Predicting the effect of substance abuse treatment on probationer recidivism
Predicting the effect of substance abuse treatment
on probationer recidivism
PAMELA K. LATTIMORE*Department of Criminology and Criminal Justice, University of South Carolina, Columbia,
SC 29208, USA
Health, Social, and Economics Research, RTI International**, 3040 Cornwallis Road, Research
Triangle Park, NC 27709-2194, USA
*corresponding author: E-mail: [email protected]
CHRISTOPHER P. KREBS, WILLEM KOETSE, CHRISTINE LINDQUIST and
ALEX J. COWELLHealth, Social, and Economics Research, RTI International**, 3040 Cornwallis Road, Research
Triangle Park, NC 27709-2194, USA
Abstract. Support for the effectiveness of substance abuse treatment to reduce substance use and
recidivism among populations supervised by the criminal justice system continues to grow in substance
abuse and criminal justice literature. Recent studies show that a variety of programs including the
Breaking the Cycle program and drug courts appear to result in improved outcomes for offenders. In this
paper, we examine the effect of non-residential substance abuse treatment on arrest. Our data are for
almost 134,000 Fdrug-involved_ individuals sentenced to probation in Florida between July 1995 and
June 2000. Nearly 52,000 of these individuals received non-residential substance abuse treatment, while
81,797 did not. Our approach is a methodologically simple one that entails stratifying our data by
treatment status, estimating logit and negative binomial models of arrest for each of the two datasets,
and then applying each model to both datasets. This approach, which requires that both groups include
subjects for whom treatment is appropriate, is analogous to using regression models to predict outcomes
for new values of independent variables. For each observation in the dataset, we use the models to
predict the expected outcomes for each individual under two scenarios Y receiving non-residential
treatment and receiving no treatment. Summing over these individual estimates provides an estimate of
the total numbers of arrests that would be expected under different levels of population exposure to
treatment. Results suggest that non-residential treatment reduced both the expected numbers of
individuals who recidivated (i.e., were arrested) and the expected total numbers of arrests in the 12 and
24 months following placement on supervision.
Key words: drug-involved offender, drug treatment, non-residential substance abuse treatment,
probation, recidivism
Introduction
As of December 31, 2003, there were nearly 6.9 million offenders under
correctional supervision in the United States Y nearly 4.8 million of whom were
**RTI is an independent organization dedicated to conducting innovative, multidisciplinary research
that improves the human condition.
Journal of Experimental Criminology (2005) 1: 159–189 # Springer 2005
being supervised by state or Federal probation (Glaze 2004). A large proportion
of these offenders are drug-involved. For example, the Arrestee Drug Abuse
Monitoring (ADAM) Program reported that approximately 63% of arrestees
in 2001 tested positive for at least one illicit drug (ADAM 2001). Providing
substance abuse treatment to drug-involved offenders is one of the most com-
mon approaches to preventing both drug use and associated crime, and extant
literature suggests that treatment can reduce recidivism and a variety of other
social consequences (Lurigio 2000; Pelissier et al. 2001; Banks and Gottfredson
2003).
The high prevalence of drug use among offender populations and the increase
in the proportion of offenders who are drug-involved have been well-documented
Y 24% of the 4 million adults on probation have a drug law violation (BJS
2003a) and approximately 20% of state prisoners and 55% of Federal inmates are
being held for drug offenses (BJS 2003b). The criminal justice system has
responded to the large numbers of drug-involved offenders through institution-
based strategies such as self-help groups and therapeutic communities, as well as
community-based strategies such as providing offenders with treatment through
drug courts, TASC (Treatment Alternatives to Street Crimes, also known as
Treatment Approaches for Safer Communities) models, or specialized caseloads
for supervising drug-involved probationers. Forty-one percent of adults on
probation in 1995 had drug or alcohol treatment imposed as a condition of their
supervision, and 17% reported having participated in a drug treatment program
during their probation sentence (BJS 1995). Although drug treatment is intended
to result in a variety of positive outcomes for individual offenders and society,
recidivism is the primary outcome of interest to the criminal justice system because
treatment provided or coordinated by the criminal justice system is intended to
reduce substance use and, therefore, the number of offenders re-entering the
system with drug or drug-related charges. Therefore, methodologically rigorous
evaluations of the impact of drug treatment on offender populations constitute
important contributions to the field.
This study examines the impact of non-residential substance abuse treatment on
the recidivism of drug-involved probationers under supervision in Florida.1 The
study focuses on nearly 134,000 individuals admitted to community supervision in
Florida between July 1, 1995 and June 30, 2000. Data describing the characteristics
of the offenders were obtained from the Florida Department of Corrections
(FLDC) and were augmented by up to six years of post-admission arrest data
provided by the FLDC and the Florida Department of Law Enforcement (FDLE).
Our analytic methods control for differences between those who did and those who
did not receive treatment and are used to predict the expected outcomes under
alternative treatment assumptions: the total numbers of individuals arrested and the
total numbers of arrests.
The next section provides a brief summary of the literature on the impact of
substance abuse treatment on recidivism, focusing on some important method-
ological issues. Subsequently, we provide a description of our sample, our study
methodology and our findings. In the concluding section, we discuss our findings
PAMELA K. LATTIMORE ET AL.160
and their policy relevance. Our results suggest that substance abuse treatment
reduces recidivism and that states might be able to dramatically reduce arrests
and, we presume, crime by providing treatment to a larger percentage of drug-
involved offenders who are under correctional supervision. The results also imply
a simple approach to costYbenefit estimation, allowing examination of the point(s)
at which investment in treatment may provide positive returns in terms of Farrest
savings._
Previous research
Several evaluations of drug abuse treatment have illustrated the effectiveness of
drug treatment in reducing crime. Much of this research has focused on the ef-
fectiveness of prison-based residential treatment programs, with several studies
finding reduced recidivism among treatment participants (Pelissier et al. 2001;
Rhodes et al. 2001), particularly when prison-based therapeutic communities are
combined with community-based aftercare (Martin et al. 1995; Wexler et al. 1995;
Prendergast et al. 1996). Because the current analysis considers the effectiveness of
community-based treatment among probationers, we have prioritized methodolog-
ically strong evaluations of treatment delivered to offenders under community su-
pervision in the review that follows. The treatment evaluated in these studies
includes a number of modalities and approaches; reviews of specific approaches to
delivering treatment to offenders under community supervision have been con-
ducted previously (Prendergast et al. 1995) and are beyond the scope of the current
paper.
Among the earliest evaluations of community-based drug treatment were
NIDA-funded studies that reported declines in recidivism rates as a result of
community-based drug treatment (representing a variety of treatment modalities),
including the Drug Abuse Reporting Program (DARP), the Treatment Outcome
Prospective Study (TOPS), and the Drug Abuse Treatment Outcome Study
(DATOS) (reviewed in Lurigio 2000).
One widely used approach for community-based treatment among offenders is
the TASC model, which establishes linkages between the criminal justice system
and community-based substance abuse treatment and emphasizes referral, moni-
toring, and case management. Although some studies classified by Chanhatasilpa,
Mackenzie, and Hickman in their 2000 review as methodologically weak have
found positive results for TASC-like programs when comparing completers to non-
completers (Van Stelle et al. 1994; Vito et al. 1993), a rigorous 1996 evaluation of
the effectiveness of TASC did not find evidence of reduced recidivism among
TASC participants compared to non-participants (Anglin et al. 1996). Other meth-
odologically strong evaluations of programs modeled after TASC have generated
similarly negative results (Rhodes and Gross 1997).
Given the increasing popularity of drug treatment courts as a strategy for
treating drug-involved offenders, much of the recent literature on the effectiveness
PREDICTING THE EFFECT OF SUBSTANCE ABUSE TREATMENT 161
of drug treatment in reducing recidivism focuses on treatment provided through
drug courts. However, because drug courts include components in addition to drug
treatment (e.g., increased monitoring, intensive case management and provision of
other social services, and judicial contact), it is not possible to isolate the
independent effects of treatment on recidivism. Although not all drug court
evaluations report reductions in recidivism for drug court participants (Miethe
et al. 2000; Wolfe et al. 2002), a recent summary of drug court evaluation studies
conducted between 1999 and April 2001 (Belenko 2001) identified several studies
reporting significantly lower recidivism rates among drug court participants than
comparison offenders (Brewster 2001; Deschenes et al. 2001; Truitt et al. 2000;
Gottfredson and Exum 2002). A study on the effectiveness of both graduated
sanctions and drug testing in the Washington, DC Adult Drug Court program
(Harrell 1998) found that the arrest rates among sanctions program participants
were substantially lower than arrest rates among defendants in a control group.
Several more recently published studies have also reported significantly lower
recidivism rates for drug court participants (Fielding et al. 2002; Turner et al. 2002;
Banks and Gottfredson 2003).
Among previous evaluations of substance abuse treatment, numerous predictors
of treatment success have been considered, with duration of, or retention in,
treatment being identified as an important factor (Simpson et al. 1997a, b, c; Flynn
et al. 1997; Hubbard et al. 1997). Several studies have reported that outpatient
treatment participants who remain in treatment for at least 90 days have
significantly lower arrest rates than those who terminate treatment earlier (Falkin
et al. 1999; Hubbard et al. 1984). However, the utility of retention indicators is
qualified by potential selection bias if those participants who stay in treatment
longer are those less likely to recidivate whether or not they received treatment.
Although many of the evaluations discussed above are methodologically
rigorous, design limitations in many offender-based drug treatment evaluations
limit the conclusions one is able to draw from the findings. These limitations
include short follow-up periods, small samples, limited populations, and in-
appropriate comparison groups. We briefly discuss each of these.
First, many drug treatment evaluations are characterized by short follow-up
periods, including, in many cases, follow up that does not extend beyond treat-
ment participation. Although it is necessary to assess whether treatment has
positive effects during or shortly after provision, studies that examine effec-
tiveness over multiple years are needed to determine whether the investment in
treatment provides benefits at the individual and societal level over an extended
period of time.
Second, many evaluations include small samples Y often for pragmatic reasons
of limited program capacity or evaluation funds. Although smaller studies may
have an advantage in monitoring treatment fidelity and minimizing variability
(Weisburd et al. 1993), the cost of small samples is limited statistical power to
detect small effects (e.g., a 0.1 or 0.2 reduction). Small effects may be the best
that current treatment approaches can obtain with criminal justice populations in
which substance abuse is often co-occurring with other problems. With small
PAMELA K. LATTIMORE ET AL.162
sample sizes, we reduce the likelihood that we will find treatment effects if they
exist, leading to false conclusions of treatment ineffectiveness. From a policy
perspective, however, even small effects may be substantial Y and worth the
investment of public resources Y if they can be extended to large populations, such
as supervised offenders.
Third, studies often focus on programs provided in a single jurisdiction (e.g.,
one city or county). Extrapolating results from these single-population and/or
single-program studies requires confidence that there is no unobserved factor
associated with success that is unique to the population or program.
Finally, many quasi-experimental evaluations use inappropriate comparison
groups and/or methods. Of particular concern are evaluations that use treatment
dropouts as comparison subjects. Since those who drop out of treatment are likely
to be Fself selecting_ not only into the dropout group but the outcome failure group,
selection bias poses a threat to the validity of findings.
Although many studies of the impact of treatment on recidivism suffer from
methodological limitations, several studies have addressed selection bias issues by
using a variety of strategies, including the Heckman approach (Pelissier et al.
2001; Rhodes et al. 2001), instrumental variable approach (Rhodes et al. 2001),
and propensity modeling (Banks and Gottfredson 2003). These studies constitute
improvements in methodological rigor, but additional studies that address the
issues that complicate research of this kind are needed to create a better un-
derstanding of the potential impact of treatment on the likelihood of offenders to
recidivate.
The current analysis builds on the analytic approach of Linster (2000) and
overcomes several of the methodological limitations already discussed. Specifi-
cally, the analyses use a very large dataset, thus assuring substantial statistical
power to find treatment effects if they exist. As the data include cases from across
Florida, the generalizability of the findings are greater than those based, e.g., on a
single program in one location. We examine results at both 12 and 24 months
following assignment to supervision, incorporating the period in which most who
fail are likely to fail (see, for example, Banks and Gottfredson (2003), in which
failure by re-arrest for drug court and control subjects stabilized at 16 months).
Finally, the analytic approach allows us to control Y in an approach that is
conceptually simpler and from an estimation standpoint Y for potential differences
between the treatment and comparison groups in determining the impact of non-
residential substance abuse treatment on recidivism. As noted by Linster (2000:
28), the approach provides B. . .one way by which estimated treatment effects may
be disentangled from risk-related population differences.^ In so doing, we are able
to answer the Fwhat-if_ question that distinguishes impact assessments from
prediction (see, e.g., Berk 1987). The output of the analyses provides estimates of
Farrest savings_ as a result of treatment that, in turn, can be used to identify when
treatment may be cost effective.
In addition to making methodological contributions to the field of substance
abuse treatment research, we fill a gap in the literature by focusing exclusively on
non-residential treatment. Very few studies, none of which was classified as
PREDICTING THE EFFECT OF SUBSTANCE ABUSE TREATMENT 163
methodologically strong in Chanhatasilpa et al.’s (2000) review, have evaluated
the effectiveness of outpatient treatment among offender populations. Although
outpatient treatment appears to be promising with regard to reducing recidivism
among offenders (Finigan 1996; Oregon Department of Corrections 1994),
rigorous studies are necessary in order to contribute to our understanding of its
effectiveness. The following section describes the subjects included in the analysis;
subsequent sections detail our methods and findings.
Subjects
We obtained administrative data from the Florida Department of Corrections
(FLDC) and the Florida Department of Law Enforcement (FDLE) on the 293,156
offenders who entered community supervision in the State of Florida between
July 1, 1995 and June 30, 2000. Supervision includes felony probation, drug-
offender probation, community control and community release. The FLDC data
included a variety of indicators of prior criminal history, the instant offense and
sentence, the nature and content of supervision, and substance use and treatment.
The FDLE data included all arrests reported by local and state police departments
for the subjects.
Because the focus of our research is the impact of drug treatment, we needed to
identify those on supervision who were most likely to be drug involved. There was
no direct measure of current drug use, such as an assessment, in the data; however,
there were a variety of measures that we could use as proxies for drug involvement.
Specifically, we classified an offender as Bdrug involved^ if he or she met one or
more of the following criteria:
� Ever arrested for a drug-related offense� Ever participated in a drug court program� Ever enrolled in drug-offender probation� Ever tested positive on a criminal justice system-administered drug test� Ever referred to substance abuse treatment by the criminal justice system
The resulting sample of drug-involved offenders included 148,076. We then
excluded those who received residential treatment and those with missing or
ambiguous data. The decision to exclude those who began residential treatment
(n = 11,319) during the instant supervision was based primarily on two consid-
erations. First, those in residential treatment facilities face a systematically
different risk (presumably lower) of recidivism (arrest) than those who are free
in the community. Second, those in residential placement are a heterogeneous
group including those whose initial placement is residential and those who have
Ffailed_ non-residential treatment. Of the 1,721 observations dropped because of
missing values on one or more variables, most (1,597) were excluded because they
could not be linked to the arrest file obtained from FDLE (i.e., to the file from
PAMELA K. LATTIMORE ET AL.164
which we generated our dependent variables, as well as some of our covariates).
Finally, we excluded 627 cases that were assigned to administrative probation, as
their supervision and contact requirements were minimal and 633 cases that were
classified as Bother race.^ The resulting data set contained data on 133,776 drug-
involved offenders who entered supervision in the State of Florida between July 1,
1995 and June 30, 2000.2
The characteristics of our subjects are shown in Table 1. We provide summary
measures for the complete sample of 133,776 and for our two subsamples Y the
51,979 who received non-residential substance abuse treatment and the 81,797 who
did not. We also include p-values for comparisons between those receiving and not
receiving treatment. The two groups are statistically different (at the 0.01 level) on
all measures except gender (about 81% of both groups are male) and one of the 21
judicial circuit indicator variables. Of course, with samples as large as ours, even
small differences reach statistical significance at conventional levels, but closer
examination suggests that the differences between the two groups are not only
statistically significant but meaningfully large on some variables that we would
expect to be related to recidivism outcomes.
Thus, we observe in our data what we might expect from non-experimental
data Y differences between our treatment and no-treatment groups along dimen-
sions that we might reasonably expect to be related to our outcome measures (e.g.,
arrest). Specifically, members of the no-treatment group have more prior arrests,
are less likely to be white, and are more likely to have prior supervisions/prison
terms Y all factors generally associated with higher recidivism rates. On the other
hand, our treatment group was supervised more intensely, which could be asso-
ciated with a higher recidivism rate, especially if supervision was not coupled with
treatment (Fulton et al. 1997; Petersilia 1998; Gendreau et al. 2000; Paparozzi
1999).
The average age of the sample is 31 years and there is little meaningful
difference between the average ages of our two subsamples. About 58% of the
treatment group was white, compared to 50% of the group that did not receive
treatment.
Current offense is the most serious offense associated with the supervision
term that led to inclusion in the sample. Current offense was categorized as
violent, property, drug, or Fother._ Comparing the two groups over this dis-
tribution, we see little difference in the proportions whose current offense is
violent or Fother_ Y 17% versus 16% with a violent offense and 7% versus 9%
with an Fother_ offense for the treatment and no-treatment groups, respectively.
There are substantial differences in the proportions with property and drug
offenses. Those in the treatment group were much more likely than those in the
no-treatment group to have a current drug offense (58% versus 47%, respectively),
while those in the no-treatment group were much more likely than the treatment
group to have a current property offense (28% versus 18%, respectively). We
should note, however, that since property Ftrumps_ drugs in the seriousness
hierarchy, those identified based on the most serious charge as property offenders
may also have a drug charge. Overall, the no-treatment group had a slightly
PREDICTING THE EFFECT OF SUBSTANCE ABUSE TREATMENT 165
Table 1. Subject characteristics.
Variables All subjects
(N = 133,776)
Treatment
(N = 51,979)
No treatment
(N = 81,797)
t-test,
p-value
n % n % n %
Demographic characteristics
Age 31.002 31.089 30.946 0.0084
Male 107,929 80.68% 42,036 80.87% 65,893 80.56% 0.1557
African American 52,799 39.47% 18,189 34.99% 34,610 42.31% G0.0001
Hispanic 10,383 7.76% 3,764 7.24% 6,619 8.09% G0.0001
White 70,594 52.77% 30,026 57.77% 40,568 49.60% G0.0001
Criminal history
Current violent offense 21,830 16.32% 8,820 16.97% 13,010 15.91% G0.0001
Current property offense 32,403 24.22% 9,503 18.28% 22,900 28.00% G0.0001
Current other offense 11,008 8.23% 3,713 7.14% 7,295 8.92% G0.0001
Current drug offense 68,534 51.23% 29,942 57.60% 38,592 47.18% G0.0001
Prior arrests 4.6244 3.9892 5.028 G0.0001
Prior drug arrests 1.0331 0.8593 1.1435 G0.0001
# prior supervisions 0.8438 0.7241 0.9198 G0.0001
# prior prison terms 0.3306 0.2591 0.3759 G0.0001
Supervision
DOP sup 24,901 18.61% 10,364 19.94% 14,537 17.77% G0.0001
FP sup 92,955 69.49% 35,285 67.88% 57,670 70.50% G0.0001
CC sup 15,920 11.90% 6,330 12.18% 9,590 11.72% 0.0125
Field contacts/year 2.346 3.0938 1.8712 G0.0001
Home contacts/year 6.864 8.4178 5.8762 G0.0001
Office contacts/year 8.951 11.441 7.3683 G0.0001
UA tests/year 3.593 5.2257 2.4666 G0.0001
Positive UAs/year 0.8534 1.0515 0.7275 G0.0001
Positive UA first month 10,778 8.06% 5,053 9.72% 5,725 7.00% G0.0001
Geographic indicator
Judicial circuit 1 5,995 4.48% 3,102 5.97% 2,893 3.54% G0.0001
Judicial circuit 2 3,637 2.72% 1,302 2.50% 2,335 2.85% 0.0001
Judicial circuit 3 2,493 1.86% 1,349 2.60% 1,144 1.40% G0.0001
Judicial circuit 4 5,344 3.99% 1,241 2.39% 4,103 5.02% G0.0001
Judicial circuit 5 5,903 4.41% 2,322 4.47% 3,581 4.38% 0.4384
Judicial circuit 6 13,260 9.91% 6,835 13.15% 6,425 7.85% G0.0001
Judicial circuit 7 5,686 4.25% 2,454 4.72% 3,232 3.95% G0.0001
Judicial circuit 8 2,946 2.20% 1,079 2.08% 1,867 2.28% 0.0121
Judicial circuit 9 11,525 8.62% 3,976 7.65% 7,549 9.23% G0.0001
Judicial circuit 10 6,329 4.73% 2,621 5.04% 3,708 4.53% G0.0001
Judicial circuit 11 11,181 8.36% 3,125 6.01% 8,056 9.85% G0.0001
Judicial circuit 12 4,334 3.24% 2,330 4.48% 2,004 2.45% G0.0001
Judicial circuit 13 13,202 9.87% 5,694 10.95% 7,508 9.18% G0.0001
Judicial circuit 14 3,130 2.34% 1,487 2.86% 1,643 2.01% G0.0001
Judicial circuit 15 5,423 4.05% 2,735 5.26% 2,688 3.29% G0.0001
Judicial circuit 16 1,645 1.23% 526 1.01% 1,119 1.37% G0.0001
Judicial circuit 17 15,819 11.82% 4,227 8.13% 11,592 14.17% G0.0001
Judicial circuit 18 5,384 4.02% 1,889 3.63% 3,495 4.27% G0.0001
Judicial circuit 19 3,499 2.62% 1,099 2.11% 2,400 2.93% G0.0001
Judicial circuit 20 5,196 3.88% 1,722 3.31% 3,474 4.25% G0.0001
Judicial circuit none 1,845 1.38% 864 1.66% 981 1.20% G0.0001
PAMELA K. LATTIMORE ET AL.166
more active criminal history than those who received treatment Y more prior arrests
(an average of 5 versus 4, respectively), more previous community supervision
terms (and average of 0.9 versus 0.7, respectively), and more prior prison terms (an
average of 0.4 versus 0.3, respectively). As criminal history is one of the most
reliable predictors of recidivism (Bonta et al. 1998; Gendreau et al. 1996), we
would expect that the no-treatment group would recidivate at higher rates than the
treatment group, other things being equal.
The subjects were supervised via one of three supervision mechanisms, which
in increasing order of intensity are felony probation, drug offender probation, and
community control. About 68% of the treatment group and 70% of the no-
treatment group were on felony probation (FP). Twenty percent of the treatment
group and 18% of the no-treatment group were on drug offender probation
(DOP), an intensive supervision that is supposed to include frequent contacts,
urine testing, and drug treatment. About 12% of both groups were on community
control (CC), which is the most intensive supervision status and includes curfew
and/or house arrest, along with other conditions. In addition to type of
supervision, we also had some measures of supervision intensity. Although the
proportions of each group assigned to these three types of supervision were about
the same, we see that those in the treatment group were supervised much more
intensively than those in the no-treatment group Y having more contacts with
probation officers in the field, at home, and at FLDC’s offices, as well as more
drug tests. As the treatment group was Fwatched_ more closely, we might expect
higher failure rates for this group consistent with previous research on intensive
supervision (Fulton et al. 1997; Petersilia 1998; Gendreau et al. 2000; Paparozzi
1999).
The data contain 21 judicial circuit indicators that are used to control for
important differences within the state that could otherwise confound the es-
timated relationship between treatment and recidivism.3 Judicial circuits vary in
law enforcement practice, criminal opportunity, culture, urbanicity, population
size, geographic size, and treatment availability. The judicial circuits also vary
with respect to receipt of treatment; the percentage receiving treatment ranged
from 23.2% to 54.1% across the judicial circuits. In three of the circuits, less
than 30% of the probationers went to nonresidential treatment, while more
than 50% participated in five circuits. Five circuits contain almost half (48.6%)
of our sample but a slightly smaller proportion of those who received treatment
(46.9%).
Although there are differences between the groups, Table 1 also suggests that
there is considerable overlap between the groups in terms of the measured
characteristics, which is important for our methodology (as we discuss in the next
section). Specifically, there is evidence to suggest that (at least some of ) those in
the no-treatment group were currently drug involved Y 18% were in the drug
offender probation (DOP) program, a supervision status that requires treatment;
7% had a positive UA during their first month of supervision; and 47% had a drug
offense related to their current community supervision. Those in the no-treatment
group had more prior drug arrests, on average, than those in the treatment group
PREDICTING THE EFFECT OF SUBSTANCE ABUSE TREATMENT 167
(1.1 versus 0.86, respectively). Finally, the no-treatment group averaged 0.73
positive drug tests per year during their supervision.4
As can be seen in Table 2, those who did not receive treatment were arrested
at a higher rate. During the first 12 months following placement on supervision,
33% of the no-treatment group was arrested for a felony compared with 21% of
the treatment group. Two years following placement on supervision, nearly 46%
of the no-treatment group had been arrested at least once compared with 33%
of the treatment group. A similar pattern is observed if we look only at arrests
for a felony drug charge Y at 12 months, 17% and 10% arrest rates were observed
for our no-treatment group and treatment group, respectively; and, at 24 months, the
rates were 25% and 16%.5 Members of the no-treatment group also had greater
numbers of arrests when compared with the treatment group Y an average of 0.49
felony arrests during the first 12 months of supervision compared with 0.29,
respectively.
The data in Table 2 provides our first potential indication that treatment may
reduce the recidivism of drug-involved probationers in Florida. But, because of the
differences in the two groups, we cannot be certain that the lower recidivism rates
are due to treatment and not to differences in the characteristics of the groups. In
other words, the treatment group may have had lower recidivism rates even
without treatment. In the next section, we describe our approach to disentangling
these issues.
Table 2. Outcome variables.
Variable All subjects
(N = 133,776)
Nonresidential
treatment
(N = 51,979)
No treatment
(N = 81,797)
t-test, p
n (%) n (%) n (%)
Any felony arrest past
12 months
37,915 28.3% 10,979 21.1% 26,936 32.9% G0.0001
Any felony arrest past
24 months
54,199 40.5% 16,957 32.6% 37,242 45.5% G0.0001
Any felony drug arrest past
12 months
18,740 14.0% 5,196 10.0% 13,544 16.6% G0.0001
Any felony drug arrest past
24 months
29,193 21.8% 8,550 16.5% 20,643 25.2% G0.0001
Mean # felony arrests past
12 months
0.4095 0.2860 0.4881 G0.0001
Mean # felony arrests past
24 months
0.7147 0.5206 0.8380 G0.0001
Mean # felony drug arrests
past 12 months
0.1732 0.1183 0.2081 G0.0001
Mean # felony drug arrests
past 24 months
0.3106 0.2170 0.3701 G0.0001
PAMELA K. LATTIMORE ET AL.168
Methodology
We estimated the likelihood and numbers of arrests associated with treatment using
models that controlled for the rich set of covariates presented. The covariates in the
models include indicators of criminal history, demographics, and judicial circuit, as
well as eight measures of supervision. We present results for any felony arrest, any
drug-related arrest, total numbers of felony arrests, and total numbers of drug-
related arrests within 12 and 24 months of placement on supervision. We
considered treatment in two ways. First, we examined Fany treatment,_ which
represents the mix of treatment exposures across the entire treatment group.
Second, we created two treatment groups based on whether time in treatment was
less than 90 days or more. These models address the finding in the literature that
outpatient treatment is particularly effective beyond 90 days.
The approach is similar to that of Linster (1999), Lattimore (1999), and Linster
et al. (1998). To understand this approach, consider that we are interested in
predicting the number of arrests one would expect if treatment were provided to
those who were not treated Y in other words, to answer the question BWhat if
everyone received drug treatment?^ (or, conversely, BWhat if no one had received
drug treatment?^). We address these questions by stratifying our sample into
subsamples Y treatment and no-treatment Y and estimating separate models of
arrest for each of the subsamples.
We used logistic regression to estimate the likelihood of arrest and negative
binomial models to generate estimates of numbers of arrests for each individual.
The simplest models are the logistic regressions for the two groups, treatment and
no treatment. For these models, we estimated a logistic regression model of the
probability of arrest using only the data for those who received treatment Y in other
words, we estimated the probability of arrest conditional on treatment. We then
combined the coefficient estimates from that model with the corresponding data for
those who did not receive treatment to generate an estimate (or prediction) of the
probability of arrest under the assumption of treatment for each untreated person.
By summing over these probability estimates, we identify the expected number of
individuals arrested for the no-treatment group under the assumption of treatment.
The difference between this number and the observed number of arrests incurred
by the no-treatment group provides an estimate of the effect of treatment on felony
arrest within 12 months.
Secondly, we estimated a logistic regression model using only the data for those
who received no treatment Y i.e., we estimated the probability of arrest conditional
on not receiving treatment. By combining the coefficients from this model with the
data for the treatment group we generate an estimate for each individual of the
probability of arrest under the assumption of no treatment. The sum of these
individual predicted probabilities provides the expected number of individuals
arrested for the treatment group under the assumption of no treatment.
This approach is analogous to that of a forecaster who estimates a regression
equation using historical data and subsequently applies the resulting model to
answer Fwhat if_ questions for different values of the independent variables (e.g.,
PREDICTING THE EFFECT OF SUBSTANCE ABUSE TREATMENT 169
BWhat is the impact on GDP if the price of oil increases?^). Such predictions are
valid as long as the estimation sample includes values for the independent
variables that are similar to those in the data from which predictions will be made.
As noted earlier, we have overlap in our two groups on our independent variables.
Thus, we can predict for each individual in the no-treatment group the effect of
treatment on recidivism.
In addition to modeling the probability of any arrest, we also used this approach
to model the number of felony arrests per individual in a two-year period using
negative binomial models. By applying the results from these models to the data
using the same approach described previously, the negative binomial models allow
us to compare expected differences in total numbers of felony arrests per in-
dividual under alternative assumptions about treatment.
In total, we conducted simulations for four groups of probationers: those who
received any treatment, the two sub-groups of treatment (less than 90 days and
more than 90 days of treatment), and no treatment. The six outcomes used in the
models were: arrested for a felony within 12 months, arrested for a felony within
24 months, arrested for a drug felony within 12 months, arrested for a drug felony
within 24 months, number of felony arrests within 12 months, and number of
felony arrests within 24 months. The logistic regression models are included in the
Appendices 1; the negative binomial models are available from the authors. In the
next section, we present the results of our analyses, which provide estimates of
the impact of treatment on felony and drug felony arrests.
Findings
In this section, we present the Barrests savings^ that could be expected as a result
of treatment. We compare no treatment with any non-residential treatment and
with either less than 90 days or more than 90 days of treatment. We first consider
the felony and drug arrest outcomes derived from the logistic regressions and then
examine the expected impact on the total number of felony arrests derived from the
negative binomial models.
Any felony arrest
Tables 3Y6 summarize the implications of the logistic regression models for the
impact of non-residential drug treatment on recidivism, as measured by arrest for a
felony within 12 and 24 months, respectively, of placement on supervision. Each
row in the tables presents the estimated numbers of failures (i.e., probationers who
were arrested after entering supervision) for the specified group under the
alternative, hypothetical treatment scenarios and the differences between the
estimates and the actual, observed numbers of those arrested.
Looking first at Table 3, we see the results for no-treatment versus any
treatment (which is actually Fsome treatment_ or the mix of treatment delivered;
PAMELA K. LATTIMORE ET AL.170
i.e., the status quo). Across all subjects, 37,915 were arrested at least once for a
felony during the first 12 months of supervision. Specifically, 26,936 of the no-
treatment group were arrested during the first year. If this group had received
treatment, the treatment model predicts that only 20,735 or 6,201 fewer would
have been arrested Y a 23% reduction in the expected number of probationers in
this group who were arrested. We see a similar effect in the expected number of
failures among the treatment group predicted in the absence of treatment. In this
case, 10,979 of the Fany treatment group_ were arrested during the first 12 months.
The no-treatment model suggests that without treatment, 13,504 members of this
group Y or an additional 2,525 Y would have been arrested Y a 23% increase.
The expected benefit of more extensive treatment is apparent in Table 4. If all
subjects had received less than 90 days of treatment, the expected total number of
Table 3. Expected numbers of subjects with at least one felony arrest, first 12 months of supervision,
under alternative treatment scenarios Y no treatment versus some treatment.
Group Number in
group
No treatment Any treatment
Expected from
model
Difference
from actual
Expected from
model
Difference
from actual
No treatment 81,797 26,936a 0 20,735 j6,201
Any treatment 51,979 13,504 +2,525 10,979a 0
Total 133,776 40,440 +2,525 31,714 j6,201
a Estimates from model equal the observed number of subjects with at least one felony arrest during the
initial 12 months of supervision. Overall, 37,915 subjects had a felony arrest.
Table 4. Expected numbers of subjects with at least one felony arrest, first 12 months of supervision,
under alternative treatment scenarios Y no treatment versus less than or 90 days or more.
Group Number in
group
No treatment G90 Days treatment Q90 Days treatment
Expected
from
model
Difference
from
actual
Expected
from
model
Difference
from
actual
Expected
from
model
Difference
from
actual
No
treatment
81,797 26,936a 0 23,301 j3,635 17,538 j9,398
G90 Days
treatment
23,896 6,740 +723 6,017 a 0 4,594 j1,423
Q90 Days
treatment
28,083 6,764 +1,802 6,262 +1,300 4,962 a 0
Total 133,776 40,440 +2,525 35,581 j2,335 27,094 j10,821
aEstimates from model equal the observed number of subjects with at least one felony arrest during the
initial 12 months of supervision. Overall, 37,915 subjects had a felony arrest.
PREDICTING THE EFFECT OF SUBSTANCE ABUSE TREATMENT 171
probationers who would be arrested across the three groups is 35,581 Y a reduction
of 2,334 failures (6.1%) over the 37,915 that were observed. The expected total of
35,581 includes an expected increase of 1,300 failures among the group who
actually received at least 90 days of treatment. Finally, we see that if all subjects
were given at least 90 days of treatment, the model predicts that only 27,093
probationers would have been arrested Y a reduction of 10,822 failures (28.5%).
The results in Tables 5 and 6 show similar expected effects extending over a
two-year period. More than 40% of the subjects (54,199) had been arrested at least
once for a felony within two years of placement on supervision. The model
suggests that by providing treatment to the no-treatment group, the number of
probationers arrested could have been reduced by 6,603 from 54,199 to 48,136 Yan 11% reduction. If everyone had received at least 90 days of treatment, the
Table 5. Expected numbers of subjects with at least one felony arrest, first 24 months of supervision,
under alternative treatment scenarios Y no treatment versus some treatment.
Group Number
in group
No treatment Any treatment
Expected from
model
Difference from
actual
Expected from
model
Difference from
actual
No treatment 81,797 37,242a 0 31,179 j6,063
Any treatment 51,979 19,639 +2,682 16,957 a 0
Total 133,776 56,882 +2,682 48,136 j6,063
a Estimates from model equal the observed number of subjects with at least one felony arrest during the
initial 24 months of supervision. Overall, 54,199 subjects had a felony arrest.
Table 6. Expected numbers of subjects with at least one felony arrest, first 24 months of supervision,
under alternative treatment scenarios Y no treatment versus less than or 90 days or more.
Group Number in
group
No treatment G90 Days treatment Q90 Days treatment
Expected
from
model
Difference
from
actual
Expected
from
model
Difference
from
actual
Expected
from
model
Difference
from
actual
No
treatment
81,797 37,242a 0 33,188 j4,054 28,651 j8,591
G90 Days
treatment
23,896 9,613 + 903 8,710a 0 7,625 j1,085
Q90 Days
treatment
28,083 10,026 +1,779 9,197 +950 8,247a 0
Total 133,776 56,882b +2,682 51,095 j3,104 44,523 j9,676
aEstimates from model equal the observed number of subjects with at least one felony arrest during the
initial 24 months of supervision. Overall, 54,199 subjects had a felony arrest.bTotal does not sum perfectly due to rounding of estimates.
PAMELA K. LATTIMORE ET AL.172
model predicts that 44,523 probationers (or 9,676 fewer) would have been arrested
within the 24-month period Y an 18% reduction.
Any felony drug arrest
Tables 7Y10 summarize the implications of the models of non-residential drug
treatment on recidivism, as measured by felony arrest on a drug charge within 12
and 24 months of placement on supervision. The observed numbers of those
probationers who were arrested on a felony drug charge within 12 and 24 months
were 18,740 and 29,194, respectively. As with the earlier tables, each row presents
the expected numbers of failures for the specified group under the alternative,
Table 7. Expected numbers of subjects with at least one felony drug arrest, first 12 months of
supervision, under alternative treatment scenarios Y no treatment versus some treatment.
Group Number in
group
No treatment Any treatment
Expected from
model
Difference
from actual
Expected from
model
Difference
from actual
No treatment 81,797 13,544a 0 9,581 j3,963
Any treatment 51,979 6,741 +1,545 5,196a 0
Total 133,776 20,285 +1,545 14,777 j3,963
a Estimates from model equal the observed number of subjects with at least one felony arrest during the
initial 12 months of supervision. Overall, 18,740 subjects had a felony drug arrest.
Table 8. Expected numbers of subjects with at least one felony drug arrest, first 12 months of
supervision, under alternative treatment scenarios Y no treatment versus less than or 90 days or more.
Group Number
in group
No treatment G90 Days treatment Q90 Days treatment
Expected
from
model
Difference
from
actual
Expected
from
model
Difference
from
actual
Expected
from
model
Difference
from
actual
No
treatment
81,797 13,544a 0 11,146 j2,398 7,634 j5,910
G90 Days
treatment
23,896 3,404 +435 2,969a 0 2,091 j878
Q90 Days
treatment
28,083 3,336 +1,109 3,037 +810 2,227a 0
Total 133,776 20,285 +1,544 17,152 j1,588 11,952 j6,788
a Estimates from model equal the observed number of subjects with at least one felony arrest during the
initial 12 months of supervision. Overall, 18,740 subjects had a felony drug arrest.
PREDICTING THE EFFECT OF SUBSTANCE ABUSE TREATMENT 173
hypothetical treatment scenarios. Looking first at the results for no treatment
versus any treatment, we see in Table 7 that 13,544 of the no-treatment group were
arrested during the first year. If this group had received the same treatment, or mix
thereof, as the treatment group, the treatment model predicts only 9,581 would
have been arrested Y a 29% reduction in the expected number of failures for this
group. Of those in the Fany treatment group,_ 5,196 were arrested for a felony drug
offense Y the no-treatment model suggests that without treatment 6,741 members
of this group would have been arrested Y a 30% increase over what was observed.
The expected benefit of more extensive treatment is also apparent in Table 8. If all
subjects had received less than 90 days of treatment, the expected total number of
failures across the three groups drops from 18,740 to 17,152 Y a reduction of 1,588
(8.5%), which includes an expected 810 additional arrestees in the group who
Table 9. Expected numbers of subjects with at least one felony drug arrest, first 24 months of
supervision, under alternative treatment scenarios Y no treatment versus some treatment.
Group Number in
group
No treatment Any treatment
Expected from
model
Difference
from actual
Expected from
model
Difference
from actual
No treatment 81,797 20,643a 0 15,671 j4,972
Any treatment 51,979 10,690 +2,140 8,550a 0
Total 133,776 31,333 +2,140 24,220 j4,972
aEstimates from model equal the observed number of subjects with at least one felony arrest during the
initial 24 months of supervision. Overall, 29,193 subjects had a felony arrest.
Table 10. Expected numbers of subjects with at least one felony drug arrest, first 24 months of
supervision, under alternative treatment scenarios Y no treatment versus less than or 90 days or more.
Group Number
in group
No treatment G90 Days treatment Q90 Days treatment
Expected
from
model
Difference
from
actual
Expected
from
model
Difference
from
actual
Expected
from
model
Difference
from
actual
No
treatment
81,797 20,643a 0 17,120 j3,523 13,910 j6,733
G90 Days
treatment
23,896 5,277 +758 4,519a 0 3,794 j725
Q90 Days
treatment
28,083 5,413 +1,382 4,680 +649 4,031a 0
Total 133,776 31,333 +2,140 26,319 j2,874 21,734 j7,458
a Estimates from model equal the observed number of subjects with at least one felony arrest during the
initial 24 months of supervision. Overall, 29,193 subjects had a felony arrest.
PAMELA K. LATTIMORE ET AL.174
actually received at least 90 days of treatment. If all subjects were given at least 90
days of treatment, the model predicts that 11,952 would have been arrested Y a
reduction of 6,788 over the 18,740 (36.2%) who were arrested for a felony drug
offense under the status quo treatment.
The results in Tables 9 and 10 show similar expected effects over a two-year
period. More than 22% of the subjects (29,193) had been arrested at least once for
a felony drug offense within two years of placement on supervision. The model
suggests that by providing treatment to the no-treatment group, the number
arrested in this group could have been reduced from 20,643 to 15,671 Y a 24.1%
reduction (see Table 9). If everyone had received at least 90 days of treatment,
Table 10 shows that the model predicts that 21,734 would have been arrested
within the 24-month period Y a 25.6% reduction over what was observed.
Total number of felony arrests
The previous findings suggest that current policy with respect to providing
substance abuse treatment reduced the number of probationers arrested during the
first two years of supervision over what would have been expected if non-
residential drug treatment had not been provided to some of these subjects. The
results also suggest that providing treatment to those who did not receive it or that
providing more treatment could reduce the expected number of arrests. Of course
some offenders are arrested multiple times over the course of two years, and the
probationers in our data were no exception. Overall, during the first 12 months of
supervision, the group experienced an average of 0.41 felony arrests, with a range
of 0 to 15. During the first 24 months, the average was 0.71 arrests and the range
was 0 to 18. Just as the likelihood of arrest varied by group, the groups varied
regarding the numbers of felony arrests, where again those who received more
treatment were arrested fewer times. Mean numbers (and maximum numbers) of
arrests during the first 12 months were 0.49 (maximum of 15), 0.29 (maximum of
10), 0.35 (maximum of 8), and 0.23 (maximum of 10) for the no treatment, any
treatment, less than 90 days treatment, and at least 90 days treatment groups,
respectively. Mean numbers (and maximum numbers) of arrests during the first
24 months were 0.84 (maximum of 18), 0.52 (maximum of 17), 0.61 (maximum of
17), and 0.44 (maximum of 15) for the no treatment, any treatment, less than
90 days treatment, and at least 90 days treatment groups, respectively.
We used negative binomial regression, following the same procedures described
previously, to estimate Fcount models_ for each of our four groups.6 For these mo-
dels, the outcome was number of felony arrests during the 12 (or 24 month) period.7
Tables 11 and 12 summarize the implications of the negative binomial models.
These tables contain the expected numbers of arrests for each group under each of
the alternative treatment scenarios for 12 and 24 months, respectively, following
placement on supervision. The 133,776 subjects were arrested for 54,787 felonies
during the initial 12 months and a total of 95,606 during the first 24 months of
supervision.
PREDICTING THE EFFECT OF SUBSTANCE ABUSE TREATMENT 175
As can be seen in Table 11, our models suggest that in the absence of treatment,
we would have expected to see during the first 12 months a total of 59,317 arrests
or 4,530 additional arrests (an 8.3% increase). If status quo treatment had been
provided to everyone, the models suggest there would have been 45,323 felony
arrests Y a 17% reduction over what was observed. The most substantial effects,
however, are observed when we examine the results of providing at least 90 days
Table 12. Expected numbers of felony arrests, first 24 months of supervision, under alternative treatment
scenarios.
Group Number in
group
Treatment
No
treatment
Any
treatment
G90 Days
treatment
Q90 Days
treatment
No treatment 81,797 69,256a 54,746
Any treatment 51,979 33,664 27,476a
Total 133,776 102,920 82,221
No treatment 81,797 69,256a 59,878 47,872
G90 Days
treatment
23,896 16,900 14,924 a 12,105
Q90 Days
treatment
28,083 16,764 15,023 12,550 a
Total 133,776 102,920 89,825 72,527
a Estimates from model during the initial 24 months of supervision. Overall, the subjects had 95,606
actual felony arrests.
Table 11. Expected numbers of felony arrests, first 12 months of supervision, under alternative treatment
scenarios.
Group Number in
group
Treatment
No
treatment
Any
treatment
G90 Days
treatment
Q90 Days
treatment
No treatment 81,797 40,193a 30,287
Any treatment 51,979 19,125 15,036a
Total 133,776 59,317 45,323
No treatment 81,797 40,193a 34,227 24,987
G90 Days
treatment
23,896 9,694 8,501a 6,256
Q90 Days
treatment
28,083 9,431 8,543 6,529a
Total 133,776 59,317 51,271 37,772
a Estimates from model during the initial 12 months of supervision. Overall, the subjects had 54,787
actual felony arrests.
PAMELA K. LATTIMORE ET AL.176
of treatment to everyone. Under this scenario, the models predict 37,772 arrests Y a
31% reduction over what was observed and a 36% reduction over what we would
expect in the absence of any treatment.
Table 12 presents the findings for the total number of felony arrests expected
during the first 24 months of supervision. We observed a total of 95,606 felony
arrests for the 133,776 subjects during this period. Table 12 shows that if no
treatment had been provided, we would have expected to see 102,920 arrests
during the first 24 months, about 8% more than were actually observed and
substantially more than we would expect to see if some treatment had been
provided to everyone. Under the Fany treatment_ scenario, the models suggest a
total of 82,221 felony arrests, 13,385 fewer than were observed (a 14% decrease).
As was true with the other models, the most substantial Farrest savings_ are
observed when everyone is provided at least 90 days in treatment. Under this
scenario, the expected number of felony arrests drops to 72,527 Y 24% less than
was observed under the provided or status quo mix of treatment. Overall, the
models identify a difference of 30,393 felony arrests over 24 months between the
no treatment and at least 90 days of treatment options.
Discussion and conclusions
We have used a large administrative database provided by the FLDC and FDLE
to examine the expected effects of different amounts of non-residential substance
abuse treatment on drug-involved probationers. The four non-residential
treatment alternatives were no treatment, any treatment (non-residential treatment
of any length as experienced by those who received at least one day of
treatment), less-than-90-days treatment, and at least 90 days treatment. To
identify the expected effects, we stratified our data by treatment and estimated
separate models for each of our treatment groups for any (and total numbers of)
felony arrest(s) and felony drug arrest(s) within 12 and 24 months of placement
on supervision. We used these models to predict the expected likelihood (and
numbers) of failures for all groups and were able to quantify the impact of
non-residential substance abuse treatment on arrest. Specifically, we were able
to identify the effect of current treatment policy compared with no treatment,
and to examine the differential effect of less than versus at least 90 days of
treatment. The results suggested that the treatment provided reduced the number
of recidivists and the numbers of arrests that would have been expected by these
arrestees over the 24 months following admission to community supervision.
Further, the results showed that additional Farrest savings_ were possible by
providing treatment to those drug-involved offenders who did not receive
treatment.
The analyses that we conducted are only reasonable if we assume that not all
of those who would benefit from treatment actually received treatment. In other
words, it is necessary that the no-treatment group includes many who should be
PREDICTING THE EFFECT OF SUBSTANCE ABUSE TREATMENT 177
in treatment and, thus, some who would benefit. Is this assumption a reasonable
one? We believe so for the following two reasons: First, our data include
indicators of current (as well as past) drug use by many of those in the no-
treatment group. To assume that these individuals would not benefit from
treatment is to assume that there are numerous individuals under community
supervision who are using drugs but for whom treatment is deemed unnecessary
and/or inappropriate. Second, using the same criteria statewide for inclusion in
our sample, there was considerable variability across judicial districts in terms
of the proportions of individuals who were in treatment. This variability, again,
suggests that many members of the no-treatment group could have been in
treatment. Otherwise, we would need to assume, for example, that in one
district only about one-fourth of those identified by our criteria should be in
treatment while in another the proportion exceeds one-half. For this to be true,
we would have to assume that many more of those in the first district as
compared to the second were either (a) former but not current drug users and/or
(b) drug sellers/distributors rather than drug users. In fact, we believe that it is
more reasonable to assume that treatment availability, attitudes about treatment
(among probation officers, offenders and, perhaps, judges), and Fthe luck of the
draw_ contribute to whether or not someone began treatment Y creating a quasi-
experiment on which the methods described here can be used to estimate
effects.
The results of the models are interesting in that they yield the Farrest savings_that can accrue to treatment. We can take these savings one step further and
comment on the implications of these Fsavings_ for economic analysis. We will
focus on the implication of the results of the models for total numbers of felony
arrest within 24 months and, in particular, the no- and any-treatment models (see
Table 12). These models suggest that the drug treatment provided to 51,979
offenders generated savings in numbers of arrests of 6,188. The simplest economic
model identifies the Fbreakeven_ point in terms of investment in drug treatment
with the only returns being a reduction in the (costs associated with) arrests. In
other words, it will be worthwhile to invest in drug treatment as long as the
following is true:
NT:CT � NA
:CA
From our data we know (or have assumed) the number treated, NT, and our
models provided us with an estimate of the reduction in the total number of arrests,
NA. Given that treatment was provided to 51,979 individuals and if the (average)
cost of treatment, CT, is $1,000 per individual, the above equation suggests that
that such an investment is worthwhile as long as the (average) cost of an arrest
(and all related criminal justice processing), CA, is at least $8,400. If we extend the
analysis and assume that everyone is treated at an (average) cost of $1,000 per
person, it will be cost effective to provide treatment as long as the (average) cost of
arrest (and all related criminal justice processing and corrections) exceeds about
PAMELA K. LATTIMORE ET AL.178
$6,463.8 For comparison, the annual average cost of incarceration in Florida in
2001 was $20,190 (Stephan 2004).
Results of our analyses suggest that the nonresidential substance abuse
treatment provided to about 52,000 of the nearly 134,000 probationers in our
sample reduced both the expected number of individuals who were arrested and the
expected total number of felony arrests during the initial 12 and 24 months
following placement on supervision. More than 28% (37,915) of the 133,776
subjects had been arrested on a felony charge within 12 months of placement on
supervision. Within 24 months, this percentage grew to more than 40% (54,199 of
133,776). In total, this group had 54,787 and 95,606 felony arrests within 12 and
24 months, respectively, of placement on supervision.
Our results suggest that for the 51,979 who received some non-residential
treatment, 10,979 were arrested at least once within one year Y about 2,500
fewer than expected in the absence of treatment. Those arrested had a total of
15,036 arrests, compared with an expected number of arrests of 19,125 under the
no-treatment scenario. Thus, the results suggest that the treatment that was
provided reduced the number arrested and the number of arrests by 18.7% and
21.4%, respectively, for the group. Results were similar when we examined the
impact of treatment on arrest over 24 months: 2,682 fewer probationers with a
felony arrest and 6,188 fewer arrests. More substantial reductions in number
arrested and numbers of arrests occurred under the scenario of lengthier treat-
ment (i.e., at least 90 days).
We also looked at the impact of providing treatment to the 81,797 who did
not receive treatment. As discussed earlier, this group had a 12-month felony
arrest rate of 32.9% (26,936 with at least one arrest) and a 24-month felony
arrest rate of 45.5%. The models suggest that providing this group access to
treatment as it was provided to the group who received it could have reduced the
number arrested within one year from 26,936 to 20,735 (23% reduction) and the
number arrested within two years from 37,242 to 31,179 (16.3% reduction). This
group experienced 40,193 and 69,256 felony arrests within 12 and 24 months,
respectively. The models suggest with treatment, we could expect 30,287 and
54,746 arrests within 12 and 24 months, respectively. These numbers represent
reductions in the expected numbers of arrests of 24.6% and 21.0%, respectively.
Once again, greater reductions were forecast under the lengthier treatment
scenario.
Many offenders who are arrested and supervised in the community are drug-
involved. Some portion of these offenders are referred to and receive substance
abuse treatment, which appears to reduce the likelihood, as well as the number of
times, they will recidivate. A substantial proportion of state (20%) and Federal
(55%) prison inmates are being held on drug charges, and the costs associated
with imprisonment are substantial. At a time when state and Federal budgets are
stretched, it seems that the provision of treatment might yield a number of
societal and criminal justice system benefits Y both in terms of improvements in
public safety and potential cost savings. Treatment, although not free, is, on
average, substantially less expensive than incarceration and the financial and
PREDICTING THE EFFECT OF SUBSTANCE ABUSE TREATMENT 179
personal (e.g., victim impact) benefits of avoiding future crimes are difficult to
overestimate.
The policy implications of our findings are straightforward. Providing non-
residential substance abuse treatment to offenders being supervised in the community
appears to be a feasible approach to reducing the likelihood and collateral impacts of
recidivism, and thus enhancing public safety. The provision of treatment, to the
extent that it reduces recidivism and the related need for incarceration, might also
result in valuable public resource savings.
Many quasi-experimental studies of the impact of substance abuse treatment on
outcomes like recidivism do not adequately control for the biases that differentiate
treatment and comparison group members. Failing to control for these differences
often invalidates the findings and reduces the credibility of the literature. This
study employs an analytic approach that controls for meaningful differences
between the treatment and comparison groups, utilizes a large administrative
dataset, and produces findings suggesting that the provision of non-residential
substance abuse treatment can substantially reduce the likelihood and impact of
recidivism.
The approach presented here provides an alternative to the other approaches that
have been used to address selection bias issues.9 The approach does require that the
untreated group contains subjects that could potentially benefit from the
intervention. If this requirement is met, the approach has the added benefit of
being simpler to estimate (and explain) than more complex models based on, for
example, Heckman’s approach or an instrumental variable approach.
Acknowledgements
This study was supported by grant DA13086 from the National Institute on Drug
Abuse (NIDA). Points of view are those of the authors and do not necessarily
represent the views of the NIDA. The authors would like to acknowledge Bill
Bales of Florida State University (formerly of the Florida Department of
Corrections), Kristine Dougherty of the Florida Department of Corrections, and
Sue Burton of the Florida Department of Law Enforcement for their assistance in
developing the data, as well as their guidance and support. We would also like to
thank the Journal of Experimental Criminology editor and reviewers for their
insights and recommendations.
Appendix A: Model
Tables A-1 and A-2 show the odds ratio estimates from the logit models for anyfelony arrest and any felony drug arrest, respectively, within 12 and 24 months.Shown are the estimates of the ratio of the odds of failure (i.e., felony arrest within
PAMELA K. LATTIMORE ET AL.180
the specified time period) to the odds of success Y values greater than one implyincreasing odds of arrest and values less than one imply decreasing odds of arrest,other things being equal. Values significantly greater than one (Wald test, 0.05significance level) are shaded. Two measures of goodness of fit, the R2 and themaximum-rescaled R2 are also included.
Notes
1 The non-residential substance abuse treatment provided is contracted by the FLDC
throughout the state. There are several different models supported (details available from
authors or by going to the FLDC website, http://www.dc.state.fl.us), but overall the fact
that the contracting is done at the state level suggests at least some comparability among
the treatment provided across the state.2 For the cases with missing data, the non-residential treatment participation rate was
40.7% Y about equal to the rate for those for whom we had complete data (38.9%).3 Florida has 20 judicial circuits. The judicial circuit reflects the circuit in which the
individual was sentenced. An individual was identified with the judicial circuit of
sentencing. We have 21 indicators for judicial circuit because we included 1,845
individuals in our analysis who either had a missing value for judicial circuit or who had
been sentenced out of state. Those in this group comprise the Bjudicial circuit none.^ We
estimated the models excluding these individuals and found results (not shown) very
similar to those presented here.4 We also examined the reasons why members of each group were included in our drug-
involved sample. Again, although there are differences between the groups, it is important
to note that both groups were represented in each of the criteria. For the treatment and no-
treatment groups, respectively, (1) 37.1% versus 55.6% had at least one drug-related
arrest; (2) 2.7% versus 2.0% had any history of drug court participation; (3) 22.0% versus
21.4% had ever been on DOP supervision; (4) 46.6% versus 53.2% had any FLDC record
of a positive drug test ever; and (5) 100% versus 21.1% had ever attended treatment. (By
definition, of course, all of the treatment group would have had to meet this last criterion.)5 Members of the no-treatment group were also more likely than the treatment group to be
arrested for violent or property offenses and to have a probation violation (data not
shown).6 The ability of a subject to accumulate arrests depends partly on being free in the
community. If the first arrest results in confinement, the likelihood of a subsequent arrest
during that confinement period is substantially reduced, including only additional arrests
for offenses committed prior to the incarceration or to arrests stemming from incidents
that occur while incarcerated. For our purposes, we are only interested in predicting the
numbers of arrests. We are not interested in predicting how many arrests would have been
expected over the period in the absence of any incarceration.7 Negative binomial model results are available from the authors upon request.8 This estimate is based on the estimate of 82,221 total arrests with all receiving some
treatment differenced from the 102,920 arrests estimated under the no treatment scenario.
This results in a total reduction in arrests of 20,699 (NA) following from the treatment of
133,776 (NT).9 We are currently completing work of a series of propensity score-based analyses that
provide treatment Y effect results similar to those presented here (Lattimore et al. 2004).
PREDICTING THE EFFECT OF SUBSTANCE ABUSE TREATMENT 181
Ta
ble
A-1
.O
dd
sra
tio
esti
mat
esfr
om
felo
ny
arre
stre
cid
ivis
mm
od
els.
Va
ria
ble
Fel
on
ya
rres
tw
ith
in1
2m
on
ths
Fel
on
ya
rres
tw
ith
in2
4m
on
ths
An
y
trea
tmen
t
(N=
51
,97
9)
Tre
atm
ent
G9
0d
ays
(N=
23
,89
6)
Tre
atm
ent
Q9
0d
ays
(N=
28
,08
3)
No
trea
tmen
t
(N=
81
,79
7)
An
y
trea
tmen
t
(N=
51
,97
9)
Tre
atm
ent
G9
0d
ays
(N=
23
,89
6)
Tre
atm
ent
Q9
0d
ays
(N=
28
,08
3)
No
trea
tmen
t
(N=
81
,79
7)
Dem
og
rap
hic
cha
ract
eris
tics
Ag
e(y
ears
)0
.966
0.9
61
0.9
71
0.9
64
0.9
63
0.9
58
0.9
67
0.9
59
Mal
e1
.095
1.0
75
1.1
10
1.0
79
1.1
03
1.0
79
1.1
18
1.0
92
Bla
ck*
1.2
77
1.2
63
1.2
41
1.2
81
1.3
64
1.3
59
1.3
35
1.4
22
His
pan
ic*
0.9
71
0.9
81
0.9
51
1.1
00
0.9
99
1.0
02
0.9
91
1.1
37
Cri
min
al
his
tory
Curr
ent
vio
lent
off
ense
00
.931
0.9
28
0.9
43
1.0
07
0.9
17
0.9
08
0.9
32
1.0
71
Curr
ent
pro
per
tyoff
ense
01
.470
1.3
87
1.5
25
1.4
78
1.4
63
1.4
46
1.4
54
1.6
08
Curr
ent
oth
ero
ffen
se0
0.7
77
0.6
96
0.8
52
0.8
15
0.8
07
0.7
80
0.8
26
0.9
24
#p
rio
rar
rest
s1
.061
1.0
64
1.0
58
1.0
62
1.0
73
1.0
78
1.0
67
1.0
76
#p
rio
rd
rug
off
ense
s1
.070
1.0
81
1.0
60
1.0
55
1.0
82
1.0
90
1.0
76
1.0
67
#p
rio
rsu
per
vis
ion
s0
.991
0.9
91
0.9
90
0.9
97
0.9
85
0.9
87
0.9
82
0.9
97
#p
rio
rp
riso
nte
rms
0.9
71
0.9
68
0.9
73
0.9
83
0.9
65
0.9
50
0.9
78
0.9
69
Su
per
visi
on
DO
Psu
per
vis
ion-
1.5
15
1.4
40
1.6
29
1.3
81
1.3
64
1.4
14
1.3
25
1.3
52
FP
sup
erv
isio
n1
.278
1.1
85
1.4
78
1.1
08
1.1
68
1.1
76
1.1
97
1.1
54
Fie
ldco
nta
cts/
yea
r1
.007
1.0
01
1.0
10
1.0
06
1.0
03
1.0
04
1.0
02
1.0
00
Ho
me
con
tact
s/y
ear
1.0
12
1.0
13
1.0
12
1.0
07
1.0
12
1.0
13
1.0
11
1.0
09
Offi
ceco
nta
cts/
yea
r0
.967
0.9
67
0.9
72
0.9
68
0.9
68
0.9
65
0.9
72
0.9
69
UA
test
s/yea
r0
.982
0.9
78
0.9
96
0.9
41
0.9
86
0.9
81
0.9
95
0.9
60
Po
siti
ve
UA
s/yea
r1
.068
1.0
53
1.0
74
1.1
00
1.0
82
1.0
73
1.0
89
1.0
85
Po
siti
ve
UA
firs
tm
on
th1
.083
1.1
17
1.0
27
1.0
70
0.9
87
1.0
03
0.9
60
1.0
38
PAMELA K. LATTIMORE ET AL.182
Geo
gra
ph
icin
dic
ato
r
Jud
icia
lci
rcu
it1.
0.9
57
0.8
43
1.0
21
0.8
98
1.0
66
0.9
73
1.1
44
0.8
99
Jud
icia
lci
rcu
it2.
0.8
51
0.7
55
0.9
12
0.9
30
0.8
90
0.7
57
0.9
83
0.9
43
Jud
icia
lci
rcu
it3.
1.0
55
0.9
02
1.1
57
0.8
27
1.0
67
0.9
79
1.1
23
0.8
01
Jud
icia
lci
rcu
it4.
0.9
94
0.9
87
0.8
57
1.1
02
1.1
09
1.1
64
0.9
59
1.1
44
Jud
icia
lci
rcu
it5.
0.9
09
1.0
56
0.6
76
0.9
53
1.0
37
1.2
13
0.8
35
0.9
96
Jud
icia
lci
rcu
it6.
1.0
57
1.1
12
0.8
96
1.1
91
1.1
18
1.1
86
1.0
02
1.1
78
Jud
icia
lci
rcu
it7.
1.0
63
1.0
43
0.9
11
1.0
85
1.1
28
1.1
71
0.9
83
1.0
72
Jud
icia
lci
rcu
it8.
1.6
25
1.7
32
1.5
43
1.6
50
1.7
35
1.7
90
1.6
93
1.6
97
Jud
icia
lci
rcu
it9.
0.9
56
0.9
92
0.7
76
1.0
98
0.9
88
1.0
05
0.9
06
1.1
11
Jud
icia
lci
rcu
it1
0.
1.0
92
1.2
11
0.8
96
1.3
12
1.1
95
1.3
38
1.0
35
1.4
29
Jud
icia
lci
rcu
it1
1.
1.1
50
1.2
15
1.1
34
1.1
08
1.2
63
1.3
30
1.2
40
1.1
60
Jud
icia
lci
rcu
it1
2.
1.1
17
1.0
65
1.0
86
0.9
66
1.1
27
1.0
88
1.1
16
0.9
51
Jud
icia
lci
rcu
it1
3.
1.3
49
1.4
54
1.0
66
1.3
83
1.2
52
1.3
56
1.0
78
1.3
08
Jud
icia
lci
rcu
it1
4.
1.1
03
1.1
79
1.0
07
0.8
08
1.1
01
1.1
04
1.0
85
0.7
98
Jud
icia
lci
rcu
it1
5.
0.7
74
0.8
50
0.6
66
0.8
61
0.8
11
0.9
24
0.7
07
0.9
29
Jud
icia
lci
rcu
it1
6.
1.7
12
1.7
61
1.3
12
1.1
58
1.6
55
1.7
07
1.4
14
1.0
67
Jud
icia
lci
rcu
it1
8.
1.0
08
0.9
93
0.9
67
0.9
75
1.0
30
1.0
88
0.9
58
0.9
90
Jud
icia
lci
rcu
it1
9.
1.0
71
1.0
07
1.0
53
1.0
05
1.1
68
1.1
68
1.1
36
0.9
75
Jud
icia
lci
rcu
it2
0.
0.9
64
1.1
00
0.8
34
1.0
19
1.0
28
1.1
70
0.9
22
1.0
02
Jud
icia
lci
rcu
itn
on
e.0
.513
0.4
26
0.5
64
0.6
32
0.4
95
0.4
67
0.5
07
0.6
52
R2
0.0
66
80
.08
45
0.0
48
40
.097
60
.095
20
.11
46
0.0
75
40
.122
6
Max
imu
m-r
esca
led
R2
0.1
03
80
.12
49
0.0
79
70
.135
90
.132
80
.15
29
0.1
07
40
.163
9
Ital
odds
rati
os
init
alic
sar
est
atis
tica
lly
signifi
cant
atth
e0.0
5le
vel
.
*W
hit
eis
refe
rence
cate
gory
..Ju
dic
ial
circ
uit
17
isth
ere
fere
nce
cate
gory
.-C
om
munit
yco
ntr
ol
isth
ere
fere
nce
cate
gory
.0C
urr
ent
dru
goff
ense
isth
ere
fere
nce
cate
gory
.
PREDICTING THE EFFECT OF SUBSTANCE ABUSE TREATMENT 183
Ta
ble
A-2
.O
dd
sra
tio
esti
mat
esfr
om
felo
ny
dru
gar
rest
reci
div
ism
mo
del
s.
Va
ria
ble
Fel
on
yd
rug
arr
est
wit
hin
12
mo
nth
sF
elo
ny
dru
ga
rres
tw
ith
in2
4m
on
ths
An
y
trea
tmen
t
(N=
51
,979
)
Tre
atm
ent
G9
0d
ays
(N=
23
,89
6)
Tre
atm
ent
Q9
0d
ays
(N=
28
,08
3)
No
trea
tmen
t
(N=
81
,797
)
Tre
atm
ent
(N=
51
,97
9)
Tre
atm
ent
G9
0d
ays
(N=
23
,896
)
Tre
atm
ent
Q9
0d
ays
(N=
28
,083
)
No
trea
tmen
t
(N=
81
,797
)
Dem
og
rap
hic
cha
ract
eris
tics
Ag
e(y
ears
)0
.96
40
.96
20
.967
0.9
70
0.9
64
0.9
62
0.9
67
0.9
66
Mal
e1
.15
21
.09
21
.228
0.9
97
1.1
75
1.1
25
1.2
24
1.0
33
Bla
ck*
1.5
53
1.6
05
1.4
21
1.5
89
1.6
50
1.6
69
1.5
90
1.6
62
His
pan
ic*
0.9
93
0.9
78
0.9
86
1.1
33
1.0
46
0.9
79
1.0
99
1.1
69
Cri
min
al
his
tory
Curr
ent
vio
lent
off
ense
00
.49
00
.49
00
.490
0.5
57
0.4
86
0.4
96
0.4
78
0.5
93
Curr
ent
pro
per
tyoff
ense
00
.71
40
.66
90
.752
0.6
62
0.7
63
0.7
37
0.7
79
0.7
33
Curr
ent
oth
eroff
ense
00
.59
30
.59
70
.576
0.6
38
0.6
05
0.6
29
0.5
77
0.7
37
#p
rio
rar
rest
s1
.01
91
.01
81
.018
1.0
21
1.0
21
1.0
19
1.0
22
1.0
26
#p
rio
rd
rug
off
ense
s1
.19
51
.20
71
.188
1.1
61
1.2
12
1.2
20
1.2
09
1.1
72
#p
rio
rsu
per
vis
ion
s0
.99
00
.99
50
.981
0.9
95
0.9
87
0.9
95
0.9
77
0.9
91
#p
rio
rp
riso
nte
rms
0.9
61
0.9
64
0.9
58
0.9
72
0.9
56
0.9
50
0.9
63
0.9
69
Su
per
visi
on
DO
Psu
per
vis
ion-
1.5
92
1.4
99
1.8
06
1.4
22
1.4
55
1.4
60
1.4
75
1.3
26
FP
sup
erv
isio
n1
.32
51
.16
91
.730
1.2
08
1.2
53
1.1
78
1.4
02
1.2
26
Fie
ldco
nta
cts/
yea
r1.0
00
0.9
96
1.0
04
1.0
06
0.9
96
0.9
98
0.9
95
1.0
01
Hom
eco
nta
cts/
yea
r1
.01
21
.01
01
.013
1.0
08
1.0
12
1.0
12
1.0
11
1.0
09
Offi
ceco
nta
cts/
yea
r0
.97
30
.97
30
.978
0.9
72
0.9
73
0.9
70
0.9
78
0.9
74
UA
test
s/yea
r0
.97
60
.97
60
.989
0.9
27
0.9
81
0.9
78
0.9
90
0.9
52
Po
siti
ve
UA
s/yea
r1
.09
41
.06
71
.115
1.1
05
1.0
94
1.0
78
1.1
11
1.0
75
Po
siti
ve
UA
firs
tm
on
th1
.12
51
.19
31
.011
1.0
14
1.0
61
1.1
00
1.0
03
1.0
07
PAMELA K. LATTIMORE ET AL.184
Geo
gra
ph
icin
dic
ato
r
Jud
icia
lci
rcu
it1.
0.6
34
0.6
12
0.6
07
0.5
91
0.6
64
0.6
25
0.6
78
0.5
44
Jud
icia
lci
rcu
it2.
0.6
73
0.6
06
0.7
17
0.6
36
0.7
43
0.6
67
0.7
91
0.6
77
Jud
icia
lci
rcu
it3.
0.7
78
0.6
53
0.8
70
0.5
54
0.6
52
0.4
89
0.7
82
0.5
16
Jud
icia
lci
rcu
it4.
0.8
83
0.7
82
0.8
93
0.7
53
0.8
95
0.8
46
0.8
49
0.8
16
Jud
icia
lci
rcu
it5.
0.6
92
0.7
68
0.4
95
0.6
58
0.7
48
0.7
62
0.6
60
0.6
59
Jud
icia
lci
rcu
it6.
0.8
88
0.9
06
0.7
28
1.0
01
0.8
96
0.8
53
0.8
55
0.9
11
Jud
icia
lci
rcu
it7.
0.8
98
0.8
51
0.7
35
0.9
17
0.9
13
0.9
15
0.7
46
0.8
43
Jud
icia
lci
rcu
it8.
1.4
14
1.2
80
1.5
05
1.2
23
1.3
72
1.3
14
1.3
96
1.2
15
Jud
icia
lci
rcu
it9.
0.8
10
0.8
00
0.6
65
0.9
10
0.8
63
0.8
34
0.7
93
0.8
95
Jud
icia
lci
rcu
it1
0.
0.8
42
0.8
98
0.6
82
0.9
49
0.9
69
0.9
96
0.8
70
0.9
52
Jud
icia
lci
rcu
it1
1.
0.9
01
0.9
06
0.9
19
0.8
61
0.9
39
0.9
36
0.9
45
0.8
72
Jud
icia
lci
rcu
it1
2.
0.8
61
0.8
53
0.7
76
0.9
10
0.8
89
0.8
46
0.8
66
0.8
73
Jud
icia
lci
rcu
it1
3.
1.1
03
1.1
35
0.8
61
1.1
56
0.9
92
0.9
83
0.8
89
1.0
77
Jud
icia
lci
rcu
it1
4.
0.8
40
0.9
18
0.7
18
0.6
47
0.7
89
0.7
43
0.7
96
0.6
18
Jud
icia
lci
rcu
it1
5.
0.7
21
0.7
38
0.6
48
0.8
02
0.7
70
0.7
95
0.7
09
0.8
16
Jud
icia
lci
rcu
it1
6.
1.5
11
1.4
97
1.1
04
1.0
75
1.5
45
1.4
42
1.3
77
0.9
54
Jud
icia
lci
rcu
it1
8.
0.8
76
0.8
80
0.7
90
0.8
78
0.8
50
0.8
17
0.8
33
0.8
27
Jud
icia
lci
rcu
it1
9.
0.8
34
0.8
53
0.7
34
0.8
54
0.8
76
0.9
57
0.7
67
0.8
27
Jud
icia
lci
rcu
it2
0.
0.8
62
0.8
76
0.7
78
0.8
18
0.9
59
0.9
63
0.9
04
0.7
81
Jud
icia
lci
rcu
itn
on
e.0
.506
0.3
62
0.6
10
0.5
74
0.4
86
0.3
93
0.5
54
0.5
78
R2
0.0
52
20
.06
78
0.0
36
80
.068
00
.07
63
0.0
90
60
.061
90
.086
1
Max
imum
-res
cale
dR
20
.109
30
.12
84
0.0
86
60
.114
80
.12
91
0.1
45
90
.110
30
.127
2
Ital
od
ds
rati
os
init
alic
sar
est
atis
tica
lly
sign
ifica
nt
atth
e0
.05
lev
el.
*W
hit
eis
refe
ren
ceca
teg
ory
..Ju
dic
ial
circ
uit
17
isth
ere
fere
nce
cate
gory
.-C
om
munit
yco
ntr
ol
isth
ere
fere
nce
cate
gory
.0C
urr
ent
dru
go
ffen
seis
the
refe
ren
ceca
teg
ory
.
PREDICTING THE EFFECT OF SUBSTANCE ABUSE TREATMENT 185
References
Anglin, D. M., Longshore, D., Turner, S., McBride, D., Inciardi, J. & Prendergast, M.
(1996). Studies of the functioning and effectiveness of Treatment Alternatives to Street
Crime (TASC) programs: Final report.
Banks, D. & Gottfredson, D. C. (2003). The effects of drug treatment and supervision on
time to rearrest among drug treatment court participants. Journal of Drug Issues 33(2),
385Y412.
Belenko, S. (2001). Research on drug courts: A critical review 2001 update. The National
Center on Addiction and Substance Abuse.
Berk, R. A. (1987). Causal inference as a prediction problem. In D. M. Gottfredson &
M. Tonry (Eds.), Prediction and classification: Criminal justice decision making (pp.
183Y200). Chicago: The University of Chicago Press.
Bonta, J., Law, M. & Hanson, K. (1998). The prediction of criminal and violent recidivism
among mentally disordered offenders: A meta-analysis. Psychological Bulletin 123,
123Y142.
Brewster, M. P. (2001). An evaluation of the Chester County (PA) Drug Court Program.
Journal of Drug Issues 31(1), 177Y206.
Bureau of Justice Statistics. (1995). Characteristics of adults on probation, 1995. (Publica-
tion No. NCJ 164267). Washington, DC: Department of Justice, Bureau of Justice Statistics.
Bureau of Justice Statistics. (2003a). Probation and parole in the United States, 2002.
(Publication No. NCJ 201135). Washington, DC: Department of Justice, Bureau of Justice
Statistics.
Bureau of Justice Statistics. (2003b). Prisoners in 2002. (Publication No. NCJ 200248).
Washington, DC: Department of Justice, Bureau of Justice Statistics.
Chanhatasilpa, C., MacKenzie, D. L. & Hickman, L. J. (2000). The effectiveness of
community-based programs for chemically dependent offenders: A review and assessment
of the research. Journal of Substance Abuse Treatment 19, 383Y393.
Deschenes, E. P., Moreno, K. & Condon, C. (2001). Success of drug court participants:
Central and South Justice Centers Superior Court of Orange County, California.
Falkin, G. P., Strauss, S. & Bohen, T. (1999). Matching drug-involved probationers to appro-
priate drug interventions: A strategy for reducing recidivism. Federal Probation 63(1), 3Y8.
Fielding, J. E., Tye, G., Ogawa, P. L., Imam, L. J. & Long, A. M. (2002). Los Angeles County
drug court programs: Initial results. Journal of Substance Abuse Treatment 23(3), 217Y224.
Finigan, M. (1996). Societal outcomes and cost savings of drug and alcohol treatment in the
state of Oregon. Report prepared for Office of Alcohol and Drug Abuse Programs, Oregon
Department of Human Resources, and Governor’s Council on Alcohol and Drug Abuse
Programs.
Flynn P. M., Craddock, S. G., Hubbard, R. L., Anderson, J. & Etheridge, R. M. (1997).
Methodological overview and research design for DATOS. Psychology of Addictive
Behaviors 11(4), 230Y243.
Fulton, B., Latessa, E., Stichman, A. & Travis, L. (1997). The state of ISP: Research and
policy implications. In R. P. Corbett, Jr. & M. K. Harris (Eds.), Federal Probation 61(4),
65Y75.
Gendreau, P., Goggin, C. & Little, T. (1996). Predicting adult offender recidivism: What
works! (User Report No. 1996-07). Ottawa: Department of the Solicitor General of
Canada.
Gendreau, P., Goggin, C., Cullen, F. T. & Andrews D. A. (2000). The effects of community
sanctions and incarceration on recidivism. Forum on Corrections Research 12(3), 10Y13.
PAMELA K. LATTIMORE ET AL.186
Glaze, L. (2004). Probation and parole in the United States, 2003. (Publication No. NCJ
205336). Washington, DC: Department of Justice, Bureau of Justice Statistics. http://
www.ojp.usdoj.gov/bjs/pub/pdf/ppus03.pdf.
Gottfredson, D. C. & Exum, M. L. (2002). The Baltimore city drug treatment court: One-
year results from a randomized study. Journal of Research in Crime and Delinquency
39(3), 337Y356.
Harrell, A. (1998). Drug courts and the role of graduated sanctions. (Publication No. NCJ
169597). Washington, DC: National Institute of Justice.
Hubbard, R. L., Rachael, J. V., Craddock, S. G. & Cavanaugh, E. (1984). Treatment
outcome prospective study (TOPS): Client characteristics and behaviors before, during,
and after treatment (DHHS Publication No. [ADM] 84-1349). In F. Tims & J. Ludford
(Eds.), Drug abuse treatment evaluation: Strategies, progress, and prospects (NIDA
Monograph No. 51) (pp. 29Y41). Rockville, MD: National Institute on Drug Abuse.
Hubbard, R. L., Craddock, S. G., Flynn, P. M., Anderson, J. & Etheridge, R. M. (1997).
Overview of 1-year follow-up outcomes in the Drug Abuse Treatment Outcome Study
(DATOS). Psychology of Addictive Behaviors 11(4), 261Y278.
Lattimore, P.K. (July 1999). Effectiveness of residential drug treatment for Florida
probationers. Paper presented at the Annual Conference on Criminal Justice Research
and Evaluation: Enhancing Policy and Practice, Washington, DC.
Lattimore, P. K., Krebs, C. P., Cowell, A. J. & Koetse, W. (2004). Alternative approaches to
the analysis of large, administrative criminal justice databases. Paper presented at the
2004 American Society of Criminology Conference, Nashville.
Linster, R. L. (1999). Evaluation of Florida’s residential treatment program prison diversion
program, Final report. Washington, DC: Unpublished final report submitted to the
National Institute of Justice.
Linster, R. L., Lattimore, P. K. & Dougherty, K. (1998). Residential drug treatment for
Florida probationers: Preliminary findings. Paper presented by P. Lattimore at the
American Probation and Parole Association Winter Training Institute, Orlando, FL.
Lurigio, A. J. (2000). Drug treatment availability and effectiveness: Studies of the general
and criminal justice populations. Criminal Justice and Behavior 27, 495Y528.
Martin, S. S., Butzin, C. A. & Inciardi, J. (1995). Assessment of a multi-state therapeutic
community for drug involved offenders. Journal of Psychoactive Drugs 27, 109Y116.
Miethe, T., Lu, H. & Reese, E. (2000). Reintegrative shaming and recidivism risks in
drug court: Explanations for some unexpected findings. Crime and Delinquency 46(4),
522Y541.
Oregon Department of Corrections. (1994). Comparison of outcomes and costs: Residential
and outpatient treatment programs for inmates Y alcohol and drug, mental health, sex
offender, and social skills treatment. Salem, OR: Oregon Department of Corrections,
Research and Evaluation Unit.
Paparozzi, M. (1999). Evaluation of intensive supervision programs (ISP) in Montreal and
Toronto. Unpublished manuscript.
Pelissier, B., Wallace, S., O’Neil, J. A., Gaes, G. G., Camp, S., Rhodes, W. & Saylor, W.
(2001). Federal prison residential drug treatment reduces substance use and arrests after
release. American Journal of Drug and Alcohol Abuse 27(2), 315Y337.
Petersilia, J. (1998). A decade of experimenting with intermediate sanctions: What have we
learned? Federal Probation 62(2), 3Y9.
Prendergast, M. L., Anglin, M. D. & Wellisch, J. (1995). Up to speed: Treatment for drug-
abusing offenders under community supervision. Federal Probation 59(4), 66Y75.
Prendergast, M. L., Wellisch, J. & Wong, M. M. (1996). Residential treatment for women
PREDICTING THE EFFECT OF SUBSTANCE ABUSE TREATMENT 187
parolees following prison-based drug treatment: Treatment experiences, needs, and
service outcomes. Prison Journal 76, 253Y274.
Rhodes, W. & Gross, M. (1997). Case management reduces drug use and criminality among
drug-involved arrestees: An experimental study of an HIV prevention intervention. A final
summary report presented to the National Institute of Justice and the National Institute on
Drug Abuse.
Rhodes, W., Pelissier, B., Gaes, G. G., Saylor, W., Camp, S. & Wallace, S. (2001).
Alternative solutions to the problem of selection bias in an analysis of federal residential
drug treatment programs. Evaluation Review 25(3), 331Y369.
Simpson, D., Joe, G., Broome, K., Hiller, M., Knight, K. & Rowan-Szal G. (1997a).
Program diversity and treatment retention rates in the Drug Abuse Treatment Outcome
Study (DATOS). Pyschology of Addictive Behaviors 1194, 279Y293.
Simpson, D., Joe, G. W., Rowan-Szal, G. A. & Greener, J. M. (1997b). Drug abuse
treatment process components that improve retention. Journal of Substance Abuse
Treatment 14(6), 565Y572.
Simpson, D., Joe, G. W. & Rowan-Szal, G. A. (1997c). Drug abuse treatment retention and
process effects on follow-up outcomes. Drug and Alcohol Dependence 47(3), 227Y235.
Stephan, J. J. (2004). State prison expenditures, 2001. (Publication No. NCJ 202949).
Washington, DC: Department of Justice, Bureau of Justice Statistics. http://www.ojp.
usdoj.gov/bjs/pub/pdf/spe01.pdf.
Truitt, L., Rhodes, W. M., Seeherman, A. M., Carrigan, K. & Finn, P. (March 2000). Phase
I: Case studies and impact evaluations of Escambia County, Florida and Jackson County,
Missouri drug courts. Cambridge, MA: Abt Associates.
Turner, S., Longshore, D., Wenzel, S., Deschenes, E., Greenwood, P., Fain, T., Harrell, A.,
Morral, A., Taxman, F., Igushi, M., Greene, J. & McBride, D. (2002). A decade of drug
treatment court research. Substance Use and Misuse 37(12Y13), 1489Y1527.
Van Stelle, K. R., Mauser, E. & Moberg, D. P. (1994). Recidivism to the criminal justice
system of substance-abusing offenders diverted into treatment. Crime and Delinquency
40, 175Y196.
Vito, G. F., Wilson, D. G. & Holmes, S. T. (1993). Drug testing in community corrections:
Results from a four-year program. Prison Journal 73, 343Y354.
Weisburd, D., Petrosino, A. & Mason, G. (1993). Design sensitivity in criminal justice
experiments. In M. Tonry (Ed.), Crime and justice: A review of research, (Vol. 17,
pp. 337Y379). Chicago: University of Chicago Press.
Wexler, H. K., Graham, W. F., Koronowski, R. & Lowe, L. (1995). Evaluation of amity in-
prison and post-release substance abuse treatment programs. Washington, DC: National
Institute of Drug Abuse.
Wolfe, E., Guydish, J. & Termondt, J. (2002). A drug court outcome evaluation comparing
arrests in a two year follow-up period. Journal of Drug Issues 32(4), 1155Y1172.
About the authors
Pamela K. Lattimore is Professor and Director of the Center for the Management of Risk
Behavior, Department of Criminology and Criminal Justice at the University of South
Carolina. Dr. Lattimore is affiliated as a senior research social scientist with Research
Triangle Institute and has directed a number of research and evaluation projects, including
currently serving as co-principal investigator of the Multi-site Evaluation of the Serious and
PAMELA K. LATTIMORE ET AL.188
Violent Offender Reentry Initiative (SVORI), a five-year evaluation of a federal initiative to
improve outcomes for offenders returning to the community from prison. Her research
focuses on the evaluation of interventions, investigation into the causes and correlates of
criminal behavior, particularly among young offenders and probationers, and development
of approaches to improve criminal justice operations. She served as Chair of the Division on
Corrections and Sentencing of the American Society of Criminology from 2001 to 2003 and
serves on the editorial boards of several professional journals. She received her Ph.D. in
economics from the University of North Carolina in 1987.
Christopher P. Krebs is a Senior Research Social Scientist at the Research Triangle
Institute. He earned his B.A. in Sociology from Emory University and his M.S. and Ph.D. in
Criminology from Florida State University. His research interests include research
methodology, juvenile justice and delinquency, corrections and correctional policy, inmate
and offender behavior, and substance use and abuse treatment.
Willem Koetse is a Statistician and Research Analyst at Research Triangle Institute who
works exclusively on criminal justice, substance abuse treatment and violence related
projects. He has a solid background in statistical methodology, experimental design and
using software to manage, model and analyze data. He has a B.A. in Psychology and
Philosophy from SUNY Binghamton, and a master`s degree in Statistics from North
Carolina State University.
Christine Lindquist is a Research Sociologist at Research Triangle Institute. She earned a
Ph.D. in Medical Sociology from the University of Alabama at Birmingham. Her research
interests and experience include multisite evaluation design, violence and substance abuse
prevention, alternative treatment for drug-involved criminal offenders, prisoner reentry, and
violence against women.
Alex J. Cowell is an economist with the Behavioral Health Economics Program at RTI
International. Examples of his work include cost analyses in the areas of substance use,
mental health and criminal justice. Using data for these populations, Dr. Cowell also has
conducted analyses focusing on econometric issues and on topics in services research.
PREDICTING THE EFFECT OF SUBSTANCE ABUSE TREATMENT 189