A Randomized Clinical Trial of a Therapeutic Workplace for Chronically Unemployed, Homeless,
Mikhail N. Koffarnus1, Conrad J. Wong1, Karly Diemer1, Mick Needham1, Jacqueline Hampton1, Michael Fingerhood1,
Dace S. Svikis1, George E. Bigelow2and Kenneth Silverman1,*
1Center for Learning and Health, Johns Hopkins University School of Medicine, MFL W142, 5200 Eastern Ave., Baltimore, MD 21224, USA and2Behavioral
Pharmacology Research Unit, Johns Hopkins University School of Medicine, 5510 Nathan Shock Drive., Baltimore, MD 21224, USA
*Corresponding author: Tel.: +1-410-550-2694; Fax: +1-410-550-7495; E-mail: email@example.com
(Received 2 February 2011; in revised form 18 April 2011; accepted 27 April 2011)
Abstract — Aims: To assess the efficacy of the Therapeutic Workplace, a substance abuse intervention that promotes abstinence while
simultaneously addressing the issues of poverty and lack of job skills, in promoting abstinence from alcohol among homeless alco-
holics. Methods: Participants (n=124) were randomly assigned to conditions either requiring abstinence from alcohol to engage in
paid job skills training (Contingent Paid Training group), offering paid job skills training with no abstinence contingencies (Paid
Training group) or offering unpaid job skill training with no abstinence contingencies (Unpaid Training group). Results: Participants in
the Contingent Paid Training group had significantly fewer positive (blood alcohol level≥0.004 g/dl) breath samples than the Paid
Training group in both randomly scheduled breath samples collected in the community and breath samples collected during monthly
assessments. The breath sample results from the Unpaid Training group were similar in absolute terms to the Contingent Paid Training
group, which may have been influenced by a lower breath sample collection rate in this group and fewer reported drinks per day con-
sumed at intake. Conclusion: Overall, the results support the utility of the Therapeutic Workplace intervention to promote abstinence
from alcohol among homeless alcoholics, and support paid training as a way of increasing engagement in training programs.
Alcoholism and chronic unemployment are some of the most
prevalent problems facing the homeless. Studies estimated
that between 30 and 50% of homeless adults have current
alcohol-use disorders (Breakey et al., 1989; Johnson and
alcohol-use disorders are between 2 and 15 times more
common in the homeless than in housed individuals (Fischer
and Breakey, 1991).
While many substance abuse treatments in this population
are ineffective at reducing alcohol use or abuse (Braucht
et al., 1995; Burnam et al., 1995; Lapham et al., 1995; Smith
et al., 1995; Stahler et al., 1995; Wright and Devine, 1995),
a few treatments have been shown to be effective (Miller,
1975; Lam et al., 1995; Sosin et al., 1995; Milby et al.,
1996). One effective and promising intervention for homeless
substance abusers used the principles of contingency man-
agement, a system where a desirable incentive is provided to
the patient contingent on some therapeutically relevant be-
havior. In a series of studies, Milby et al. (1996 , 2000,
2005) provided housing and work therapy to homeless
substance-abusing individuals who could remain in the
housing and attend work therapy contingent upon periodic
urinalysis tests confirming the absence of drug use. This
intervention decreased substance use significantly. Alcohol
use in homeless adults with frequent arrests for public drun-
kenness was targeted by a contingency management inter-
vention that provided access to goods and services including
temporary work or housing contingent upon objective
measures of reductions in alcohol use (Miller, 1975). This
intervention was shown to decrease arrests for public drun-
kenness and decrease alcohol use.
Other contingency management interventions have shown
some promise in treating alcohol-use disorders (for review,
see Wong et al., 2008) and substance use in general (Higgins
et al., 2008). However, these abstinence reinforcement con-
tingencies alone do not often address the range of problems
associated with extreme poverty that homeless people face.
To address some of these problems, our research group has
employment-based substance abuse treatment intervention
that may have considerable potential in promoting sustained
abstinence from alcohol and drugs in homeless alcohol-
dependent individuals, while simultaneously addressing
some of their interrelated and critical problems of poverty,
unemployment and homelessness. This intervention, the
Therapeutic Workplace, integrates the voucher-based absti-
nence reinforcement contingencies of proven efficacy into an
employment program. In this program, substance-abuse
patients are hired and paid to work each day, performing
data entry jobs. Persons who lack prerequisite academic or
job skills are given intensive and individualized computer-
ized training. To reinforce abstinence, participants in an
experimental group are allowed to work and earn wages only
when they remain abstinent, and abstinence is biologically
verified. This intervention has been shown to be effective at
treatment-resistant young mothers (Silverman et al., 2001,
2002) and abstinence from cocaine in unemployed injection
drug and crack cocaine users enrolled in methadone treat-
ment (Silverman et al., 2007; Donlin et al., 2008; DeFulio
et al., 2009).
The present randomized, controlled trial was conducted to
evaluate the efficacy of the Therapeutic Workplace interven-
tion in homeless, alcohol-dependent adults and to assess the
contribution of the abstinence reinforcement component of
the intervention. A control condition offering job skills train-
ing was compared with two experimental conditions—one
assessing the impact of payment for participation in the train-
ing and one assessing the impact of requiring alcohol
abstinence as a daily pre-condition of access to paid training.
heroin andcocaine in
Alcohol and Alcoholism Vol. 46, No. 5, pp. 561–569, 2011
Advance Access Publication 27 May 2011
© The Author 2011. Published by Oxford University Press on behalf of the Medical Council on Alcohol. All rights reserved
Setting and participant selection
This study was conducted at the Center for Learning and
Health, a treatment-research unit at the Johns Hopkins
Bayview Medical Center (Baltimore, MD) and was approved
by theJohns Hopkins Institutional
Participants were enrolled in this study from December 2001
to October 2005.
Participants were recruited by one of the two methods.
First, individuals were recruited from an inpatient detoxifica-
tion unit located on the Johns Hopkins Bayview Medical
Center, the Chemical Dependency Unit, while undergoing
detoxification from alcohol. While on the inpatient detoxifi-
cation unit, interested participants completed a full screening
interview to determine study eligibility. Individuals were also
recruited from local community agencies that provide services
to the homeless, including food kitchens and homeless shel-
ters. Eligible participants recruited from community agencies
were then assessed by a physician for risk of experiencing
severe adverse effects from alcohol withdrawal. Participants
determined to be at risk for experiencing severe adverse
effects from alcohol withdrawal were required to receive inpa-
tient alcohol detoxification at the Johns Hopkins Bayview
Chemical Dependency Unit prior to study enrollment.
Full screening interview
The full screening included urine and breath samples col-
lected under observation and tested for cocaine, opiates,
benzodiazepines, amphetamines and alcohol; the Addiction
Severity Index (McLellan et al., 1980), a subsection of the
psychoactive substance-abuse sections of the Diagnostic
and Statistical Manual of Mental Disorders, fourth edition
(DSM-IV) Checklist (the questions relating to alcohol,
cocaine, opiate, and benzodiazepines) (Hudziak et al., 1993);
the Risk Assessment Battery (Metzger et al., 1993) and the
Timeline Follow-Back interview (Sobell and Sobell, 1996).
Questionnaires developed in our research program on demo-
graphics and homelessness also were administered during the
intake interview. The Wide Range Achievement Test
(WRAT-3; Wilkinson, 1993) was administered to assess
reading and spelling ability. Participants were paid $50 in
vouchers for completing the full intake interview.
To be eligible, a participant had to be at least 18 years of
age, report being homeless (stayed in a shelter, on the street
or in an abandoned house at least one night over the past 30
days; lost public housing assistance recently or are at risk of
losing residence; or slept in more than two places over the
past 30 days), report being unemployed and meet DSM-IV
criteria for alcohol dependence. Participants were excluded if
they reported current suicidal ideation or reported hallucina-
tions. Eligible participants were invited to sign a main study
Experimental design and study groups
Stratification and random assignment
Of the 321 participants screened, 125 met eligibility criteria.
One participant was removed from the study 2 days after
signing consent when it was found out that he lied at intake,
and was actually employed in a position that could
remaining participants (n =124) were enrolled and randomly
assigned to the Unpaid Training (n=39), Paid Training (n =
42) or Contingent Paid Training (n=43) group. Participants
were randomized to study conditions by a study coordinator,
using a stratification procedure (similar to Silverman, 2004)
based on whether the participant received inpatient detoxifi-
cation for alcohol dependence prior to randomization,
whether full screen urine sample tested positive for cocaine,
and whether full screen urine sample tested positive for
opiates. Note that the 125 randomized participants are fewer
than the intended 156 participants called for by a power
analysis conducted prior to the study. Additional participants
could not be randomized without extending the study dur-
ation beyond that allowed by available funding.
All participants were invited to receive training in the
Therapeutic Workplace for 4 h every weekday throughout a
26-week intervention period, and were required to provide
‘daily’ and ‘random’ breath samples that were tested for
Daily breath samples
After signing into the workplace each morning and after the
daily lunch break, all participants were required to provide a
breath sample under observation. Breath samples were col-
lected with the Alco-Sensor III (AlcoPro, Inc., Knoxville,
Randomly scheduled breath assessments
To attempt to capture instances of alcohol use outside the
workplace and to add an unpredictable element to breath
sample measurements, randomly scheduled breath samples
were also collected. On average, all participants were sched-
uled to provide two breath samples on a random schedule
during a 7-day week between the hours of 9 a.m. and 5 p.m.
Each day was split into four 2-h periods, and each participant
was randomly assigned to be contacted during an average of
twoperiods per week (minimum
Participants were provided with a pager or cell phone, and
were informed that they would be randomly paged or
phoned so that staff could collect a breath sample from them.
Participants were informed that if they answered their pager
or cell phone and allowed research staff to collect a breath
sample within 60 min, they would receive a $35 voucher.
Random breath samples scheduled to occur while a partici-
pant was in the workplace were still collected as scheduled,
and were in addition to the daily breath samples described
one each week).
Independent of Therapeutic Workplace attendance, partici-
pants were contacted once each month and offered $30 in
vouchers for the completion of an assessment, except the
6-month assessment for which they were offered $50 in vou-
chers. These assessments included the collection of a breath,
blood and urine samples, as well as the administration of
some or all of the questionnaires collected at intake by a staff
person blinded to the group assignment of the participants.
562 Koffarnus et al.
Unpaid training group
Participants assigned to the Unpaid Training group were
invited to receive training independent of their daily or ran-
domly collected breath sample results, and they did not earn
monetary vouchers for their participation in the workplace.
Vouchers were still earned for providing randomly scheduled
breath samples and completing assessments.
Paid training group
Participants in the Paid Training group were invited to attend
the workplace every weekday to receive stipend-supported
keyboarding training. Participants in this group could earn a
base pay hourly wage in vouchers for attending the work-
place and productivity pay for performance on the training
programs. These participants were allowed to work and earn
vouchers independent of whether their daily or randomly
scheduled breath samples were positive for alcohol (blood
alcohol level (BAL) ≥0.004g/dl).
Contingent paid training group
Participants in the Contingent Paid Training group were
invited to attend the workplace to receive stipend-supported
keyboarding training similar to participants in the Paid
group. However, for Contingent Paid participants, access to
the workplace and to earn voucher pay was contingent upon
the alcohol content of the daily and randomly scheduled
breath samples. A Contingent Paid participant who provided
an alcohol-positive breath sample (BAL ≥0.004 g/dl) was
not permitted access to the workplace on that day and
received a temporary decrease in pay on subsequent days
Therapeutic workplace training programs
The Therapeutic Workplace is delivered via a web-based
application, which allows staff to administer and electroni-
cally monitor treatment and training for each trainee. Aspects
of the treatment most relevant to the keyboarding training are
described below in detail. Other details of the web-based
Therapeutic Workplace treatment are described in detail else-
where (Silverman et al., 2005).
In the workplace, participants were taught keyboarding
skills using two computer-based training programs (see
Dillon et al., 2004 for a detailed description of the training
programs). Trainees worked on the training programs for 2 h
in the morning (10:00 a.m.–12:00 p.m.) and for 2 h in the
afternoon (1:00 p.m.–3:00 p.m.).
All participants earned $50 in vouchers for completing the
intake and 6-month assessments, $30 in vouchers for com-
pleting other monthly assessments and $35 in vouchers for
each random assessment. Participants in the Contingent Paid
and Paid groups could earn an hourly wage as well as pay
for performance on the training programs. All voucher earn-
ings were automatically added to each participant’s voucher
account and displayed on the participant’s computer screen.
Voucher earnings were exchangeable for goods and services
in the community that were purchased for participants
Hourly base pay began at an initial low rate of $1.00/h and
increased by $.10 to a maximum of $5.00/h for each day a
participant arrived on time (i.e. 10 a.m.) and completed a
work shift (≥3.5 out of 4 h in attendance). The base pay rate
could be reset to $1.00/h if the participant failed to complete
a work shift, or arrived late to the workplace (Paid and
Contingent Paid groups). Base pay was also reset for
Contingent Paid participants if a breath sample was positive
or missed. Once the base pay hourly rate was reset, it
increased again by $.10 per hour for each day the participant
met the attendance and abstinence requirements. After nine
consecutive days of meeting each of the requirements, base
pay was restored to the value in place before the reset.
Productivity pay was not affected by a reset.
To provide some flexibility, participants started training
with 5 ‘late-not-reset days’ and 5 ‘personal days’. In
addition, participants earned 1 ‘late-not-reset day’ for every
10 completed work shifts and 1 ‘personal day’ for every 5
completed work shifts. Participants could use ‘late-not-reset’
or ‘personal’ days to prevent a reset for being late or failing
to work a complete work shift, respectively.
Paid and Contingent Paid participants were able to earn
additional voucher pay for performance on the training pro-
grams. First, participants could earn and lose voucher money
for correct and incorrect characters, respectively. Second,
participants could earn bonuses for each step they passed.
On most steps, participants earned 0.03 cents for every 10
correct characters and lost 0.01 cent for every incorrect char-
acter. The bonuses began at $1.00 and increased in value as
trainees progressed through the program.
Standard treatment services
Standard treatment services were available to all participants.
Motivational enhancement therapy
Motivational Enhancement Therapy was offered to all par-
ticipants in this study according to the procedures specified
in the Project MATCH MET Manual (Miller et al., 1992) by
a Masters-level clinical social worker. In this therapy, a
therapist provides feedback to participants of the risks associ-
ated with alcohol and drug use, utilizes empathy, emphasizes
personal responsibility, gives participants a menu of options
for change and attempts to facilitate self-efficacy (Miller
et al., 1992). Motivational Enhancement Therapy was
provided in four individual sessions.
HIV transmission risk reduction counseling
A masters-level clinical social worker provided HIV risk
reduction counseling following the National Institute on
Drug Abuse HIV Counseling and Education Intervention
Model (Coyle, 1993).
Employment case management
A clinical social worker used the Job Club approach devel-
oped by Azrin and Besalel (1980), a manualized behavioral
approach to vocational counseling.
Therapeutic Workplace for Alcohol-Dependent Adults563
Clinical case management
A clinical social worker provided clinical case management
services, including referrals to address patient needs in the
areas of employment, and mental/medical health. The social
worker also assisted participants in finding shelter/housing
and encouraged them to use voucher earnings to pay for
shelter. The clinical social worker guided participants to seek
available social service benefits, and to utilize Baltimore
City’s network of emergency shelters, transitional shelters
and motel placements, as needed.
The breath samples collected from the participants during the
random and monthly assessments represented the primary
outcome of interest. These samples were analyzed with two
cutoff values: BAL ≥ 0.004 g/dl (which was the cutoff used
to deny access to the workplace or reset base pay for the
Contingent Paid group) and BAL ≥ 0.05 g/dl (which rep-
resents a level with more alcohol-related impairment).
Self-reports of heavy drinking was also collected at monthly
and random assessments. A participant was considered to
have engaged in heavy drinking if he/she reported consum-
ing more than four drinks (females) or five drinks (males) in
the 24 h prior to a random assessment or during at least one
day in the month prior to a monthly assessment. Separate
analyses were conducted with missing breath samples and
reportsof drinking behavior
(missing-missing, MM) or positive (missing-positive, MP).
Between-group comparisons on single measurement baseline
variables were conducted to detect whether group adjust-
ments were needed. Fisher’s exact tests were performed for
dichotomous variables, and one-way analysis of variances
were performed for continuous variables.
Dichotomous outcomes assessed repeatedly over time
within the baseline and intervention periods were analyzed
with an exchangeable correlation structure using general esti-
mating equations (GEE) (Liang and Zeger, 1986). GEE is
particularly suited for analyses of longitudinal data, and
allows for correlations among observations within an individ-
ual subject. GEE was chosen over generalized linear mixed
models as the better of two imperfect options. GEE assumes
that data are missing completely at random (Hu et al., 1998),
an assumption that may have been violated in the MM ana-
lyses (this was not a concern for the MP analyses as there
were no missing data). However, GEE is preferable to
alternative subject-specific approaches such as generalized
linear mixed models that make less stringent assumptions
about missing data for two reasons: (i) GEE makes group-
level inferences about results instead of individual-level
inferences (Hu et al., 1998), which is preferable when inter-
ested in group effects, and (ii) the interpretation of group-
level regression estimates differs between the two approaches
when the outcome is binary, and GEE generates estimates
that are more applicable to the treatment-effect-oriented
experimental questions of the current study (Neuhaus et al.,
1991). Group comparisons were considered both with no
covariates, and with reported drinks consumed per day at
intake and breath sample collection rate included as
covariates. Drinks consumed per day at intake was chosen as
a covariate because the groups reported marginally different
levels of this variable (P =0.08), and previous reports show
a consistent association of this variable with treatment out-
comes (for review, see Adamson et al., 2009). Collection
rate was included as a covariate when it significantly differed
among groups. Two-tailed tests were used and α=0.05 was
used for statistical significance. GEE analyses were con-
ducted in SPSS Statistics v. 17.02 (IBM Corporation,
Somers, NY, USA).
Overall, participants in all three groups reported high rates of
alcohol use, unemployment and homelessness at intake. In
the 30 days prior to intake, participants reported drinking
alcohol to intoxication on 23.85 (SD =8.01) days, working
on average 3.35 (SD=6.53) days and earning $112 (SD =
$369) in net income. No statistically significant differences
among the Contingent Paid, Paid and Unpaid groups were
detected on any of the characteristics collected at intake
(Table 1). Some characteristics did approach statistical
significance, including the racial composition of the groups
(P =0.06), the number of drinks consumed per day at intake
(P =0.08), lifetime treatment episodes for alcohol (P =0.08)
and other drugs (P=0.08), and likelihood of living in a resi-
dence at least one day out of the past 30 (P= 0.09).
Workplace attendance and the result of the daily breath
samples for each participant are shown in Fig. 1. It is appar-
ent by visual examination that the Contingent Paid partici-
pants attended with no alcohol use more regularly than the
Paid participants, although this difference was not statistically
significant (Table 2). Workplace attendance was significantly
lower in the Unpaid group than each of the other two groups
(19 vs. 39% and 45%; P< 0.001; Table 2). The two groups
receiving payment for attendance and performance on the
training programs arrived at the workplace more frequently
were granted access to the workplace more frequently, and
attended with no detected alcohol use more frequently than
the Unpaid group (Table 2). There were no significant differ-
ences between the Paid and Contingent Paid group on these
Breathalyzer outcomes from random assessments
The Contingent Paid group provided significantly fewer
breath samples with BAL ≥0.004 g/dl than the Paid group
(Fig. 2 top; Table 2). Random breath samples with BAL≥
0.004 g/dl in the Unpaid group did not significantly differ
from either of the other groups. The percentage of breath
samples meeting or exceeding 0.05 g/dl and self-reports of
heavy drinking followed a similar pattern as the data for
breath with BAL ≥0.004 g/dl, but neither of these measures
differed significantly among groups (Table 2). If missing
breathalyzer samples or self-reports of drinking were con-
sidered positive, no significant differences were observed on
percent of samples ≥0.004 g/dl, percent of samples≥ 0.05 g/
dl or self-reported heavy drinking.
564 Koffarnus et al.
The collection rate of random breath samples differed sig-
nificantly among groups, with the highest rate in the Paid
group, lowest in the Unpaid group and an intermediate rate
in the Contingent Paid group (Table 2). Exploratory analyses
of abstinence results when collection rate and the number of
drinks consumed per day at baseline were entered as covari-
ates are presented in Supplementary Table S1. Collection
rate significantly influenced abstinence outcomes in the MP
analyses, and number of drinks consumed per day at baseline
influenced abstinence rates with all outcome measures. The
different levels of these covariates among the groups had an
impact on the obtained abstinence rates, since when they
were included in the analyses, the estimated marginal means
of alcohol use among the Unpaid group was somewhat
higher than the Contingent Paid group. This is in contrast to
the unadjusted means (Table 2), which show no consistent
difference between these two groups. The inclusion of these
covariates did not change the overall pattern of group effects,
however (Supplementary Table S1). The largest difference
was between the Contingent Paid and Paid groups, which
met or approached statistical significance, depending on the
outcome measure. The other groups did not differ statistically
from one another.
Breathalyzer outcomes from monthly assessments
The results from the monthly assessment breath samples
largely mirror those from the randomly collected breath
Table 1. Participant characteristics by group
(Fisher’s exact test)
Age, mean (SD)
Gender, % Male
High School Diploma or GED, %
Employment and income
Usually unemployed past 3 years prior to intake, %
Days worked in past 30 days, mean (SD)
Past 30 days income, mean (SD)
Pension, benefits or social security
Mate, family, friends
Drug use and treatment
Days used in past 30 days, mean (SD)
Number of drinks consumed per day in past 30 days, mean (SD)
Money spent in past 30 days, mean (SD)
Times in life treated, mean (SD)
Urinalysis and breath at intake
Opiates, % negative
Cocaine, % negative
BAL, % negative
DSM-IV diagnosis, %
Lived in the following places ≥1 day of past 30 days, %
In a residence
In recovery/treatment program
In a shelter
In a hotel or motel
In a hospital
Grade level reading, mean (SD), no.
Grade levels spelling, mean (SD), no.
Therapeutic Workplace for Alcohol-Dependent Adults565
samples. Percent samples with BAL≥ 0.004 g/dl (Fig. 2
bottom; Table 2) and percent samples with BAL≥0.05 g/dl
(Table 2) were lowest in the Contingent Paid group, signifi-
cantly higher in the Paid group and at an intermediate level
in the Unpaid group that did not differ statistically from
either of the other groups. While a similar pattern of results
was seen with MP analyses of these variables, the differences
were not statistically significant. In addition, self-reported
heavy drinking did not differ among groups.
Unlike with the randomly collected samples, the collection
rate of samples collected during monthly assessments did not
differ among groups (Table 2), and therefore no covariate
analysis is reported with this variable. The number of drinks
consumed per day at intake was considered as a covariate in
an exploratory analysis, however (Supplementary Table S1).
This covariate was significantly associated with verified and
reported alcohol use in both the MM and MP analyses, and
like with the randomly collected breath samples, increased
the spread between the Contingent Paid and Unpaid group
somewhat when included in the analysis. With this covariate,
the Contingent Paid group had significantly lower rates of
verified alcohol use than the Paid group and the Unpaid
group in the MM analyses (BAL ≥ 0.05 g/dl only), with the
Paid and Unpaid groups not significantly different. No sig-
nificant difference among groups was noted in the MP ana-
lyses of verified or reported alcohol use (Supplementary
The current study demonstrates two things: (i) that pay
dependent adults, and (ii) that making access to the work-
place contingent on negative breathalyzer results improves
alcohol abstinence without reducing workplace attendance. It
is perhaps not surprising that pay improves workplace attend-
ance, but the observation that requiring abstinence as a daily
precondition to workplace access significantly improves
abstinence without decreasing attendance is a new finding
that has important implications for how contingency manage-
ment procedures can be clinically implemented to improve
of homeless alcohol-
Fig. 1. Daily attendance and breathalyzer results. Workplace attendance for each participant (ordinate) on each consecutive workday (abscissa) is indicated by
a box if the participant was in attendance. Filled boxes represent days when the participant was in attendance and did not have a positive breath sample, and
open boxes represent days when the participant was in attendance and had a positive breath sample in that day. Open boxes in the Contingent Paid group
represent days in which the participant was denied access to the workplace due to a positive breath sample. The small dots indicate the length of treatment for
each individual, which differed among participants due to holidays and closings. The top, middle and bottom panels represent data from the Contingent Paid,
Paid and Unpaid participants, respectively. Within each panel, participants are sorted so that participants will the most attendance appear at the top of the
566 Koffarnus et al.
alcohol treatment outcomes. The homeless adults in this
study faced many barriers to overcoming their alcohol depen-
dence. Participants were undereducated, and the vast
majority was usually unemployed for 3 years prior to intake.
Housing instability was a serious issue, with a large majority
spending at least one day in the past 30 outdoors, in a
vehicle or in an abandoned house, and 30–40% spending at
least one day in a hospital. Many also lacked social support,
and only one participant out of 124 was married. In addition,
concurrent cocaine or opioid dependence was a common
problem. These data suggest a treatment approach that
addresses multiple barriers simultaneously could be useful
for this population. The current study demonstrated that pro-
viding vouchers contingent upon job skills training attend-
ance can increase attendance, and vouchers contingent on
alcohol abstinence can promote abstinence from alcohol.
While the Contingent Paid group had less verified and
reported alcohol use than the Paid group on a number of
measures, the Unpaid group had an intermediate level of use
that was not statistically more than the Contingent Paid
group in any of the analyses of the raw data without covari-
ates included in the analysis. The relatively low level of veri-
fied and reported alcohol use in the Unpaid group was in no
instance statistically less than the Paid group, but warrants
discussion due to the similar level of abstinence in absolute
terms as the Contingent Paid group. The similar levels of
abstinence in the Contingent Paid and Unpaid groups may
have been an artifact of the collection rate of random breath
samples, as the Unpaid group had the lowest collection rate.
It seems plausible that missing samples would be more
likely to be positive than successfully collected samples (an
assumption that is the basis of MP analyses), raising the
possibility that measured alcohol use in the Unpaid group
was artificially low due to the higher rate of uncollected
samples. Second, one of the variables on which the groups
Fig. 2. Positive breath sample results. Positive breath samples (BAL≥
0.004g/dl) during random breath assessments (top panel) and during
monthly assessments (bottom panel), expressed as a percent of collected
breathalyzer samples. Points represent individual participants in each group
and horizontal lines indicate group means.
Table 2. Comparison of study groups on measures of alcohol abstinence and workplace attendance using GEE results with no covariates
Group means Group GEE resultsBonferonni-corrected post hoc P-value
Cont. (%)Paid (%)Unpaid (%)Chi (df=2)
P-value Cont. vs. Paid Cont. vs. UnpaidPaid vs. Unpaid
Days arrived at workplace
Days worked at least 1 min
Days attended with no alcohol use
Collection rate (%)
Percent BAL≥0.004 (MM)
Percent BAL≥0.004 (MP)
Percent BAL≥0.05 (MM)
Percent BAL≥0.05 (MP)
Self-reported heavy drinking (MM)
Self-reported heavy drinking (MP)
Collection rate (%)
Percent BAL≥0.004 (MM)
Percent BAL≥0.004 (MP)
Percent BAL≥0.05 (MM)
Percent BAL≥0.05 (MP)
Self-reported heavy drinking (MM)
Self-reported heavy drinking (MP)
Group means for each dependent variable are shown, as well as the results from the GEE analyses comparing these groups with no covariates entered in the
analyses. The P values of post hoc tests are also displayed. Bold values indicate statistical significance.
MM, missing samples were treated as missing (missing-missing); MP, missing samples were treated as positive for alcohol use (missing-positive); Cont.,
Therapeutic Workplace for Alcohol-Dependent Adults567
were stratified was the number of self-reported days out of
30 preceding intake with any level of drinking. The average
number of drinks per day was not a stratification variable,
however, and the Unpaid group had a lower number of
reported drinks per day at intake, a difference that
approached statisticalsignificance (P= 0.08; Table 1).
Amount of alcohol consumption has consistently predicted
poorer treatment outcomes across a range of alcohol treat-
ment methodologies, with no known report showing the
opposite association (e.g. Solomon and Annis, 1990;
Duckert, 1993; Sobell et al., 1995; Kavanagh et al., 1996;
Breslin et al., 1997; Kranzler et al., 1999; Greenfield et al.,
2002; Haver, 2003; Staines et al., 2003). It is possible that
the relatively low level of measured alcohol use in the
Unpaid group was partially due to these factors, a supposi-
tion that is supported by the estimated marginal means
displayed in Supplementary Table S1. Accounting for collec-
tion rate (random samples only) and drinks per day at intake
increased the level of positive breath samples in the Unpaid
group relative to the Contingent Paid group for both random
and monthly breath samples, as evidenced by the estimated
marginal means displayed in Table 2 and Supplementary
A factor that complicated the execution of the current
experiment was the collection procedure for the random
breath samples. Due to the relatively rapid elimination rate of
alcohol, standard breath analyses require frequent testing to
obtain an accurate representation of alcohol use. Across the
three groups, the collection rate of the random samples in
this study was ~60%. In addition, no random breath samples
were collected before 9 a.m. or after 5 p.m., leaving open the
possibility that much alcohol use was not captured by these
assessments. These factors might contribute to artificially
low measured rates of alcohol use, and explain why rates of
self-reported heavy drinking were typically higher than rates
of positive breath samples. These limitations do not necess-
arily eliminate the possibility of using contingency manage-
biological assays that have a longer detection window could
be used, as could measurement techniques that reduce the
costs associated with the random assessment procedure in
the current experiment. For example, internet-based verifica-
tion techniques have been developed for measuring carbon
monoxide levels for smoking behavior (Dallery et al., 2007).
Similar techniques could allow for obtaining and verifying
breath samples remotely, although these techniques would
not be practical for homeless populations that lack internet
poverty is a difficult task with many obstacles, but this study
suggests two features of treatment programs that could be
important when addressing this population. First, payment
for attendance in training or assistance programs can signifi-
cantly increase attendance and utilization of these services.
Second, abstinence-contingent access to paid training can
significantly increase abstinence from alcohol.
Supplementary data are available at Alcohol and Alcoholism
Acknowledgements — The authors thank Christopher Taylor for his efforts in random
breath sample collection and Jeannie-Marie Leoutsakos for statistical advice.
Funding — This work was supported by the National Institutes of Health (R01
AA12154, T32 DA007209). The content is solely the responsibility of the authors and
does not necessarily represent the views of the National Institutes of Health.
Adamson SJ, Sellman JD, Frampton CMA. (2009) Patient predic-
tors of alcohol treatment outcome: a systematic review. J Subst
Abuse Treat 36:75–86.
Azrin NH, Besalel VA. (1980) Job Club Counselor’s Manual: A
Behavioral Approach to Vocational Counseling. Baltimore, MD:
University Park Press.
Braucht GN, Reichardt CS, Geissler LJ et al (1995) Effective ser-
vices for homeless substance abusers. J Addict Dis 14:87–109.
Breakey WR, Fischer PJ, Kramer M et al. (1989) JAMA-J Am Med
Breslin FC, Sobell MB, Sobell LC et al. (1997) Toward a stepped
care approach to treating problem drinkers: the predictive utility
of within-treatment variables and therapist prognosis ratings.
Burnam MA, Morton SC, McGlynn EA et al. (1995) An exper-
imental evaluation of residential and nonresidential treatment for
dually diagnosed homeless adults. J Addict Dis 14:111–34.
Coyle SL. (1993) The NIDA HIV Counseling and Education
Intervention Model. Rockville, MD: National Institute on Drug
Dallery J, Glenn IM, Raiff BR. (2007) An internet-based abstinence
reinforcement treatment for cigarette smoking. Drug Alcohol
DeFulio A, Donlin WD, Wong CJ et al. (2009) Employment-based
abstinence reinforcement as a maintenance intervention for the
treatment of cocaine dependence: a randomized controlled trial.
Dillon EM, Wong CJ, Sylvest CE et al. (2004) Computer-based
typing and keypad skills training outcomes of unemployed injec-
tion drug users in a therapeutic workplace. Subst Use Misuse
Donlin WD, Knealing TW, Needham M et al. (2008) Attendance
rates in a workplace predict subsequent outcome of employ-
ment-based reinforcement of cocaine abstinence in methadone
patients. J Appl Behav Anal 41:499–516.
Duckert F. (1993) Predictive factors for outcome of treatment for
alcohol problems. J Subst Abuse 5:31–44.
Fischer PJ, Breakey WR. (1991) The epidemiology of alcohol,
drug, and mental disorders among homeless persons. Am
Greenfield SF, Kolodziej ME, Sugarman DE et al. (2002) History
of abuse and drinking outcomes following inpatient alcohol
treatment: a prospective study. Drug Alcohol Depen 67:227–34.
Haver B. (2003) Comorbid psychiatric disorders predict and influ-
ence treatment outcome in female alcoholics. Eur Addict Res
Higgins HT, Silverman K, Heil SH (eds). (2008) Contingency
Management in Substance Abuse Treatment. New York: The
Hu FB, Goldberg J, Hedeker D et al. (1998) Comparison of
population-averaged and subject-specific approaches for analyz-
ing repeated binary outcomes. Am J Epidemiol 147:694–703.
Hudziak JJ, Helzer JE, Wetzel MW et al. (1993) The use of the
DSM-III-R checklist for initial diagnostic assessments. Compr
Johnson TP, Barrett ME. (1995) Substance use and treatment needs
among homeless persons in Cook County, Illinois. Int J Addict
Kavanagh DJ, Sitharthan T, Sayer GP. (1996) Prediction of results
Kranzler HR, Mulgrew CL, Modesto-Lowe V et al. (1999) Validity
of the Obsessive Compulsive Drinking Scale (OCDS): does
568 Koffarnus et al.
craving predict drinking behavior? Alcohol Clin Exp Res Download full-text
Lam JA, Jekel JF, Thompson KS et al. (1995) Assessing the value
of a short-term residential drug treatment program for homeless
men. J Addict Dis 14:21–39.
Lapham SC, Hall M, Skipper BJ. (1995) Homelessness and sub-
stance use among alcohol abusers following participation in
project H&ART. J Addict Dis 14:41–55.
Liang K, Zeger SL. (1986) Longitudinal data analysis using gener-
alized linear models. Biometrika 78:13–22.
McLellan AT, Luborsky L, Woody GE et al. (1980) An improved
diagnostic evaluation instrument for substance abuse patients.
The Addiction Severity Index. J Nerv Med Dis 168:26–33.
Metzger D, Woody GE, Navaline H et al. (1993) The risk assess-
ment battery (RAB): validity and reliability. Paper presented at
the Sixth Annual Meeting of National Cooperative Vaccine
Development Group for AIDS.
Milby JB, Schumacher JE, Raczynski JM et al. (1996) Sufficient
conditions for effective treatment of substance abusing homeless
persons. Drug Alcohol Depen 43:39–47.
Milby JB, Schumacher JE, McNamara C et al. (2000) Initiating
abstinence in cocaine abusing dually diagnosed homeless
persons. Drug Alcohol Depen 60:55–67.
Milby JB, Schumacher JE, Wallace D et al. (2005) To house
or not to house: the effects of providing housing to homeless
substance abusersin treatment.
Miller PM. (1975) A behavioral intervention program for chronic
public drunkenness offenders. Arch Gen Psychiatry 32:915–8.
Miller WR, Zweben A, DiClemente CC et al. (1992) Motivational
enhancement therapy manual: a clinical research guide for thera-
pists treating individuals with alcohol abuse and dependence.
Project MATCH Monograph Series, Vol. 2. Rockville MD:
National Institute on Alcohol Abuse and Alcoholism.
Neuhaus JM, Kalbfleisch JD, Hauck WW. (1991) A comparison of
cluster-specific and population-averaged approaches for analyz-
ing correlated binary data. Int Stat Rev 59:25–35.
Robertson MJ, Zlotnick C, Westerfelt A. (1997) Drug use disorders
and treatment contact among homeless adults in Alameda
County, California. Am J Public Health 87:221–8.
Silverman K. (2004) Exploring the limits and utility of operant con-
ditioning in the treatment of drug addiction. Behav Analyst
AmJ Public Health
Silverman K, Svikis D, Robles E et al. (2001) A reinforcement-based
therapeutic workplace for the treatment of drug abuse: six-month
abstinence outcomes. Exp Clin Psychopharmacol 9:14–23.
Silverman K, Svikis D, Wong CJ et al. (2002) A reinforcement-
based therapeutic workplace for the treatment of drug abuse:
three-year abstinence outcomes. Exp Clin Psychopharmacol
Silverman K, Wong CJ, Grabinski MJ et al. (2005) A web-based
therapeutic workplace for the treatment of drug addiction and
chronic unemployment. Behav Modif 29:417–63.
Silverman K, Wong CJ, Needham M et al. (2007) A randomized
trial of employment-based reinforcement of cocaine abstinence
in injection drug users. J Appl Behav Anal 40:387–410.
Smith EM, North CS, Fox LW. (1995) Eighteen-month follow-up
data on a treatment program for homeless substance abusing
mothers. J Addict Dis 14:57–72.
Sobell LC, Sobell MB. (1996) Timeline Follow Back: A Calendar
Method for Assessing Alcohol and Drug Use (User’s Guide).
Toronto: Addiction Research Foundation.
Sobell MB, Sobell LC, Gavin DR. (1995) Portraying alcohol treat-
ment outcomes: different yardsticks of success. Behav Ther
Solomon KE, Annis HM. (1990) Outcome and efficacy expectancy
in the prediction of post-treatment drinking behaviour. Br J
Sosin MR, Bruni M, Reidy M. (1995) Paths and impacts in the pro-
gressive independence model: a homelessness and substance
abuse intervention in Chicago. J Addict Dis 14:1–20.
Stahler GJ, Shipley TF, Jr, Bartelt D et al. (1995) Evaluating
alternative treatments for homeless substance-abusing men: out-
comes and predictors of success. J Addict Dis 14:151–67.
Staines G, Magura S, Rosenblum A et al. (2003) Predictors of
drinking outcomes among alcoholics. Am J Drug Alcohol Ab
Wilkinson GS. (1993) WRAT3: Wide Range Achievement Test
Administration Manual. Wilmington, DE: Wide Range Inc.
Wong CJ, Silverman K, Bigelow GE. (2008) Alcohol. In Higgins
ST, Silverman K, Heil SH (eds). Contingency Management in
Substance Abuse Treatment. New York: The Guilford Press,
Wright JD, Devine JA. (1995) Factors that interact with treatment to
predict outcomes in substance abuse programs for the homeless.
J Addict Dis 14:169–81.
Therapeutic Workplace for Alcohol-Dependent Adults 569