University of Denver
Digital Commons @ DU
Electronic Theses and Dissertations
Graduate Studies
1-1-2015
Comparing Short-term Outcomes of Three
Problem Gambling Treatments: A Mulit-group
Propensity Score Analysis
Adam David Soberay
University of Denver, [email protected]
Follow this and additional works at: http://digitalcommons.du.edu/etd
Recommended Citation
Soberay, Adam David, "Comparing Short-term Outcomes of Three Problem Gambling Treatments: A Mulit-group Propensity Score
Analysis" (2015). Electronic Theses and Dissertations. Paper 615.
This Dissertation is brought to you for free and open access by the Graduate Studies at Digital Commons @ DU. It has been accepted for inclusion in
Electronic Theses and Dissertations by an authorized administrator of Digital Commons @ DU. For more information, please contact
[email protected].
COMPARING SHORT-TERM OUTCOMES OF THREE PROBLEM GAMBLING
TREATMENTS: A MULTI-GROUP PROPENSITY SCORE ANALYSIS
____________ A Dissertation
Presented to
The Faculty of the Morgridge College of Education
University of Denver
__________
In Partial Fulfillment
of the Requirements for the Degree
Doctor of Philosophy
__________
by
Adam D. Soberay
March 2015
Advisor: Dr. Antonio Olmos
©Copyright by Adam D. Soberay 2015
All Rights Reserved
Author: Adam D. Soberay
Title: COMPARING SHORT-TERM OUTCOMES OF THREE PROBLEM
GAMBLING TREATMENTS: A MULTI-GROUP PROPENSITY SCORE ANALYSIS
Advisor: Dr. Antonio Olmos
Degree Date: March 2015
Abstract
This study applied a multi-group form of propensity score analysis to the study of
outcomes related to problem gambling treatment. Across various treatment settings, it is
often unfeasible or unethical to randomly assign participants to different treatment
conditions, particularly when one of the conditions involves not receiving treatment.
Additionally, evaluative practices often involve assessing outcomes from a primarily
treatment focused setting, in which case clients are likely not randomly assigned to
treatment. Consequently, where randomization does not exist, methods such as
propensity score matching need to be implemented to separate what part of the observed
outcomes is attributable to treatment and what part may be due to preexisting differences
between the comparison groups. Traditional propensity score matching procedures
involve matching and comparing across two groups, typically a treatment and a control
group. This study applied newly developed methods for matching participants on
propensity scores across three groups.
This study uses archival treatment data to compare three psychotherapeutic
problem gambling treatments (cognitive-behavioral therapy, solution-focused brief
therapy, and time-limited dynamic psychotherapy) where outcomes were likely
influenced by self-selection of form of therapy. Specifically, this study looked at whether
ii
participants improved their psychosocial functioning through five weeks of treatment,
and, if so, are the three forms of treatment equally effective.
The results of this study support the utility of multi-group propensity score
matching procedures. Covariate imbalance was improved through each of the four
implemented matching procedures, though two of the matching procedures (caliper
matching and 3:2:n matching) were more effective in reducing bias. The matching
procedures also indicate that there may be a difference between treatment effects that was
not observed through an unmatched analysis. The matching procedures consistently
estimated the treatment effect for cognitive-behavioral therapy to be greater than that of
the time-limited dynamic psychotherapy. This difference was found to be statistically
significant on two of the four matching methods. Limitations of this study and
recommendations for future research are also discussed.
iii
Table of Contents
Chapter One: Introduction ...................................................................................................1
Background ..............................................................................................................1
Statement of the Problem .........................................................................................5
Purpose of the Study ................................................................................................6
Research Questions ..................................................................................................7
Definitions................................................................................................................7
Review of the Literature ..........................................................................................8
Propensity Score Analysis ...........................................................................8
Problem Gambling .....................................................................................18
Summary ................................................................................................................52
Chapter Two: Method ........................................................................................................54
Participants.............................................................................................................54
Procedure ...............................................................................................................54
Instruments.............................................................................................................58
URICA .......................................................................................................58
NODS.........................................................................................................59
OQ-45 ........................................................................................................60
WAI-S ........................................................................................................61
HANDS ......................................................................................................62
MDQ ..........................................................................................................62
CDGAD .....................................................................................................63
SPRINT-4 ..................................................................................................64
Analytic Strategy ...................................................................................................64
Missing Values Analysis............................................................................64
Research Question One ..............................................................................65
Research Question Two .............................................................................66
Research Question Three ...........................................................................66
Chapter Three: Results .......................................................................................................70
Missing Values Analysis........................................................................................70
Characteristics of the Sample.................................................................................72
Assessing Treatment Selection ..............................................................................72
Comparing Treatment Selection Groups ...............................................................73
Comparing Overall Treatment Groups ..................................................................75
Relationships Between Covariates and Outcomes .................................................76
Unmatched Analysis ..............................................................................................77
Propensity Score Analysis .....................................................................................78
Specifying the Propensity Score Model.....................................................78
Covariate Balance Across Propensity Score Estimation Models ..............79
Maximum Treat Matching .........................................................................80
Caliper Matching .......................................................................................86
iv
2:1:n Matching ...........................................................................................91
3:2:n Matching ...........................................................................................95
Comparing Matching Procedures ............................................................100
Chapter Four: Discussion.................................................................................................102
Overview ..............................................................................................................102
Research Question One ........................................................................................104
Research Question Two .......................................................................................107
Research Question Three .....................................................................................109
Limitations ...........................................................................................................117
Recommendations for Future Research ...............................................................119
References ........................................................................................................................122
Appendix A ......................................................................................................................170
Appendix B ......................................................................................................................175
Appendix C ......................................................................................................................180
Appendix D ......................................................................................................................185
v
List of Tables
Table 1
Description of Therapy Choices ................................................................57
Table 2
Missing Values...........................................................................................71
Table 3
Observed Therapy Selections ....................................................................73
Table 4
Baseline Covariates Across Treatment Selection Groups .........................74
Table 5
Comparing Covariate Balance Across Treatment Groups .........................76
Table 6
Bivariate Correlations Between Baseline Characteristics and Outcome
Measure .....................................................................................................77
Table 7
Pairwise Differences of Covariates Prior to Matching ..............................79
Table 8
Unmatched Participants Within Maximum Treat Matching ......................81
Table 9
Covariate Values Following Maximum Treat Matching ...........................81
Table 10
Inferential Testing of Maximum Treat Post-Matching Covariate Balance
....................................................................................................................82
Table 11
Pairwise Differences of Covariate Balance Following Maximum Treat
Matching ....................................................................................................83
Table 12
Post-Hoc Comparisons of Outcome Following Maximum Treat Matching
....................................................................................................................84
Table 13
Sensitivity Analysis for CBT-TLDP Comparison Following Maximum
Treat Matching ...........................................................................................86
Table 14
Unmatched Participants Within Caliper Matching ....................................86
Table 15
Covariate Values Following Caliper Matching .........................................87
Table 16
Inferential Testing of Caliper Post-Matching Covariate Balance ..............87
Table 17
Pairwise Differences of Covariate Balance Following Caliper Matching .88
Table 18
Post-Hoc Pairwise Comparisons of Outcome Following Caliper Matching
....................................................................................................................89
vi
Table 19
Sensitivity Analysis for SFBT-TLDP Comparison Following Caliper
Matching ....................................................................................................91
Table 20
Sensitivity Analysis for CBT-TLDP Comparison Following Caliper
Matching ....................................................................................................91
Table 21
Unmatched Participants Within 2:1:n Matching........................................92
Table 22
Covariate Values Following 2:1:n Matching .............................................92
Table 23
Inferential Testing of 2:1:n Post-Matching Covariate Balance .................93
Table 24
Pairwise Differences of Covariate Balance Following 2:1:n Matching ....93
Table 25
Unmatched Participants Within 3:2:n Matching........................................96
Table 26
Covariate Values Following 3:2:n Matching .............................................97
Table 27
Inferential Testing of 3:2:n Post-Matching Covariate Balance .................97
Table 28
Pairwise Differences of Covariate Balance Following 3:2:n Matching ....98
Table 29
Comparing Matching Procedures ............................................................101
vii
List of Figures
Figure 1
Covariate Balance from the Three Propensity Scores Estimation Models
....................................................................................................................80
Figure 2
Boxplots of OQ-45 Change for Each Group After Maximum Treat
Matching ....................................................................................................84
Figure 3
Boxplots of Pairwise Differences of OQ-45 Change After Maximum Treat
Matching ....................................................................................................85
Figure 4
Boxplots of OQ-45 Change for Each Group After Caliper Matching .......89
Figure 5
Boxplots of Pairwise Differences of OQ-45 Change After Caliper
Matching ....................................................................................................90
Figure 6
Boxplots of OQ-45 Change for Each Group After 2:1:n Matching...........94
Figure 7
Boxplots of Pairwise Differences of OQ-45 Change After 2:1:n Matching
....................................................................................................................95
Figure 8
Boxplots of OQ-45 Change for Each Group After 3:2:n Matching...........99
Figure 9
Boxplots of Pairwise Differences of OQ-45 Change After 3:2:n Matching
....................................................................................................................99
viii
Chapter One: Introduction
Background
Quasi-experimental or observational methods are utilized frequently in social
science research, as randomized experimental methods are often unfeasible or unethical.
For instance, it may be unethical in certain circumstances to randomly assign participants
to a no-treatment control group, particularly when denying treatment would put the
individuals at risk. Consequently, such studies often involve self-selection of participants
into treatment groups. This self-selection comes at the price of the balancing of
confounding variables across groups that is associated with randomization (Heckman,
Ichimura, Smith, & Todd, 1998). In addition, when self-selection occurs systematic
differences may exist between the groups that may confound the relationship between
treatment and outcome. Therefore, treatment assignment is likely to violate the ignorable
treatment assignment assumption (ITAA), which states that assignment to any treatment
condition is independent of outcome (Rosenbaum, 1984). Consequently, violation of the
ITAA constitutes a selection bias that can pose a significant threat to the internal validity
of the research findings. Moreover, this selection bias, if left unchecked, weakens the
ability to draw conclusions on the causal counterfactual, which is the outcome of
participants had they been in a different treatment condition.
1
Traditional approaches to controlling for these potentially confounding influences
often involve multivariate analyses, which statistically control for the influence of
covariates on the outcome. This approach will lead to unbiased estimates of the treatment
effect so long as the ITAA is not violated and the model is appropriately specified
(Rosenbaum, 1994). However, the ITAA is analogous to the ordinary least squares
regression assumption of independence of the error term from all independent variables.
When treatment is included in a multivariate analysis as a dichotomous independent
variable, this assumption of independence is violated when ITAA does not hold, because
it results in dependence between the error term and an independent variable.
Consequently, estimates of treatment effect will be biased (Guo & Fraser, 2010).
Therefore, this traditional method of statistical control of covariates is often inappropriate
for quasi-experimental and observational designs. Propensity score analysis was
developed as a means to address such a problem (Rosenbaum & Rubin, 1983). It is a set
of procedures that allow the researcher to account for the differences between groups that
are typically associated with observational and quasi-experimental research.
Within the general public, and even within the treatment field, addiction is a
concept that virtually everyone has an opinion on, yet where agreement is lacking
regarding what it truly is. Accordingly, the varying conceptions of addiction reflect the
underlying assumptions regarding the factors contributing to the origins, continuation,
and elimination of the addictive behavior. There also does not seem to be consistent
agreement on which behavioral disorders can and cannot constitute an addiction.
2
The National Association of Alcoholism and Drug Abuse Counselors defines
addiction as a brain disease (National Association of Alcoholism and Drug Abuse
Counselors, 2013). Such a conceptualization reflects a growing body of research that
shows the role of, and effect on the brain associated with addiction.
The American Society of Addiction Medicine also focuses on the role that brain
circuitry plays in addictive behaviors (American Society of Addiction Medicine, 2013).
They indicate that addiction brain circuitry manifests itself in biological, psychological,
and social dysfunction, which leads an individual to seek reward and/or relief through
their addictive behavior. This definition suggests that addiction has physiological,
cognitive, and emotional, as well as social components. Kranzler and Li (2008) indicates
that due to the multiple components of addiction, it crosses disciplines, including biology,
psychology, sociology, and pharmacology.
Various twelve-step organizations, such as Alcoholics Anonymous and Gamblers
Anonymous, define addiction as an illness (Alcoholics Anonymous , 1972; Gamblers
Anonymous, 2013). These organizations hold that the addictive behavior will continue to
progressively worsen unless the behavior is completely abstained from. This view is in
contrast to others that hold that the addictive behavior can become responsibly
moderated.
Addiction is sometimes viewed as an umbrella term that covers disordered usage
of drugs and alcohol as well as other behaviors such as gambling, shopping, and sexual
activity. However, in other instances addiction can be more narrowly defined as only
pertaining to drugs and alcohol. The American Psychological Association defines
3
addiction as a condition in which a person must take a substance to avoid psychological
and physical withdrawal (American Psychological Association, 2013). This
conceptualization also includes issues such as a physical dependence, as well as building
up a tolerance to the substance, which are concepts that are generally associated with
alcohol and substance abuse. However, issues of withdrawal have also been associated
with other behavioral addictions, such as pathological gambling (Rosenthal & Lesieur,
1992).
The American Psychiatric Association’s Diagnostic and Statistical Manual of
Mental Disorders (DSM) through its iterations has shown an evolution of the
conceptualization of addictive disorders. Prior to the latest edition, the DSM-5, the DSM
has emphasized the dependence aspects of addictive disorders; in fact, it used the term
dependence in lieu of the term addiction (Kranzler & Li, 2008). However, the fifth
edition of the DSM has seen significant changes to these diagnoses. There has been a
growing disenchantment with the choice of the term dependence in the DSM, as physical
dependence and its associated tolerance and withdrawal is a phenomenon often also
found with appropriate usage of medication (Courtwright, 2011; O’Brien, Volkow, & Li,
2006; O’Brien, 2011). Moreover, this conceptualization largely excluded non-substance
behavioral addictions, such as gambling. Among the major changes in the DSM-5 was
the introduction of a new category of behavioral addictions, to which gambling was the
first addition (American Psychiatric Association, 2013). This move is often viewed as the
first step in the inclusion of other behavioral addictions, such as internet addiction
(Holden, 2010).
4
When viewing addiction as a phenomenon that transcends alcohol and substance
misuse researchers have identified elements that seem to be hallmarks of addiction.
Shaffer (2011) identified three aspects of a behavior that go towards identifying it as an
addiction. According to Shaffer, for an addiction to exist, the following three
characteristics must be present: the addictive behavior is motivated by emotions, the
behavior is continued despite negative consequences, and the person is experiencing
some level of loss of control over their behavior.
Statement of the Problem
Until very recently, the implementation of propensity score matching has dealt
with matching across two groups. This lent itself to studies that involved the comparison
of a treatment and a control group. However, these methods were not fully conducive to
the comparison of multiple treatment groups. Recent advances in statistics have provided
approaches to generate propensity score matching procedures that can address
comparisons of more than two groups (Bryer, 2013; Rassen et al., 2013). As these
methods are relatively new, the current literature lacks examples of practical
implementations of these methods.
Across the social sciences, research related to the treatment of problem gambling
continues to lag far behind the evidence base surrounding other addictive disorders,
particularly alcohol and substance abuse. Moreover, gambling addiction is a problem that
can have significant adverse consequences, individually, interpersonally, as well as
socially. Therefore, treatment must be available that utilizes evidence-based practices, so
as to ensure that this problem is addressed in an efficacious manner.
5
Purpose of the Study
As an application of a multi-group propensity score analysis, this study seeks to
compare the effectiveness of three common psychotherapeutic orientations, solution
focused brief therapy (SFBT), time limited psychodynamic therapy (TLPD), and
cognitive-behavioral therapy (CBT) for the treatment of gambling problems. CBT
treatment methods have well-documented effects in the treatment of gambling problems.
However, very little empirically based knowledge exists on the application of SFBT and
TLPD in this area. Consequently, in addition to adding to the literature regarding a threegroup propensity score analysis, this study also seeks to add to the literature by
determining whether these relatively unstudied problem gambling treatments can be
equally as effective as the well-studied cognitive-behavioral approach.
The first component of this study will explore whether various factors help to
explain the particular therapeutic selections of the participants. The study will include the
following variables in relation to treatment selection: age, gender, stages of change, cooccurring psychological disorders (depression, post-traumatic stress disorder, anxiety,
and mood disorder), initial level of psychosocial functioning, and severity of gambling
problems. Moreover, it will explore whether the problem gamblers in this study are
disproportionately selecting any of the treatment options.
The second component will analyze whether each of the treatments is effective
and the relative effectiveness of each treatment. Particularly, the study will examine
short-term improvement in psychosocial functioning through the first five sessions of
treatment.
6
Research Questions
The following three research questions will be addressed:
1) Is therapeutic selection explained by various demographic and psychological
indicators? Are there certain subgroups of problem gamblers that may prefer one type of
therapy relative to others?
2) Overall, do problem gambling therapy clients have particular therapeutic
preferences? Are they selecting each of the three therapies in equal proportions?
3) Are these three psychological interventions for problem gambling effective in
reducing overall psychosocial distress? Are the interventions equally effective?
Definitions
Propensity scores are the conditional probabilities of receiving a particular level
of treatment given a set of pre-treatment covariates. Propensity score analysis refers to
any of a set of statistical procedures involving the utilization of propensity scores in the
calculation of a treatment effect.
Pathological gambling is a disorder from the Diagnostic and Statistical Manual
IV-Text Revisions (DSM-IV TR) and is assigned according to a set of criterion therein
(American Psychiatric Association, 2000). Gambling disorder is the diagnosis in the
current, 5th edition of the DSM (American Psychiatric Association, 2014). Problem
gambling is defined as gambling behavior that is having significant adverse consequences
on the psychosocial functioning of the individual engaging in the behavior. Therefore,
problem gambling is a term that encompasses both diagnosable as well as sub-clinical
levels of disordered gambling behavior.
7
For the purposes of this paper, addiction will be viewed as a psychological
disorder with multiple epidemiological influences. It will also be viewed as having
psychological, interpersonal, physical, and social consequences. It is also viewed as a
concept that applies to behaviors beyond substance abuse, which share common
characteristics, as outlined by Shaffer (2011). Further, addiction will be held as a
treatable disorder. Therefore, gambling addiction is used to describe the underlying
psychological condition that leads to disordered gambling behavior.
Psychotherapy refers to any of various forms of treatment consisting primarily of
talk-based interventions for psychological disorders, including problem gambling.
Review of the Literature
Propensity Score Analysis
Propensity score matching is a set of procedures designed to statistically
counteract the selection bias associated with non-experimental research (Rosenbaum &
Rubin, 1983). The procedure generally consists of two or three steps (Guo & Fraser,
2010). The first step in either case is the estimation of the propensity scores, which is
typically achieved through logistic regression. The logistic regression produces a
conditional probability that an individual will be in the treatment, as opposed to the
comparison group. These probabilities become the propensity scores. Propensity scores
reduce multidimensional covariates into a single value. Therefore, a treatment and a
comparison participant with comparable propensity scores are considered to be balanced
on the vector created by the combination of the covariates. It should also be noted that in
addition to logistic regression, propensity scores can also be estimated through
8
discriminant function analysis or by OLS regression where the propensity score is the
predicted value when a binary, dummy outcome representing treatment group is
regressed upon the observed covariates (Holmes, 2014).
A critical component of this step of the process is adequate specification of the
logistic regression model. In other words, the selection of variables to include in this
model plays a large role in determining how well the propensity scores will balance
groups. Data simulations have shown that propensity score matching can exacerbate the
bias produced by unmeasured covariates, so long as these unmeasured covariates vary
independently of those being measured and included in the propensity score estimation
model (Brooks & Ohsfeldt, 2013). There is some debate as to whether to include in the
propensity score estimation model all measured covariates, only those related to
treatment selection, only those related to outcomes, or only those related to assignment
and outcome (Austin, 2011). However, it has been demonstrated that propensity scores
can be estimated more effectively when including variables associated with the outcome
variable, and not only those that are believed to simultaneously impact selection and
outcome (Cuong, 2013). Further, it has also been argued that a characteristic that varies
between treatment groups and has no influence over the outcome(s) will not bias the
estimates of treatment effect (Tanner-Smith & Lipsey, 2014).
After estimating the propensity scores, determining if balance has been achieved
is another important step in the process. There are a number of ways to check balance,
including: significance tests, standardized differences, percent reduction, and graphical
methods (Holmes, 2014). Significance tests, for continuous variables, include F and t
9
tests, or the Mann-Whitney test can be used if the distribution is non-normal. For
categorical covariates, chi-square tests of independence can determine if significant
differences exist. Standardized mean differences of each covariate between groups can
reveal which covariates significantly differ between groups, as well as be used to
determine whether or not post-matched groups are balanced on the observed covariates
(Kuss, 2013). Additionally, an advantage of assessing standardized differences is the
ability to determine relative imbalance among the covariates. Percent reduction deals
with calculating the proportion of the differences in observed covariates between groups
that is reduced through matching. Finally, graphical procedures, such as Q-Q plotting
determine balance through comparing the distribution of the propensity scores across
groups.
If through matching the systematic differences in the baseline covariates have
been removed then the propensity score model is considered to be adequately specified
(Austin, 2011). Sensitivity analysis can also be conducted to search for hidden bias in the
propensity score estimation that may indicate the omission of a significant covariate
(Rosenbaum, 2002). If satisfactory balance is not achieved, Holmes (2014) recommends
adding relevant covariates or assessing and correcting for non-linear relationships, which
may require data transformations.
For the three-step procedures, the second step is the selection and implementation
of a matching procedure. Once propensity scores have been estimated, there are various
categories of strategies for matching participants once the propensity scores have been
10
calculated, including: greedy matching, optimal matching, and fine balancing (Guo &
Fraser, 2010).
Greedy matching procedures include: Mahalanobis metric, nearest neighbor, and
caliper matching (Guo & Fraser, 2010). Mahalanobis metric matching procedures predate
propensity score matching (Cochran & Rubin, 1973). These initial procedures involved
matching participants based on Mahalanobis distances, or the multivariate distance of a
participant from a common point. With the introduction of propensity scores,
Mahalanobis metric matching was expanded to include propensity scores as an additional
covariate. Nearest neighbor matching procedures involve matching participants across
groups based on minimum absolute difference in propensity scores. These procedures can
either involve a 1:1 match, or a 1:n match where a participant in the treated group can be
matched to multiple controls. Caliper matching extends nearest neighbor procedures by
placing a constraint on the maximum allowable distance between propensity scores. In
traditional nearest neighbor approaches, a match will still be made even if there is a
relatively large discrepancy between propensity scores so long as it is the closest possible
match. This combination of nearest neighbor and caliper matching begins by randomly
ordering treatment and comparison participants. Then, in order, treatment participants are
matched with the closest control so long as the match is within the predetermined caliper.
The caliper can vary, though 0.25 standard deviations of the propensity scores is a typical
value, with tighter calipers resulting in reduced bias but increased unmatched participants
(Lunt, 2014). Further, an analysis of propensity score matching among three groups
compared calipers in increments of 0.1 and found that 0.2 was optimal in yielding
11
estimates of treatment effect (Wang et al., 2013). Related to this idea of increasing the
precision of matches, it has been shown that the more precisely propensity score
matching balances groups on the observed covariates, the greater the risk for loss of data,
which can be both a statistical and financial issue (Golinelli, Ridgeway, Rhoades, Tucker,
& Wenzel, 2012). This nearest neighbor within a caliper procedure has gained popularity
as the matched samples can be analyzed using virtually any multivariate statistical
procedure. In fact, a major advantage of these procedures is the ability to implement
relatively simple post-matching analyses, such as a paired samples t-test, which further
makes these procedures accessible to most clinical researchers (Cotton, Cuerden, &
Cook, 2011). Finally, greedy matching procedures can also include a combination of the
nearest neighbor within a caliper and Mahalanobis metric matching procedures.
One of the primary drawbacks to the greedy matching procedures is the
requirement of significant overlap in propensity scores between the treated and
comparison participants (Heckman, Ichimura, Smith, & Todd, 1996). Without this
common support region of scores there can be a significant loss in data through
unmatched participants.
As opposed to greedy matching procedures, which match each treatment
participant sequentially with its closest match without regard for future matches, optimal
matching procedures do not take this sequential approach, which can result in more
precise overall matching (Rosenbaum, 1989). In other words, optimal matching
procedures are focused upon creating matches that contribute to the minimum total
distance in propensity scores, as opposed to minimum distances within individual
12
matches. There are three general strategies to optimal matching: pair matching, variable
matching, and full matching. Similar to nearest neighbor matching, pair matching and
variable matching create 1:1 and 1:n matches, respectively. With full matching, not only
are treated participants matched to one or more control participants, but control
participants are matched to one or more treated participants (Rosenbaum, 2002).
Fine balancing is a relatively new approach to matching (Rosenbaum, Ross, &
Silber, 2007). Unlike greedy and optimal matching, which use a single propensity score
to balance groups, fine balancing involves balancing groups based on a nominal variable.
Fine balancing is a procedure that is used in conjunction with propensity score matching,
and involves balancing a particular covariate across groups rather than between
individual participants.
In addition to the aforementioned matching procedures, there are also
nonparametric approaches to propensity score matching. A primary advantage of
nonparametric statistical procedures is their ability to provide accurate statistical
estimations when the assumptions regarding the normality of the underlying populations
cannot be upheld (Hollander, Wolfe, & Chicken, 2013). Kernel-based matching is an
approach to propensity score matching that utilizes nonparametric regression (Heckman,
Ichimura, & Todd, 1998). Kernel-based matching produces one-to-many matches, which
involves weighting matches based on closeness of propensity scores. Kernel-based
matching has been shown to be equally effective as 1:1 matching when there are
sufficient control participants but more effective when number of control participants are
equal to or less than the number of treated participants (Berg, 2011). Semiparametric
13
propensity score estimation techniques have also demonstrated value when assumptions
related to normality are violated (Lehrer & Kordas, 2013). Random forest classification is
a nonparametric procedure designed to predict group membership based on a series of
covariates, which has successfully been applied to propensity score estimation (Watkins
et al., 2013).
The final of the three steps is the post-matching statistical analysis. As discussed
earlier, a primary benefit to greedy matching is that multivariate analyses may be applied
to determine whether treatment effects truly exist after balancing the groups on the
observed covariates. Another analytical approach using propensity scores involves the
stratification of participants. Through this approach, participants are divided into strata,
often quintiles, based on propensity score. Mean differences are then calculated within
each stratum, which are then averaged together to estimate the overall treatment effect.
With optimal matching, unlike greedy matching, special considerations must be taken as
to which post-matching analyses are appropriate (Rosenbaum, 1989). For full and
variable matching procedures, the Hodges-Lehmann aligned rank test can be used to
estimate the average treatment effect. Finally, optimal match pairs can be subject to a
special type of regression adjustment whereby the difference on the outcome variable
between a matched treatment and comparison participant is regressed on the difference
between the observed covariate(s). There is also an analytic technique that combines the
Hodges-Lehman procedure with the regression adjustment (Fraser & Guo, 2010).
However, not all propensity score techniques involve a matching procedure. For
the two-step procedures, statistical analyses are conducted using propensity scores as
14
regression weights (Freedman & Berk 2008). However, if the causal model and/or the
propensity score estimation model are not adequately specified then this procedure can be
counterproductive through increasing random error.
Another facet of propensity score matching that has been explored is the usage of
bootstrapping procedures. A bootstrap procedure, which uses the mean propensity score
across bootstrap samples, has been shown to reduce the bias resulting from multiple
propensity score matching procedures (Bai, 2013).
In terms of practical applicability, propensity score matching has demonstrated
the ability to result in relatively unbiased estimates of treatment effect even for sample
sizes as small as forty (Pirracchio, Resche-Rigon, & Chevret, 2012). Therefore, this
procedure can be conducted in various settings where large amounts of data are not
available.
If conducted properly, propensity score analyses can be very beneficial to quasiexperimental and observational research. It is a sophisticated yet elegant approach to
counteracting the selection bias that exists in the absence of randomization. By removing
this selection bias, the researcher is able to increase the internal validity of their findings
through better isolation of the effect of the treatment on the outcome, which is a critical
component of drawing conclusions regarding the causal effects of the treatment
(Campbell, 1957).
One of the major practical drawbacks of traditional propensity score analysis
procedures is the limitation of comparing only two treatment conditions, typically a
treatment and a control group. However, in practical research settings, comparisons are
15
often drawn between multiple treatment groups. One approach to comparing multiple
groups using binary propensity score approaches is through the implementation of
pairwise comparisons of each treatment condition. However, simulation study has
demonstrated a significant bias associated with this procedure when significant treatment
effect heterogeneity exists (Rassen et al., 2013).
Imbens (2000) is generally credited with the initial work in extending propensity
score analysis to applications for multiple treatment groups. Imbens’ definition of the
generalised propensity score is the conditional probability of receiving a particular level
of the treatment given the observed covariates. This extension of the definition of
propensity scores allows for multi-group comparisons. Propensity scores in this context
are generally calculated through multinomial logistic regression, particularly when the
levels of treatment are qualitatively different (Baser, 2008). However, implementation of
multiple group propensity score analyses is relatively limited, particularly with regard to
practical applications.
As with binary propensity score analysis, multiple strategies can be used in the
implementation of the propensity score. One approach to utilizing multiple propensity
scores is to include them as independent variables within a multiple regression equation
along with dummy variables related to treatment received (Spreeuwenberg et al., 2010).
Additionally, both propensity score regression adjustment and propensity score weighting
procedures have been demonstrated to be effective in multiple group comparisons (Feng,
Zhou, Zou, Fan, & Li, 2012).
16
Recently, statistical software packages have been created to conduct propensity
score matching across multiple groups (Bryer, 2013; McCaffrey et al., 2013; Rassen et
al., 2013). In particular, the TriMatch package was created for the R statistical software
specifically to compare three non-equivalent treatment groups (Bryer, 2013). TriMatch
begins by using logistic regression to calculate propensity scores. Several matching
strategies then exist based on minimizing the total standardized distance between
propensity scores within a given matched triplet. Matching options include matching
within a specified caliper and with specified ratios (e.g. 3:2:n or 2:1:n). The TriMatch
package also provides multiple means for checking the covariate balance achieved
through the matching process, including multiple graphical outputs. Finally, this package
performs analysis on the matched groups, using repeated measures ANOVA and a
Friedman Rank Sum Test. If either of these tests is found to be significant, a Wilcoxon
Signed Rank Test and three dependent sample t-tests are performed to assess differences
between individual groups.
Utilizing one of these multi-group matching procedures may result in more
precise estimates of treatment effect. Simulation has shown that such matching
procedures that utilize an algorithm to create matches across all comparison groups have
been shown to be preferable to pairwise approaches of comparing multiple groups in
terms of reducing bias and increasing covariate balance, particularly when treatment
effect heterogeneity exists (Rassen et al., 2013). Therefore, the TriMatch procedures have
been chosen for this application.
17
Problem Gambling
Addiction, in a broad sense, affects a relatively large proportion of the general
population. Recent estimates indicate that over 10% of the adult American population
will develop a drug use disorder at some point during their lifetime (Compton, Thomas,
Stinson, & Grant, 2007). However, when expanding the definition of addiction to include
behavioral addictions, such as gambling, sex, exercise, and shopping, it has been
estimated that approximately 47% of the U.S. adult population may exhibit addiction-like
behavior (Sussman, Lisha, & Griffiths, 2011). This underscores the need for adequate
identification and treatment of addictive disorders, particularly given the profoundly
negative consequences these behaviors can have on the individuals experiencing them,
those around them, as well as society as a whole.
Gambling addiction is also a relatively common phenomenon. In fact, given the
often hidden nature of the disorder, rates of gambling addiction are likely much higher
than the general public perceives. Current estimates suggest that from 2% to 6% of the
population will develop significant gambling related problems during their lifetime;
moreover, an estimated additional 2% will develop diagnosable levels of gambling
problems during their lifetime (Kessler et al., 2008; Park et al., 2010; Penfold et al.,
2006a, b; Shaffer & Hall, 2001). Therefore, as many as approximately one in twelve
people will experience significant problems as a result of gambling at some point during
their lifetime. Moreover, these estimates are likely to increase with the introduction of the
changes to the gambling diagnosis in the DSM-V, most notably the decreasing of the
threshold for diagnosis (Weinstock et al., 2013).
18
Gambling has become ubiquitous in the United States over the past several
decades. Casino gambling, in particular, has seen a dramatic increase. Annual revenue
from legal casinos in the United States has increased substantially over the past several
decades as more and more states have legalized casino gambling (Eadington, 1999). In
fact, between 1991 and 2006, casino revenue in the United States increased from 9 billion
dollars annually to 32 billion (Ozurumba, 2009). Although the total gross gambling
revenues in the United States decreased approximately 8% between 2007 and 2009,
coinciding with an overall downturn in the American economy, the revenue was still
greater in 2009 than in 2004 indicating a continuing upward trend in gambling
(Eadington, 2011). This dramatic increase in gambling revenues is being observed
outside of the United States as well, particularly in Europe, Asia, South Africa, and
Australia (Wynne & Shaffer, 2003).
Casino gambling, of course, is not the only form of gambling that is generating a
tremendous amount of revenue in the United States. State sponsored lotteries continue to
be the largest source of gambling revenue. According to the North American Association
of State and Provincial Lotteries (2014), in 2011 total lottery revenues in the United
States exceeded 63 billion dollars. Further, charitable gambling activities, those
conducted by non-profit organizations, generated approximately 1.3 billion dollars in
gambling revenues in 1997; however, it has been shown that as the availability of forprofit gambling opportunities increases, the ability of non-profit organizations to generate
revenue through charitable gambling decreases (Dolan & Landers, 2006).
19
Another major element related to the increase in overall gambling is the
proliferation of online gambling opportunities. Online gambling has increased
dramatically over the past few years, and internet gamblers may be particularly
susceptible to developing gambling problems (Wood & Williams, 2007). Furthermore,
internet gambling may be particularly attractive to younger people, which may suggest
that it will continue to grow as a preferred means of gambling (Wood & Williams, 2011).
Regardless of the particular form it takes, gambling in the United States is an
activity that people are spending a tremendous amount of money on each year. Further,
gambling behavior and the monies associated with it continue to increase. Therefore,
gambling is an activity that must be taken seriously, particularly as it relates to the
potential for a continued increase in associated problems.
Until very recently, pathological gambling was listed in the Diagnostic and
Statistical Manual of Mental Disorders within the category of impulse-control disorders,
along with other behavioral disorders such as kleptomania, pyromania, and intermittent
explosive disorder (American Psychiatric Association, 2000). However, clinical and
research experience have shown that disordered gambling behavior shares many common
characteristics with drug and alcohol misuse, including psychological and treatment
components (Ashley & Boehlke, 2012; El-Guebaly et al., 2012; Potenza, 2006). Further,
similar neurological pathways and vulnerabilities have been implicated for both
substance abuse and gambling disorders (de Ruiter, Oosterlaan, Veltman, van den Brink,
& Goudriaan, 2012; Frascella, Potenza, Brown, & Childress, 2010; Koehler et al., 2013).
The alignment of gambling addiction with alcohol and substance use disorders will
20
hopefully have the impact of further legitimizing gambling disorder as a valid condition
that will receive the recognition and resources it deserves.
Proposed changes have been enacted whereby the disorder was renamed from
“pathological gambling” to “disordered gambling.” It has also become re-categorized
with substance-related disorders (Holden, 2010; Petry, 2010). Additionally, the criterion
related to the committing of illegal acts has been eliminated from the list of diagnostic
criteria, and the threshold for diagnosis has been reduced from five of ten to four of nine
criteria (Mitzner et al., 2011). Other proposed changes, however, were not accepted into
the revisions of the DSM-V. For instance, Cunningham-Williams et al. (2009) advocated
for the expansion of criteria related to the withdrawal-like symptoms associated with
gambling.
Initial investigation of the impact of the removal of the illegal acts criterion and
reduction of the threshold for diagnosis indicates an improvement in diagnostic
classification without a reduction in the psychometric properties of existing measures,
such as the National Opinion Research Center DSM-IV Screen for Gambling Problems
(NODS; Petry et al., 2012). An initial investigation using gambling helpline callers
estimates that the changes will result in an increase in individuals meeting the diagnostic
criteria of as much as 11% depending on the setting (Petry et al., 2012; Weinstock et al.,
2013).
The legalization and growing availability of gambling is often viewed as having
various and far reaching social consequences (Barmaki, 2010). The most obvious
consequence is that as gambling opportunities increase so too will gambling behaviors.
21
The opening of casinos has been associated with an increased in gambling frequency,
amounts of money lost gambling, as well as problem gambling behavior in the
surrounding areas (Jacques, Ladouceur, & Ferland, 2000; Room, Turner, & Ialomiteanu,
1999). The fallout from the increased availability of gambling may occur relatively
quickly, and it will likely continue to increase over time. It has been estimated that a
majority of treatment seeking problem gamblers had developed significant gamblingrelated problems within 2 years of beginning gambling (Lahti, Halme, Pankakoski,
Sinclair, & Alho, 2012). Further, the prevalence of pathological gamblers has been found
to continue to increase in areas where legal gambling has been available for extended
periods of time. Rates of pathological gambling may be as much as three times higher in
areas where legal gambling has been available for more than 20 years compared to areas
where legal gambling has been available for less than 10 years (Volberg, 1994).
This then raises the issue of whether states or countries that allow legalized
gambling have a social obligation to address issues of gambling addiction. Unfortunately,
governmental support for research and treatment are far from commensurate with the
growth in legal gambling (Pavalko, 2004). At the mental health agency level, there may
be a general lack of preparation by the agency for the anticipated increase in gambling
addiction that accompanies gambling expansion (Engel, Rosen, Weaver, & Soska, 2010).
Economic factors, such as high unemployment rates, are commonly cited as
reasons for the legalization of gambling (Argusa, Lema, Asage, Maples & George, 2010;
Richard, 2010). It is assumed that legalized gambling will generate much needed
revenues that can be allocated towards social betterment. However, the expansion of
22
legalized gambling has been linked to negative financial impacts, such as a decreased
ability for families to financially save, particularly for low-income, low-education, and
urban families (MacDonald, McMullan, & Perrier, 2004). Further, legal casino gambling
has been found to be associated with increased local bankruptcy rates (Barron, Staten, &
Wilshusen, 2008; Nichols, Stitt, & Giacopassi, 2000). It has been estimated that the
availability of casino gambling increases bankruptcy rates by between 2% and 9%
(Boardman & Perry, 2007; Daraban & Thies, 2011). Research has also demonstrated a
link between states that have larger proportions of residents who visit out-of-state casinos
and increased bankruptcy rates (Garrett & Nichols, 2008). Gambling, in particular state
lotteries have been associated with greater social economic inequity (Freund & Morris,
2006). However, in certain jurisdictions, the introduction of legal casino gambling has
been associated with positive social and economic changes, including: increased funding
for public education, increased availability of health care services, and a decreased rate of
households relying on government financial assistance (Long, Johnson, & Oakley, 2011).
Therefore, though gambling may in fact generate revenues that are put towards good
causes, it is often the case that these revenues come at the cost of the overall financial
well-being of the constituent populations.
Research has also focused on how legalized gambling is perceived by local
residents. However, these studies on the perceptions of those living near casinos have
yielded mixed results. Despite the empirically demonstrated negative consequences of
available legalized gambling, it has been shown that people living near casinos generally
view them positively (Chhabra, 2007). However, other studies have demonstrated that
23
following the introduction of a casino the perceptions of residents became less favorable
compared to prior to the casino’s introduction, as well as less favorable as the availability
of gambling remains in the community (Jacques, Ladouceur, & Ferland, 2000).
Individual perceptions of whether legalized gambling has had a positive or negative
impact seem to be linked to the individual’s perception of whether crime has increased
and whether the community has economically benefitted as a result of the introduction of
legalized gambling (Hsu, 2000). To this issue of increased crime, research has also
demonstrated that the opening of legal casinos may be associated with an increase in the
number of individuals with criminal histories coming to the area (Piscitelli & Albanese,
2000). Similarly, quality of life for those residents of area with casino gambling was
found to be related to perceived social impacts of the casinos. The less educated and
those living in urban areas reported perceiving more negative social impacts of the
casinos (Roehl, 1999). Therefore, the individual perception of the legalized gambling
seems to be related to the individual experience, which may relate to issues of social
equity.
As the extreme proliferation of gambling is a relatively recent phenomenon, the
full extent of the impact may not yet be fully understood. However, it seems relatively
clear that it leads to increased rates of gambling problems and the associated social and
economic difficulties. It also appears that these negative impacts of gambling are serving
to increase existing social inequities.
24
For the individual, problem gambling is thought of as progressing through a series
of phases: winning, losing, and desperation (Lesieur & Custer, 1984). In the winning
phase the individual is beginning gambling behavior and often experiences varying
degrees of winning. The losing phase, as the name suggests, is characterized by the loss
of money, often lost in an attempt to recoup previous losses. In the desperation phase the
gambling has caused significant life problems, and the gambler is often driven to extreme
measures, such as illegal acts. Recently, these phases have been expanded to include a
fourth phase, hopelessness (Denure, Ford, Hillyard, Moore, & Scherer, 2006). At this
point, the problem gambler has become so overwhelmed by the consequences of their
gambling behavior that suicide often becomes a likely outcome.
Gambling addiction can have a variety of profoundly negative consequences.
Most obviously, the individual exhibiting the gambling behavior can struggle in a
multitude of ways, including financially and psychologically. Additionally, interpersonal
functioning is often significantly impacted. Also, individuals with whom the gambler has
a personal relationship are in many ways affected by this behavior.
Interpersonal difficulties are often associated with gambling addiction, as both a
precipitant and a consequent of the behavior (Callan et al., 2011; Downs & Wollrych,
2010; Reid et al., 2011). These interpersonal difficulties may relate to a number of
factors. Gambling has been referred to as the hidden addiction, as the lack of
recognizable physical signs as is seen with alcohol or drug use can facilitate concealment
of the behavior (Ladouceur, 2004). Furthermore, internet gambling, with its ease of
access and relative anonymity, may increase the extent to which gambling behavior can
25
be concealed from significant others (Valentine & Hughes, 2012). Therefore, the secrecy
associated with gambling, both in terms of the behavior itself as well as the financial
difficulties the behavior has created, can be a contributing factor to interpersonal
difficulties. In fact, lying and concealing gambling behavior is a major factor associated
with diagnosing gambling disorder, as it is one of the criteria used in its diagnosis
(American Psychiatric Association, 2013). It is estimated that approximately 78% of
pathological gamblers lie to those around them regarding the extent of their gambling
(Toce-Gerstein et al., 2003). Pathological gambling has also been associated with
increased guilt related to the consequent deteriorating quality of interpersonal
relationships (Locke, Shillkret, Everett, & Petry, 2013). These interpersonal difficulties
may further contribute to the significant psychological difficulties faced by the
pathological gambler, which will be discussed later.
Another major issue related to gambling addiction is the large sums of financial
debt that can be amassed (e.g., Ciarrochi, 2002; Downs & Woolrych, 2010; Yip et al.,
2007). One study of pathological gamblers receiving inpatient treatment found that,
excluding home mortgages, 50% were more than $25,000 in debt, 10% were between
$50,000 and $100,000 in debt, and 18% were more than $100,000 in debt; furthermore,
pathological gamblers are nearly five times more likely to have declared bankruptcy than
individuals with no gambling problems (Ciarrocchi, 2002). In another study of
pathological gamblers within treatment, Teo and colleagues (2007) found a mean
reported debt of $102,735.45, with the largest debt being $1.5 million. Problem gamblers
often reported borrowing money from friends and family, thereby receiving what is
26
termed a “bailout” (Tang et al., 2007). In addition to borrowing money directly from
friends and family, gamblers often take out loans from financial institutions, often in
response to the debt that they have incurred from their gambling (Brown, Dickerson,
McHardy, & Taylor, 2012; Tang et al., 2007).
Debt among pathological gamblers has been linked to significant mental health
issues. Individuals in debt have been found to exhibit more psychological disorders if
they are experiencing gambling problems than if they are non-gamblers (Meltzer,
Bebbington, Brugha, Farrell, & Jenkins, 2013). Further, these financial losses resulting
from gambling have been linked to a strong sense of shame that is often experienced by
pathological gamblers (Yi & Kanetkar, 2011). Research has also found a significant link
between gamblers’ financial problems and suicidality, which is another major issue with
gambling addiction that will be discussed later (Meltzer et al., 2011; Wong et al., 2010a,
b). In fact, from posthumous studies of gamblers who have died by suicide, it is estimated
that between 47.2% and 100% had amassed gambling related debt at the time of their
death (Wong et al., 2010a, b). Further, among callers to a problem gambling hotline,
those that reported a risk of suicide (25.6%) were more likely to acknowledge financial,
as well as family, legal, mental, and substance abuse problems (Ledgerwood et al., 2005).
Therefore, financial issues can play a major role in the development of, and fallout from
gambling addiction.
Pathological gambling has also been associated with increased rates of criminal
behavior, most commonly crimes designed to acquire revenue (Meyer & Stadler, 1999).
Problem and pathological gamblers often resort to stealing in order to continue their
27
gambling behavior, and often this stealing is from sources familiar to the individual, such
as friends, family, and the workplace (Cheah et al., 2008; Meyer & Stadler, 1999). In a
study of pathological gamblers, Meyer and Stadler (1999) found the most common forms
of illegal behavior admitted to were: fraud (37.7%), theft from the workplace (23.3%),
embezzlement (21.7%), and theft from family (21%). Another study estimates that
between 20.7% and 23.5% of pathological gamblers admit to a history of illegal acts (Lee
et al., 2011). Estimates of histories of illegal acts among treatment-seeking gamblers have
been found to be lower (11.6%), which may indicate that those with criminal histories
may be less likely to seek treatment (Martin, MacDonald, & Ishiguro, 2013). The
prevalence and nature of this criminal behavior seems to speak to the desperation of the
situation that gamblers find themselves in, particularly as it relates to financial difficulty.
However, it has been found that of the diagnostic criteria for pathological gambling in the
DSM-IV-TR, the one related to the committing of illegal acts was the least commonly
endorsed (Molde et al., 2010). Further, as previously discussed, the current set of criteria
for gambling disorder in the DSM-V no longer contains a criterion related to illegal acts
(American Psychiatric Association, 2013).
Those engaging in problematic levels of gambling behavior are often divided into
two sub-groups based upon how they orient themselves towards the behavior. For many,
gambling is a coping strategy whereby the individual engages in the behavior to avoid
dealing with a negative emotional state (Wood & Griffiths, 2007). Such individuals, who
are often labeled as “escape” gamblers, generally engage in the behavior as a way to
numb themselves against some sort of underlying emotional dysregulation. These
28
individuals generally tend to favor games of chance, such as slot machines. Conversely,
those labeled as “action” gamblers crave the high that they can receive from gambling,
and they generally tend to favor games of perceived skill, such as poker. This distinction
is often related to the gender of the gambler, with women and the elderly being more
commonly associated with the escape-type gambling and men with the action-type (Li,
2007; Odlaug, Marsh, Kim, & Grant, 2011; Potenza et al., 2001). However, this
distinction, also referred to as strategic and non-strategic gambling, is not always entirely
clear-cut, as one study found that over 40% of pathological gamblers regularly engage in
both types of gambling (Odlaug et al., 2011). However, identifying and understanding
this distinction may have implications for the proper treatment of gambling addiction
(Tang et al., 2007).
Certain demographic groups have been associated with an increased risk for
developing gambling-related problems. Also, various demographic factors have been
associated with differential gambling presentations. Moreover, factors have also been
identified that can help to understand the manifestations and course of gambling
addiction, as well as having implications for treatment.
Gender is a factor that has received much attention for its role in gambling.
Typical gender differences are found in patterns of gambling activities, types of gambling
related problems, and treatment seeking (e.g., Nelson et al., 2006; Nower &
Blaszczynski, 2006; Tang et al., 2007). Compared to male gamblers, female gamblers
begin gambling later in life, progress more quickly from first gambling experience to
problematic levels of gambling, have less gambling-related debt, and tend to prefer
29
gambling on slot machines, as opposed to sports betting or card games (Crisp et al., 2004;
Nower & Blaszczynski, 2006; Tang, Wu, & Tang, 2007). Female gamblers have also
been found to bet in smaller denominations compared to men, and female pathological
gamblers have been found to earn significantly less income than their male counterparts
(Nower & Blaszczynski, 2006).
Gender also has implications for gambling treatment. Women have been found to
have begun gambling later in life, first attempted to quit gambling later in life, and have
gambled for a shorter duration by the time they enter treatment (Ladd & Petry, 2002).
Women have also been associated with greater gambling and financial problems at the
time of accessing treatment, as well as a greater readiness to change (Ledgerwood,
Wiedeman, Moore, & Arfken, 2012). Women pathological gamblers have been found to
be more likely to have accessed mental health treatment, and have been found to access
treatment later in life (Potenza et al., 2001; Tang et al., 2007). This may indicate a need to
more actively seek male pathological gamblers who are in need of treatment.
Women have also been found to present with differing gambling symptoms,
particularly men were more likely to meet the DSM criteria related to preoccupation,
illegal acts (DSM-IV-TR), and lost/hurt relationships/jobs, whereas women were more
likely to meet the criterion related to escaping personal problems (Crisp et al., 2000).
However, another study of adolescents found differences in symptom patterns in a
community sample but failed to find any differences in symptom endorsement among
male and female pathological gamblers in treatment (Faregh & Derevensky, 2011). In
either case, clinicians may need to be aware that men and women may present for
30
treatment with different symptom patterns. Other clinical issues may also vary between
men and women. Women have been found to be more likely to indicate a desire to
prevent suicide as their reason for self-excluding from casinos, which is a process by
which an individual can bar themselves from entering a casino (Nower & Blaszczynski,
2006). Also, female pathological gamblers have been found to attempt suicide more
frequently than males; however, pathological gamblers who die by suicide have been
found to have an increased likelihood of being male (Martins et al., 2004; Potenza et al.,
2001; Wong et al., 2010a; Wong et al., 2010b). Further, women have been found to be
more likely to report resolution of their problems through treatment than men (Crisp et
al., 2000).
Ethnicity may also play a role in gambling addiction. A study of callers to a
problem gambling helpline found that Caucasian callers were more likely than AfricanAmerican callers to have previously sought mental health treatment; however, the study
found no significant difference in proportions reporting financial problems related to
gambling (Barry et al., 2008). One study looking at minority acculturation to the
dominant culture indicates that successfully assimilating into the dominant culture can be
a protective factor against problem gambling (Oei & Raylu, 2009). Also, ethnic
minorities are greatly underrepresented in gambling treatment (Volberg, 1994).
Age also seem to play a significant role in gambling behaviors. Age of gambling
behavior onset has been found to have an impact on gambling manifestation. Individuals
who began gambling in pre/early adolescence reported greater severity of psychiatric,
family/social, substance abuse problems, and suicidal ideation but no difference in
31
suicide attempts (Burge et al., 2006). Individuals who experience late onset of
pathological gambling, after age 55, have been found to be less likely to have declared
bankruptcy, less likely to have a parent with a gambling problem, and more likely to
exhibit an anxiety disorder (Grant, Kim, Odlaug, Buchanan, & Potenza, 2009). Older
onset of gambling problems has also been associated with more severe psychopathology,
while younger onset has been associated with more severe gambling problems (JimenezMurcia et al., 2010).
Current age is also a factor in gambling manifestation. Older adult pathological
gamblers have been found to have accumulated significantly higher debt (Kennedy et al.,
2005). However, recreational gambling among older adults has been found to have no
association with overall health and well-being (Desai, Maciejewski, Dausey, Caldarone,
& Potenza, 2004). Older adults have also been found to more often cite suicide
prevention as a reason for self-exclusion from casinos (Nower and Blaszczynski, 2008).
Rates of gambling among adolescents are estimated to be far greater than in the general
population (Gupta & Derevensky, 1998; LaBrie & Shaffer, 2007). Further, other
researchers have found that younger individuals who gamble may actually be at a
significantly higher risk for gambling-related problems, including suicidal ideation and
attempts, and low rates of seeking help for their gambling problems (Afifi et al., 2007;
Dowling, Clark, Memery, & Corney, 2005; Froberg et al., 2012; Martins et al., 2004;
Nower et al., 2004). One particular age-related subset of individuals, student athletes,
have been found to be associated with increased likelihood to engage in gambling
behaviors and have gambling-related debt (Bovard, 2008; Stuhldreher et al., 2007). The
32
interplay between age and gambling is, therefore, obviously a complex one whose
relationship may not necessarily be linear given the evidence that both the young and the
elderly may be at increased risk.
Research has also implicated exposure to gambling as a risk factor in the
development of gambling problems, particularly as it relates to early childhood exposure
to gambling (Reith & Dobbie, 2011). Complicating this issue somewhat is the research
that has implicated inherited genetic factors in the development of gambling addiction
(Slutske, Zhu, Meier, & Martin, 2010). Therefore, it is likely a combination of exposure
and genetic predispositions that contribute to the development of gambling problems.
Various personological variables have been associated with gambling addiction,
including impulsivity, interpersonal sensitivity, interpersonal distrust, poor coping skills,
lack of self-discipline, and a tendency to make decisions hastily (Reid et al., 2011).
Impulsivity, in particular, may play a moderating and a mediating role in the relationship
between life stress and gambling problems (Tang & Wu, 2012). Other personological
variables associated with gambling problems are neuroticism, low esteem from others,
and low family support (Taormina, 2009). Anger problems are also commonly seen
amongst pathological gamblers (Korman et al., 2008).
Gambling behavior has also been strongly associated with incarcerated
individuals, which may be an issue during incarceration as well as upon re-entry into
society (Williams & Walker, 2009). A study of incarcerated individuals found that nearly
16% had moderate to severe gambling problems (Turner, Preston, Saunders, McAvoy,
33
Jain, 2009). Another study of adolescents in youth detention centers found that 18% met
the criteria for pathological gambling (Magoon, Gupta, & Derevensky, 2007).
Religion may also play a role in the development, or prevention of gambling
problems. Individuals that frequently attend religious services have been found to less
often develop gambling problems (Hoffman, 2000). Therefore, active religious affiliation
may be a protective factor for developing gambling problems.
Co-occurring disorders, both psychological and substance-related, have been
strongly associated with gambling addiction. The relationship between the two might be
very complicated, which is to say that co-occurring disorders might influence the
development and course, as well as be a product of gambling addiction. It has been
estimated that as many as 96.3% of pathological gamblers will develop at least one
World Health Organization Composite International Diagnostic Interview (CIDI)/DSMIV-TR disorder(s) during their lifetime (e.g., Chan et al., 2009; Gill et al., 2006; Kessler
et al., 2008; Lorains et al., 2011). Further, individuals who engage in non-pathological, or
social gambling have been found to have higher rates of psychiatric disorders compared
to non-gamblers (Westermeyer, 2008).
There is debate over whether mental health issues are generally a cause or a
consequence of problem gambling; however, the literature seems to suggest that the
mental health issues are likely more of a precipitant to the gambling (O’Brien, 2011).
From this perspective, the gambling is used as a coping mechanism for the preexisting
psychological difficulties. It has been estimated that for individuals who experience
pathological levels of gambling along with another lifetime mental health disorder,
34
74.3% had at least one disorder that preceded the onset of the pathological gambling
(Kessler et al., 2008). This supports the idea that the gambling disorder is developed as an
attempt to cope with a mental health condition.
There is a broad range of disorders that are found to co-occur with gambling
addiction. A meta-analysis of research on gambling and co-occurring disorders found that
the following were the most commonly observed co-occurring mental health conditions:
any type of mood disorder (37.9%) and any type of anxiety disorder (37.4%) (Lorains,
Cowlishaw, & Thomas, 2011). In the National Comorbidity Study replication, Kessler et
al. (2008) found that among pathological gamblers, 38.6% also had major depressive
disorder or dysthymia, 55.6% also had a type of mood disorder, 60.3% also had a type of
anxiety disorder, and 14.8% also had Posttraumatic Stress Disorder (PTSD). These are
compared to rates of 19.1%, 20.8%, 28.8%, and 6.8%, respectively, for these four
psychological disorders found within the general population (Kessler et al., 2005). This
indicates that for the disorders identified by Kessler (2005; 2008), rates may be
approximately twice as high for pathological gamblers compared to the general
population. Among anxiety disorders, obsessive-compulsive disorder, in particular, has
been found to commonly co-occur with gambling addiction (Gonzalez-Ibanez et al.,
2003). Personality disorders have also been linked to pathological gambling. It has been
estimated that over 60% of pathological gamblers have at least one personality disorder,
with obsessive-compulsive, paranoid, and antisocial personality disorders being the most
common (Petry, Stinson, & Grant, 2005). Attention Deficit/ Hyperactivity Disorder
(ADHD) is another disorder that is gaining increasing amounts of attention for its
35
relationship to gambling addiction, in part, because of the impulsive characteristics
commonly associated with both ADHD and pathological gambling (Duven, Unterrainer,
& Wölfling, 2012). Multiple studies have linked disordered gambling and ADHD, with
lifetime rates of ADHD have been estimated between 13% and 20% among those with
lifetime pathological gambling (Crockford & el-Guebaly, 1998; Kessler et al., 2008;
Rugle & Melamed, 1993; Specker et al., 2005). Furthermore, individuals experiencing
pathological levels of gambling are at a significantly increased risk for experiencing subclinical levels of psychiatric symptoms other than those included in the diagnostic criteria
for pathological gambling (Boudreau, Labrie, &Shaffer, 2009).
These co-occurring mental health conditions have implications for clinical
presentation and treatment. Research supports that the rate of psychiatric disorders
increase with severity of gambling problems (Park et al., 2010). Similarly, among
individuals seeking psychiatric treatment, those exhibiting pathological gambling
presented with greater numbers of psychiatric disorders (Zimmerman, Chelminski, &
Young, 2006). It has also been found that as the number of co-occurring disorders with
which a client presents for treatment increases so too does the severity of gambling
problems increase (Soberay et al., 2013). Consequently, clinical assessment that focuses
on comorbid psychiatric conditions may lead to a better understanding of the gambling
problems with which a client presents.
In terms of substance use and misuse with gambling addiction, drug, alcohol, and
tobacco use have all been associated with the gambling disorder (Walther, Morgenstern,
& Hanewinkel, 2012). Lorains, Cowlishaw, and Thomas (2011) found that approximately
36
58% of pathological gamblers also have a co-occurring substance use disorder. Similar to
mental health issues, substance use disorders may play a significant role in the
development as well as the perpetuation of gambling. In terms of treatment issues,
individuals in treatment for gambling problems who were also currently using alcohol
frequently presented with more severe gambling behavior, yet were equally responsive to
treatment (Stinchfield, Kushner, & Winters, 2005). Similarly, within treatment, substance
dependence is a major factor in differentiating those who present with pathological levels
of gambling from those presenting with sub-clinical levels (Namrata & Oei, 2009). Even
among substance abusers, those with gambling problems had greater rates of psychiatric
distress (Petry, 2000). Therefore, the presence of multiple addictive disorders may have
strong implications for problem gambling treatment.
Studies have repeatedly shown that a relatively large proportion of individuals
exhibiting gambling addiction have considered, attempted, or completed suicide, which
makes this a critical clinical issue for this population. Lifetime rates of suicidal ideation
and attempts among pathological gamblers have been estimated between 24.7% to 81.4%
and 6.3% to 32.7%, respectively (e.g., Battersby et al., 2006, Bu & Skuttle, 2012; GrallBronnec et al., 2012; Ledgerwood & Petry, 2004; Park et al., 2010). Therefore, as many
as nearly one in three pathological gamblers will attempt suicide during their lifetime.
When looking only at the past twelve months, a study of problem gamblers found that
10.2% reported suicidal ideation and 2.4% reported suicide attempts during this time
period (Afifi et al., 2010). Research also supports that increased gambling severity is
37
associated with increased risk for suicide (Brooker et al., 2009; Hodgins et al., 2006;
Langhinrichsen-Rohling et al., 2004; Ledgerwood et al., 2005).
Psychotherapy is the most widely studied and most commonly applied approach
to gambling treatment; moreover, psychotherapy has demonstrated effectiveness in the
treatment of gambling addiction on a number of gambling, psychological, and
interpersonal measures (e.g., Pallensen et al., 2005; Sharma & Sharma, 2012; Sylvain et
al., 1997). Also, studies have shown that individuals who have received treatment for
their gambling behavior have a decreased risk for suicide (Kennedy et al., 2005). A metaanalysis of studies examining the outcome effect of psychological treatment of
pathological gambling found that of the 22 targeted studies, the meta-analysis found
psychological treatments were far more effective than no treatment at termination as well
as at follow-up periods beyond the termination of treatment (Pallesen et al., 2005). This
meta-analysis also found that cognitive-behavioral therapy was the most commonly
applied form of psychotherapy. Furthermore, Crisp et al. (2001) found that
psychotherapeutic treatment of problem gamblers can demonstrate effectiveness after as
few as five treatment sessions. In fact, one study found that a single-session psychoeducation group was associated with reductions in gambling problems (Petry, Weinstock,
Ledgerwood, & Morasco, 2008). However, it has been found that longer retention in
treatment is associated with more positive outcomes and greater satisfaction with
treatment (Toneatto & Dragonetti, 2008).
Delivery methods of therapy may be varied. Online-based support for problem
gambling has various advantages, including cost-effectiveness and easy access, and can
38
range from psycho-education, peer-support networks, to professionally delivered
intervention (Griffiths & Cooper, 2003). For instance, cognitive-behavioral treatments
utilizing an internet-based format have demonstrated effectiveness in the treatment of
gambling problems (Carlbring, Degerman, Jonsson, Andersson, 2012; Castren et al.,
2013). Another study demonstrated that telephone counseling might be equally as
effective as face-to-face counseling interventions (Tse et al., 2013). Self-help workbooks
have also demonstrated effectiveness, but not as effective as face-to-face therapy (Petry
et al., 2006). These findings may be particularly relevant for those who live in
geographically isolated areas where treatment may not be available or for those who for
any other reason may not be able to utilize in-person treatment.
Psychotherapy is not the only form of treatment for gambling addiction. Various
medications have been identified as possible treatments for pathological gambling,
particularly those that target the neurotransmitters serotonin and dopamine (Potenza,
2008). However, in a study of treatment preferences among pathological gamblers,
medication was found to be the least desired of the available treatments, which included
numerous forms of psychotherapy (Najavits, 2011).
Other, informal methods may also be helpful for individuals to overcome
gambling problems. For instance, journaling has been found to be beneficial to problem
gamblers (Dwyer, Piquette, Buckle, & McCaslin, 2013). Self-exclusion is a nontherapeutic approach whereby a gambler can voluntarily bar him or herself from entering
casinos. Self-exclusion has been associated with reducing gambling problems,
39
particularly when coupled with formal treatment or self-help groups (Nelson,
Kleschinsky, LaBrie, Kaplan, & Shaffer, 2010).
However, it is unclear whether varying forms of psychotherapy are more or less
effective. Studies comparing a cognitive-behavioral treatment to a twelve-step
facilitation treatment found that each was effective compared to a control, yet neither
treatment was significantly more effective than the other (Marceaux & Melville, 2011;
Toneatto & Dragonetti, 2008). Another study comparing cognitive-behavioral and
motivational interviewing based group treatments found each to be superior to no
treatment but equivalent to each other (Carlbring, Jonsson, Josephson, & Forsberg, 2010).
Not all people that overcome gambling problems do so as a result of any
identifiable treatment steps. So-called natural, or spontaneous recovery may be a
relatively common phenomenon among individuals who experience gambling problems
at some point during their lifetime (Hodgins & El-Guebaly, 2000). In fact, it has been
estimated that approximately one third of those that develop pathological gambling
during their lifetime will overcome their gambling problems without seeking any form of
treatment (Slutske, 2006).
In the treatment of gambling addiction, both cognitive and behavioral counseling
strategies are commonly utilized forms of treatment; additionally, these two approaches
are often used in conjunction, in what is referred to as Cognitive-Behavioral Therapy
(CBT) (Ladouceur et al., 2002). Cognitive models of addiction treatment are grounded on
the idea that gambling addiction is grounded in, and perpetuated by erroneous beliefs and
maladaptive patterns of thinking, particularly regarding the ultimate outcomes of
40
gambling behavior, conceptions of personal luck, and erroneously perceived control over
outcomes of random events (Johnson & Dixon, 2009; Ladoceur et al., 2002; Wohl,
Young, & Hart, 2007). Moreover, irrational beliefs regarding the outcomes of gambling
behavior have been linked to the development and perpetuation of gambling problems
(Tang & Wu, 2010). Irrational thoughts about gambling behavior have also been
positively associated with severity of gambling problems with which individuals present
for treatment, as well as an increase in the number of co-occurring psychiatric disorders
(Houndslow, Smith, Battersby, & Morefield, 2011; Xian et al., 2008). Metacognitions, or
the ways we think about our thoughts, have also been implicated in gambling addiction,
particularly negative beliefs regarding thoughts related to the uncontrollability or need to
control gambling have been associated with problem gambling behavior (Lindberg,
Fernie, & Spada, 2011). Therefore, cognitive interventions for gambling problems,
through a process termed cognitive reappraisal, are designed to help challenge and,
ultimately, alter these faulty and erroneous thinking patterns (Tolchard & Battersby,
2013). This process of cognitive correction can focus on various common gambling
misconceptions, including: concepts of randomness, understanding of independence of
events, and illusions of control over gambling outcomes (Ladouceur et al., 2003).
Behavioral models of addiction treatment focus on the degree to which addicted
behaviors are learned and perpetuated through a process of reinforcement of the behavior
(Ladouceur et al., 2002). Therefore, behavioral treatments place importance on the extent
to which an addiction is a learned behavior and attempt to identify and modify sources of
reinforcement and punishment in the development, maintenance, and treatment of
41
addictions. Behavioral intervention techniques, such as activity scheduling, contingency
management, and in-vivo or imaginal desensitization, have also been found to be
effective interventions for the treatment of pathological gambling (Dowling, Jackson, &
Thomas, 2008; Rawson, McCann, Flammino, Shoptaw, Miotto, Reiber & Ling, 2006;
Tavares, Zilberman, & el-Guebaly, 2003). Exposure therapy, a behavioral technique
whereby a problem gambler is exposed to gambling cues without initiating gambling
behavior, has also demonstrated effectiveness for gambling addiction (Riley, Smith, &
Oakes, 2011).
CBT treatments often consist of the following aspects: (1) correcting negative
beliefs regarding randomness and outcomes of gambling behavior, (2) development of
problem-solving skills, (3) social skills training, and (4) relapse prevention, including
ways of coping with “high risk” situations (Sylvain, Ladouceur, & Boisvert, 1997). The
development of coping skills may be a critical component to the success of CBT
treatment. The attainment of adequate coping skills has been shown to mediate the
relationship between treatment and outcomes (Petry, Litt, Kadden, & Ledgerwood,
2007). CBT interventions have also demonstrated effectiveness in reducing gambling
behavior among at-risk, sub-clinical gamblers (Larimer et al., 2012). Further, a metaanalysis of CBT interventions for gambling addiction found effect sizes to be significant
as far out as 24 months from cessation of treatment (Gooding & Tarrier, 2009). It seems
likely that therapeutic gains may last even longer; however, studies that follow clients for
greater than two years following treatment are lacking.
42
CBT interventions are commonly delivered in both individual and group formats.
Both group and individual delivery methods of CBT interventions have demonstrated
effectiveness, although individual CBT may have greater therapeutic effectiveness,
particularly at follow-up (Dowling, Smith, & Thomas, 2007). Further, combining
individual and group CBT interventions has also demonstrated effectiveness in reducing
gambling problems and improving overall functioning (Oakes, Laughlin, McLaughlin, &
Battersby, 2012).
Dialetical Behavior Therapy (DBT), an offshoot of CBT, uses a focus on
developing such skills as mindfulness, distress tolerance, and emotional regulation, and it
has also been demonstrated to be helpful in the treatment of gambling problems
(Christensen et al., 2013). In particular, mindfulness based interventions may be helpful
in coping with the distressing thoughts and emotional states associated with gambling
addiction (de Lisle, Dowling, & Sabura, 2011). Similarly, Acceptance and Commitment
Therapy (ACT), another emerging form of CBT that focuses on accepting current
experiences and committing to value-driven action has also shown initial evidence that it
may be helpful in the reduction of gambling problems (Hayes, Luoma, Bond, Masuda, &
Lillis, 2006; Nastally & Dixon, 2012; Weatherly, Montes, Peters, & Wilson, 2012).
Given that CBT interventions are the most studied form of psychotherapy for the
treatment of gambling problems, its effectiveness has been researched among special
populations, particularly those that may be most susceptible. CBT interventions have
demonstrated effectiveness in treating pathological gamblers with chronic schizophrenia,
those with Parkinson’s disease, as well as those with acquired brain injury (Echeburua,
43
Gomez, & Freixa, 2011; Guercio, Johnson, & Dixon, 2012; Jimenez-Murcia et al., 2012).
CBT has also shown to be adaptable for congruence with various cultural beliefs and
behaviors (Okuda, Balan, Petry, Oquendo, & Blanco, 2009).
Solution-Focused Therapy (SFT) is a strengths-based, goal-directed approach to
psychotherapy that originated in part as a response to the problem-focused treatments that
characterize most of psychotherapy (de Shazer & Isebaert, 2004). SFT focuses more on
helping clients work towards an idealized goal state that represents what they would like
their lives to be like, rather than focusing primarily on their current difficulties. Common
SFT techniques include the “miracle question”, whereby clients are asked to describe
their lives if all their problems were miraculously solved. SFT also commonly uses
scaling questions to assess client progress towards their goal state (de Shazer & Isebaert,
2004). Since the client is in control of determining their own goal-state, SFT is well
suited to those seeking abstinence as well as those seeking to moderate their problem
behavior (Nelle, 2005).
SFT has recently been applied more to, and has shown effectiveness in, the
treatment of addictive disorders. Specifically it has demonstrated effectiveness in the
treatment of substance abuse (Smock et al., 2008). However, the literature does not
address whether this effectiveness will generalize to the treatment of problem gambling
(Pallesen et al., 2005).
From the psychodynamic perspective, addictive disorders are an issue of
difficulty regulating behaviors, relationships, and emotions (Khantzian, 2012). The
addictive behavior then is an attempt to cope with this dysregulation. From this
44
perspective, addiction is also often viewed as a relational problem where the addict
develops an attachment to their addicted behavior in an attempt to cope with previous
attachment problems (Khantzian & Weegmann, 2009). Psychodynamic treatments of
addiction often focus on the psychological transference that occurs in the therapeutic
relationship (Sweet, 2012). This relationship can help serve to develop adaptive relational
patterns, which in turn help to decrease the dependence on the addiction (Potik, Adelson,
& Schreiber, 2007).
Psychodynamic models of therapy have demonstrated effectiveness in the
treatment of various substance addictions, including nicotine, opiate, alcohol, and cocaine
dependences (Crits-Cristoph et al., 1999; Gregory et al., 2008; Sandahl, Herlitz, Ahlin, &
Ronnberg, 1998; Woody, McLellan, Luborsky, & O’Brien, 1995; Zernig et al., 2008). As
is also the case with SFT, large-scale, methodologically sound studies specifically
looking at the effectiveness of psychodynamic treatments for pathological gambling are
virtually non-existent (Pallesen et al., 2005; Raylu, Loo, & Oei, 2013). However,
particularly given the effectiveness of psychodynamic therapy on the treatment of
personality disorders, it may be a viable treatment option for pathological gambling given
the large proportion of pathological gamblers with comorbid personality disorders
(Leichsenring & Leibing, 2003; Petry, Stinson, & Grant, 2005).
In contrast to strictly psychodynamic treatments, studies investigating the
interpersonal elements related to problem gambling treatment have yielded more
promising results. Given the interpersonal and familial antecedents and consequents of
gambling behavior, an increasing number of clinicians conceive of treatment as an
45
interpersonal process, which focuses on improving interpersonal functioning and,
thereby, improving overall functioning. This, in turn, can contribute to controlling
gambling problems (McComb, Lee, & Sprenkle, 2009). Such approaches to addiction
treatment have demonstrated increased motivation to change as well as personal
empowerment (Wood, Englander-Golden, Golden, & Pillai, 2010). Interpersonal models
of therapy that focus specifically on working with gamblers and their significant others
have also demonstrated effectiveness (Lee, 2007). Further, research indicates that
gambling treatment outcomes are more positive and retention is greater for those whose
significant others are involved in the treatment process (Ingle, Marotta, McMillan, &
Wisdom, 2008).
The working relationship between the psychotherapy client and their counselor
has been demonstrated to be a moderate predictor of outcomes across treatment settings
(Martin, Garske, & Davis, 2000). Research has also demonstrated the importance that the
therapeutic alliance plays in the treatment of a wide range of addictions. Through a
retrospective study of individuals having received a problem gambling treatment, client
ratings of the working alliance were found to be a significant predictor of problem
resolution (Smith, Thomas, & Jackson, 2010). The findings related to what source of
ratings of the therapeutic alliance, whether the client, counselor, or an outside observer,
best predicts outcomes have been mixed. Research has demonstrated that client, observer,
and therapist perceptions and ratings of the therapeutic alliance may vary (Fenton,
Cecero, Nich, Frankforter, & Carroll, 2001). Some findings suggest that therapist ratings
of the alliance may be a better predictor of outcomes (Meier, Donmall, McElduff,
46
Barrowclough, & Heller, 2006). However, in a study of problem gamblers, client-rated
alliance predicted both gambling and general functioning outcomes, while therapist-rated
alliance measures only predicted general functioning outcomes; moreover, client
satisfaction with treatment was found to have a mediating relationship between alliance
and outcomes (Dowling & Cosic, 2011). Several factors have been implicated in whether
or not an addiction client forms a strong alliance with their therapist, including: social
support, attachment style, readiness to change, external pressure to change, counselor
qualifications, and whether the counselors were themselves “ex-addicts” (Meier,
Donmall, Barrowclough, McElduff, & Heller, 2005).
Current understandings of behavioral change conceptualize it as a cyclical process
that traverses a series of stages, which range from a denial of any problem and an absence
of any motivation for change to maintaining behavioral changes that have already been
achieved (DiClemente, Schlundt, & Gemmell, 2004; Prochaska & Norcross, 2001). Each
stage of change presents unique characteristics and clinical challenges, and an
understanding of a client’s current stage of change may have implications for appropriate
interventions. A problem gambler’s readiness for change has been found to increase with
higher levels of emotional awareness, previous experience with GA, and higher levels of
depression (Gomes & Pascual-Leone, 2008).
Models differ on exactly how many stages of change exist that a client may go
through. Precontemplation, Contemplation, Preparation, Action, Maintenance, and
Relapse/Termination are generally considered an exhaustive list of the stages; however,
not all models include all six stages (Norcross, Krebs, & Prochaska, 2011; Prochaska &
47
Norcross, 2001; Prochaska, DiClemente, & Norcross, 1992). Precontemplation is
characterized by a lack of awareness or complete denial that a problem exists; therefore,
there is no motivation to change at this stage. At the contemplation stage, the individual
has become aware that a problem exists but are still very much dealing with their
ambivalence to change the behavior. At the preparation stage, the individual has resolved
to change their problem behavior, and they may have begun to make minor modifications
to their behaviors. In the action stage, the individual is actively changing their problem
behavior. An individual is generally considered to enter the maintenance stage when they
have sustained successful change of their problem behavior for at least six months.
Relapse is generally considered a normal part of the addiction change process, and it
signifies the point at which an individual begins to cycle back through the previous
stages. If an individual does not experience relapse, the termination stage is viewed as
being achieved when the individual has sustained change long enough that they no longer
have to actively work at modifying their behavior.
An individual’s stage of change at the time of entering treatment has been
implicated in the severity of problems with which they present. Awareness that a problem
exists has been found to be associated with more severe clinical presentations, while
those that have begun to modify their behaviors have been associated with less severe
presentations (Gossop, Stewart, & Marsden, 2007). This finding indicates that the
individuals associated with the contemplation stage may enter treatment reporting the
most severe problems. Petry (2005), in a sample of pathological gamblers, found that
individuals associated with the precontemplation stage reported less severe gambling
48
problems, which may be indicative of the denial associated with this stage. Petry (2005)
also found that those associated with the action and maintenance stages reported more
severe gambling problems, which, again, may be indicative of the increased awareness of
the problems of their behaviors at these stages.
Viewing a client in terms of stages of change may also aid in understanding how
long they may remain in treatment and/or whether they are at risk for dropping out of
treatment prematurely. Clients exhibiting traits consistent with the precontemplation
stage have been found to prematurely terminate from therapy (Callaghan et al., 2005).
Similarly, those not exhibiting traits of the contemplation stage have also been found to
prematurely terminate from therapy (Derisley & Reynolds, 2000). These findings suggest
the role that acknowledgement of a problem plays in remaining in treatment. Conversely,
clients and those with higher levels of readiness to change, such as those exhibiting
characteristics of the maintenance stage, have been associated with greater treatment
retention (George et al., 1998; Henderson, Saules, & Galen, 2004). In a sample of
pathological gamblers, stages of change as measured by the University of Rhode Island
Change Assessment (URICA) were a moderate predictor of treatment dropout; however
this study found that no single stage of change was significantly related to retention
(Gomez-Pena et al., 2012). Therefore, stages of change have been implicated in client
retention in, and dropout from treatment. However, like many other areas of treatment,
the research specific to gambling addiction remains mixed and incomplete.
Stages of change have also been associated with the extent to which an individual
improves through the course treatment. Within addiction treatment, the precontemplation
49
stage has been associated with less symptom improvement through treatment, the
contemplation stage has been associated with moderate levels of symptom improvement,
and the action and maintenance stages have been associated with relatively higher levels
of symptom improvement (Henderson, Saules, & Galen, 2004; Norcross, Krebs, and
Prochaska, 2011; Rochlen, Rude, & Baron, 2005). These findings mirror those related to
treatment retention in that individuals more characterized by denial improve less while
those characterized by the actively working on change benefit more from treatment.
Likewise, a study of the relationship between stages of change and improvement through
treatment for a sample of pathological gamblers found similar results: precontemplation
was associated with lower levels of therapeutic change, while action and maintenance
were associated with higher observed levels of change (Petry, 2005). However, another
study using a sample of pathological gamblers failed to find an association between
stages of change and clinical improvement post-treatment or at treatment follow-up
(Gomez-Pena, et al., 2012).
Assessing a client’s readiness for change may play a central role in the treatment
of gambling addiction. However, relatively few studies have been conducted using
problem gambling samples. Consequently, these relationships need to be further
investigated using problem and pathological gambling samples.
As previously discussed, co-occurring mental health and substance use disorders
are common, if not typical among those experiencing gambling addiction (e.g. Kessler et
al., 2008). Research has shown that the assessment and treatment of any co-occurring
conditions may be beneficial in a number of ways. Treatment outcomes have been found
50
to be more favorable when the treatment included a focus on co-occurring disorders
compared to treatment with a sole focus on the addictive behavior (Kim et al., 2006;
Penney et al., 2012). Also, individual who have undergone treatment that addressed cooccurring conditions have been shown to be more resilient to relapse (Brown et al.,
1997). Moreover, addressing co-occurring disorders has been associated with increased
client satisfaction with treatment (Schulte et al., 2011). The body of research indicates
that it is best practices for gambling treatment to expand its scope beyond solely
gambling behavior to include other disorders that the client is likely experiencing that
may be playing a role in perpetuating the gambling.
Certain personological characteristics may also be related to treatment outcomes.
Sensation seeking-traits have been associated with dropping out of treatment and,
consequently, less favorable treatment outcomes (Smith et al., 2010). High levels of
impulsiveness have also been associated with treatment dropout (Alvarez-Moya et al.,
2011). Similarly, individuals exhibiting neurotic personality traits and those exhibiting
low levels of conscientiousness have been found to be at risk for treatment drop out and
relapse (Ramos-Grille, Aragay, Gomá-i-Freixanet, Valero, & Vallès, 2013).
Another element related to treatment outcomes is the severity of the symptoms
with which the client presents for treatment. In psychotherapy in general, greater
symptom severity at the beginning of treatment has been associated with less
improvement through treatment (Lambert & Anderson, 1996). Accordingly, in studies of
pathological gamblers, those presenting for treatment with more severe psychological
distress were found to have less positive outcomes following treatment (Dowling, 2009;
51
Jimenez-Murcia et al., 2007). Previous experience with treatment may also be related to
outcomes. Individuals re-entering treatment for gambling problems have been found to
have more positive outcomes compared to those seeking treatment for the first time
(Jackson et al., 2008).
There seem to be two general approaches to the measurement of gambling
treatment outcomes. The first approach is to explicitly measure gambling behavior.
However, this approach may be incomplete, particularly in the short term. Considering
poor overall functioning as an indicator of risk for relapse, it may be more telling to
consider taking the second approach, which is to measure the person’s psychosocial
functioning (Sander & Peters, 2009). Moreover, a study of treatment outcomes found a
very strong correlation between gambling-related outcomes and general functioning
outcomes (Dowling & Cosic, 2011).
Summary
In summary, propensity score analyses are a set of procedures that were designed
to address the selection bias that often exists within social science research. Traditional
propensity score techniques were developed to balance two groups across a set of
observed covariates, typically a treatment and a control. Recently, propensity scores have
been extended to allow for the comparison of multiple treatment groups. However,
examples of the practical implementation of these multi-group procedures are relatively
scarce. This study seeks to add to the current literature through implementing one of
these strategies, TriMatch, in the comparison of three different therapeutic approaches to
the treatment of problem gambling.
52
Problem gambling was selected as the subject matter of this analysis for several
reasons. Problem gambling is a behavioral disorder that has a profound negative impact
on a significant portion of the general population. It is also relatively understudied
compared with other addictive behaviors, such as alcohol and substance use disorders.
Also, with the expansion of legalized gambling in the United States, problem gambling is
likely to receive an increased amount of attention over the coming years. Finally, the
current body of literature deals very little with two of the three therapeutic approaches
contained in this study: solution-focused brief therapy and time-limited dynamic
psychotherapy.
53
Chapter Two: Method
Participants
Participants in this study were drawn from consecutive admissions of individuals
seeking outpatient counseling services for problems related to their gambling behaviors.
Inclusion criteria for participation included age: all were adults, at least 18 years of age
with no upper bound for age. Participants also must have been appropriate recipients of
outpatient psychotherapy. In particular, individuals were referred out of outpatient
treatment and removed from the study if the severity of their psychopathology warranted
more intensive treatment. A detailed description of the sample is contained in the results
section (Table 2; Table 4; Table 5).
Procedure
This study utilized archival data from a university-based counseling clinic, which
provided both gambling-specific and general counseling services to members of the
community. Consecutive admissions for outpatient counseling services for gambling
problems were invited to allow their clinical information to be used for research
purposes.
Prior to treatment, participants were administered a series of psychological
instruments. Each participant was administered the University of Rhode Island Change
Assessment (URICA) to gather data on pre-treatment stages of change. At this time, The
54
Hands Depression Screen, The Mood Disorder Questionnaire, the Carroll-Davidson
Generalized Anxiety Disorder Screen, and the Sprint-4 PTSD Screen were also
administered to assess for the presence of four commonly observed co-occurring
disorders. Pre-treatment severity of gambling problems was assessed through the NORC
Diagnostic Screen for Gambling Problems (NODS). The NODS also allowed for
classification of participants as pathological, problem, or at-risk gamblers. Therefore,
prior to treatment data were gathered related to the severity of gambling problems, the
presence of common co-occurring disorders, and participant’s readiness to change.
Also prior as a part of the pre-treatment paperwork, clients provided basic
demographic information, including age and gender. Clients were also asked for
racial/ethnic identity. However, this was not included in the analyses due the relatively
homogeneity of the clients at this treatment setting. Specifically, the vast majority,
approximately 85 percent, of the clients at this setting was non-hispanic/white, which fits
with the tendency for ethnic and racial minorities to be underrepresented in gambling
treatment settings (Volberg, 1994). Therefore, all other racial and ethnic groups
composed only about one seventh of the total group, which would be insufficient for
inferential testing of this factor.
Prior to each therapy session, including the first, each participant was
administered the Outcome Questionnaire 45 (OQ-45) to track psychosocial functioning
through the course of therapy. Therefore, the general functioning of the participants was
continually assessed during treatment.
55
Treatment was delivered at a university-based outpatient counseling clinic, which
has a focus on the treatment of problem and pathological gambling. However, treatment
is available to all community members, not only members of the university. The clinic is
a research and training facility for masters and doctoral level graduate students within the
Counseling Psychology program at the university. Therefore, treatment for this study was
provided by graduate students. These students received direct supervision related to the
therapeutic approaches being administered from licensed psychologists. Further, the
students were directly observed during therapy sessions through the use of closed circuit
television systems.
Treatment consisted of one hour of individual psychotherapy per week, except
when the university was closed. Treatment began with a semi-structured intake interview,
which typically took one to two sessions to complete. Immediately following the intake
interview, participants were allowed to select which therapeutic approach they desired to
receive. The participants could choose between Cognitive-Behavioral Therapy (CBT),
Time-Limited Dynamic Psychotherapy (TLDP), and Solution-Focused Brief Therapy
(SFBT). Prior to making their selection, participants were provided with a description of
the basic tenets of each form of therapy (Table 1). However, for a period, it had been the
practice of the treatment center to also randomly assign clients to a form of therapy. As
such, a smaller proportion of the clients in this study were not given the opportunity to
select their treatment.
Following the intake interview and therapy selection, psychotherapy began in
accordance with the form of therapy the participant chose or were assigned to. Treatment
56
plans were personalized in accordance with the therapeutic approach and the participants’
particular goals for therapy, as no standardized treatment protocols were utilized. There
was no limit on the number of sessions of therapy a participant could attend. Similarly,
there was no expectation or mandate regarding how many sessions a participant was to
attend. Treatment continued until there was a mutual agreement between the client and
therapist that therapy was no longer required or until the client discontinued attending
therapy sessions.
Table 1
Description of Therapy Choices
Therapy
Description
Cognitive-Behavioral
Cognitive Behavioral Therapy (CBT) is a goal oriented
Therapy
therapy that is active and directive in nature. The purpose of
this therapy is to explore thoughts and behaviors that may
cause you to engage in problematic behaviors. You and
your counselor work together to develop new ways of
thinking about problems, and you will learn new skills to
deal with them. To help identify patterns, thought logs are
used frequently in your session and between sessions. Your
counselor will ask you to complete assignments and try
change techniques that may be practiced throughout your
week.
Time-Limited
Dynamic Therapy’s goals including improved relationships,
Dynamic
attunement to feelings, and/or a resolution of a conflict.
Psychotherapy
Your therapist focuses on the expression of emotions, and
explores wishes, attitudes, and behaviors. Your therapist
will help you to talk about yourself and your relationships to
57
identify your expectations and repetitive patterns in your
life and your relationships. The focus is often on resolving
past experiences and prior traumas, and identifying
expectations you have for yourself and others. You will be
asked to think about yourself and relationships between
sessions.
Solution-Focused
Solution-Focused Therapy is a goal-oriented therapy that
Brief Therapy
focuses on helping you to clarify what is important to you,
changes you would like to have in your life, and steps you
might take to achieve your goals. This is an active therapy
where your counselor and you will be working to identify
your strengths and successes and will search with you for
solutions to your present dilemma. There often is discussion
on what small changes and steps will improve your life, and
what to pay attention to and what to think about doing
differently between sessions. You will be asked to notice
any progress.
Instruments
URICA
The University of Rhode Island Change Assessment (URICA) was developed as a
scale to measure an individual’s readiness to, or stage of change (McConnaughy,
Prochaska, & Velicer, 1983). The URICA comprises 32-items, with 8 items relating to
each of the four measured stages of change: precontemplation, contemplation, action, and
maintenance. Items include: “As far as I’m concerned, I don’t have any problems that
need changing” and “I am finally doing some work on my problems.” The participant
assigns a rating of 1 to 5 (1= strong disagreement, 5 = strong agreement) to each item
58
based on the relevance of the statement to their particular situation. Therefore, scores for
each stage can range from 8 to 40, with higher scores reflecting greater endorsement of
the characteristics of that stage. The instrument then assigns each participant to one of the
four stages of change.
Confirmatory factor analysis has indicated that the questions load onto the
theorized four-factor structure, which supports the validity of the instrument (Pantalon et
al., 2002). The URICA has also demonstrated concurrent validity, as it has been found to
correlate strongly with the Contemplation Ladder, another instrument that measures
readiness to change (r=0.41) (Amodei & Lamb, 2004). The URICA has demonstrated
acceptable to strong internal consistency estimates for each of the four subscales (α’s =
.74–.88) during administrations with treatment seeking pathological gamblers (Petry,
2005).
NODS
The National Opinion Research Center (NORC) DSM-IV Diagnostic Screen for
Gambling Problems (NODS) is a 17-item instrument designed to measure gamblingrelated problems. The NODS directly assesses the ten diagnostic criteria for Pathological
Gambling found in the Diagnostic and Statistical Manual of Mental Disorders, fourth
edition with text revisions (DSM-IV-TR) (American Psychiatric Association, 2000;
Gerstein et al., 1999). The NODS is comprised of yes-no questions such as: “Have you,
during the past year, tried to stop, cut down or control your gambling?” and “Have there
been periods during the past year when you needed to raise your bets in order to get the
same feeling of excitement?” Consequently, scores on the NODS range from 0 to 10,
59
depending on how many criteria are endorsed. Higher scores indicate greater gamblingrelated problems. A score of 5 or higher indicates pathological gambling in accordance
with the DSM-IV-TR. The NODS has demonstrated the ability to identify subclinical
levels of disordered gambling behavior (Hodgins, 2004; Volberg, 2002). Scores of 3 or 4
indicate problem gambling.
Fager (2006) in a study of people seeking treatment for gambling problems found
strong evidence of test-retest reliability of (r=.77). Wickwire (2008) found the NODS to
have strong internal consistency estimates (α=.88). The NODS has also demonstrated
high concurrent validity (r=.85) with the South Oaks Gambling Screen, another common
instrument designed to measure problematic gambling behavior (Wickwire, 2008).
OQ-45
The Outcome Questionnaire 45 (OQ-45) was designed as a measure of mental
healthcare outcomes (Lambert et al., 1996). The instrument measures across three
domains of psychosocial functioning: subjective discomfort, interpersonal relations, and
social role performance. The OQ-45 provides a score for each of the three domains, in
addition to a total score representing overall psychosocial functioning. The Outcome
Questionnaire is a 45 item instrument, which uses a Likert-scale of response options. The
instrument contains items such as: “I feel no interest in things,” “I feel irritated,” and “I
feel lonely.” Total scores on the OQ-45 can range from 0 to 180. Lower scores on the
OQ-45 indicate higher levels of psychosocial functioning. Conversely, higher scores
indicate greater levels of distress. Scores at or above 63 are considered to indicate
psychosocial distress that is of clinical significance.
60
Lambert et al. (1996) found the OQ-45 to have sound psychometric properties,
including: high test-retest reliability estimate (r = 0.84), strong overall internal
consistency (α = 0.93), and moderate to strong concurrent validity estimates (r = 0.60 to r
= 0.88) across a number of measures of psychosocial distress. Further analysis of the OQ45 by Vermeersch, Lambert, and Burlingame (2000) has demonstrated the OQ-45 to be
sensitive to an individual’s change in level of psychosocial functioning.
WAI-S
The Working Alliance Inventory- Short (WAI-S) is an abbreviated version of the
Working Alliance Inventory, which was created to measure the three specific aspects of
the therapeutic alliance: agreement on tasks, agreement on goals, and interpersonal bond
between client and therapist (Horvath & Greenberg, 1989). The 12 items on the WAI-S
are designed to measure how well the client and therapist are working together in
therapy, including: “We agree on what is important for me to work on”, “I believe the
way we are working with my problem is correct”, and “I believe _____ likes me”. The
WAI-S is a 12-item instrument, the items on which are answered on a 7-point rating scale
(1= never, 7 = always). A higher score on the WAI-S, therefore, indicates a stronger
therapeutic alliance.
Evaluation of implementations of the WAI-S used to measure the client’s
perceptions of the therapeutic alliance has demonstrated moderate to strong internal
consistency estimates (α’s = .74–.90; Busseri & Tyler, 2003). The WAI-S has also
demonstrated evidence for concurrent validity, as it correlates highly with previously
established instruments for measuring the therapeutic alliance, such as the Penn Helping
61
Alliance Questionnaire (r = 0.75) and the California Psychotherapy Alliance Scale (r =
0.80) (Hatcher & Gillaspy, 2006).
HANDS
The Harvard Department of Psychiatry/National Depression Screening Day Scale
(HANDS) was designed to assess for a major depressive episode in accordance with the
DSM-IV-TR criteria (Baer et al., 1999). The instrument enquires into the functioning of
the individual during the preceding two weeks. Items on the instrument assess how often
an individual is experiencing certain depressive symptoms, including: poor appetite,
decreased interest, and suicidal ideation. This 10-item scale uses a 4-point response scale,
with response options including: none or little of the time, some of the time, most of the
time, and all of the time. Possible scores can range of 0 to 30. Higher scores indicate a
greater number of symptoms and/or higher frequency of symptom occurrence. A score of
9 points or higher indicates a positive screen for a major depressive episode.
Through psychometric evaluation, the HANDS has demonstrated strong internal
consistency (α = 0.87; Baer et al., 1999). Validity investigation found that the HANDS
has a sensitivity of 0.95 and a specificity of 0.94, which was found in that study to be
superior to the Beck Depression Inventory-II (Baer et al., 1999).
MDQ
The Mood Disorder Questionnaire (MDQ) assesses lifetime history of bipolar
disorder (Hirschfeld et al., 2000). It contains 13 yes/no questions related to manic and
hypomanic symptoms associated with bipolar disorder I and II. The instrument assesses if
there has ever been a period during the individual’s life when they have exhibited
62
hypomanic symptoms, including: hyperactivity, racing thoughts, and decreased need for
sleep. Possible scores range from 0 to 13, depending on number of endorsed symptoms.
Scores of 7 points or higher constitute a positive screen for a mood disorder.
The MDQ has demonstrated high internal consistency (α=0.90; Hirschfield et al.,
2000). The MDQ has also demonstrated acceptable levels of sensitivity (0.73) and strong
specificity (0.90) in the original validation study (Hirschfeld, 2010). In a review of
numerous studies implementing the MDQ, it was found that the overall sensitivity was
0.61 and overall specificity was 0.88; however, the sensitivity was found to be
significantly higher in studies that implemented the MDQ to clinical samples as opposed
to the general population (Zimmerman & Galione, 2011).
CDGAD
The Carroll-Davidson Generalized Anxiety Disorder Screen (CDGAD) was
designed to detect symptoms of generalized anxiety from the previous six months in
accordance the DSM-IV-TR criteria for Generalized Anxiety Disorder (Carroll &
Davidson, 2000). This 12-item instrument asks individuals if for most days during the
past six months they have experienced anxious symptoms, including: nervous feeling,
excessive worry, irritability, and difficulty sleeping. Each item has yes/no response
options. Therefore, total scores range from 0 to 12, with higher scores indicating greater
symptom endorsement. Scores of 6 or higher constitute a positive screen for Generalized
Anxiety Disorder. An evaluation of this instrument found it to exhibit moderate internal
consistency (α = 0.82; Leyton-Armakan et al., 2012). This instrument has also
63
demonstrated a sensitivity of 0.64 and specificity of 0.90 (Carroll & Davidson, 2000;
Screening for Mental Health, 2014).
SPRINT-4
The SPRINT-4 is an abbreviated version of the Short Posttraumatic Stress
Disorder Rating Interview (SPRINT; Connor & Davidson, 2001). The instrument begins
by asking if the participant has experienced or witnessed a traumatic event that involved
serious injury, loss of life, or significant threat of injury. Then, four yes/no questions
assess whether the individual, as a consequence of that event, is currently experiencing
symptoms of Posttraumatic Stress Disorder (PTSD), such as intrusive memories of the
event. Moreover, each item corresponds to one of the four PTSD symptom clusters
(intrusive, avoidance, numbing, and hyperarousal). Affirmative responses to 2 or more of
the items qualify as a positive screen for PTSD. The SPRINT has demonstrated strong
concurrent validity with the Clinician-Administered Scale for DSM-IV (CAPS) with
correlations between the two instruments found to range between 0.64 and 0.79
(Vaishnavi, Payne, Connor, & Davidson, 2006). A Romanian version of the SPRINT
demonstrated very strong internal consistency within a clinical sample (r = 0.90), as well
as strong sensitivity (.88) and specificity (.90; Herta, Nemes, & Cozman, 2013).
Analytical Strategy
R version 3.0.2 and SPSS 22 were used for the statistical analyses.
Missing Values Analysis
Given the nature of this study as an implementation of propensity score analysis
with a relatively small sample, missing values were imputed rather than listwise deleting
64
incomplete cases. Prior to imputation, missing values had to be tested for whether or not
they were missing completely at random. Little’s Missing Completely At Random
(MCAR) Test was used to determine if missing values were occurring completely at
random (Little, 1988). When data is MCAR the missing values have no relationship with
any of the observed values within the dataset, which means that the missing values can be
imputed without introducing bias to the eventual analyses (McKnight, McKnight, Sidani,
Figueredo, 2007).
Expectation Maximization (EM) was utilized to replace the missing values. EM is
an iterative procedure that utilizes maximum likelihood (ML) estimation. This process
consists of estimating missing values using ML, generating parameter estimates, and this
process iterates until finally converging upon a solution (McKnight et al., 2007). EM
techniques have been shown effective in calculating relatively unbiased estimates for
missing values (Gold & Bentler, 2000; Liu & Brown, 2013)
Research Question One
The first component of this research sought to explore which variables were
contributing to the participants’ particular therapeutic selections. Particularly given the
exploratory nature of this research question, the following variables were analyzed in
individual multinomial logistic regression models to determine their relationship with
therapeutic selection: age, gender, stages of change, co-occurring psychological disorders
(depression, post-traumatic stress disorder, anxiety, and mood disorder), and severity of
gambling problems. To explore which client characteristics might be related to treatment
selected groups based on treatment selected were compared for equivalence across these
65
variables. For the continuous variables, one-way ANOVAs were conducted to test
whether means of these variables were equivalent between groups. For the categorical
variables, chi-square tests of independence were run to determine if proportions varied
between groups. Variables whose p-value was less than 0.25 were retained and included
in the propensity score estimation model.
Research Question Two
This component of the research sought to determine if the participants as a whole
were selecting particular forms of therapy disproportionately to the other forms.
Therefore, a chi-square goodness of fit test was run to assess whether or not the three
therapies were selected in equal proportions. A chi-square of 5.991 with 2 degrees of
freedom (p<0.05) was used as the critical value at which differences in proportions of
therapeutic selections was to be considered statistically significantly different.
Research Question Three
For the second component of the research, to determine whether or not the three
therapeutic approaches demonstrate equally efficacious short-term outcomes, change in
psychosocial functioning, as measured by the OQ-45, from the first to the fifth treatment
session was compared across the three treatment groups. Since the majority of
participants were allowed to choose the type of treatment they receive, propensity score
matching was implemented to correct for any potential selection bias within the sample.
The TriMatch statistical package within R was designed to allow for matching of three
groups, as opposed to traditional propensity score matching techniques that are designed
to only match participants from two groups (Bryer, 2013). Similar to other propensity
66
score matching methods, the TriMatch algorithm uses logistic regression to estimate
propensity scores. In the estimation of propensity scores, three logistic regression models
are calculated. The first model estimates the probability of being in the CBT group, as
opposed to SFBT. The second model estimates the probability of being in the CBT group,
as opposed to TLDP. The third model estimates the probability of being in the SFBT
group, as opposed to TLDP. The matching system then finds trios of participants, one
from each group, based on minimizing the within-trio differences in propensity scores.
The Trimatch statistical package includes various matching options (Bryer, 2013).
The first procedure is analogous to matching without replacement in two-group
propensity score matching. This procedure, which is referred to Maximum Treat
matching within Trimatch, will only match a treatment 1 unit for a second time if a
treatment 2 unit would be otherwise left unmatched. The second procedure, the caliper
matching, allows for matching of all possible matched triplets within a specified caliper.
In other words, the program will create all matches that meet a certain criteria for
closeness of match. The default caliper is 0.25 standard deviations of the propensity
scores. The third and fourth procedures are analogous to the 1:n procedures in two-group
propensity score matching. These procedures utilize 2:1:n and 3:2:n matching,
respectively. These final two procedures specify how many times each of the first two
groups may be reused within the matching. It should be noted that the default in Trimatch
is to order the groups in terms of sample size. Therefore, the 3:2:n allowed participants in
the largest group to be matched up to three times and participants in the second largest
group to be matched up to two times. Each of these matching procedures were utilized
67
and compared in terms of level of effectiveness in balancing the groups across the
identified covariates.
After the matching has occurred, this package conducted a repeated measures
ANOVA comparing mean OQ-45 scores across groups. If the main effect of time was
found to be significant at p<0.05, the package automatically allowed for dependent
samples t-tests to explore specific group differences. When significant pairwise
differences were detected, sensitivity analyses were conducted to test the robustness of
these findings to potential confounding influences of unobserved covariates (Rosenbaum,
1987).
To compare matching strategies, three criteria were assessed: proportion of
subjects used in the matching, number of matched triplets created, and balance of the
covariates from the propensity score estimation models. Balance will be assessed through
calculating statistical significance of covariate differences across groups following
matching. Balance will also be assessed through an inspection of the reduction in bias
through matching (Cochran & Rubin, 1973). Bias, in this context, was defined as the sum
of the absolute values of the pairwise comparisons between groups on a given covariate.
Percent bias reduced is defined as the percentage of the pre-match bias that is reduced
through the matching procedure. For instance, a bias reduced value of 0.75 indicates that
the post-matching bias is 75 percent less than the pre-match bias. It was calculated by
subtracting the post-matching bias from the pre-match bias, then dividing that amount by
the pre-match bias. The final way in which balance was assessed was through the
calculation of standardized post-matching bias (Caliendo & Kopeinig, 2008). Following
68
matching, the pairwise standardized differences were calculated by dividing each
pairwise difference for a given covariate divided by the pooled standard deviation for the
groups being compared. These standardized pairwise differences were then averaged to
achieve an overall standardized bias for each covariate. These values were then averaged
across each covariate within a given matching strategy. These overall standardized biases
were also used to compare the matching procedures.
69
Chapter Three: Results
Missing Values Analysis
Prior to data imputation, missing values were assessed for appropriateness for
replacement. Table 2 contains the proportions of missing values for each variable
included in subsequent analyses. Proportions of missing values ranged from 0 to 6.9
percent. The most commonly missing variables were drawn from the questionnaire
related to the co-occurring mental health disorders. Data were also imputed using OQ-45
values from the first six sessions. Each of the three cases missing the OQ-45 score from
the fifth session provided an OQ-45 score from a sixth session.
The results of the Little’s MCAR test were not statistically significant (X2(175) =
183.04, p = 0.323; Little, 1988). This indicates that the missing values are likely to be
occurring completely at random.
70
Table 2
Missing Values
Variable
Mean (SD)
N
% Missing
Imputed
Values
Mean (SD)
Age
45.91 (11.90)
101
0
45.91 (11.90)
Gender
59.8% Male
101
0
59.8% Male
URICA1
10.16 (1.71)
96
5.0
10.12 (1.70)
OQ first
68.25 (20.27)
100
1.0
68.10 (20.22)
57.51 (22.36)
98
3.0
58.23 (22.60)
WAI-S3
10.02 (2.23)
98
3.0
10.00 (2.20)
HANDS4
55.32% positive
94
6.9
56.44%
session2
OQ fifth
session2
positive
CD-GAD5
55.32% positive
94
6.9
56.44%
positive
SPRINT-46
47.87% positive
94
6.9
49.50%
positive
MDQ7
35.11% positive
94
6.9
34.65%
positive
NODS8
1
8.09 (2.12)
101
0
8.09 (2.12)
:University of Rhode Island Change Assessment; 2: Outcome Questionnaire-45; 3: Working Alliance
Inventory- Short Form; 4: Harvard Department of Psychiatry National Depression Day Screening Scale(%
positive screen); 5: Carroll-Davidson Generalized Anxiety Disorder Screen (% positive screen); 6: Short
Post Traumatic Rating Interview(% positive screen); 7: Mood Disorder Questionnaire(% positive screen);
8
: NORC DSM-IV Diagnostic Screen for Gambling Problems.
71
Characteristics of the Sample
The characteristics of the sample, following the expectation maximization
imputation are provided in Table 2. Of the 101 individuals included in the analyses, 61
identified as male. The average age for the sample was 45.9 years. The mean NODS
score (8.09) indicates that overall this sample consisted of individuals with relatively
severe gambling problems. As measured by the NODS, 93.1 percent of the sample met
the DSM-IV-TR criteria for pathological gambling, and each participant in the sample
endorsed at least one criterion for pathological gambling.
When comparing gender differences within the sample, the female participants
were significantly older than their male counterparts (t(96.56)=3.28, p=0.001). Females
presented with significantly more severe gambling problems, as measured by the NODS
(t(96.01)=3.17, p=0.002). Females also presented with more severe psychosocial
difficulties, as measured by their first session OQ-45 score (t(99)=2.20, p=0.03).
Also notable is the prevalence of co-occurring disorders in this sample. Only 17.8
percent of the sample did not screen positively for any of the co-occurring mental health
disorders. In fact, 60.4 percent of the sample screen positively for at least two of the cooccurring disorders. Depression and Generalized Anxiety were the two most commonly
observed co-occurring mental health disorders.
Assessing Treatment Selection
The first two research questions sought to identify characteristics of the problem
gambling participants that may be influencing their selection of treatments, as well as
identify which, if any, of the treatments are being selected disproportionately to the
72
others. To address this latter question, a chi-square goodness of fit test was run to
determine if the treatments were being selected in equal proportions.
Of the 101 participants included in the analysis of outcomes, 79 (78.2%) were
allowed to select their own form of therapy as described within the method section of this
paper. The following set of analyses was conducted on this subset of the sample in order
to assess treatment preferences, as well as the factors contributing to treatment selection.
Table 3 contains the observed proportions of therapies that were selected. Both
the cognitive-behavioral therapy and the solution-focused brief therapy were selected
more often than would be expected, while the time-limited dynamic psychotherapy was
selected less often than expected. However, the results of the chi-square goodness of fit
test was not statistically significant (X2(2)=4.13, p=0.13). Therefore, the data indicate
that the therapies were selected in statistically equivalent proportions.
Table 3
Observed Therapy Selections
Observed proportion Observed Frequency Expected
Frequency
TLDP
22.8%
18
26.3
SFBT
36.7%
29
26.3
CBT
40.5%
32
26.3
Comparing Treatment Selection Groups
The comparisons between groups based on treatment selection are summarized in
Table 4. As the table indicates, none of the individual observed baseline characteristics
statistically varied statistically significantly between groups at the p<0.05 level.
73
However, when viewed as a whole the results suggest that groups may vary in terms of
overall clinical presentation. Specifically, the SFBT group had the lowest mean scores on
initial OQ-45 and NODS, as well as had the lowest proportion screening positively for
depression, generalized anxiety, and mood disorder. The selection of TLDP appears to be
favored by younger, male therapy clients.
Table 4
Baseline Covariates Across Treatment Selection Groups
CBT
SFBT
TLDP
Test Statistic
n=32
n=29
n=18
Age
45.51
47.17
39.61
F(2,76) = 2.24, p = 0.11
Gender
56.25
58.62
77.78
X2(2) = 2.49, p = 0.29
URICA
9.88
10.27
10.38
F(2,76) = 0.61, p = 0.55
OQ-45 first
70.16
62.24
68.72
F(2,76) = 1.24, p = 0.30
WAI-S
9.63
9.92
10.85
F(2,76) = 1.72, p = 0.19
HANDS
62.50
51.72
55.56
X2(2) = 0.74, p = 0.69
62.50
37.93
61.11
X2(2) = 4.28, p = 0.12
56.25
44.83
44.44
X2(2) = 1.02, p = 0.60
34.38
20.69
44.44
X2(2) = 3.08, p = 0.21
8.44
7.55
8.11
F(2,76) = 1.16, p = 0.32
(% Male)
session
(% positive)
CD-GAD
(% positive)
SPRINT-4
(% positive)
MDQ
(% positive)
NODS
74
Comparing Overall Treatment Groups
Selecting variables for the propensity score estimation model involved two steps.
The first step involved comparisons of baseline characteristics between the overall
treatment groups. Similar to the previous analysis, one-way ANOVAs were be used to
compare continuous variables, and chi-square tests of independence were used for
categorical covariates. The clinical and demographic covariates were also tested for their
relationship with the outcome variable. These analyses include both those that were
randomly assigned to a form of treatment (n=22) and those that selected their own form
of treatment (n=79).
The results of these comparisons are summarized in Table 5. Given that the
majority of participants selected their own treatment, it is not surprising that the results
are similar to those contained in Table 4. Once again, no covariates were found to
significantly vary between groups at the p<0.05 level. Considered in totality, the
Solution-Focused Brief Therapy group again had a less severe clinical presentation
compared to the other treatments. Also, the Time-Limited Dynamic Psychotherapy group
was again found to be marginally different in both age and gender.
The largest difference between the comparison of selection groups and the
comparison of overall treatment groups is with the Working Alliance Inventory measure.
The analysis of selection groups found this variable to marginally significantly differ
across groups (p = 0.19; Table 4). However, in the comparison of overall treatment
groups, this difference becomes virtually non-existent (p = 0.60; Table 5).
75
Table 5
Comparing Covariate Balance Across Treatment Groups
CBT
SFBT
TLDP
Test Statistic
n=45
n=32
n=24
Age
46.53
47.81
42.25
F(2,98) = 1.61, p = 0.21
Gender
53.33
59.38
75.00
X2(2) = 3.09, p = 0.21
URICA
10.06
10.25
10.06
F(2,98) = 0.13, p = 0.88
OQ first
70.67
63.03
70.04
F(2,98) =1 .49, p = 0.23
WAI-S
9.90
9.85
10.41
F(2,98) = 0.52, p = 0.60
HANDS
60.00
50.00
58.33
X2(2) = 0.81, p = 0.67
62.22
43.75
62.50
X2(2) = 3.07, p = 0.22
53.33
46.88
45.83
X2(2) = 0.48, p = 0.79
33.33
28.13
45.83
X2(2) = 1.96, p = 0.38
8.38
7.56
8.25
F(2,98) = 1.49, p = 0.23
(% Male)
session
(% positive)
CD-GAD
(% positive)
SPRINT-4
(% positive)
MDQ
(% positive)
NODS
Relationships Between Covariates and Outcome
The second phase of identifying covariates for the propensity score model was to
identify baseline characteristics that are related to the outcome. Table 6 contains the
bivariate correlations between each covariate and the outcome variable (the difference
between OQ-45 scores from the first and fifth sessions) for the total sample (N=101).
None of the ten observed characteristics was significantly related to the outcome variable
76
at the p<0.05 level. Only one variable, OQ-45 score from the first session, was related at
p<0.25.
Table 6
Bivariate Correlations Between Baseline Characteristics and Outcome Measure
correlation (r)
p-value
Age
0.09
0.38
Gender
0.07
0.48
URICA
0.09
0.38
OQ first session
0.16
0.11
WAI-S
-0.02
0.85
HANDS (positive screen)
-0.01
0.89
CD-GAD (positive screen)
0.04
0.67
SPRINT-4 (positive screen)
-0.04
0.66
MDQ (positive screen)
-0.03
0.73
NODS
0.09
0.39
Unmatched Analysis
For comparison, an analysis was run comparing the three treatment groups on the
outcome variable without matching or covariate adjustment. A one-way ANOVA was
used to compare the means of the dependent variable, amount of change in OQ-45 scores
from the first to the fifth treatment sessions, across treatment groups. The results indicate
that this analysis was unable to detect any differences between groups (F(2,98) = 0.93, p
= 0.40).
77
Propensity Score Analyses
The following sections include the tests involved in specifying the propensity
score estimation models, as well as the results from each of the four matching procedures.
Specifying the Propensity Score Model
In accordance with the a priori criteria set in the proposed analytic strategy,
covariates that differed between groups at the p<0.25 level were retained for the
propensity score estimation model. Additionally, covariates correlated with the outcome
at the p<0.25 level were included in the propensity score model. From the comparison of
covariate balance across treatment groups the following covariates were identified: age,
gender, first session OQ-45 score, CD-GAD diagnostic classification, and NODS score.
The only variable identified for its correlation with the outcome, first session OQ-45
score, was already selected for inclusion based on the comparison of treatment groups.
Therefore, propensity scores were estimated using these five covariates.
The propensity score matching procedures would then seek to balance the
treatment groups, specifically as related to these five covariates. In order to assess the
effectiveness of the balance created by each matching method, a baseline level of bias
was calculated for each covariate. Bias, in accordance with the method utilized by Bai
(2013), was calculated by summing the absolute value of the pairwise differences
between each group for each covariate (Table 7).
78
Table 7
Pairwise Differences of Covariates Prior to Matching
CBT-SFBT
CBT-TLDP
SFBT-TLDP
Bias
Age
-1.28
4.28
5.56
11.12
Gender
-0.06
-0.22
-0.16
0.44
Initial OQ-45
7.64
0.63
-7.01
15.28
Anxiety
-0.18
0.00
-0.19
0.37
NODS
0.82
0.13
-0.69
1.64
Covariate Balance Across Propensity Estimation Models
In the estimation of propensity scores, three logistic regression models are
calculated. The first model estimates the probability of being in the CBT group, as
opposed to SFBT. The second model estimates the probability of being in the CBT group,
as opposed to TLDP. The third model estimates the probability of being in the SFBT
group, as opposed to TLDP.
Prior to any matching procedures, covariate balance was tested using the
propensity scores from these three models stratified across quintiles. Figure 1 illustrates
the post matching covariance balance for all three propensity score models. Column 1 on
Figure 1 illustrates the logistic regression model created between CBT and SFBT,
column 2 of Figure 1 illustrates the model created between the CBT and TLDP groups,
and column 3 illustrates the balance created from the model containing SFBT and TLDP.
The red line illustrates the un-adjusted effect sizes of the covariate imbalance, whereas
the blue line illustrates the adjusted biases resulting from the stratified comparison. As
can be seen, the balance is improved for all covariates across each treatment combination,
79
except in one instance. The proportions of positive anxiety screens were marginally less
balanced when stratified by propensity scores, but this covariate was already well
balanced between the CBT and TLDP groups prior to matching (Table1, column 2).
Figure 1
Covariate Balance From the Three Propensity Score Estimation Models
Column 1: the model containing CBT and SFBT clients
Column 2: the model containing CBT and TLDP clients
Column 3: the model containing SFBT and TLDP clients
Maximum Treat Matching
Following the matching procedure, 28 matched triplets were created. Overall, 61
(60.4%) of the subjects were utilized with this matching procedure. Table 8 summarizes
80
the unmatched individuals by treatment group. Only those cases involved in the matching
process were included in the following set of analyses.
Table 8
Unmatched Participants Within Maximum Treat Matching
Number unmatched
% unmatched
CBT
26
57.8
SFBT
9
28.1
TLDP
5
20.8
Total
40
39.6
The values for each covariate by group are contained in Table 9. The results of the
inferential tests of covariate balance are contained in Table 10. As was also the case prior
to matching, none of the covariates significantly differed at p<0.05 after matching.
Table 9
Covariate Values Following Maximum Treat Matching
CBT
SFBT
TLDP
44.21
44.79
45.64
0.43
0.36
0.43
Initial OQ-45
67.25
66.04
72.50
Anxiety2
0.64
0.54
0.64
7.82
8.54
8.43
Age
Gender
1
NODS
1
2
: proportion male; : proportion screen positively for generalized anxiety by the CD-GAD.
81
Table 10
Inferential Testing of Maximum Treat Post-Matching Covariate Balance
Test Statistic
p-value
Age
F(2,54) = 0.20
p = 0.82
Gender
X2(2) = 0.80
p = 0.67
OQ
F(2,54) = 0.85
p = 0.43
Anxiety
X2(2) = 1.06
p = .59
NODS
F(2,54) = 1.70
p = 0.19
The reduction in bias was calculated for each covariate following the Maximum
Treat matching (Table 11). The largest bias reduction was observed for age and gender
with 74% and 68%, respectively. The matching was less effective for the other
covariates, particularly initial OQ-45 scores and NODS scores, with 9% and 12% bias
reductions, respectively. The statistical significance of the between-group differences of
NODS scores remained marginally significant despite a modest reduction in bias (Table
10). Boxplots of each covariate by group, as well as boxplots of the pairwise comparisons
following Maximum Treat matching are contained in Appendix A.
82
Table 11
Pairwise Differences of Covariate Balance Following Maximum Treat Matching
CBT-
CBT-
SFBT-
SFBT
TLDP
TLDP
Age
-0.58
-1.43
-0.85
Gender
0.07
0.00
Initial
1.21
Anxiety
NODS
Bias
Stand.
% Bias
Bias
reduced
2.86
0.09
0.74
-0.07
0.14
0.09
0.68
-5.25
-6.46
13.92
0.20
0.09
0.1
0.00
-0.1
0.20
0.14
0.46
-0.72
-0.61
0.11
1.44
0.28
0.12
OQ-45
Following the matching procedure, a repeated measures ANOVA was run
comparing the three treatment groups on the dependent variable, the amount of change in
OQ-45 scores between the first and fifth sessions. The results of the test were statistically
significant (F(2,54) = 3.37, p = 0.042. Given the statistical significance, a series of
dependent sample t-tests were automatically conducted through Trimatch to assess
pairwise comparisons of treatment effectiveness (Table 12). These results suggest that the
CBT treatment was significantly more effective than the TLDP treatment. The outcome
values for each treatment group are presented in Figure 2. The boxplots in Figure 2
contain a box and whisker plot containing the median and quartile ranges, as well as a red
circle indicating the mean and green bars illustrating the 95% confidence interval about
the mean. The pairwise comparisons are illustrated in Figure 3 in the same fashion as the
plots in Figure 2, except the standard error is illustrated by a solid green box and the
pairwise differences are listed on the graph.
83
Table 12
Post-hoc Pairwise Comparisons of Outcome Following Maximum Treat Matching
Value
Test Statistic
p-value
SFBT-TLDP
4.9
t(27) = 1.28
0.21
SFBT-CBT
-4.3
t(27) = -1.31
0.20
TLDP-CBT
-9.2
t(27) = -2.62
0.01
Figure 2
Boxplots of OQ-45 Change for Each Group After Maximum Treat Matching
84
Figure 3
Boxplots of Pairwise Differences of OQ-45 Change After Maximum Treat Matching
A sensitivity analysis was also conducted to test the robustness of the CBT-TLDP
pairwise comparison against bias introduced by unobserved variables to the propensity
score estimation models (Rosenbaum, 1987; Table 13). The results suggest the findings
remain robust to Gamma < 1.4. In other words, an unobserved variable(s) would have to
change the odds of assignment by a factor of 1.4 before the finding would become nonsignificant.
85
Table 13
Sensitivity Analysis for CBT-TLDP Comparison Following Maximum Treat Matching
Gamma
p-value
1.0
0.01
1.1
0.02
1.2
0.03
1.3
0.04
1.4
0.06
Caliper Matching
The caliper matching resulted in the largest number of matched triplets (126). It
also resulted in the largest portion of the overall sample being utilized in the matching
procedure (79.2%). The unmatched participants by group are summarized in Table 14.
Only those cases involved in the matching process were included in the following set of
analyses.
Table 14
Unmatched Participants Within Caliper Matching
Number unmatched
% unmatched
CBT
7
15.6
SFBT
9
28.1
TLDP
5
20.8
Total
21
20.8
The values for each covariate by treatment group following the caliper matching
are contained in Table 15. None of the covariates differed significantly across groups at
86
p<0.05; however, the covariates for positive anxiety screens and NODS scores
approached statistical significance (Table 16). The fact that between-group differences
became closer to statistically significant following a matching procedure that reduced
bias for the anxiety and NODS covariates by 35% and 51%, respectively, is likely a result
of the greatly increased statistical power resulting from the large number of matched
triplets created with this procedure (Table 17). Boxplots of each covariate by group, as
well as boxplots of the pairwise comparisons following caliper matching are contained in
Appendix B.
Table 15
Covariate Values Following Caliper Matching
CBT
SFBT
TLDP
Age
45.47
44.47
46.13
Gender
0.49
0.49
0.53
Initial OQ-45
71.74
70.71
73.62
Anxiety
0.78
0.66
0.69
NODS
8.51
8.76
8.36
Table 16
Inferential Testing of Caliper Post-Matching Covariate Balance
Age
Test Statistic
p-value
F(2,250) = 1.35
p = 0.26
2
Gender
X (2) = 1.43
p = 0.49
OQ
F(2,250) = 0.85
p = 0.43
Anxiety
X2(2) = 5.66
p = 0.06
NODS
F(2,250) = 2.83
p = 0.06
87
Bias reduced for each covariate following the caliper matching is outlined in
Table 17. For four of the five covariates, bias was reduced by more than 50%. The only
exception was proportion of positive anxiety screens, as measured by the CD-GAD,
which had a bias reduction of 35%.
Table 17
Pairwise Differences of Covariate Balance Following Caliper Matching
CBT-
CBT-
SFBT-
SFBT
TLDP
TLDP
Age
1.00
-0.66
-1.66
Gender
-0.04
0.00
Initial
1.03
Anxiety
NODS
Bias
Stand.
Bias
Bias
reduced
3.32
0.11
0.70
0.04
0.08
0.05
0.81
-1.88
-2.91
5.82
0.09
0.62
0.12
0.09
-0.03
0.24
0.18
0.35
-0.25
0.15
0.40
0.80
0.18
0.51
OQ-45
The repeated-measures ANOVA main effect of treatment on the outcome variable
was statistically significant (F(2,250) = 9.37, p < 0.001), indicating significant variability
in treatment effectiveness. The post-hoc comparisons are contained in Table 18. As with
the Maximum Treat matching, CBT was found to be more effective than TLDP. Unlike
the previous analysis, SFBT was also found to be more effective than TLDP. Figure 4
illustrates the differences in outcome between groups, while Figure 5 illustrates the
pairwise comparisons.
88
Table 18
Post-hoc Pairwise Comparisons Following Caliper Matching
Value
Test Statistic
p-value
SFBT-TLDP
6
t(125) = 3.69
< 0.001
SFBT-CBT
-1.4
t(125) = -0.74
0.46
TLDP-CBT
-7.3
t(125) = -3.89
< 0.001
Figure 4
Boxplots of OQ-45 Change for Each Group After Caliper Matching
89
Figure 5
Boxplots of Pairwise Differences of OQ-45 Change After Caliper Matching
Two sensitivity analyses were run for the caliper matching data, one for each of
the significant pairwise comparisons of the outcome (Table 19; Table 20). For each of
these sensitivity analyses, the results were found to be robust against hidden biases at
Gamma < 1.5. This suggests that these findings are slightly more robust than that found
through the Maximum Treat matching.
90
Table 19
Sensitivity Analysis for SFBT-TLDP Comparison Following Caliper Matching
Gamma
p-value
1.0
< 0.001
1.1
< 0.01
1.2
< 0.01
1.3
0.02
1.4
0.04
1.5
0.08
Table 20
Sensitivity Analysis for CBT-TLDP Comparison Following Caliper Matching
Gamma
p-value
1.0
< 0.001
1.1
< 0.01
1.2
< 0.01
1.3
0.02
1.4
0.04
1.5
0.07
2:1:n Matching
Following the 2:1:n matching, 44 matched triplets were created. A total of 62, or
61.4% of the total sample was utilized by this matching procedure. The unmatched
participants are summarized by group in Table 21. Only those cases involved in the
matching process were included in the following set of analyses.
91
Table 21
Unmatched Participants Within 2:1:n Matching
Number unmatched
% unmatched
CBT
22
48.9
SFBT
9
28.1
TLDP
8
33.3
Total
39
38.6
The values for each covariate by treatment group following the 2:1:n matching are
summarized in Table 22. Boxplots of each covariate by group, as well as boxplots of the
pairwise comparisons following 2:1:n matching are contained in Appendix C. The
covariates were each statistically balanced across groups at p<0.05 (Table 23).
Table 22
Covariate Values Following 2:1:n Matching
CBT
SFBT
TLDP
Age
44.16
44.73
44.57
Gender
0.50
0.39
0.50
Initial OQ-45
66.55
65.70
69.05
Anxiety
0.61
0.50
0.64
NODS
7.82
8.36
8.18
92
Table 23
Inferential Testing of 2:1:n Post-Matching Covariate Balance
Test Statistic
p-value
Age
F(2,86) = 0.06
p = 0.95
Gender
X2(2) = 2.94
p = 0.23
OQ
F(2,86) = 0.32
p = 0.73
Anxiety
X2(2) = 2.30
p = 0.32
NODS
F(2,86) = 1.39
p = 0.25
Further inspection of the balance created by the 2:1:n matching indicates that for
three of the covariates (age, gender, and initial OQ-45 score), at least 50% of the bias was
reduced through this procedure (Table 24). The bias in age is largely eliminated, with
90%of the overall bias reduced.
Table 24
Pairwise Differences of Covariate Balance Following 2:1:n Matching
CBT-
CBT-
SFBT-
SFBT
TLDP
TLDP
Age
-0.57
-0.41
0.16
Gender
0.11
0.00
Initial
0.85
Anxiety
NODS
Bias
Stand.
Bias
Bias
reduced
1.14
0.04
0.90
-0.11
0.22
0.15
0.50
-2.50
-3.35
6.70
0.10
0.56
0.11
-0.03
-0.14
0.28
0.18
0.24
-0.54
-0.36
0.18
1.08
0.21
0.34
OQ-45
The results of the repeated-measures ANOVA on the outcome variable returned
nonsignificant results (F(2,86) = 2.81, p = 0.06). Given that this test was nonsignificant at
93
p < 0.05, no post-hoc dependent sample t-tests were returned. Neither was sensitivity
analysis warranted. Boxplots of the dependent variable by treatment group, as well as
boxplots of the pairwise comparisons of the dependent variable are contained in Figure 6
and Figure 7, respectively.
Figure 6
Boxplots of OQ-45 Change for Each Group After 2:1:n Matching
94
Figure 7
Boxplots of Pairwise Differences of OQ-45 Change After 2:1:n Matching
3:2:n Matching
The final matching procedure, the 3:2:n matching, resulted in 67 matched triplets.
This procedure utilized 69.3% of the total sample. A summary of the unmatched
participants by treatment group is contained in Table 25. Only those cases involved in the
matching process were included in the following set of analyses.
95
Table 25
Unmatched Participants Within 3:2:n Matching
Number unmatched
% unmatched
CBT
16
35.6
SFBT
9
28.1
TLDP
6
25
Total
31
30.7
Values for each covariate by treatment group following the 3:2:n matching are
contained in Table 26. Though no covariate significantly differed between groups at p <
0.05, NODS score was only marginally equivalent across groups (Table 27). Similar to
what was observed with the caliper matching, the between-group difference on the
NODS covariate was closer to statistically significant than prior to matching despite
having bias reduced by 23% with the 3:2:n matching (Table 28). This is likely due to the
power increase associated with the large number of triplets created through this
procedure. Boxplots of each covariate by group, as well as boxplots of the pairwise
comparisons following 3:2:n matching are contained in Appendix D.
96
Table 26
Covariate Values Following 3:2:n Matching
CBT
SFBT
TLDP
Age
44.23
44.69
44.90
Gender
0.42
0.40
0.42
Initial OQ-45
67.97
66.58
68.95
Anxiety
0.66
0.53
0.58
NODS
7.94
8.57
8.16
Table 27
Inferential Testing of 3:2:n Post-Matching Covariate Balance
Test Statistic
p-value
Age
F(2,122) = 0.11
p = 0.89
Gender
X2(2) = 0.09
p = 0.96
OQ
F(2,122) = 0.24
p = 0.79
Anxiety
X2(2) = 2.65
p = 0.26
NODS
F(2,122) = 2.74
p = 0.07
Pairwise comparisons and bias reductions for each covariate are contained in
Table 28. For two of the covariates, age and gender, bias was greatly reduced, with 88%
and 91% respectively. However, the 3:2:n matching also had two covariates where the
bias reduced was at or less than 30%.
97
Table 28
Pairwise Differences of Covariate Balance Following 3:2:n Matching
CBT-
CBT-
SFBT-
SFBT
TLDP
TLDP
Age
-0.46
-0.67
-0.21
Gender
0.02
0
Initial
1.39
Anxiety
NODS
Bias
Stand.
Bias
Bias
reduced
1.34
0.04
0.88
-0.02
0.04
0.02
0.91
-0.98
-2.37
4.74
0.07
0.69
0.13
0.08
-0.05
0.26
0.18
0.30
-0.63
-0.22
0.41
1.26
0.25
0.23
OQ-45
The results of the repeated-measure ANOVA on the 3:2:n matched groups was
not statistically significant (F(2,122) = 2.98 , p = 0.055). Given this nonsignificance,
post-hoc analyses, as well as sensitivity analyses were not conducted. Boxplots for the
outcome variable for each group and boxplots for the pairwise comparisons of the 3:2:n
matched groups are contained in Figure 8 and Figure 9, respectively.
98
Figure 8
Boxplots of OQ-45 Change for Each Group After 3:2:n Matching
Figure 9
Boxplots of Pairwise Differences of OQ-45 Change After 3:2:n Matching
99
Comparing Matching Procedures
Each matching procedure is compared across four elements: number of matched
triplets created, proportion of the overall sample that were involved in the matching,
proportion of bias reduced through matching, and the p-value of the overall comparison
of treatment effectiveness (Table 29).
In terms of creating matches, the caliper matching procedure far outperformed the
other three methods. The caliper matching produced nearly twice as many matches as the
next highest, the 3:2:n matching, and over four times as many matches as the lowest, the
Maximum Treat matching. The Maximum Treat was the only procedure utilizing fewer
data points than the unmatched analysis, which used the total of 101 participants. The
Maximum Treat utilized only 28 matched triplets, or 84 total data points.
For proportion of the overall sample utilized in the matching, three of the four
procedures, the Maximum Treat, the 2:1:n, and the 3:2:n, all performed relatively
equivalently. Each of these three procedures matched between 60 and 70% of the total
sample. The caliper matching again outperformed the other methods, matching 79.2% of
the total sample.
In order to compare the effectiveness of each matching procedure at reducing
covariate imbalance, total proportion of bias reduced was calculated by averaging the
percentage of bias reduced for each covariate within each matching procedure (Table 29).
Two of the procedures, the caliper and the 3:2:n matching, had virtually identical
proportions of total bias reduced through matching, at 60 percent. The 2:1:n matching
reduced 51 percent of the pre-matching bias, and the Maximum Treat matching reduced
100
the least amount of bias, at 42 percent. Balance achieved was also calculated as average
standardized bias for each covariate within each matching procedure (Table 29). The
caliper and the 3:2:n matching again outperformed the other procedures, though the 3:2:n
matching slightly outperformed the caliper matching on this metric.
The caliper matching had the lowest p-value (p < 0.001) among the post-matching
repeated-measures ANOVAs of the outcome variable, which may at least in part relate to
the larger sample size utilized in that analysis. The Maximum Treat matching also
returned a significant repeated-measures ANOVA (p = 0.04) despite utilizing the smallest
sample size of all the analyses. The remaining, two procedures, the 2:1:n and 3:2:n,
returned p-values that were close to but not statistically significant at p<0.05. Therefore,
taken from a strict p-value approach, only two of the procedures lead to the conclusion of
differential treatment effect.
Table 29
Comparing Matching Procedures
Max Treat
Caliper
2:1:n
3:2:n
Triplets
28
126
44
67
% Matched
60.4
79.2
61.4
69.3
Stand. Bias
0.16
0.12
0.14
0.11
PBR
0.42
0.60
0.51
0.60
p-value
0.04
< 0.001
0.06
0.055
101
Chapter Four: Discussion
Overview
This study sought to expand current knowledge on several statistical and clinical
issues, on which the current literature base is as of yet insufficient. Propensity score
analyses are a relatively new set of procedures designed to counteract selection bias in
non-randomized studies (Rosenbaum & Rubin, 1983). These procedures have only
recently been adapted to studies involving more than two treatment groups (e.g. Bryer,
2013). One of the aims of this study was provide a practical application of propensity
score matching across three treatment groups. Also, there is a relatively small amount of
literature regarding the implementation of propensity score analyses with smaller sample
sizes. This study utilized a sample that may well represent the number of individuals
commonly available for analysis from a treatment setting.
In terms of clinical contributions, this study sought to expand the current
understanding related to problem gambling treatment. This was primarily achieved
through the implementation of two forms of therapy that are not represented in the
current problem gambling literature: Solution-Focused Brief Therapy and Time-Limited
Dynamic Psychotherapy. These therapies were compared with Cognitive-Behavioral
Therapy to answer three questions: 1) are the problem gambling clients differentially
selecting these treatments, 2) what factors contribute to which therapy is selected, 3) do
102
these three treatments exhibit comparable short-term effects. Each of these three
questions was formulated to yield substantive contributions to the field of problem
gambling treatment.
Prior to the primary analyses, the missing data were screened for appropriateness
for imputation. Given that this study already involved a relatively small sample size for
propensity score analysis, data imputation was considered a superior strategy to listwise
deletion of cases with missing values given the consequent loss of subjects and power
associated with the removal of cases with missing data. Overall, the proportion of
missing values for each variable was relatively small, ranging from 0 to 6.9%, which
indicated that expectation maximization imputation could be appropriate (Gold &
Bentler, 2000). Also prior to imputation, missing data were assessed for whether it was
random occurrence. Little’s MCAR test was designed to test whether missing values
within a multivariate dataset are occurring completely at random (Little, 1988). The
results of this test suggested that data were missing completely at random and, therefore,
further demonstrated that the missing data were eligible for imputation. Following
imputation, not surprisingly given the relatively small proportion of missing values, postimputation values did not differ substantially from pre-imputation values (Table 2, pp.
71).
Assessing the sample revealed that the individuals included in this study were
relatively representative of treatment-seeking problem gamblers in general. The gender
breakdown of this study was very similar to that found by Potenza et al. (2007) in a study
of gambling helpline callers. The mean NODS score (8.09) indicates that, overall, this
103
sample represents individuals with relatively severe gambling problems. The high
prevalence of co-occurring disorders in this sample is also indicative of what the
literature suggests is typical among this population (e.g., Kessler et al., 2008). The gender
differences observed in this study are also similar to those suggested in the literature. The
female subjects in this study were found to be significantly older, which likely echoes
previous findings that women begin gambling later in life (Ladd & Petry, 2002). The
women in this study also presented with more severe gambling problems, which may
help confirm the notion that there is a social stigma against women gambling that
prevents them from seeking treatment until their problems become more severe
(Ledgerwood et al., 2012). Overall, this sample seems to well reflect what is generally
expected in a representative problem gambling sample, which helps to bolster the
external validity of the conclusions drawn from this study.
Research Question One
The first research question focused on what clinical and demographic factors
might be related to treatment selection. The reasons for this were twofold: identifying
factors related to treatment selection is an important component in propensity score
estimation, and identifying which clients might prefer a particular therapy may have
implications for matching clients with an appropriate treatment (Brooks & Ohsfeldt,
2013). However, this issue became somewhat complicated for this study, because not all
participants were given the opportunity to select their treatment. Of the total sample of
101, 22 participants were randomly assigned to treatment, so it would have been
misleading to assess group differences of the overall sample and represent that as
104
assessing factors related to treatment selection. Consequently, separate analyses were run
on only those individuals that selected a treatment so as to determine what types of
problem gambling clients might be preferring particular therapies.
The current body of literature provided very little guidance in the formulation of
hypotheses regarding treatment selection. In fact, only one study could be located that
addressed the issue of treatment preferences among problem gambling therapy clients.
Najavits (2011) studied the influence of co-occurring PTSD on treatment preferences and
found that preferences did vary between gamblers with and those without PTSD.
Interestingly, they found the group with PTSD ranked treatments as more likely to be
helpful. However, this trend went across the board, as opposed to the comorbid group
preferring only particular treatments more than the PG only group. Though this study did
not provide any specific information regarding how the selection of the three treatments
in this study might be influenced by a co-occurring disorder, it did provide evidence that
treatment preference and selection might be related to comorbid mental health conditions.
This study assessed the impact of four common co-occurring mental health disorders,
including PTSD, on treatment selection and outcome. In addition this study also sought to
explore the role of certain demographic and clinical indicators that have been implicated
in problem gambling treatment outcomes.
For each covariate tested, the groups did not differ significantly at p < 0.05.
However, some interesting overall trends emerged from these comparisons. The SFBT
group had the least severe presentation. This group had the lowest initial OQ-45 scores,
lowest proportion screening positively for three of the four co-occurring disorders
105
(depression, generalized anxiety, and mood disorder), as well as lowest pre-treatment
NODS scores. These characteristics seem to logically fit with the strengths-based nature
of SFBT. The description of SFBT that is provided to the clients mentions that SFBT
‘[works] to identify your strengths and successes...’ (Table 1, pp. 64). It seems that those
with more severe presentations may be less willing to and/or capable of identifying
strengths on which to build.
The TLDP group had the highest ratings on the WAI-S, was the youngest group, and
had the highest proportion of males. The high ratings on the WAI-S seem to fit with
TLDP’s focus on relationships. Clients with higher levels of interpersonal functioning
may be more willing to select TLDP and may also form better working alliances with
their counselor. The lower age found in the TLDP group may reflect a generation gap in
the willingness of younger gamblers to work on the interpersonal aspects of disordered
gambling. The gender imbalance found with TLDP came as a surprise. Given the gender
stigma often experienced by males regarding psychotherapy, it seemed like men would
be less likely to select a therapy that focuses on interpersonal relationships and
expressing emotions (Schomerus, Matschinger, & Angermeyer, 2009). Yet nearly fourfifths of the clients that selected TLDP were male.
The CBT group appeared to have the most severe overall presentations. This group
had the highest initial OQ-45 scores, highest proportion screening positively for three of
the four co-occurring disorders (depression, generalized anxiety, and PTSD), as well as
the highest pre-treatment NODS score. The CBT group also had the lowest scores on the
URICA and WAI-S. These trends for the CBT group may best be understood by
106
considering what CBT is not. More severe clients, as discussed, may be less likely to
select a strengths-based treatment (SFBT). More severe clients may also be such as a
result of poor interpersonal functioning and social support, which may preclude them
from preferring TLDP. This latter aspect may be supported by the low WAI-S scores
associated with the CBT group.
Though none of the single covariates significantly differed between the treatment
selection groups, the evidence suggests that the selection process is not random. In other
words, there are meaningful differences between the groups related to overall clinical
presentations. This would then lead to the necessity to balance groups across these
characteristics associated with the treatment selection process.
Research Question Two
The second aspect of the research focused on whether the problem gamblers, as a
whole, were differentially selecting from among the three treatment options. The results
of the chi-square goodness of fit test returned a marginal p-value (p = 0.13), which
warranted a closer investigation of the specific proportions of the treatment selections.
Among those clients that were allowed to select their treatment (n=79), 32
(40.5%) selected CBT. In contrast, only 18 (22.8%) selected TLDP, and 29 (36.7%)
selected SFBT. The expected frequency for each group was 26.3.
An overall preference for CBT may fit with the traditional view of problem
gambling as a brain disease that has its origins in 12-step approaches to recovery
(Tangenberg, 2005). CBT might then be seen as directly addressing the thought
dysfunction that this paradigm views as foundational to disordered gambling behavior.
107
CBT may also seem the most targeted intervention to individuals who are entering
treatment control one particular set of behaviors. The opposite may be perceived of
TLDP. This sample, which is likely indicative of typical treatment-seeking problem
gamblers, presented for treatment with relatively severe gambling problems. Therefore, it
is reasonable to assume a level of desperation to control their gambling. In the description
of CBT that was provided to the clients there is a reference to addressing ‘problematic
behaviors.’ Conversely, the description of TLDP describes the goals of treatment as
‘including improved relationships, attunement to feelings, and/or a resolution of a
conflict.’ For these individuals seeking to control their gambling behavior, it may be
difficult to perceive the indirect effect that problematic relationships have on the
perpetuation of their gambling problems.
CBT may also be appealing to problem gamblers for not seeming personally
invasive. The high rates of psychological disorders, including PTSD, among this
population coupled with a desire to focus specifically on the gambling behavior may
preclude many individuals for desiring a treatment that is perceived to focus on past
experiences and underlying emotional content (Kessler et al., 2008). This may further
explain the trend away from TLDP and toward selecting CBT.
These overall treatment selection trends may also be better understood in the
context of comparing the preferences of problem gambling and non-problem gambling
therapy clients. Soberay and Faragher (2011) made just such a comparison using the
same three treatment options as presented in this study. Their results suggested that there
was a significant difference in the proportions of these treatments selected by problem
108
gamblers compared to non-problem gambling clients. Specifically, the general therapy
clients actually had a slight preference for the TLDP treatment while it was the least
selected among the problem gambling clients. This may further strengthen the argument
that problem gambling clients are particularly seeking treatment to change their gambling
behaviors without exploring underlying relationships, feelings, and conflicts. However,
these results suggest that there may be sub-groups within the problem gambling
population that may prefer TLDP treatment, such as younger clients. Any comparison of
outcomes between these various treatments would need to control for these client
differences.
Research Question Three
The initial step to the propensity score matching was to ensure that the model
estimating the propensity scores was adequately specified. This study chose a two-step
approach: identify variables that differ between treatment groups and identify variables
that are related to the dependent variable. Cuong (2013) suggested that also including the
latter element would lead to more effective estimation of the propensity scores.
To identify variables to include in the propensity score estimation model the
entire sample was utilized, including both those that selected their form of therapy and
the smaller portion that were assigned to a treatment type. Though these analyses also
included individuals randomly assigned to a treatment, the results were very similar to
those found when assessing only those clients that selected their treatment. This was not
surprising given that the majority of the participants (78.2 %) selected their treatment.
The SFBT group was again differentiated by having the least severe overall clinical
109
presentation, and the TLDP group again tended to be younger and with a larger
proportion of males. Though none of the individual variables differed at p<0.05, multiple
variables differed at the a priori inclusion criteria of p < 0.25. These variables were age,
gender, initial OQ-45 score, positive screen for generalized anxiety (CD-GAD), and
severity of gambling problems (NODS).
Perhaps the most unexpected results of this study were in assessing the
relationships of the covariates with the dependent variable, the change in OQ-45 scores
through treatment. Only one variable, initial OQ-45 score, was found to be related at the
inclusion criteria of p < 0.25. However, previous studies would have suggested that this
relationship would have been in the opposite direction. Dowling (2009) and JimenezMurcia (2007) both suggested that the more severe the problem gambler’s presentation,
the less they improve through treatment. In this study, though, there was a positive
relationship between severity of initial psychosocial distress and improvement through
treatment. The existing literature would have also suggested that other variables,
including gender, readiness to change, and strength of therapeutic alliance would be
associated with treatment outcomes (Crisp et al., 2000; Martin et al., 2000; Petry, 2005).
Consequently, the empirically derived propensity score estimation model used in this
study excluded several variables that the literature would have suggested to include.
From the described approach, the following variables were selected for the
propensity score estimation model: age, gender, initial OQ-45 score, whether the
individual screened positive for generalized anxiety (CD-GAD), and NODS score. These
variables seemed to provide good coverage for the differences in overall presentation
110
observed between groups. OQ-45, the anxiety screen, and the NODS should have
adequately captured the variability in severity between groups. Also, age and gender
should have adequately captured the demographic differences that were primarily
observed with the TLDP group. Therefore, achieving balance on these variables should
go a long way in providing overall balance to the three treatment groups.
Analogous to how adding independent variables to an OLS regression equation
will always increase the proportion of variance explained, the propensity scores would
have been, at least marginally, more accurately estimated had all the available variables
been included in the estimation model. However, the issue of sample size precluded this
approach. Pirracchio (2012) demonstrated that propensity scores could be effectively
estimated with group sizes as small as twenty utilizing four covariates. This ratio of one
covariate per five participants was used as a general guideline for this study.
Coincidentally, the criteria set for identification of variables to include in the propensity
score estimation for this study resulted in a nearly identical ratio of one variable per every
five participants in the smallest treatment group. Further consideration of this issue would
have been warranted had a substantially larger number of covariates been identified as
important for the propensity score estimation through the aforementioned inclusion
process.
Following the propensity score estimation, four matching procedures were
utilized that are available through the Trimatch statistical package: 1) a form of matching
without replacement called Maximum Treat, 2) a with replacement caliper matching, 3) a
2:1:n matching, 4) a 3:2:n matching. The procedures were compared in terms of three
111
values: number of matched triplets created, proportion of overall sample utilized in the
matching procedure, and proportion of bias reduced through matching.
As expected, the caliper matching, since it had no limitations of the reutilization
of cases to make matches so long as they fit within the specified caliper, created the
largest number of matches, 126. This number was nearly twice as many as any of the
other procedures. It also utilized the largest proportion of the overall sample, 79.2%. This
latter aspect may be particularly salient to small group propensity score analyses where
excluding cases may be more problematic than when dealing with larger samples.
Conversely, the Maximum Treat method, since it only reused a participant if in doing so
it would create a match with another participant who would have otherwise been
unmatched, resulted in the smallest amount of matched triplets, 28.
Two procedures, the caliper matching and the 3:2:n matching, outperformed the
others in terms of reducing covariate imbalance. Both of these procedures reduced the
pre-matching bias by 60%. However, the profiles of the individual balances achieved
through these two procedures differed. The 3:2:n matching had two covariates that each
had approximately 90% of the bias reduced through matching. On the other hand, the
3:2:n procedure also had two variables that had less bias reduced than the least balanced
variable from the caliper matching. The caliper matching had a more consistent pattern of
bias reduction across covariates. It may be important to remember that the proportion of
bias reduced statistic that was utilized to compare matching procedures involved
averaging the bias reduced for each individual variable. In essence, this metric assumes
that balance on each covariate is equally important. Yet, as Cuong (2013) indicated, it
112
may be most important to balance across covariates that are also related to the outcome
variable. Therefore, it may be preferable to have a matching procedure that creates good
balance for all covariates, as opposed to a procedure that creates strong balance for some
and leaves others less balanced. For this reason, the caliper matching may have more
utility than the 3:2:n matching in balancing groups across all covariates. However, the
3:2:n matching did marginally outperform the caliper matching in terms of standardized
bias post-matching.
It should also be noted that even though the 3:2:n and caliper matching performed
the other procedures in terms of reducing bias, each of these two procedures had a
covariate that approached statistical significance for differing between groups following
matching. Since no covariates were near statistical significance at p < 0.05 prior to
matching and all variables experienced a reduction in bias through matching, this would
appear to be largely a power issue. These two procedures, in addition to creating the best
balance, created the largest numbers of matched triplets, which contributed to the postmatching analyses having more power to find between-group differences. This would
also suggest that had the treatment groups been larger in size, the pre-matching
differences for at least some covariates might have been statistically significant.
Prior to any matching procedure, a one-way ANOVA was run to test whether an
unadjusted analysis would indicate a significant difference in treatment effect. The results
of this test indicated no difference (p = 0.40). In stark contrast, each of the four matching
procedures returned a test of overall differential treatment effect that was either
significant or marginally significant at p < 0.05. Initially, it may be assumed that the
113
significant results post-matching would be due primarily to increasing the sample size
and statistical power through matching participants several times. However, the
Maximum Treat matching created only 28 triplets, which is a total of 84 participants, yet
it returned a significant differential overall effect. This is likely primarily due to the
Maximum Treat matching resulting in the largest difference in mean OQ-45
improvement between any two groups (a 9.2 unit difference between CBT and TLDP).
Given that this study utilized actual clinical information, not simulated data, it
cannot be said which method best represented the true differences in treatment effects.
Though, given the purpose of propensity score analysis, these findings do lend support to
the practical utility of these approaches. Due to the fact that this study involved
participants who self-selected their treatment, it could be assumed that some level of bias
was being introduced. The utilization of propensity score matching with this data would,
in theory, reduce this bias, thereby resulting in a more accurate estimate of treatment
effects. The fact that each matching procedure returned at least a marginally significant
result, p ≤ 0.06 for each matching procedure, suggests that there are in fact differences in
treatment effects that could only be detected once this selection bias was accounted for.
Specifically, the TLDP treatment may have less utility in short term outcomes,
particularly relative to CBT treatment. This finding would seem to fit with the overall
purpose of these treatments. The TLDP treatment is designed to address relational
difficulties and conflict that likely underlie the gambling problems. It may take an
extended period of time before an individual is able to resolve these conflicts and,
thereby, fully control their gambling behavior. The CBT and SFBT treatments, though
114
ultimately addressing underlying issues, begin with more of a focus on incremental and
immediate changes. Therefore, TLDP may have more utility in a treatment setting where
there is more of an expectation of prolonged treatment engagement, such as a residential
treatment facility, though longer-term effectiveness of this treatment for this population
has yet to be established.
One of the fundamental issues in propensity score analysis is the inclusion of all
relevant covariates in the propensity score estimation (Brooks & Ohsfeldt, 2013). To
address this issue, Rosenbaum (2002) devised a sensitivity analysis for use with
propensity score analyses that would assess the robustness of the findings to the influence
of unobserved covariates. The significant pairwise comparisons for the Maximum Treat
and the caliper matching were found to remain significant at Gamma < 1.4 and Gamma <
1.5, respectively. This means that the significant results would remain robust unless an
unobserved covariate changed the odds of treatment assignment by a factor of 1.4 or 1.5
respectively. In other words, this study could be considered sensitive to a hidden
confounding bias. That, of course, is not to say that the results of this study should be
disregarded. Rather, an element of caution should be exercised. The variables tested for
inclusion into the propensity score estimation model reflect a broad range of clinical
indicators, including several that are commonly associated with the dependent variable
from this study. However, given that there is very little knowledge regarding what factors
underlie treatment selection, particularly among problem gamblers, it is plausible that an
influential covariate may have been omitted.
115
This study reflects the potential for conducting propensity score matching
procedures with relatively small sample sizes. Previous research has suggested propensity
score analyses utilizing matching procedures may not be as effective as other uses of
propensity scores (e.g. as regression weights) with smaller samples (Holmes & Olsen,
2010). However, that study was attempting to balance large covariate differences and
with a relatively restrictive caliper setting (0.05). The results of this study seem to suggest
that with small to moderate pre-matching covariate imbalance, matching procedures may
not result in a problematic loss of participants and, therefore, may be effective in
reducing bias and isolating treatment effects.
One other issue, not explored by this study, was the utilization of bootstrapping
procedures. Bai (2013) found evidence that the utilization of bootstrapping could help
improve the accuracy of propensity score estimation. However, among the matching
procedures implemented in that study, caliper matching was found to benefit the most
from the bootstrapping. Given that this procedure might benefit only one of the four
matching procedures in this study, it was not applied.
The results of these propensity score matching analyses contain implications for
problem gambling treatment, as well as the ways in which the associated data are
analyzed. Given the ubiquity of non-randomized studies, as well as studies including
more than two treatment groups, multi-group propensity score analyses appear to be a
viable option for reducing bias and estimating treatment effects. Specifically, the
Trimatch package for R appears capable of balancing potentially confounding covariates
across three treatment groups, as well as detecting treatment effects where unadjusted
116
analyses may fail to do so. Clinically, the results of these analyses suggest that particular
treatments, particularly CBT, may be more effective in reducing psychosocial distress for
short-term problem gambling interventions.
Limitations
In terms of the propensity score analysis, this study had several limitations.
Perhaps most notably, propensity score analyses, as previously discussed, were designed
to correct for selection biases found in observational studies, particularly as they relate to
individuals selecting their own form of treatment. However, in this study not all the
participants were given the opportunity to choose their form of therapy. Twenty-two of
the participants in this sample were randomly assigned to a form of treatment. This may
have limited the ability to detect systematic between-group differences when using the
complete sample.
Another major limitation in terms of assessing the utility of this multi-group
propensity score matching approach relates to the magnitude of the observed differences
between groups. The between-group differences on the observed covariates were all
below the threshold for statistical significance at p < 0.05, which may indicate one of
several possibilities. There may be unobserved variables that explain treatment selection,
which would result in a misspecification of the propensity score estimation models.
Alternatively, treatment selection may be a complex process that is difficult to explain
through the assessment of any single characteristic. The differences in overall clinical
presentations, as previously discussed, may support this notion. In either case, the lack of
large observed between-group differences may have hindered the ability to demonstrate
117
the effectiveness of these matching procedures. With relatively small group differences,
this study did not have the opportunity to demonstrate that these matching procedures are
effective in reducing relatively large covariate imbalances.
For the treatment aspects of the study, there are also multiple limitations. This
study contained no formal means of validating treatment fidelity. Specific supervision in
each of the treatment types was provided for the counselors, and closed-circuit television
was used to observe treatment sessions live. Both of these measures served to increase
treatment fidelity; however, there was no systematic review of the interventions utilized.
More to this point, the therapists for this study were all graduate students, which
introduced an element of variability in clinical experience, especially as it relates to the
particular forms of treatment. These limitations impact the ability to attribute the
outcomes to a strict view of what each treatment type should consist of.
Other limitations relate to the duration of treatment assessed by this study. Given
that treatment selection occurred as a part of the treatment process, the specific treatments
were not applied for the entirety of the observed timeframe. Often, the treatment selection
occurred during the second session, so each client did not receive all sessions in the form
of treatment they selected. This may have limited the ability to see differences between
treatments. This time frame was selected as a practical representation of the amount of
time that clients are likely to remain in treatment before attrition becomes a major factor.
However, other, particularly residential treatment settings may be more interested in
outcomes from a longer-term treatment.
118
Recommendations for future research
Future research would benefit from building upon the statistical and clinical
ramifications of this study. Propensity score analyses for multiple treatment groups utilize
statistical applications that have only recently been developed (e.g. Bryer, 2013;
Ridgeway, McCaffrey, Morral, Burgette, & Griffen, 2014). Therefore, future research
would benefit from a more in-depth study of these applications and the circumstances
under which they may be more or less effective.
This study also uses a sample that may be of typical size for many treatment
settings. As previously mentioned, Holmes and Olsen (2010) found that matching may
not be the ideal approach with small sample sizes, multiple treatment groups, and large
between-group covariate imbalance. This would then lead to the question of what sets of
circumstances would contribute to a given usage of propensity scores being more or less
effective when dealing with multiple treatment groups. Therefore, future research may
benefit from further investigation of propensity score analyses with smaller sample sizes,
particularly as it relates to multi-group propensity score techniques.
The propensity scores for this study were estimated using empirical criteria for
covariate inclusion. The result was the exclusion of several variables, such as the strength
of the working alliance, which the existing literature has implicated in treatment
outcomes. It may be useful to compare the results of this empirically derived model to
one that uses a theoretical approach to specifying the propensity score estimation model.
The specification of the propensity score estimation model was also limited by the
sample size. Given the sample size requirements of the logistic regression models used to
119
estimate the propensity scores, the number of covariates selected had to be limited.
Future research may want to focus on whether larger samples and the inclusion of more
covariates would result in better matching in a similar study. Also, it may be valuable to
look at alternative propensity score estimation techniques that don’t have the sample size
requirements when using smaller samples, such as the predicted values from a multiple
regression.
This study sought to explore factors related to treatment selection among problem
gambling therapy clients, the results of which suggested that overall clinical presentation
may be related to therapeutic preference. However, this study does not address the
clinical utility of allowing problem gambling clients to select a form of treatment and/or
trying to match a client with a particular form of therapy. Future research may want to
assess whether the outcomes might be different for clients that are assigned to a treatment
compared to those whose treatment is based on individual preference. Particularly if there
is a utility found in tailoring treatment to the individual, future research may also want to
further explore what demographic, clinical, and personological variables may be related
to treatment selection.
The clients at this treatment facility were relatively homogenous in terms of racial
and ethnic composition, which is commonly observed in problem gambling treatment
settings (Volberg, 1994). Future research may want to take a more proactive approach to
recruiting minority clients so as to better understand the role of diversity as it relates to
the treatment of problem gambling.
120
This study also provided evidence that forms of psychotherapy, other than CBT,
may be effective in the treatment of this population but that not all forms may be equally
efficacious, at least not in the short term. Future research may benefit from exploring
whether these forms of therapy may be differentially effective over a longer course of
treatment and/ or differentially effective at periods following the termination of
treatment. Given that this study suggests that a broader range of therapies than previously
researched may be effective in treating this population, additional forms of therapy may
be tested with this population.
121
References
Afifi, T.O., Cox, B.J., & Katz, L.Y. (2007). The associations between health risk
behaviours and suicidal ideation and attempts in a nationally representative
sample of young adolescents. The Canadian Journal of Psychiatry, 52(10), 666674.
Afifi, T.O., Cox, B.J., Martens, P.J., Sareen, J., & Enns, M.W. (2010). The relationship
between problem gambling and mental and physical health correlates among a
nationally representative sample of Canadian women. Canadian Journal of Public
Health, 101(2), 171-175.
Alcoholics Anonymous (1972). A brief guide to alcoholics anonymous. New York:
Author.
Alexander, P., Mazmanian, D., Jamieson, J., & Black, N. (2012). Factors associated with
recent suicide attempts in clients presenting for addiction treatment. International
Journal of Mental Health and Addiction, 10(1), 132-140.
Alvarez-Moya, E. M., Ochoa, C., Jiménez-Murcia, S., Aymamí, M., Gómez-Peña, M.,
Fernández-Aranda, F., Santamaria, T., Moragas, L., Bove, F., & Menchón, J. M.
(2011). Effect of executive functioning, decision-making and self-reported
impulsivity on the treatment outcome of pathologic gambling. Journal Of
Psychiatry & Neuroscience, 36(3), 165-175.
American Psychiatric Association. (2000). Diagnostic and statistical manual of mental
disorders (4th ed., text rev.). Washington, DC: Author.
American Psychiatric Association. (2013). Diagnostic and statistical manual of
122
mental disorders (5th ed.) Washington, DC: Author.
American Psychological Association (2013). Addictions. Retrieved from
http://www.apa.org/topics/addiction/
American Society of Addiction Medicine (2013). Definition of Addiction. Retrieved
from http://www.asam.org/for-the-public/definition-of-addiction
Amodei, N. & Lamb, R.J. (2004). Convergent and concurrent validity of the
Contemplation Ladder and URICA scales. Drug and Alcohol Dependence,
73(3), 301-306.
Agrusa, J., Lema, J., Asage, S., Maples, A., & George, B. (2010). Introduction of casino
gaming in Okinawa, Japan: A case study of challenges and opportunities. Journal
Of Asia Pacific Studies, 1(3), 570-590.
Apodaca, T.R., Longabaugh, R. (2009). Mechanisms of change in motivational
interviewing: a review and preliminary evaluation of the evidence. Addiction,
104(5), 705-715.
Ashley, L. L., & Boehlke, K. K. (2012). Pathological gambling: A general overview.
Journal Of Psychoactive Drugs, 44(1), 27-37.
Austin, P. C. (2011). An introduction to propensity score methods for reducing the
effects of confounding in observational studies. Multivariate Behavioral
Research, 46(3), 399-424.
Baer, L., Jacobs, D., Meszler-Reizes, J., Blais, M., Fava, M., Kessler, R., Magruder, K.,
Murphy, J., Kopans, B., Cukor, P., Leahy, L., O'Laughlen, J. (1999).
Development of a Brief Screening Instruments: The HANDS. Psychotherapy and
123
Psychosomatics, 69(1), 35.
Bai, H. (2013). A Bootstrap Procedure of Propensity Score Estimation. Journal Of
Experimental Education, 81(2), 157-177.
Barmaki, R. (2010). Gambling as a social problem: On the social conditions of gambling
in Canada. Journal Of Youth Studies, 13(1), 47-64.
Barron, J. M., Staten, M. E., & Wilshusen, S. M. (2002). The impact of casino gambling
on personal bankruptcy filing rates. Contemporary Economic Policy, 20(4), 440455.
Barry, D.T., Steinberg, M.A., Wu, R., & Potenza, M.N. (2008). Characteristics of black
and white callers to a gambling helpline. Psychiatric Services, 59(11), 1347-1350.
Barry, D.T., Steinberg, M.A., Wu, R., & Potenza, M.N. (2009). Differences in
characteristics of Asian American and White problem gamblers calling a
gambling helpline. CNS Spectrums, 14(2), 83-91.
Baser, O. (2008). Propensity score matching with multi-level categories: An
application. Official News & Technical Journal of the International Society for
Pharmacoeconomics and Outcomes Research, 14(2), 4-8.
Battersby, M., Tolchard, B., Scurrah, M., & Thomas, L. (2006). Suicide ideation and
Behavior in people with pathological gambling attending treatment service.
International Journal of Mental Health and Addiction, 4(3), 233-246.
Berg, G. D. (2011). An application of kernel-based versus one-to-one propensity score
matching for a nonexperimental causal study: example from a disease
management program evaluation. Applied Economics Letters, 18(5), 439-447.
124
Biddle, D., Hawthorne, G., Forbes, D. & Coman, G. (2005). Problem gambling in
Australian PTSD treatment-seeking veterans. Journal of Traumatic Stress, 28(6),
759-767.
Boardman, B., & Perry, J. J. (2007). Access to gambling and declaring personal
bankruptcy. Journal Of Socio-Economics, 36(5), 789-801.
Boudreau, A., Labrie, R., & Shaffer, H.J. (2009). Towards DSM-V: 'Shadow Syndrome'
symptom patterns among pathological gamblers. Addiction Research & Theory,
17(4), 406-419.
Boutin, C., Dumont, M., Ladouceur, R., & Montecalvo, P. (2003). Excessive gambling
and cognitive therapy: How to address ambivalence. Clinical Case Studies, 2(4),
259.
Bovard, R.S. (2008). Risk behaviors in high school and college sport. Current Sports
Medicine, 7(6), 359-366.
Brooker, I.S., Clara, I.P., & Cox, B.J. (2009). The Canadian Problem Gambling Index:
Factor structure and associations with psychopathology in a nationally
representative Sample. Canadian Journal of Behavioural Science, 41(2), 109-114.
Brooks, J. M., & Ohsfeldt, R. L. (2013). Squeezing the balloon: Propensity scores and
unmeasured covariate balance. Health Services Research, 48(4), 1487-1507.
Brown, S., Dickerson, A., McHardy, J., & Taylor, K. (2012). Gambling and credit: an
individual and household level analysis for the UK. Applied Economics, 44(35),
4639-4650.
Bryer, J.M. (2013). TriMatch: An R package for propensity score matching of non-binary
125
treatments. Retrieved from http://cran.r-project.org/web/packages/TriMatch/
vignettes/TriMatch.pdf
Bu, E.T.H. & Skuttle, A. (2013). After the ban on slot machines in Norway: a new group
of treatment seeking pathological gamblers?. Journal of Gambling Studies, 29(1),
37-50.
Burge, A.N., Pietrzak, R.H., & Petry, N.M. (2006). Pre/early adolescent onset of
gambling and psychosocial problems in treatment-seeking pathological gambling.
Journal of Gambling Studies, 22(3), 263-274.
Busseri, M.A. & Tyler, J.D. (2003). Interchangeability of the Working Alliance Inventory
and Working Alliance Inventory, Short Form. Psychological Assessment, 15(2),
193-197.
Callaghan, R.C., Hathaway, A., Cunningham, J.A., Vettese, L.C., Wyatt, S., & Taylor, L.
(2005). Does stage-of-change predict dropout in a culturally diverse sample of
adolescents admitted to inpatient substance abuse treatment facility? A test of the
Transtheoretical Model. Addictive Behaviors, 30, 1834-1847.
Callan, M.J., Shead, N.W., & Olson, J.M. (2011). Personal relative deprivation, delay
discounting, and gambling. Journal of Personality and Social Psychology, 101(5),
955-973.
Carlbring, P., Degerman, N., Jonsson, J., & Andersson, G. (2012). Internet-based
treatment of pathological gambling with a three-year follow-up. Cognitive
Behaviour Therapy, 41(4), 321-334.
Carlbring, P., Jonsson, J., Josephson, H., & Forsberg, L. (2010). Motivational
126
interviewing versus cognitive behavioral group therapy in the treatment of
problem and pathological gambling: A randomized controlled trial. Cognitive
Behaviour Therapy, 39(2), 92-103.
Caroll, B.J., Davidson, J.R.T. (2000). Screening Scale for DSM-IV GAD. Copyright
2000.
Castrén, S., Pankakoski, M., Tamminen, M., Lipsanen, J., Ladouceur, R., & Lahti, T.
(2013). Internet-based CBT intervention for gamblers in Finland: Experiences
from the field. Scandinavian Journal Of Psychology, 54(3), 230-235.
Chan, S.S., Chiu, H.F., Chen, E.Y., Chan, W.S., Wong, P.W., Chan, C.L., Law, Y.W., &
Yips, P.S. (2009). Population-attributable risk of suicide conferred by axis I
psychiatric diagnoses in a Hong Kong Chinese population. Psychiatric Services,
60(8), 1135-1138.
Cheah, D., Schmitt, G., & Pridmore, S. (2008). Suicide, misappropriation and
Impulsivity. Australian and New Zealand Journal of Psychiatry, 42(6), 544 -546.
Chhabra, D. (2007). Exploring Social Exchange Theory dynamics in Native American
casino settings. UNLV Gaming Research & Review Journal, 11(2), 31-48.
Christensen, D. R., Dowling, N. A., Jackson, A. C., Brown, M., Russo, J., Francis, K. L.,
& Umemoto, A. (2013). A proof of concept for using brief Dialetical Behavior
Therapy as a treatment for problem gambling. Behaviour Change, 30(2), 117-137.
Ciarrocchi, J.W. (2002). Counseling problem gamblers. San Diego, CA: Academic Press.
Cochran, W. G., & Rubin, D. B. (1973). Controlling bias in observational studies: A
review. Sankhyā: The Indian Journal of Statistics, 35(4), 417-446.
127
Compton, W.M., Thomas, Y.F., Stinson, F.S., Grant, B.F. (2007). Prevalence, correlates,
disability, and comorbidity of DSM-IV drug abuse and dependence in the United
States. Archives of General Psychiatry, 64 (5), 566-576.
Connor, K.M., Davidson, J.R.T. (2001). SPRINT: a brief global assessment of
posttraumaticstress disorder. International Clinical Pharmacology, 16(5), 279284.
Cotton, C. A., Cuerden, M. S., & Cook, R. J. (2011). Causal inference in nonrandomized
studies via propensity score methods. Transfusion, 51(12), 2536-2539.
Courtwright, D. T. (2011). Language use disorder: comment on DSM-V’s proposed
‘addiction and related disorders’ and Charles O’brien’s ‘addiction and
dependence in DSM-V’. Addiction, 106(5), 878-879.
Crisp, B. R., Thomas, S. A., Jackson, A. C., Thomason, N., Smith, S., Borrell, J., Ho, W.,
& Holt, T. A. (2000). Sex differences in the treatment needs and outcomes of
problem gamblers. Research on Social Work Practice, 10(2), 229-242.
Crisp, B.R., Jackson, A.C., Thomas, S.A., Thomason, N., Smith, S., Borrell, J., Ho,
W.Y., Holt, T.A. (2001). Is More Better? The relationship between outcomes
achieved by problem gamblers and the number of counselling sessions attended.
Australian Social Work, 54, 83-92.
Crisp, B. R., Thomas, S. A., Jackson, A. C., Smith, S., Borrell, J., Ho, W., & Holt, T.A.,
Thomason, N. (2004). Not the same: A comparison of female and male clients
seeking treatment for problem gambling counseling services. Journal Of
Gambling Studies, 20(3), 283-299.
128
Crits-Christoph, P., Siqueland, L., Blaine, J., Frank, A., Luborsky, L., Onken, L. S., ... &
Beck, A. T. (1999). Psychosocial treatments for cocaine dependence: National
Institute on Drug Abuse collaborative cocaine treatment study. Archives of
General Psychiatry, 56(6), 493-502.
Crockford, D.N. & el-Guebaly, N. (1998). Psychiatric comorbidity in pathological
gambling: A critical Review. Canadian Journal of Psychiatry, 43(1), 43-50.
Cuong, N. (2013). Which covariates should be controlled in propensity score matching?
Evidence from a simulation study. Statistica Neerlandica, 67(2), 169-180.
Daraban, B., & Thies, C. (2011). Estimating the effects of casinos and of lotteries on
bankruptcy: A panel data set approach. Journal Of Gambling Studies, 27(1), 145154.
de Lisle, S. M., Dowling, N. A., & Sabura Allen, J. J. (2011). Mindfulness-based
cognitive therapy for problem gambling. Clinical Case Studies, 10(3), 210-228.
de Ruiter, M. B., Oosterlaan, J., Veltman, D. J., van den Brink, W., & Goudriaan, A. E.
(2012). Similar hyporesponsiveness of the dorsomedial prefrontal cortex in
problem gamblers and heavy smokers during an inhibitory control task. Drug &
Alcohol Dependence, 121(1), 81-89.
de Shazer, S., & Isebaert, L. (2004). The Bruges model: A solution-focused approach to
problem drinking. Journal of Family Psychotherapy, 14(4), 43-52.
Denure, B., Ford, J., Hillyard, P., Moore, C., & Scherer, P. (2006). Effective strategies
for treating those with the illness of pathological gambling. Journal of Groups in
Addiction & Recovery, 1(3-4), 31-57.
129
Derisley, J., & Reynolds, S. (2000). The transtheoretical stages of change as a predictor
of premature termination, attendance, and alliance in psychotherapy, British
Journal of Clinical Psychology, 39(4), 371-382.
Desai, R. A., Maciejewski, P. K., Dausey, D. J., Caldarone, B. J., & Potenza, M. N.
(2004). Health correlates of recreational gambling in older adults. American
Journal Of Psychiatry, 161(9), 1672-1679.
DiClemente, C.C., Schlundt, D., & Gemmell, L. (2004). Readiness and stages of change
in addiction treatment. American Journal on Addictions, 13(2), 103-119.
Dolan, D. A., & Landers, J. (2006). Gambling on an alternative revenue source: The
impact of riverboat gambling on the charitable gambling component of nonprofit
finances. Nonprofit Management and Leadership, 17(1), 5-24.
Dowling, N. (2009). Client characteristics associated with treatment attrition and
outcome in female pathological gambling. Addiction Research & Theory, 17(2),
205-219.
Dowling, N., & Cosic, S. (2011). Client engagement characteristics associated with
problem gambling treatment outcomes. International Journal Of Mental Health &
Addiction, 9(6), 656-671.
Dowling, N., Jackson, A.C., & Thomas, S.A. (2008). Behavioral interventions in the
treatment of pathological gambling: A review of activity scheduling and
desensitization. International Journal of Behavior Consultation and Therapy, 4,
172-187.
Dowling, N., & Smith, D. (2007). Treatment goal selection for female pathological
130
gambling: A comparison of abstinence and controlled gambling. Journal Of
Gambling Studies, 23(3), 335-345.
Dowling, N., Smith, D., & Thomas, T. (2007). A comparison of individual and group
cognitive-behavioural treatment for female pathological gambling. Behaviour
Research and Therapy, 45(9), 2192-2202.
Dowling, N., Smith, D., & Thomas, T. (2009). A preliminary investigation of abstinence
and controlled gambling as self-selected goals of treatment for female
pathological gambling. Journal Of Gambling Studies, 25(2), 201-214.
Dowling, N., Clarke, D., Memery, L., & Corney, T. (2005). Australian apprentices and
gambling. Youth Studies Australia, 24(3), 17-23.
Downs, C. & Woolrych, R. (2010). Gambling and debt: the hidden impacts on family and
work life. Community, Work, and Family, 13(3), 311-328.
Duven, E. E., Unterrainer, J. J., & Wölfling, K. K. (2012). P-110 - Gaming addiction –
disturbed impulse control in internet addicts and pathological gamblers. European
Psychiatry, doi:10.1016/S0924-9338(12)74277-3
Dwyer, S. C., Piquette, N., Buckle, J. L., & McCaslin, E. (2013). Women gamblers write
a voice: Exploring journaling as an effective counseling and research tool.
Journal of Groups in Addiction & Recovery, 8(1), 36-50.
Eadington, W. R. (1999). The economics of casino gambling. Journal Of Economic
Perspectives, 13(3), 173-192.
Eadington, W. R. (2011). After the great recession: The future of casino gaming in
America and Europe. Economic Affairs, 31(1), 27-33.
131
Echeburúa, E., Gómez, M., & Freixa, M. (2011). Cognitive-behavioural treatment of
pathological gambling in individuals with chronic schizophrenia: A pilot study.
Behaviour Research & Therapy, 49(11), 808-814.
El-Guebaly, N., Mudry, T., Zohar, J., Tavares, H., & Potenza, M. N. (2012). Compulsive
features in behavioural addictions: the case of pathological gambling. Addiction,
107(10), 1726-1734.
Elliott, K. (2006). Anthetic inner critic work as a method for relapse prevention:
Identifying and overcoming self-critical beliefs. Alcoholism Treatment Quarterly,
24(3), 109-119.
Engel, R., Rosen, D., Weaver, A., & Soska, T. (2010). Raising the stakes: Assessing the
human service response to the advent of a casino. Journal Of Gambling Studies,
26(4), 611-622.
Fager, M. (2006). How does one measure gambling problems? Reliability and validity
Of the NORC DSM-IV Screen for gambling problems. (Dissertation).
Linkopings University.
Faregh, N., & Derevensky, J. (2011). A comparative latent class analysis of endorsement
profiles of DSM-IV diagnostic criteria for problem gambling among adolescents
from a community and a treatment sample. Addiction Research & Theory, 19(4),
323-333.
Feng, P., Zhou, X. H., Zou, Q. M., Fan, M. Y., & Li, X. S. (2012). Generalized
propensity score for estimating the average treatment effect of multiple
treatments. Statistics in Medicine, 31(7), 681-697.
132
Fenton, L.R., Cecero, J.J., Nich, C., Frankforter, T.L., & Carroll, K.M. (2001).
Perspective is everything: The predictive validity of six working alliance
instruments. Journal of Psychotherapy Practice and Research, 10(4), 262-268.
Ferentzy, P., Skinner, W., & Antze, P. (2006). Rediscovering the Twelve Steps: Recent
Changes in Gamblers Anonymous. Journal Of Groups In Addiction & Recovery,
1(3/4), 59-74.
Flores, P.J. (2007). Group psychotherapy with addicted populations: An integration of
twelve- step and psychodynamic theory (3rd ed.). New York: Haworth.
Fong, T. W. (2009). Addictive disorders. Psychiatric Times, 26(9), 20-25.
Frascella, J., Potenza, M. N., Brown, L. L., & Childress, A. (2010). Shared brain
vulnerabilities open the way for nonsubstance addictions: Carving addiction at a
new joint?. Annals Of The New York Academy Of Sciences, 1187(1), 294-315.
Freidenberg, B. M., Blanchard, E. B., Wulfert, E., & Malta, L. S. (2002). Changes in
physiological arousal to gambling cues among participants in motivationally
enhanced cognitive-behavior therapy for pathological gambling: A preliminary
study. Applied Psychophysiology & Biofeedback, 27(4), 251-260.
Freund, E. A., & Morris, I. L. (2006). Gambling and income inequality in the states.
Policy Studies Journal, 34(2), 265-276.
Froberg, F., Hallqvist, J., & Tengstrom, A. (2012). Psychosocial health and gambling
problems among men and women aged 16-24 years in the Swedish National
Public Health Survey. The European Journal of Public Health, 23(3), 427-433.
Gamblers Anonymous (2013). Questions and answers about gamblers anonymous.
133
Retrieved from http://www.gamblersanonymous.org/ga/content/
questions-answers- about-gamblers-anonymous
Garrett, T. A., & Nichols, M. W. (2008). Do casinos export bankruptcy?. Journal Of
Socio-Economics, 37(4), 1481-1494.
George, W., Joe, D., Simpson, D., Broome, K.M. (1998). Effects of readiness for drug
abuse treatment on client retention and assessment process. Addiction, 93(8),
1177-1190.
Gerstein, D., Murphy, S., Toce, M., Hoffman, J., Palmer, A., Johnson, R., Larison, C.,
Chuchro, L., Bard, A., Engelman, L., Hill, M.A., Buie, T., Volberg, R., Harwood,
H., Tucker, A., Christiansen, E., Cummings, W. & Sinclair, S. (1999). Gambling
impact and behavior study: Report to the national gambling impact study
commission: Chicago, IL: National Opinion Research Center.
Gill, T., Dal Grande, E., & Taylor, A.W. (2006). Factors associated with gamblers: A
population-based cross-sectional study of South Australian adults. Journal of
Gambling Studies, 22(2), 143-164.
Glenn, M. K., Díaz, S. R., & Hawley, C. (2009). Preparing addition specialists to include
case management and vocational rehabilitation services in the treatment model for
problem gamblers. Journal of Teaching in the Addictions, 8(1-2), 27-41.
Gold, M. S., & Bentler, P. M. (2000). Treatments of missing data: A Monte Carlo
comparison of RBHDI, iterative stochastic regression imputation, and
expectation-maximization. Structural Equation Modeling, 7(3), 319-355.
Golinelli, D., Ridgeway, G., Rhoades, H., Tucker, J., & Wenzel, S. (2012). Bias and
134
variance trade-offs when combining propensity score weighting and regression:
with an application to HIV status and homeless men. Health Services & Outcomes
Research Methodology, 12(2/3), 104-118
Gomes, K., & Pascual-Leone, A. (2009). Primed for change: Facilitating factors in
problem gambling treatment. Journal Of Gambling Studies, 25(1), 1-17.
Gomez-Pena, M., Penelo, E., Granero, R., Fernandez-Aranda, F., Alvarez-Moya, E.,
Santamaria, J.J., Moragas, L., Aymami, M.N., Gunnard, K., Menchon, J.M., &
Jimenez-Murcia, S. (2012). Correlates of motivation to change in pathological
gamblers completing cognitive-behavioral group therapy. Journal of Clinical
Psychology, 68(7), 732-744.
González-Ibáñez, Á., Rosel, P., & Moreno, I. (2005). Evaluation and treatment of
pathological gambling. Journal Of Gambling Studies, 21(1), 35-42.
Gonzalez-Ibanez, A., Aymami, M. N., Jiménez, S., Doménech, J. M., Granero, R., &
Lourido-Ferreira, M. R. (2003). Assessment of pathological gamblers who use
slot machines. Psychological reports, 93(3), 707-716.
Gooding, P., & Tarrier, N. (2009). A systematic review and meta-analysis of cognitivebehavioural interventions to reduce problem gambling: Hedging our bets?.
Behaviour Research & Therapy, 47(7), 592-607.
Gossop, M., Stewart, D., & Marsden, J. (2007). Readiness for change and drug use
outcomes after treatment. Addiction, 102(2), 301-308.
Grall-Bronnec, M., Wainstein, L., Feuillet, F., Bouju, G., Rocher, B., Venisse, J.L., &
135
Sebille-Rivain, V. (2012). Clinical profiles as a function of level and type of
impulsivity in a sample group of at-risk and pathological gamblers seeking
treatment. Journal of Gambling Studies, 28(2), 239-252.
Grant, J. E., Kim, S., Odlaug, B. L., Buchanan, S. N., & Potenza, M. N. (2009). Lateonset pathological gambling: Clinical correlates and gender differences. Journal
Of Psychiatric Research, 43(4), 380-387.
Grant, J. E., Donahue, C. B., Odlaug, B. L., & Kim, S. W. (2011). A 6-month follow-up
of imaginal desensitization plus motivational interviewing in the treatment of
pathological gambling. Annals of clinical psychiatry: official journal of the
American Academy of Clinical Psychiatrists, 23(1), 3-10.
Gregory, R. J., Chlebowski, S., Kang, D., Remen, A. L., Soderberg, M. G., Stepkovitch,
J., & Virk, S. (2008). A controlled trial of psychodynamic psychotherapy for cooccurring borderline personality disorder and alcohol use disorder.
Psychotherapy: Theory, Research, Practice, Training, 45(1), 28-41.
Griffiths, M., & Cooper, G. (2003). Online therapy: Implications for problem gamblers
and clinicians. British Journal of Guidance and Counselling, 31(1), 113-135.
Guercio, J. M., Johnson, T., & Dixon, M. R. (2012). Behavioral treatment for
pathological gambling in persons with acquired brain injury. Journal Of Applied
Behavior Analysis, 45(3), 485-495.
Guo, S. & Fraser, M.W. (2010). Propensity score analysis: Statistical methods and
applications. Thousand Oaks, CA: Sage.
Gupta, R., & Derevensky, J. (1998). Adolescent gambling behavior: A prevalence study
136
and examination of the correlates associated with excessive gambling. Journal of
Gambling Studies, 14(4), 319-345.
Hatcher, R.L., Gillaspy, J.A. (2006). Development and validation of a revised short
version of the working alliance inventory. Psychotherapy Research, 16(1), 12-25.
Hayes, S. C., Luoma, J. B., Bond, F. W., Masuda, A., & Lillis, J. (2006). Acceptance and
commitment therapy: Model, processes and outcomes. Behaviour research and
therapy, 44(1), 1-25.
Heckman, J. J., Ichimura, H., Smith, J., & Todd, P. (1996). Sources of selection bias in
evaluating social programs: An interpretation of conventional measures and
evidence on the effectiveness of matching as a program evaluation method.
Proceedings of the National Academy of Sciences, 93(23), 13416-13420.
Heckman, J. J., Ichimura, H., & Todd, P. (1998). Matching as an econometric evaluation
estimator. The Review of Economic Studies, 65(2), 261-294.
Henderson, M.J., Saules, K.K., Galen, L.W. (2004). The predictive validity of the
University of Rhode Island Change Assessment questionnaire in a heroinaddicted polysubstance abuse sample. Psychology of Addictive Behaviors, 18(2),
106-112.
Hirschfeld, R. (2010). The mood disorder questionnaire: Its impact on the field
[corrected]. Depression and Anxiety, 27(7), 627-630.
Hirschfeld, R.., Williams, J., Spitzer, R., Calabrese, J., Flynn, L., Keck, P., Lewis, L.,
McElroy, S., Post, R., Rapport, D., Russell, J., Sachs, G., Zajecka, J. (2000).
Development and validation of a screening instrument for Bipolar Spectrum
137
Disorder: The Mood Disorder Questionnaire. American Journal of Psychiatry,
157, 1873-1875.
Hing, N., & Nuske, E. (2011). Assisting problem gamblers in the gaming venue: A
counselor venue. International Journal Of Mental Health & Addiction, 9(6), 696708.
Hodgins, D. C., & el-Guebaly, N. (2000). Natural and treatment-assisted recovery from
gambling problems: a comparison of resolved and active gamblers. Addiction,
95(5), 777-789.
Hodgins, D. C. (2004). Using the NORC DSM Screen for Gambling Problems as an
outcome measure for pathological gambling: psychometric evaluation. Addictive
Behaviors, 29(8), 1685-1690.
Hodgins, D.C. & el-Guebaly, N. (2004). Retrospective and prospective reports of
Precipitants to relapse in pathological gambling. Journal of Consulting and
Clinical Psychology, 72(1), 72-80.
Hodgins, D.C., Mansley, C., & Thygesen, K. (2006). Risk factors for suicide ideation and
attempts among pathological gamblers, The American Journal on Addictions,
15(4), 303-310.
Hoffmann, J. P. (2000). Religion and problem gambling in the U.S. Review Of Religious
Research, 41(4), 488.
Holden, C. (2010). Behavioral addictions debut in proposed DSM-V. Science 19,
327(5968), 935.
Hollander, M., Wolfe, D. A., & Chicken, E. (2013). Nonparametric statistical methods.
138
Hoboken, NJ: John Wiley & Sons.
Holmes, W., & Olsen, L. (2010). Using propensity scores with small samples. In annual
meetings of the American Evaluation Association. San Antonio, Texas.
Holmes, W.M. (2014). Using propensity scores in quasi-experimental designs. Thousand
Oaks, CA: Sage.
Horvath, A.O. & Greenberg, L.S. (1989). Development and validation of the Working
Alliance Inventory. Journal of Counseling Psychology, 36(2), 223-233.
Hounslow, V., Smith, D., Battersby, M., & Morefield, K. (2011). Predictors of problem
gambling severity in treatment seeking gamblers. International Journal Of Mental
Health & Addiction, 9(6), 682-695.
Hsu, C. H. (2000). Residents’ support for legalized gaming and perceived impacts of
riverboat casinos: Changes in five years. Journal of Travel Research, 38(4), 390395.
Imbens, G. W. (2000). The role of the propensity score in estimating dose-response
functions. Biometrika, 87(3), 706-710.
Ingle, P. J., Marotta, J., McMillan, G., & Wisdom, J. P. (2008). Significant others and
gambling treatment outcomes. Journal of Gambling Studies, 24(3), 381-392.
Jackson, P. (2004). Motivational enhancement and group therapy. In Reading.B.,
Weegmann, M. (Eds.). Group Psychotherapy and addiction. (59-80).
Philadelphia, PA: Whurr.
Jackson, A. C., Dowling, N., Thomas, S. A., & Holt, T. A. (2008). Treatment careers in
139
problem gambling: Factors associated with first treatment and treatment re-entry.
Addiction Research & Theory, 16(6), 618-632.
Jamieson, J., Mazmanian, D., Penney, A., Black, N., & Nguyen, A. (2011). When
problem Gambling is the primary reason for seeking addiction treatment.
International Journal Of Mental Health & Addiction, 9(2), 180-192.
Jimenez-Murcia, S., Alvarez-Moya, E., Granero, R., Aymami, M.N., Gomez-Pena, M.,
Jaurrieta, N., Sans, B., Rodriguez-Marti, J., Vallejo, J.(2007). Cognitive
behavioral group treatment for pathological gambling: analysis of effectiveness
and predictors of therapy outcome, Psychotherapy Research, 17(5), 544-552.
Jimenez-Murcia, S., Aymamí, N., Gómez-Peña, M., Santamaría, J. J., Álvarez-Moya, E.,
Fernández-Aranda, F., Granero, R., Penelo, E., Bueno, B., Moragas, L., Gunnard,
K., & Menchón, J. M. (2012). Does exposure and response prevention improve
the results of group cognitive-behavioural therapy for male slot machine
pathological gamblers?. British Journal Of Clinical Psychology, 51(1), 54-71.
Jiménez-Murcia, S., Bove, F. I., Israel, M., Steiger, H., Fernández-Aranda, F., ÁlvarezMoya, E., Granero, R., Penelo, E., Verge, B., Aymami, N., Santamaria, J.J.,
Gomez-Pena, M., Moragas, L., Savvidou, L.G., & Menchón, J. M. (2012).
Cognitive-behavioral therapy for pathological gambling in Parkinson’s disease: A
pilot controlled study. European Addiction Research, 18(6), 265-274.
Jiménez-Murcia, S., Álvarez-Moya, E. M., Stinchfield, R., Fernández-Aranda, F.,
140
Granero, R., Aymamí, N., Gomez-Pena, M., Jaurrieta, N., Bove, F., & Menchón,
J. (2010). Age of Onset in Pathological Gambling: Clinical, Therapeutic and
Personality Correlates. Journal Of Gambling Studies, 26(2), 235-248.
Jacques, C., Ladouceur, R., & Ferland, F. (2000). Impact of availability of gambling: A
longitudinal study. Canadian Journal Of Psychiatry, 45(9), 810-815.
Johnson, T. E., & Dixon, M. R. (2009). Altering respsonse chains in pathological
gamblers using a response-cost procedure. Journal Of Applied Behavior Analysis,
42(3), 735-740.
Kausch, O. (2003). Suicide attempts among veterans seeking treatment for pathological
Gambling. Journal of Clinical Psychiatry, 64(9), 1031-1038.
Kausch, O. (2004). Pathological gambling among elderly veterans. Journal of Geriatric
Psychiatry and Neurology, 17(1), 13-19.
Kausch, O., Rugle, L.,& Rowland, D.Y. (2006). Lifetime histories of trauma among
pathological gamblers. The American Journal on Addictions, 15(1), 35-43.
Kennedy, C.H., Cook, J.H., Poole, D.R., Brunson, C.L., &Jones, D.E. (2005). Review of
the first year of an overseas military gambling treatment program. Military
Medicine, 170(8), 683-687.
Kessler, R.C., Berglund, P., Demler, O., Jin, R., Merikangas, K.R., Walters, E.E. (2005).
Lifetime prevalence and age-of-onset distributions of DSM-IV disorders in the
National Comorbidity Survey Replication. Archives of General Psychiatry, 62(6),
593-602.
Kessler, R.C., Hwang, I., LaBrie, R., Petukhova, M., Sampson, N.A., Winters, K.C., &
141
Shaffer, H.J. (2008). The prevalence and correlates of DSM-IV pathological
gambling in the National Comorbidity Study Replication. Psychological
Medicine, 38(9), 1351-1360.
Khantzian, E. J. (2012). Reflections on treating addictive disorders: A psychodynamic
perspective. American Journal On Addictions, 21(3), 274-279.
Khantzian, E., & Weegmann, M. (2009). Questions of substance: Psychodynamic
reflections on addictive vulnerability and treatment. Psychodynamic Practice,
15(4), 365-380.
Kim, S.W., Grant, J.E., Eckert, E.D., Faris, P.L., & Hartman, B.K. (2006). Pathological
gambling and mood disorders: Clinical associations and treatment implications.
Journal of Affective Disorders, 92(1), 109-116.
Knezevic, B., & Ledgerwood, D. M. (2012). Gambling severity, impulsivity, and
psychopathology: Comparison of treatment- and community- recruited
pathological gamblers. American Journal On Addictions, 21(6), 508-515.
Koehler, S., Ovadia-Caro, S., van der Meer, E., Villringer, A., Heinz, A., RomanczukSeiferth, N., & Margulies, D. S. (2013). Increased functional connectivity
between prefrontal cortex and reward system in pathological gambling. Plos
ONE, 8(12), 1-13.
Korman, L., Collins, J., Littman-Sharp, N., Skinner, W., McMain, S., & Mercado, V.
(2008). Randomized control trial of an integrated therapy for comorbid anger and
gambling. Psychotherapy Research, 18(4), 454-465.
Kranzler, H.R. & Li, T.K. (2008). What is addiction? Alcohol Research and Health,
142
31(2), 93-95.
Kuss, O. (2013). The z-difference can be used to measure covariate balance in matched
propensity score analyses. Journal Of Clinical Epidemiology, 66(11), 1302-1307.
LaBrie, R., & Shaffer, H. (2007). Gambling with adolescent health. Journal of
Adolescent Health, 40(5), 387-389.
Ladd, G. T., & Petry, N. M. (2002). Gender differences among pathological gamblers
seeking treatment. Experimental And Clinical Psychopharmacology, 10(3), 302309.
Ladouceur, R., Sylvain, C., Boutin, C., & Doucet, C., (2002). Understanding and treating
the pathological gambler. New York: Wiley.
Ladouceur, R. R., Sylvain, C. C., Boutin, C. C., Lachance, S. S., Doucet, C. C., &
Leblond, J. J. (2003). Group therapy for pathological gamblers: a cognitive
approach. Behaviour Research & Therapy, 41(5), 587-596.
Ladouceur, R. (2004). Gambling: The hidden addiction. Canadian Journal of Psychiatry,
49(8), 501-503.
Lahti, T., Halme, J., Pankakoski, M., Sinclair, D., & Alho, H. (2013). Characteristics of
treatment seeking Finnish pathological gamblers: Baseline data from a treatment
study. International Journal of Mental Health and Addiction, 11(3), 307-314.
Lambert, M.J.& Anderson, E.M. (1996). Assessment for the time-limited
psychotherapies. In Dickstein, L.J., Riba, M.B., Oldham, J.M. (Eds.) Review of
Psychiatry. (pp.23-42). Washington, DC: American Psychiatric Press.
Lambert, M.J., Burlingame, G.M., Umphress, V., Hansen, N.B., Vermeersch, D.A.,
143
Clouse, G.C., Yanchar, S.C., (1996). The reliability and validity of the Outcome
Questionnaire. Clinical Psychology & Psychotherapy, 3(4), 249-258.
Langhinrichsen-Rohling, J., Rohde, P., Seeley, J.R., & Rohling, M.L. (2004). Individual,
family, and peer correlates of adolescent gambling. Journal of Gambling Studies,
20(1), 23-46.
Larimer, M. E., Neighbors, C., Lostutter, T. W., Whiteside, U., Cronce, J. M., Kaysen,
D., & Walker, D. D. (2012). Brief motivational feedback and cognitive behavioral
interventions for prevention of disordered gambling: a randomized clinical trial.
Addiction, 107(6), 1148-1158.
Ledgerwood, D. M., Arfken, C. L., Wiedemann, A., Bates, K. E., Holmes, D., & Jones,
L. (2013). Who goes to treatment? Predictors of treatment initiation among
gambling help-line callers. American Journal On Addictions, 22(1), 33-38.
Ledgerwood, D.M. & Petry, N.M. (2004). Gambling and suicidality in treatment-seeking
pathological gamblers. The Journal of Nervous and Mental Disease, 192(10),
711-714.
Ledgerwood, D.M., Steinberg, M.A., Wu, R., & Potenza, M.N. (2005). Self-reported
gambling-related suicidality among gambling helpline callers. Psychology of
Addictive Behaviors, 19(2), 175-183.
Ledgerwood, D. M., Wiedemann, A. A., Moore, J., & Arfken, C. L. (2012). Clinical
characteristics and treatment readiness of male and female problem gamblers
calling a state gambling helpline. Addiction Research & Theory, 20(2), 162-171.
Lee, B. K. (2009). Congruence Couple Therapy for pathological gambling. International
144
Journal Of Mental Health & Addiction, 7(1), 45-67.
Lee, K.M., Guo, S., Manning V., Thane, K., & Wong, K.E. (2011). Are the demographic
and clinical features of pathological gamblers seeking treatment in Singapore
changing?. Singapore Medical Journal, 52(6), 428-431.
Lehrer, S., & Kordas, G. (2013). Matching using semiparametric propensity scores.
Empirical Economics, 44(1), 13-45.
Leichsenring, F., & Leibing, E. (2003). The effectiveness of psychodynamic therapy and
cognitive behavior therapy in the treatment of personality disorders: a metaanalysis. American journal of psychiatry, 160(7), 1223-1232.
Lesieur, H. R., & Custer, R. L. (1984). Pathological gambling: Roots, phases, and
treatment. The Annals of the American Academy of Political and Social Science,
474, 146-156.
Leyton-Armakan, J., Lawrence, E., Deutsch, N., Willimas, J.L., & Henneberger, A.
(2012). Effective youth mentors: the relationship between initial characteristics of
college women mentors and mentee satisfaction and outcome. Journal of
Community Psychology, 40(8), 906-920.
Li, J. (2007). Women's ways of gambling and gender-specific research. Sociological
Inquiry, 77(4), 626-636.
Little, R. J. (1988). A test of missing completely at random for multivariate data with
missing values. Journal of the American Statistical Association, 83(404), 11981202.
Liu, Y., & Brown, S. D. (2013). Comparison of five iterative imputation methods for
145
multivariate classification. Chemometrics and Intelligent Laboratory Systems,
120, 106-115.
Lincourt, P. Kuettel, T.J., Bombardier, C.H. (2002). Motivational interviewing in a group
setting with mandated clients: A pilot study. Addictive Behaviors, 27, 381-391.
Lindberg, A., Fernie, B., & Spada, M. (2011). Metacognitions in problem gambling.
Journal Of Gambling Studies, 27(1), 73-81.
Little, R. J. (1988). A test of missing completely at random for multivariate data with
missing values. Journal of the American Statistical Association, 83(404), 11981202.
Locke, G. W., Shilkret, R., Everett, J. E., & Petry, N. M. (2013). Interpersonal guilt in
college student pathological gamblers. The American Journal of Drug and
Alcohol Abuse, 39(1), 28-32.
Long, J., Johnson, C., & Oakley, J. (2011). Changes created by the introduction of
legalized gaming: A review of the Tunica County, Mississippi experience.
Journal of Management Policy and Practice, 12(4), 156-163.
Lorains, F.K., Cowlishaw, S., & Thomas, S.A. (2011). Prevalence of comorbid disorders
in problem and pathological gambling: systematic review and meta-analysis of
population surveys. Addiction, 106(3), 490-498.
Lundahl, B. W., Kunz, C., Brownell, C., Tollefson, D., & Burke, B. L. (2010). A metaanalysis of Motivational Interviewing: Twenty-five years of empirical studies.
Research On Social Work Practice, 20(2), 137-160.
Lunt, M. (2014). Selecting an appropriate caliper can be essential for achieving good
146
balance with propensity score matching. American Journal Of Epidemiology,
179(2), 226-235.
Maccallum, F. & Blaszczynski, A. (2003). Pathological gambling and suicidality: An
Analysis of severity and lethality. Suicide and Life-Threatening Behavior, 33(1),
88-98.
MacCoun, R.J. (2009). Toward a psychology of harm reduction. In Marlatt, G.A.,
Witkiewitz, K. (Eds.). Addictive behaviors: new readings on etiology, prevention,
and treatment. (137-158). Washington, D.C.: APA.
MacDonald, M., McMullan, J. L., & Perrier, D. C. (2004). Gambling households in
Canada. Journal of Gambling Studies, 20(3), 187-236.
Magoon, M. E., Gupta, R., & Derevensky, J. (2007). Gambling among youth in detention
centers. Journal For Juvenile Justice Services, 21(1/2), 17-30.
Marceaux, J., & Melville, C. (2011). Twelve-step facilitated versus mapping-enhanced
cognitive-behavioral therapy for pathological gambling: A controlled study.
Journal Of Gambling Studies, 27(1), 171-190.
Marlatt, G.A., Witkiewitz, K. (2002). Harm reduction approaches to alcohol use: Health
promotion, prevention, and treatment. Addictive Behaviors, 27, 867-886.
Martin, D. J., Garske, J. P., & Davis, M. K. (2000). Relation of the therapeutic alliance
with outcome and other variables: a meta-analytic review. Journal of consulting
and clinical psychology, 68(3), 438-450.
Martin, G., Macdonald, S., & Ishiguro, S. (2013). The role of psychosocial characteristics
147
in criminal convictions among cocaine and gambling clients in treatment.
International Journal Of Mental Health & Addiction, 11(2), 162-171.
Martins, S.S., Tavares, H., da Silva Lobo, D.S., Galetti, A.M., & Gentil, V. (2004).
Pathological gambling, gender, and risk-taking behaviors. Addictive Behaviors,
29(6), 1231-1235.
Mathias, A.C., Vargens, R.W., Kessler, F.H., & Cruz, M.S. (2009). Differences in
Addiction severity between social and probable pathological gamblers among
substance abusers in treatment in Rio de Janeiro. International Journal of Mental
Health and Addiction, 7(1), 239-249.
McCaffrey, D. F., Griffin, B. A., Almirall, D., Slaughter, M. E., Ramchand, R., &
Burgette, L. F. (2013). A tutorial on propensity score estimation for multiple
treatments using generalized boosted models. Statistics in medicine, 32(19), 33883414.
McComb, J. L., Lee, B. K., & Sprenkle, D. H. (2009). Conceptualizing and treating
problem gambling as a family issue. Journal Of Marital & Family Therapy,
35(4), 415-431.
McConnaughy, E.A., Prochaska, J.O., Velicer, W.F. (1983). Stages of change in
psychotherapy: Measurement and sample profiles. Psychotherapy: Theory,
Research, & Practice, 20, 368-375.
McKeganey, N., Morris, Z., Neale, J., Robertson, M. (2004). What are drug users looking
for when they contact drug services: abstinence or harm reduction?. Drugs:
Education, Prevention, and Policy, 11(5),423-435.
148
McKnight, P.E., McKnight, K.M., Sidani, S., & Figueredo, A.J. (2007). Missing data: A
gentle introduction. New York: Guilford.
Meier, P.S., Donmall, M.C., Barrowclough, McElduff, P. C., & Heller, R.F. (2005).
Predicting the early therapeutic alliance in the treatment of drug misuse.
Addiction, 100(4), 500-511.
Meier, P.S., Donmall, M.C., McElduff, P., Barrowclough, C., & Heller, R.F. (2006). The
role of the early therapeutic alliance in predicting drug treatment dropout. Drug
and Alcohol Dependence, 83, 57-64.
Meltzer, H., Bebbington, P., Brugha, T., Farrell, M., & Jenkins, R. (2013). The
relationship between personal debt and specific common mental disorders. The
European Journal of Public Health, 23(1), 108-113.
Meltzer, H., Bebbington, P., Brugha, T., Jenkins, R., McManus, S., & Dennis, M.S.
(2011). Personal debt and suicidal ideation. Psychological Medicine, 41(4), 771778.
Meyer, G. & Stadler, M.A. (1999). Criminal behavior associated with pathological
Gambling. Journal of Gambling Studies, 15(1), 29-43.
Miller, W.R. & Rollnick, S. (2002). Motivational Interviewing: Preparing People for
Change. (2nd ed.). Guilford Press: New York
Mitzner, G.B., Whelan, J.P., & Meyers, A.W.(2011). Comments from the trenches:
Proposed changes to the DSM-V classification of pathological gambling. Journal
of Gambling Studies, 27, 517-521.
Molde, H., Hystad, S. W., Pallesen, S., Myrseth, H., & Lund, I. (2010). Evaluating
149
lifetime NODS using Rasch modelling. International Gambling Studies, 10(2),
189-202.
Najavits, L. (2010). Treatment utilization of pathological gamblers with and without
PTSD. Journal Of Gambling Studies, 26(4), 583-592.
Najavits, L. (2011). Treatments for PTSD and Pathological Gambling: What Do Patients
Want?. Journal Of Gambling Studies, 27(2), 229-241.
Najavits, L.M., Meyer, T., Johnson, K.M., & Korn, D. (2011). Pathological gambling and
posttraumatic stress disorder: a study of the co-morbidity versus each alone.
Journal of Gambling Studies, 27(4), 663-683.
Namrata, R., & Oei, T. S. (2009). Factors associated with the severity of gambling
problems in a community gambling treatment agency. International Journal Of
Mental Health & Addiction, 7(1), 124-137.
Nastally, B. L., & Dixon, M. R. (2012). The effect of a brief acceptance and commitment
therapyintervention on the near-miss effect in problem gamblers. Psychological
Record, 62(4), 677-690.
National Association of Alcoholism and Drug Abuse Counselors (2013). About
NAADAC. Retrieved from http://www.naadac.org/about
Nelle, A. (2005). Solution and resource-oriented addiction treatment with choices of
abstinence or controlled drinking. Journal Of Family Psychotherapy, 16(3), 5768.
Nelson, S. E., Kleschinsky, J. H., LaBrie, R. A., Kaplan, S., & Shaffer, H. J. (2010). One
150
decade of self exclusion: Missouri casino self-excluders four to ten years after
enrollment. Journal Of Gambling Studies, 26(1), 129-144.
Nelson, S.E., Laplante, D.A., Labrie, R.A., & Shaffer, H.J. (2006). The proxy effect:
Gender and gambling problem trajectories of Iowa gambling treatment program
participants. Journal of Gambling Studies, 22(2), 221-240.
Newman, S.C. & Thompson, A.H. (2007). The association between pathological
gambling and attempted suicide: findings from a national survey in Canada. The
Canadian Journal of Psychiatry, 52(9), 605-612.
Nichols, M. W., Stitt, B.G., & Giacopassi, D. (2000). Casino gambling and bankruptcy in
new United States casino jurisdictions. Journal of Socio-economics, 29(3), 247261.
Nichols, M.W., Stitt, B.G., & Giacopassi, D. (2004). Changes in suicide and divorce in
new casino jurisdictions. Journal of Gambling Studies, 20(4), 391-404.
Norcross, J.C., Krebs, P.M., Prochaska, J.O. (2011). Stages of change, Journal of
Clinical Psychology, 67(2), 143-154.
North American Association of States and Provincial Lotteries (2014). NASPL- Lottery
Sales and Transfers. Retrieved from: http://www.naspl.org/index.cfm?
fuseaction= content&menuid=17&pageid=1025
Nower, L., Gupta, R., Blaszczynski, A., & Derevensky, J. (2004). Suicidality and
depression among youth gamblers: A preliminary examination of three studies.
International Gambling Studies, 4(1), 69-80.
Nower, L. & Blaszczynski, A. (2006). Characteristics and gender differences among self151
excluded casino problem gamblers: Missouri data. Journal of Gambling Studies,
22(1), 81-99.
Nower, L. & Blaszczynski, A. (2008). Characteristics of problem gamblers 56 years of
age or older: a statewide study of casino self-excluders. Psychology and Aging,
23(3), 577-584.
Odlaug, B. L., Marsh, P. J., Kim, S., & Grant, J. E. (2011). Strategic vs nonstrategic
gambling: Characteristics of pathological gamblers based on gambling preference.
Annals Of Clinical Psychiatry, 23(2), 105-112.
O’Brien, C.P., Volkow, N., & Li, T.K. (2006). What’s in a word? Addiction versus
dependence in the DSM-V. American Journal of Psychiatry, 163(5), 764-765.
O'Brien, C. (2011). Addiction and dependence in DSM-V. Addiction, 106(5), 866-867.
O'Brien, C. (2011). Depression, cause or consequence of pathological gambling and its
implications for treatment. Counselling Psychology Review, 26(1), 53-61.
Oakes, J., Gardiner, P., McLaughlin, K., & Battersby, M. (2012). A pilot group cognitive
behavioural therapy program for problem gamblers in a rural Australian setting.
International Journal of Mental Health and Addiction, 10(4), 490-500.
Oakes, J., Pols, R., Battersby, M., Lawn, S., Pulvirenti, M., & Smith, D. (2012a). A focus
group study of predictors of relapse in electronic gaming machine problem
gambling, part 2: Factors that ‘pull’ the gambler away from relapse. Journal of
Gambling Studies, 28(3), 465-479.
Oakes, J., Pols, R., Battersby, M., Lawn, S., Pulvirenti, M., & Smith, D. (2012b). A focus
group study of predictors of relapse in electronic gaming machine problem
152
gambling, part 1: Factors that ‘push’ towards relapse. Journal of Gambling
Studies, 28(3), 451-464.
Oei, T., & Raylu, N. (2009). The relationship between cultural variables and gambling
behavior among Chinese residing in Austrailia. Journal Of Gambling Studies,
25(4), 433-445.
Oei, T., Raylu, N., & Casey, L. (2010). Effectiveness of group and individual formats
of a combined motivational interviewing and cognitive behavioral treatment
program for problem gambling: A randomized controlled trial. Behavioural and
Cognitive Psychotherapy, 38, 233-238.
Okuda, M., Balán, I., Petry, N. M., Oquendo, M., & Blanco, C. (2009). Cognitivebehavioral therapy for pathological gambling: Cultural considerations.. American
Journal Of Psychiatry, 166(12), 1325-1330.
Ozurumba, C. (2009). The impact of legalized casino gambling on state education
spending displacement. Journal Of Public Budgeting, Accounting & Financial
Management, 21(1), 83-104.
Pallesen, S., Mitsem, M., Kvale, G., Johnsen, B.H., & Molde, H. (2005). Outcome of
psychological treatments of pathological gambling: a review and meta-analysis.
Addiction, 100(10), 1412-1422.
Pantalon, M.V., Nich, C., Frankforter, T., Carroll, K.M. (2002). The URICA as a
measure of motivation to change among treatment-seeking individuals with
concurrent alcohol and cocaine problems. Psychology of Addictive Behaviors,
16(4), 299-307.
153
Park, S., Cho, M.J. Jeon, H.J., Lee, H.W., Bae, J.N., Park, J.I., Sohn, J.H., Lee, Y.R., Lee,
J.Y., & Hong, J.P. (2010). Prevalence, clinical correlates, comorbidities, and
suicidal tendencies in pathological Korean gamblers: results from the Korean
Epidemiological Catchment Area Study. Social Psychiatry and Psychiatric
Epidemiology, 45(6), 621-629.
Pasche, S. C., Sinclair, H., Collins, P., Pretorius, A., Grant, J. E., & Stein, D. J. (2013).
The effectiveness of a cognitive-behavioral intervention for pathological
gambling: A country-wide study. Annals Of Clinical Psychiatry, 25(4), 250-256.
Pavalko, R. M. (2004). Gambling and public policy. Public Integrity, 6(4), 333-348.
Penfold, A., Hatcher, S., Sullivan, S., and Collins, N. (2006a). Gambling Problems and
Attempted Suicide: Part I High prevalence amongst hospital admissions.
International Journal of Mental Health and Addiction, 4(3), 265-272.
Penfold, A., Hatcher, S., Sullivan, S., & Collins, N. (2006b). Gambling Problems and
Attempted Suicide: Part II Alcohol Abuse Increases Suicide Risk. International
Journal of Mental Health and Addiction, 4(3), 273-279.
Penney, A., Mazmanian, D., Jamieson, J., & Black, N. (2012). Factors associated with
recent suicide attempts in clients presenting for addiction treatment. International
Journal of Mental Health and Addiction, 10(1), 132-140.
Petry, N. M. (2000). Psychiatric symptoms in problem gambling and non-problem
gambling substance abusers. American Journal on Addictions, 9(2), 163-171.
Petry, N. M. (2003). Patterns and correlates of Gamblers Anonymous attendance in
154
pathological gamblers seeking professional treatment. Addictive Behaviors, 28(6),
1049.
Petry, N.M. (2005). Stages of change in treatment seeking pathological gamblers. Journal
of Consulting and Clinical Psychology, 73(2), 312-322.
Petry, N. M., Ammerman, Y., Bohl, J., Doersch, A., Gay, H., Kadden, R., Molina, C., &
Steinberg, K. (2006). Cognitive-behavioral therapy for pathological gamblers.
Journal of Consulting and Clinical Psychology, 74(3), 555-567.
Petry, N. M., Weinstock, J., Ledgerwood, D. M., & Morasco, B. (2008). A randomized
trial of brief interventions for problem and pathological gamblers. Journal Of
Consulting And Clinical Psychology, 76(2), 318-328.
Petry, N. M., Weinstock, J., Morasco, B. J., & Ledgerwood, D. M. (2009). Brief
motivational interventions for college student problem gamblers. Addiction,
104(9), 1569-1578.
Petry, N.M. (2010). Pathological gambling and the DSM-V. International Gambling
Studies, 10(2), 113-115.
Petry, N. M., Blanco, C., Stinchfield, R., & Volberg, R. (2012). An empirical evaluation
of proposed changes for gambling diagnosis in the DSM-5. Addiction, 108(3),
575-581.
Petry, N. M., Stinson, F. S., & Grant, B. F. (2005). Comorbidity of DSM-IV pathological
gambling and other psychiatric disorders: Results from the National
Epidemiologic Survey on Alcohol and Related Conditions. Journal of Clinical
Psychiatry, 66, 564-574.
155
Piquette, N., & Norman, E. (2013). An all-female problem-gambling counseling
treatment: Perceptions of effectiveness. Journal Of Groups In Addiction &
Recovery, 8(1), 51-75.
Pirracchio, R., Resche-Rigon, M., & Chevret, C. (2012). Evaluation of the Propensity
score methods for estimating marginal odds ratios in case of small sample size.
BMC Medical Research Methodology, 12(1), 70-79.
Piscitelli, F., & Albanese, J. S. (2000). Do casinos attract criminals? A study at the
Canadian-US border. Journal of Contemporary Criminal Justice, 16(4), 445-456.
Potenza, M.N., Steinberg, M.A., McLaughlin, S.D., Wu, R., Rounsaville, B.J.,&
O’Malley, S.S. (2001). Gender-related differences in the characteristics of
problem gamblers using a gambling helpline. The American Journal of
Psychiatry, 158(9), 1500-1005.
Potenza, M.N., Steinberg, M.A., & Wu, R. (2005). Characteristics of gambling helpline
callers with self-reported gambling and alcohol use problems. Journal of
Gambling Studies, 21(3), 233-254.
Potenza, M.N. (2006). Should addictive disorders include non-substance-related
conditions? Addiction, 101(s1), 142-151.
Potenza, M. N. (2008). The neurobiology of pathological gambling and drug addiction:
An overview and new findings. Philosophical Transactions of the Royal Society
B: Biological Sciences, 363(1507), 3181-3189.
Potik, D., Adelson, M., & Schreiber, S. (2007). Drug addiction from a psychodynamic
perspective: Methadone maintenance treatment (MMT) as transitional
156
phenomena. Psychology & Psychotherapy: Theory, Research & Practice, 80(2),
311-325.
Prochaska, J.O., Norcross, J.C. (2001). Stages of change. Psychotherapy: Theory,
Research, Practice, Training, (38(4), 443-448.
Prochaska, J.O., DiClemente, C.C., & Norcross, J.C. (1992). In search of how people
change: Applications to addictive behaviors. American Psychologist, 47(9), 11021114.
Ramos-Grille, I., Aragay, N., Gomá-i-Freixanet, M., Valero, S., & Vallès, V. (2013). The
role of personality in the prediction of treatment outcome in pathological
gamblers: A follow-up study. Psychological Assessment, 25(2), 599-605.
Rassen, J. A., Shelat, A. A., Franklin, J. M., Glynn, R. J., Solomon, D. H., &
Schneeweiss, S. (2013). Matching by propensity score in cohort studies with three
treatment groups. Epidemiology, 24(3), 401-409.
Rawson, R.A., McCann, M.J., Flammino, F., Shoptaw, S., Miotto, K., Reiber, C. & Ling,
W. (2006). A comparison of contingency management and cognitive-behavioral
approaches for stimulant-dependent individuals. Society for Study of Addiction.
101, 267-274.
Raylu, N., Loo, J., & Oei, T. P. (2013). Treatment of gambling problems in Asia:
Comprehensive review and implications for Asian problem gamblers. Journal of
Cognitive Psychotherapy, 27(3), 297-322.
Reid, R.C., Li, D.S., Lopez, J., Collard, M., Parhami, I., Karim, R., & Fong, T. (2011).
Exploring facets of personality and escapism in pathological gamblers. Journal of
157
Social Work Practice in the Addictions, 11, 60-74.
Reith, G., & Dobbie, F. (2011). Beginning gambling: The role of social networks and
environment. Addiction Research & Theory, 19(6), 483-493.
Richard, B. (2010). Diffusion of an economic development policy innovation: Explaining
the international spread of casino gambling. Journal Of Gambling Studies, 26(2),
287-300.
Ridgeway, G., McCaffrey, D., Morral, A., Burgette, L., & Griffin, B. A. (2014). Toolkit
for weighting and analysis of nonequivalent groups: A tutorial for the twang
package.
Riley, B., Smith, D., & Oakes, J. (2011). Exposure therapy for problem gambling in rural
communities: A program model and early outcomes. Australian Journal Of Rural
Health, 19(3), 142-146.
Rochlen, A.B., Rude, S.S., & Baron, A. (2005). The relationship of client stages of
change to working alliance and outcome in short-term counseling. Journal of
College Counseling, 8(1), 52-64.
Roehl, W. (1999). Quality of life issues in a casino destination. Journal of Business
Research, 44(3), 223-229.
Room, R., Turner, N. E., & Ialomiteanu, A. (1999). Community effects of the opening of
the Niagara casino. Addiction, 94(10), 1449-1466.
Rosenbaum, P. R., & Rubin, D. B. (1983). The central role of the propensity score in
observational studies for causal effects. Biometrika, 70(1), 41-55.
Rosenbaum, P. R. (1984). From association to causation in observational studies: The
158
role of tests of strongly ignorable treatment assignment. Journal Of The American
Statistical Association, 79(385), 41-48.
Rosenbaum, P.R. (1987). Sensitivity analysis for certain permutation inferences in
matched observational studies. Biometrika, 74(1), 13-26.
Rosenbaum, P. R. (1989). Optimal matching for observational studies. Journal Of The
American Statistical Association, 84(408), 1024-1032.
Rosenbaum, P.R. (2002). Observational Studies (2nd ed.). New York: Springer.
Rosenthal, R.J. & Lesieur, H.R. (1992). Self-reported withdrawal symptoms and
pathological gambling. The American Journal on Addictions, 1(2), 150-154.
Rugle, L. J., & Melamed, L. (1993). Neuropsychological assessment of attention
problems in pathological gamblers. Journal of Nervous and Mental Disease, 181
(2), 107–112.
Sandahl, C., Herlitz, K., Ahlin, G., & Rönnberg, S. (1998). Time-limited group
psychotherapy for moderately alcohol dependent patients: a randomized
controlled clinical trial. Psychotherapy research, 8(4), 361-378.
Sander, W., & Peters, A. (2009). Pathological gambling: Influence of quality of life and
psychological distress on abstinence after cognitive-behavioral inpatient
treatment. Journal Of Gambling Studies, 25(2), 253-262.
Schomerus, G., Matschinger, H., & Angermeyer, M. C. (2009). The stigma of psychiatric
treatment and help-seeking intentions for depression. European archives of
psychiatry and clinical neuroscience, 259(5), 298-306.
Schulte, S.J., Meier, P.S., & Stirling, J. (2011). Dual diagnosis clients’ treatment
159
satisfaction- a systematic review. BMC Psychiatry, 11, 64-75.
Screening for Mental Health (2013). National Depression Screening Day: Best Practices
Guide. Retrieved from http://www.mentalhealthscreening.org/downloads/
reserved/ndsd/ NDSD_Flash/NDSDPlanning/2013CommunityManual.pdf
Sequin, M., Boyer, R., Lesage, A., McGirr, A., Suissa, A., Tousignant, M., & Turecki, G.
(2010). Suicide and gambling: psychopathology and treatment-seeking.
Psychology of Addictive Behaviors, 24(3), 541-547.
Shaffer, H. J., & Hall, M. N. (2001). Updating and refining prevalence estimates of
disordered gambling behavior in the United States and Canada. Canadian Journal
of Public Health, 92(3), 168-172.
Shaffer, H.J. (2011). What is Addiction? A Perspective. by Howard J. Shaffer, Ph.D.,
C.A.S. Retrieved from: http:// www.divisiononaddictions.org/html/
whatisaddiction.htm.
Sharma, M. G., & Sharma, V. (2012). Effect of Psychotherapy on Pathological Gamblers.
SIS Journal Of Projective Psychology & Mental Health, 19(1), 56-60.
Slutske, W.S. (2006). Natural recovery and treatment-seeking in pathological gambling:
Results of two U.S. national surveys. The American Journal of Psychiatry,
163(2), 297-302.
Slutske, W. S., Zhu, G., Meier, M. H., & Martin, N. G. (2010). Genetic and
environmental influences on disordered gambling in men and women. Archives of
General Psychiatry, 67(6), 624 – 630.
Smith, D., Harvey, P., Battersby, M., Pols, R., Oakes, J., & Baigent, M. (2010).
160
Treatment outcomes and predictors of drop out for problem gamblers in South
Australia: A cohort study. Australian and New Zealand Journal of Psychiatry,
44(10), 911-920.
Smith, S.A., Thomas, S.A.,& Jackson, A.C. (2010). An exploration of the therapeutic
relationship and counselling outcomes in a problem gambling counselling service.
Journal of Social Work Practice, 18(1), 99-112.
Smock, S. A., Trepper, T. S., Wetchler, J. L., McCollum, E. E., Ray, R., & Pierce, K.
(2008). Solution-­‐‑Focused Group Therapy for level 1 substance abusers. Journal of
Marital and Family Therapy, 34(1), 107-120.
Sobell, M.B.& Sobell, L.C. (1973). Individualized behavior therapy for alcoholics.
Behavior Therapy, 4, 49-72.
Soberay, A., & Faragher, J. M.(2011). Differences in Therapeutic Preference Between
Problem Gambling and Non-Problem Gambling Clients at a University Based
Counseling Clinic. National Council on Problem Gambling Conference, Boston,
MA
Soberay, A., Faragher, J. M., Barbash, M., Brookover, A., & Grimsley, P. (2013).
Pathological gambling, co-occurring disorders, clinical presentation, and
treatment outcomes at a university-based counseling clinic. Journal of Gambling
Studies, DOI: 10.1007/s10899-012-9357-2
Specker, S. M., Carlson, G. A., Christenson, G. A., & Marcotte, M. (1995). Impulse
control disorders and attention deficit disorder in pathological gamblers. Annals of
Clinical Psychiatry, 7 (4), 175–179.
161
Spreeuwenberg, M. D., Bartak, A., Croon, M. A., Hagenaars, J. A., Busschbach, J. J.,
Andrea, H., Twisk, J., & Stijnen, T. (2010). The multiple propensity score as
control for bias in the comparison of more than two treatment arms: an
introduction from a case study in mental health. Medical care, 48(2), 166-174.
Stinchfield, R., Kushner, M. G., & Winters, K. C. (2005). Alcohol use and prior
substance abuse treatment in relation to gambling problem severity and gambling
treatment outcome. Journal Of Gambling Studies, 21(3), 273-297.
Straus, B. (2006). Some words about comments: How direct interpersonal
communication at Gamblers Anonymous meetings impacts GA members’
therapy. Journal Of Groups In Addiction & Recovery, 1(3), 75-111.
Substance Abuse and Mental Health Services Administration (2009). Substance Abuse
Treatment: Group Therapy. A Treatment Improvement Protocol TIP 41.
Rockville, MD: US Department of Health and Human Services.
Stuhldreher, W.L., Stuhldreher, T.J., & Forrest, K.Y.Z. (2007). Gambling as an
emerging health problem on campus. Journal of American College Health, 56(1),
75-88.
Sussman, S., Lisha, N., & Griffiths, M. (2011). Prevalence of the addictions: A problem
of the majority or the minority?. Evaluation & The Health Professions, 34(1), 356.
Suurvali, H., Hodgins, D., Toneatto, T., & Cunningham, J. (2008). Treatment seeking
among Ontario problem gamblers: Results of a population survey. Psychiatric
Services, 59(11), 1343-1346.
162
Suurvali, H., Hodgins, D., Toneatto, T., & Cunningham, J. (2012). Motivators for seeking
gambling-related treatment among Ontario problem gamblers. Journal Of
Gambling Studies, 28(2), 273-296.
Sweet, A. D. (2012). Black holes: Some notes on time, symbolization, and perversion in
the psychodynamics of addiction. International Forum Of Psychoanalysis, 21(2),
94-105.
Sylvain, C., Ladocuer, R., & Boisvert, J.M. (1997). Cognitive and behavioral treatment of
pathological gambling: A controlled study. Journal of Consulting and Clinical
Psychology, 65(5), 727-732.
Takushi, R. Y., Neighbors, C., Larimer, M. E., Lostutter, T. W., Cronce, J. M., & Marlatt,
G. (2004). Indicated prevention of problem gambling among college students.
Journal Of Gambling Studies, 20(1), 83-94.
Tanner-Smith, E. E., & Lipsey, M. W. (2014). Identifying baseline covariates for us in
propsensity scores: A novel approach illustrated for a nonrandomized study of
recovery high schools. Peabody Journal Of Education, 89(2), 183-196.
Tang, C., & Wu, A. (2010). Direct and indirect influences of fate control belief, gambling
expectancy bias, and self-efficacy on problem gambling and negative mood
among chinese college students: A multiple meditation analysis. Journal Of
Gambling Studies, 26(4), 533-543.
Tang, C., & Wu, A. (2012). Impulsivity as a moderator and mediator between life stress
and pathological gambling among Chinese treatment-seeking gamblers.
International Journal Of Mental Health & Addiction, 10(4), 573-584.
163
Tang, C.S., Wu, A.M., & Tang, J.Y. (2007). Gender differences in characteristics of
Chinese treatment-seeking problem gamblers. Journal of Gambling Studies,
23(2),145-156.
Tangenberg, K.M. (2005). Twelve-step programs and faith-based recovery: Research
controversies, provider perspectives, and practice implications. In Hilarski, C.
(Ed.).Addiction, assessment, and treatment with adolescents, adults, and
families.(19-40). New York: Haworth.
Taormina, R. J. (2009). Social and personality correlates of gambling attitudes and
behavior among Chinese residents of Macau. Journal of Social and Personal
Relationships, 26(8), 1047-1071.
Tatarsky, A., Marlatt, G.A. (2010). State of the art in harm reduction psychotherapy: an
emerging treatment for substance misuse. Journal of Clinical Psychology, 66(2),
117-122.
Tavares, H., Zilberman, M., & el-Guebaly, N. (2003). Are there cognitive and
behavioural approaches specific to the treatment of pathological gambling?.
Canadian Journal Of Psychiatry, 48(1), 22-27.
Teo, P., Mythily, S., Anantha, S., & Winslow, M. (2007). Demographic and clinical
featuresof 150 pathological gamblers referred to a community addictions
programme. Annals of the Academy of Medicine, 36(3), 165-168.
Toce-Gerstein, M., Gerstein, D.R., & Volberg, R.A. (2003). A hierarchy of gambling
disorders in the community. Addiction, 98(12), 1661-1672.
Tolchard, B., & Battersby, M. W. (2013). Cognitive behavior therapy for problem
164
gamblers: A clinical outcomes evaluation. Behaviour Change, 30(1), 12-23.
Toneatto, T. & Brennan, J. (2002). Pathological gambling in treatment-seeking substance
abusers. Addictive Behaviors, 27(3), 465-469.
Toneatto, T., & Dragonetti, R. (2008). Effectiveness of community-based treatment for
problem gambling: A qusi-experimental evaluation of cognitive-behavioral vs.
twelve-step therapy. American Journal On Addictions, 17(4), 298-303.
Tse, S., Campbell, L., Rossen, F., Wang, C., Jull, A., Yan, E., & Jackson, A. (2013).
Face-to-face and telephone counseling for problem gambling: A pragmatic
multisite randomized study. Research On Social Work Practice, 23(1), 57-65.
Turner, N. E., Preston, D. L., Saunders, C., McAvoy, S., & Jain, U. (2009). The
relationship of problem gambling to criminal behavior in a sample of Canadian
male federal offenders. Journal of Gambling Studies, 25(2), 153-169.
Vaishnavi, S., Payne, V., Connor, K., & Davidson, J. R. (2006). A comparison of the
SPRINT and CAPS assessment scales for posttraumatic stress disorder.
Depression and anxiety, 23(7), 437-440.
Valentine, G., & Hughes, K. (2012). Shared space, distant lives? Understanding family
and intimacy at home through the lens of internet gambling. Transactions Of The
Institute Of British Geographers, 37(2), 242-255.
Vermeersch, D.A., Lambert, M.J., Burlingame, G.M. (2000). Outcome Questionnaire:
Item sensitivity to change. Journal of Personality Assessment, 74(2), 242-261.
Volberg, R. A. (1994). The prevalence and demographics of pathological gamblers:
165
implications for public health. American Journal of Public Health, 84(2), 237241.
Volberg, R. (2002). Gambling and problem gambling in Nevada: Report to the Nevada
Department of Human Resources: Northampton, MA: Gemini.
Walther, B., Morgenstern, M., & Hanewinkel, R. (2012). Co-occurrence of addictive
behaviours: personality factors related to substance use, gambling and computer
gaming. European addiction research, 18(4), 167-174.
Wang, Y., Cai, H., Li, C., Jiang, Z., Wang, L., Song, J., & Xia, J. (2013). Optimal Caliper
Width for Propensity Score Matching of Three Treatment Groups: A Monte Carlo
Study. Plos ONE, 8(12), 1-7.
Watkins, S., Jonsson-Funk, M., Brookhart, M., Rosenberg, S. A., O'Shea, T., & Daniels,
J. (2013). An Empirical Comparison of Tree-Based Methods for Propensity Score
Estimation. Health Services Research, 48(5), 1798-1817.
Weatherly, J. N., Montes, K. S., Peters, D., & Wilson, A. N. (2012). Gambling behind the
walls: A behavior-analytic perspective. Behavior Analyst Today, 13(3/4), 2-8.
Weinstock, J., Rash, C., Burton, S., Moran, S., Biller, W., O'Neil, K., & Kruedelbach, N.
(2013). Examination of proposed DSM-5 changes to pathological gambling in a
helpline sample. Journal Of Clinical Psychology, 69(12), 1305-1314.
Westermeyer, J., Canive, J., Thuras, P., Thompson, J., Kim, S. W., Crosby, R. D., &
Garrard, J. (2008). Mental health of non-gamblers versus “Normal” gamblers
among American Indian Veterans: a community survey. Journal of Gambling
Studies, 24(2), 193-205.
166
Wickwire, E.M., Burke, R.S., Brown, S.A., Parker, J.D., & May, R.K. (2008).
Psychometric evaluation of the National Opionion Research Center DSM-IV
screen for gambling problems (NODS), American Journal on Addictions, 17(5),
392-395.
Williams, D. J., & Walker, G. J. (2009). Does offender gambling on the inside continue
on the outside? Insights from correctional professionals on gambling and re-entry.
Journal Of Offender Rehabilitation, 48(5), 402-415.
Wohl, M. A., Young, M. M., & Hart, K. E. (2007). Self-perceptions of dispositional luck:
Relationship to DSM gambling symptoms, subjective enjoyment of gambling and
treatment readiness. Substance Use & Misuse, 42(1), 43-63.
Wong, P.W., Chan, W.S., Conwell, Y., & Yip, P.S. (2010a). A psychological autopsy
Study of pathological gamblers who died by suicide. Journal of Affective
Disorders, 120(1), 213-216.
Wong, P.W., Cheung, D.Y., Conner, K.R., Conwell, Y., & Yip, P.S. (2010b). Gambling
and completed suicide in Hong Kong: a review of coroner court files. Primary
Care Companion to the Journal of Clinical Psychiatry, 12(6), DOI:
10.4088/PCC.09m00932blu
Wood, R. A., & Griffiths, M. D. (2007). A qualitative investigation of problem gambling
as an escape-based coping strategy. Psychology & Psychotherapy: Theory,
Research & Practice, 80(1), 107-125.
Wood, R. T., & Williams, R. J. (2007). Problem gambling on the Internet: Implications
167
for internet gambling policy in North America. New Media & Society, 9(3), 520542.
Wood, R. T., & Williams, R. J. (2011). A comparative profile of the internet gambler:
Demographic characteristics, game-play patterns, and problem gambling status.
New Media & Society, 13(7), 1123-1141.
Wood, T. E., Englander-Golden, P., Golden, D. E., & Pillai, V. K. (2010). Improving
addictions treatment outcomes by empowering self and others. International
Journal of Mental Health Nursing, 19(5), 363-368.
Woody, G. E., McLellan, A. T., Luborsky, L., & O'Brien, C. P. (1995). Psychotherapy in
community methadone programs: a validation study. The American Journal of
Psychiatry.
Wulfert, E., Blanchard, E. N., Freidenberg, B. M., & Martell, R. S. (2006). Retaining
pathological gamblers in cognitive behavior therapy through motivational
enhancement. Behavior Modification, 30(3), 315-340.
Wynne, H. J., & Shaffer, H. J. (2003). The socioeconomic impact of gambling: The
Whistler symposium. Journal Of Gambling Studies, 19(2), 111-121.
Xian, H., Shah, K. R., Phillips, S. M., Scherrer, J. F., Volberg, R., & Eisen, S. A. (2008).
Association of cognitive distortions with problem and pathological gambling in
adult male twins. Psychiatry Research, 160(3), 300-307.
Yi, S., & Kanetkar, V. (2011). Coping with guilt and shame after gambling loss. Journal
Of Gambling Studies, 27(3), 371-387.
Yip, P.S.F., Yang, K.C.T., Ip, B.Y.T., & Law, Y.W. (2007). Financial debt and suicide in
168
Hong Kong SAR. Journal of Applied Social Psychology, 37(12), 2788-2799.
Zernig, G., Wallner, R., Grohs, U., Kriechbaum, N., Kemmler, G., & Saria, A. (2008). A
randomized trial of short psychotherapy versus sustained-release bupropion for
smoking cessation. Addiction, 103(12), 2024-2031.
Zimmerman, M., Chelminski, I., & Young, D. (2006). Prevalence and diagnostic
correlates of DSM-IV pathological gambling in psychiatric outpatients. Journal
Of Gambling Studies, 22(2), 255-262.
Zimmerman, M., & Galione, J. N. (2011). Screening for Bipolar Disorder with the Mood
Disorders Questionnaire: A review. Harvard Review Of Psychiatry, 19(5), 219228.
169
Appendix A
Boxplots of Age for Each Group After Maximum Treat Matching
Boxplots of Pairwise Differences of Age After Maximum Treat Matching
170
Bar Charts for Gender for Each Group After Maximum Treat Matching
Boxplots of Pairwise Differences of Gender After Maximum Treat Matching
171
Boxplots of Initial OQ-45 Score for Each Group After Maximum Treat Matching
Boxplots of Pairwise Differences of Initial OQ-45 Score After Maximum Treat Matching
172
Bar Charts for Positive Anxiety Screens for Each Group After Maximum Treat Matching
Boxplots of Pairwise Differences of Positive Anxiety Screens After Maximum Treat
Matching
173
Boxplots of NODS Score for Each Group After Maximum Treat Matching
Boxplots of Pairwise Differences of NODS Score After Maximum Treat Matching
174
Appendix B
Boxplots of Age for Each Group After Caliper Matching
Boxplots of Pairwise Differences of Age After Caliper Matching 175
Bar Charts for Gender for Each Group After Caliper Matching
Boxplots of Pairwise Differences of Gender After Caliper Matching 176
Boxplots of Initial OQ-45 Score for Each Group After Caliper Matching
Boxplots of Pairwise Differences of Initial OQ-45 Score After Caliper Matching 177
Bar Charts for Positive Anxiety Screens for Each Group After Caliper Matching
Boxplots of Pairwise Differences of Positive Anxiety Screens After Caliper Matching 178
Boxplots of NODS Score for Each Group After Caliper Matching
Boxplots of Pairwise Differences of NODS Score After Caliper Matching 179
Appendix C
Boxplots of Age for Each Group After 2:1:n Matching Boxplots of Pairwise Differences of Age After 2:1:n Matching
180
Bar Charts for Gender for Each Group After 2:1:n Matching
Boxplots of Pairwise Differences of Gender After 2:1:n Matching 181
Boxplots of Initial OQ-45 Score for Each Group After 2:1:n Matching Boxplots of Pairwise Differences of Initial OQ-45 Score After 2:1:n Matching 182
Bar Charts for Positive Anxiety Screens for Each Group After 2:1:n Matching
Boxplots of Pairwise Differences of Positive Anxiety Screens After 2:1:n Matching 183
Boxplots of NODS Score for Each Group After 2:1:n Matching Boxplots of Pairwise Differences of NODS Score After 2:1:n Matching 184
Appendix D
Boxplots of Age for Each Group After 3:2:n Matching Boxplots of Pairwise Differences of Age After 3:2:n Matching 185
Bar Charts for Gender for Each Group After 3:2:n Matching
Boxplots of Pairwise Differences of Gender After 3:2:n Matching 186
Boxplots of Initial OQ-45 Score for Each Group After 3:2:n Matching
Boxplots of Pairwise Differences of Initial OQ-45 Score After 3:2:n Matching 187
Bar Charts for Positive Anxiety Screens for Each Group After 3:2:n Matching Boxplots of Pairwise Differences of Positive Anxiety Screens After 3:2:n Matching 188
Boxplots of NODS Score for Each Group After 3:2:n Matching Boxplots of Pairwise Differences of NODS Score After 3:2:n Matching 189