This article was downloaded by: [Georgian Court University]

On: 20 February 2015, At: 09:51

Publisher: Routledge
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered
office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

International Gambling Studies

Publication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/rigs20

Online gambling participation and

problem gambling severity: is there a
causal relationship?
a b
Kahlil S. Philander & Terri-Lynn MacKay
a
William F. Harrah College of Hotel Administration, University of
Nevada, Las Vegas, USA
b
Department of Educational and Clinical Studies, University of
Nevada, Las Vegas, USA
Published online: 31 Mar 2014.

Click for updates

To cite this article: Kahlil S. Philander & Terri-Lynn MacKay (2014) Online gambling participation
and problem gambling severity: is there a causal relationship?, International Gambling Studies,
14:2, 214-227, DOI: 10.1080/14459795.2014.893585

To link to this article: http://dx.doi.org/10.1080/14459795.2014.893585

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the
“Content”) contained in the publications on our platform. However, Taylor & Francis,
our agents, and our licensors make no representations or warranties whatsoever as to
the accuracy, completeness, or suitability for any purpose of the Content. Any opinions
and views expressed in this publication are the opinions and views of the authors,
and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content
should not be relied upon and should be independently verified with primary sources
of information. Taylor and Francis shall not be liable for any losses, actions, claims,
proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or
arising out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &
 Conditions of access and use can be found at http://www.tandfonline.com/page/terms-
and-conditions
Downloaded by [Georgian Court University] at 09:51 20 February 2015
 International Gambling Studies, 2014
Vol. 14, No. 2, 214–227, http://dx.doi.org/10.1080/14459795.2014.893585

Online gambling participation and problem gambling severity: is there

a causal relationship?
Kahlil S. Philandera* and Terri-Lynn MacKayb
a
William F. Harrah College of Hotel Administration, University of Nevada, Las Vegas, USA;
b
Department of Educational and Clinical Studies, University of Nevada, Las Vegas, USA
(Received 13 December 2013; accepted 7 February 2014)

Studies of Internet gambling have consistently shown that online gamblers are more
Downloaded by [Georgian Court University] at 09:51 20 February 2015

likely to report disordered gambling behaviour than offline gamblers. However, little
research has focused on whether this is a causal relationship or whether this risk factor
is capturing a relationship with one or more missing variables. To address whether
there is a strong causal argument for the effect of online gambling participation on
problem gambling severity, we use a secondary data method that corrects for potential
omitted variable bias. Once this issue is addressed, we find that past-year participation
in online gambling is related to a decrease in problem gambling severity, which is the
opposite of the popular view in current literature. The estimates in this study are found
to be robust to various forms of online gambling, control variables and problem
gambling measurement instruments. The findings were also consistent when using a
representative sample from the United Kingdom and when using an online research
panel from Ontario, Canada. As a primary force against the widespread adoption of
Internet gambling has been public health concern over problem gambling, this study
provides evidence that such decisions should be more closely considered by
policymakers.
Keywords: online gaming; problem gambling; addiction; policy

Introduction
Over the past 20 years, contemporary gambling expansion has been influenced strongly by
technology. While the bricks-and-mortar casino landscape has certainly evolved due to
advances in technology, the most obvious change in gambling consumption has been for
gamblers to access games via the Internet. Online gambling has grown to be a $32 billion
industry worldwide (‘The Land of eGaming Opportunity’, 2012), but in many countries
online gambling has been slow to be adopted as a legal form of gaming. This apprehension
is commonly attributed to uncertainty over the effects that expansion of the Internet
gambling industry will have on public health (e.g. Trynacity, 2011; Vardi, 2013). In
particular, public concerns over problem gambling increases have mitigated the legal
expansion of online gambling. In many cases, this has facilitated the expansion of
unregulated online gambling, with few requirements for player protection, financial
reporting or ethical behaviour.
Implications of the unregulated market aside, initial concerns over the harmful effects of
this new medium are sensible, as early research in online gambling has established a
correlation between problem gambling and online gambling participation. Empirical
studies have not been able to advance with the same accelerated pace as the industry, but one

*Corresponding author. Email: kahlil.philander@unlv.edu

q 2014 Taylor & Francis

 International Gambling Studies 215

finding that has been consistently replicated is that online gamblers are more likely to report
disordered gambling behaviour when compared to offline gamblers (Griffiths & Barnes,
2008; Griffiths, Parke, Wood, & Rigbye, 2010; Griffiths, Wardle, Orford, Sproston, &
Erens, 2009; Ladd & Petry, 2002; McBride & Derevensky, 2009; Petry, 2006; Petry &
Weinstock, 2007; Wood & Williams, 2007, 2009, 2011). A primary concern is that the
Internet, functioning as a conduit for gambling, has become a causal factor in players
developing problems. While this may be valid, empirical evidence to support such
assertions is lacking.
Shaffer and colleagues have argued that the Internet is not inherently addictive,
contending that it is the interplay of the individual with the activity that determines level of
involvement (Shaffer, 1996; Shaffer, Hall, & Bilt, 2000). A large body of empirical research
has been conducted on identifying risk and resiliency factors that influence gambling
behaviour. These include demographic factors (e.g. age, gender, minority status, education,
Downloaded by [Georgian Court University] at 09:51 20 February 2015

socio-economic status, marital status), medium factors (e.g. availability, accessibility,

exposure, sensory characteristics), cognitive factors (e.g. illusion of control, attributional
bias, gambler’s fallacy), personality factors (e.g. sensation seeking, impulsivity, arousal),
physiological/biological factors (e.g. genetics, heart rate and arousal, transmitter activity)
and co-morbid disorders (e.g. substance use, mood disorders, anxiety disorders).
Only a handful of studies have examined variables that may be associated with online
gambling. Petry (2006) found that wagering on the Internet was a significant predictor of
lower scores on mental and physical health measures, even after controlling for age,
gender and pathological gambling status. The frequency of Internet gambling has been
found as a predictor of lower scores on mental and physical health measures when
controlling for demographics and pathological gambling (Petry & Weinstock, 2007).
Griffiths et al. (2009) found that drinking at least twice the recommended amount in one
day was a significant predictor of Internet gambling. Internet gamblers who engaged in
multiple online activities have been shown to endorse problematic gambling behaviour
and have a history of mood disturbance, self-harm and substance misuse (Lloyd et al.,
2010). McCormack and Griffiths (2012) observed a sample including 15 online gamblers,
finding that the motivation to gamble online was derived from greater opportunity,
convenience, value for money, game variety and anonymity. Gainsbury, Russell, Wood,
Hing, and Blaszczynski (in press) found problem Internet gamblers more likely to be
young, less educated and have greater debts than non-problem Internet gamblers.
Using data from the 2007 British Gambling Prevalence Study, LaPlante, Nelson,
LaBrie, and Shaffer (2011) examined the relationship between specific gambling formats
and disordered gambling status. They found a general lack of significance, in particular
with Internet gambling, between gambling formats and problem gambling when gambling
involvement was also included in the model (with gambling involvement being proxied by
number of gambling activities played in the past 12 months). Welte, Barnes, Tidwell, and
Hoffman (2009) found that among a group of US young people, Internet gamblers had
high rates of problem gambling symptoms, but also high rates of overall frequency of
gambling and gambling versatility. When problem gambling symptoms were adjusted for
other gambling activity, the Internet gambling variable ceased to predict problem
gambling. Similarly, Gainsbury, Russell, Hing, Wood, and Blaszczynski (2013) found that
their results from comparing Internet and non-Internet gamblers was consistent with a
perspective that it was the additive effect of multiple forms of gambling involvement that
was impactful on severity, rather specifically engaging in online gambling. Wardle,
Moody, Griffiths, Orford, and Volberg (2011) recently identified a sample of Internet-only
gamblers and found that none of these gamblers were classified as problem gamblers.
 216 K.S. Philander and Terri-Lynn MacKay

While the aforementioned studies provide some insight into the relationship between
online gambling participation and problem gambling, no research to date has controlled
for the endogenous participation of online gamblers when attempting to identify a causal
relationship with problem gambling severity. That is, it is unclear whether online
gambling participation causes an increase in problem gambling severity, or whether
observed prevalence rates among online gamblers are a reflection of correlation with
variables that cannot be controlled in typical models. In fact, many gambling prevalence
studies have tended to focus on the determination of ‘risk factors’ rather than establish any
causal arguments. This approach may be useful for clinical assessment and responsible
gambling programme targeting, but is a potentially misleading tool for establishing public
policy.
To address whether there is a strong causal argument to be made over the effect of
online gambling participation on problem gambling, this study examines a representative
Downloaded by [Georgian Court University] at 09:51 20 February 2015

sample of gamblers and non-gamblers from a western nation with one of the most
liberalized online gambling industries, the United Kingdom (UK). Residents of the UK
have access to online gambling options though domestic gambling sites, ‘white listed’
foreign sites permitted to advertise in the UK, and other foreign sites operating offshore.
To address past issues due to omitted variable bias and spurious correlation, modelling
methods are used that estimate unbiased relationships between online gambling
participation and problem gambling severity, even when not all causal variables are
available as controls. In order to test the robustness of the findings, a similar model was
estimated on a secondary data set collected from an online research panel in Canada.

Methodology
Secondary data was obtained from the 2010 British Gambling Prevalence Study (Wardle,
Moody, Spence, et al., 2011). The data set is a nationally representative survey of 7756
adults aged 16 and older, living in private households in England, Scotland and Wales.
Among other aims, the data was collected to measure the prevalence of participation in
gambling, estimate the prevalence of problem gambling and explore the socio-
demographic factors associated with problem gambling. Respondents were asked about
participation in various gambling activities in the past 12 months (in person, online or
both) and level of involvement in those activities. Problem gambling was measured with
criteria from the Diagnostic and Statistical Manual of Mental Disorders (4th ed., DSM-IV;
American Psychiatric Association, 2000) and the Problem Gambling Severity Index
(PGSI; Ferris & Wynne, 2001). PGSI scores are categorized into non-gambler/non-
problem (92.7% of the sample), low-risk (5.2%), medium-risk (1.5%) and problem
gambler (0.6%) categories. The categories are ordinally ranked, with non-problem as the
lowest rank and problem gambling as the highest rank. We group non-gamblers with non-
problem gamblers for consistency with the reporting in Wardle, Moody, Griffiths, et al.
(2011) and because we felt the small probability that a non-gambler would qualify as
anything other than a non-problem gambler was outweighed by the increase in power from
over 2000 more observations (the majority of PGSI questions would be automatically
scored as zero for a non-gambler). However, we recognize that this inclusion may create
some small bias towards a more positive estimated coefficient size, and account for this
possibility in our interpretation of the results.
Participation in online gambling is defined by participation over the past 12 months in
any of online fruit/slots/instant wins (2.6%), online casino games (1.8%), online sports
wagering (2.1%) or online bingo (1.7%), resulting in 6.1% overall participation in at least
 International Gambling Studies 217

one form. Each of these category of games is also tested individually. Other measured
variables included gender, age, ethnicity, economic status of household (classification of
social position based on occupation of the household reference person), level of education,
self-reported general health status, smoking status and level of alcohol use. A breakdown
of demographic characteristics and gambling participation rates can be found in Wardle,
Moody, Griffiths, et al. (2011).
In order to test the robustness of the findings, a similar model was estimated on a
secondary data set survey that was conducted in Ontario, Canada. The survey was
provided to an online research panel of 3343 respondents that were stratified by age,
gender and region to measure gambling behaviour. The survey design was based on
Wiebe, Mun, and Kauffman (2006) and measured gambling behaviour (online and
offline), participation in online leisure activities (e.g. shopping) and problem gambling
severity (PGSI). Despite the data having been stratified on several variables to match
Downloaded by [Georgian Court University] at 09:51 20 February 2015

census demographic figures, the panel data should not be considered to be representative
of the general population as prior research has shown that online research panels have
higher levels of problem gambling prevalence (Williams, Volberg, & Stevens, 2012).
However, the higher incidence of gambling prevalence does provide a useful amount of
variation in the dependent variable to estimate relative impacts of online gambling
participation. The participants were categorized into PGSI non-problem/non-gambler
(81.1%), low-risk (10.9%), medium-risk (4.6%) and problem gambler (3.5%) categories,
while past-year participation in any online gambling totalled 12.8% of the sample.

Empirical estimation technique

Equation 1 defines the ordered regression model for the underlying relationship between
online gambling participation and problem gambling:

Prðoutcomej ¼ iÞ ¼ PrðK i21 , b1 ·OGj þ b2 ·x2j þ . . . þ bk ·xkj þ uj # K i Þ ð1Þ

The outcome variable is the ordered level of gambling involvement, non-problem/non-

gambler (1), low-risk (2), medium-risk (3) and problem gambler (4). OGj is a binary
variable indicating past-year participation in online gambling, where ‘1’ indicates having
participated and ‘0’ indicates having not participated; x2j . . . xkj are other explanatory
variables indicating problem gambling severity; Ki21 and Ki are respectively the lower and
upper bound cut-off criteria for a category membership; and uj is the model error term.
We recognize that we cannot fully identify all k 2 1 explanatory variables, therefore
the effects of these variables will appear in the error term vj and create bias if Cov(OGj, vj)
– 0. To correct this bias, we estimate the following system of equations, which produces a
consistent estimate of the effect of online gambling participation (Roodman, 2011;
Wooldridge, 2010):

Prðoutcomej ¼ iÞ ¼ PrðK i21 , b1 ·OGj þ b2 ·x2j þ . . . þ bk2n ·xðk2nÞj þ vj # K i Þ ð2Þ

PrðOG ¼ 1Þ ¼ Fða0 þ a1 ·z1j þ a2 ·z2j þ a3 ·x2j þ . . . þ ak2nþ1 ·xðk2nÞj þ 1j Þ ð3Þ

Where OGj is the fitted value from Equation 3, n is the number of missing explanatory
variables, and z1j and z2j are exogenous variables that must satisfy the standard
requirements of model instruments. In particular, they must satisfy two necessary
conditions: (1) have zero correlation with vj – that is, they provide no more explanatory
 218 K.S. Philander and Terri-Lynn MacKay

power for problem gambling severity than the variables already included in the model; and
(2) they must have a non-zero correlation with OGj, online gambling participation.
Put succinctly, a probit model to predict online gambling participation is estimated in a
first stage, using two instruments. The predicted values of online gambling participation
are then used in lieu of the actual values, which produces an unbiased estimate of the
model coefficient explaining the relationship between problem gambling severity and
online gambling participation.
The challenge in using this method typically centres on finding one or more
appropriate instruments. We suggest two variables as potential instruments for online
gambling participation, which are represented by z1j and z2j. The variables are two
available responses to the survey question, ‘In a month, which of the following activities,
if any, do you usually do?’ One potential response is ‘Shop online’ and another potential
response is ‘Browsing the Internet’. Both of these variables are binary, where ‘1’ indicates
Downloaded by [Georgian Court University] at 09:51 20 February 2015

having participated in the activity as recreation and ‘0’ indicates having not participated in
the activity as recreation.
Both of these instruments describe a familiarity and comfort with computers, the
Internet, and Internet commerce. Comfort with the Internet is likely to be a characteristic
that is important to choosing to gamble online, therefore the instruments should satisfy
necessary condition (2), of a non-zero correlation with online gambling participation.
These variables have no direct or established empirical relationship to problem gambling,
and it seems unlikely that general reporting of leisure behaviour taking place on the
Internet would be directly related to problem gambling severity, at the very least in so far
as they provide information beyond the other control variables – they should provide no
more prediction value of problem gambling severity than online gambling participation
does directly, thereby satisfying necessary condition (1). To further support the validity of
these necessary conditions, we also employ empirical tests in the next section.
Other independent variables used during the estimation procedure include: past-year
number of gambling activities participation (Number of gambling activities), gender
(Gender), age group 16 þ in 10-year bands (Age), ethnic group (Ethnicity), the main
economic activity of the household representative (Economic activity), highest
educational qualification (Education), general health response (General Health), smoking
status (Smoking) and category of alcohol consumption (Alcohol). Model error terms are
clustered by government office region, since these different regions will have different
gambling availability and support services. The model standard errors are otherwise robust
to arbitrary forms of heteroskedasticity using the Huber/White/sandwich estimate of
variance (Rogers, 1994).

Results
The results of the full model are provided in column 5 of Table 1, alongside regression-
adjusted models. In all models we observe that the online participation coefficient is
negative, suggesting that participation in online gambling is related to a decrease in
problem gambling severity (we note that the coefficient in model 3 is not significant, but
the significance of the other models suggests this may be an issue of power).1 Both
instruments appear to be statistically significant predictors of online gambling
participation, suggesting they may satisfy necessary condition (2).
The online gambling participation variable is noted as being substantially more
negative in models where the ‘Number of gambling activities’ variable is included. This
may be the case because inclusion of both variables suggests that one of the counted forms
 International Gambling Studies 219

Table 1. Two-stage least square ordered probit.

PGSI group (second stage) (Model 1) (Model 2) (Model 3) (Model 4) (Model 5)

Online gambling past 12 months 2 0.353* 2 0.938*** 20.0815 2 0.848*** 20.833***
(22.23) (211.06) (20.30) (2 6.16) (26.88)
Number of gambling activities 0.306*** 0.310*** 0.311***
(28.38) (23.23) (21.69)

Any online gambling past 12 months (first stage)

Browsing the Internet 0.277*** 0.307*** 0.283*** 0.296*** 0.295***

(4.52) (3.61) (4.67) (3.52) (3.57)
Shop online 0.310*** 0.316*** 0.329*** 0.315*** 0.313***
(8.91) (8.41) (9.28) (7.29) (6.95)
Number of gambling activities 0.317*** 0.319*** 0.319***
Downloaded by [Georgian Court University] at 09:51 20 February 2015

(28.74) (29.83) (33.72)

Gender Yes Yes Yes Yes Yes
Age group Yes Yes Yes Yes Yes
Ethnicity No No Yes Yes Yes
Economic activity No No Yes Yes Yes
Education No No Yes Yes Yes
General health No No Yes No Yes
Smoking No No Yes No Yes
Alcohol No No Yes No Yes
Observations 7,756 7,750 7,679 7,711 7,674
Note: t statistics in parentheses.
*p , 0.05, **p , 0.01, ***p , 0.001.

of gambling from ‘Number of gambling activities’ is online gambling, and not a more
harmful form of gambling such as fruit machines located in pubs. The addition of several
demographic variables from model 1 to model 3 causes the online gambling variable to
decrease in (absolute) size and become non-significant. This suggests that at least part of
the effect size from model 1 may be a factor related to demographic issues or other
predispositions. However, the effect sizes in models 2 and 5 are quite similar, suggesting
this difference may be statistical noise from an underspecified model. The results from the
variables that are indicated instead of fully described were either insignificant or in line
with prior research – higher levels of problem gambling severity were associated with
being male, younger, a racial minority, in poor self-reported general health, a smoker or a
heavier drinker.
Since the coefficient sizes from the ordered probit models are not directly
interpretable, we computed the average change in predicted PGSI classification, as a result
of participating in online gambling. Based on the results from model 5 of Table 1,
participation in online gambling increases the predicted probability of being a non-
problem gambler by 4.32%. As a corollary, the average predicted probability of being a
low, moderate, or problem gambler decreases by 2.67%, 1.02% and 0.81% respectively.
We note that unlike linear regression models, these predictions will differ depending on an
individual’s other characteristics. To better convey this concept, an illustrative example is
provided in Table 2, which compares scenarios for a hypothetical gambler, dependent on
whether or not the individual gambles on the Internet.
The results from Table 1 are substantially different from the results shown in Table 3,
which does not use a two-stage least squares procedure to correct for model endogeneity.
The results from these ordered probit models are representative of the results typically
 220 K.S. Philander and Terri-Lynn MacKay

Table 2. Predicted probabilities of PGSI categorization for illustrative gambler.

Non-problem Low risk Moderate risk Problem gambler

Predicted probability (Scenario 1) 89.06% 8.18% 2.12% 0.64%
Predicted probability (Scenario 2) 96.01% 3.25% 0.60% 0.13%
Scenario 1: No online gambling participation; three gambling activities.
Scenario 2: Online gambling participation; four gambling activities.
Controls: Age 45–54; male; married; white; east of England; employed in paid work; non-smoker drank 5–9
units on heaviest drinking day in past week; fair self-reported health.

found in prior studies. Online gambling participation is positively related to problem

gambling severity, with a weaker result seen in models that include the number of
Downloaded by [Georgian Court University] at 09:51 20 February 2015

gambling activities. These results highlight the danger of interpreting coefficient sizes in
the absence of a correction mechanism for endogenous correlation in the online gambling
participation variable.
In Table 4, robustness checks on the general results from the fully model are provided.
Model 1 estimates a linear two-stage least squares model on the PGSI score (not PGSI
category membership). Importantly, the results are similar for this estimation method that
uses a linear probability model in the first stage. The linear model also allows for empirical
tests of other model assumptions. The F-test of weak identification ¼ 25.97, well
exceeding the benchmark value of 10, suggesting that these variables adequately satisfied
necessary condition (2) of strong instruments (Sovey & Green, 2011; Staiger & Stock,
1997). The Sargan test statistic (also known as the Hansen J statistic), x 2 ¼ 0.98,
p ¼ 0.32, failed to reject the assumption that ‘Shop online’ and ‘Browsing the Internet’ do
not belong in the structural equation predicting PGSI membership (Sargan, 1958),
supporting the view that necessary condition (1) of the instruments is satisfied. The
empirical tests therefore support the theoretical basis for inclusion of these instruments.
The results held for the use of DSM-IV scores in lieu of PGSI scores. This is
particularly compelling, since data show limited overlap between the sets of problem
gamblers identified by the two screening tools.2 Only 0.45% of respondents qualified
as problem gamblers in both metrics, compared to 0.63% in PGSI-only and 0.83% in

Table 3. Ordered probit – no endogeneity corrections.

PGSI group (dependent var.) (Model 1) (Model 2) (Model 3) (Model 4) (Model 5)

Online gambling past 12 months 1.049*** 0.331*** 1.059*** 0.357*** 0.365***
(33.20) (5.86) (26.75) (6.75) (6.42)
Number of gambling activities 0.225*** 0.233*** 0.234***
(17.11) (17.20) (16.64)
Gender Yes Yes Yes Yes Yes
Age group Yes Yes Yes Yes Yes
Ethnicity No No Yes Yes Yes
Economic activity No No Yes Yes Yes
Education No No Yes Yes Yes
General health No No Yes No Yes
Smoking No No Yes No Yes
Alcohol No No Yes No Yes
Observations 7,748 7,745 7,675 7,710 7,673
Note: t statistics in parentheses.
*p , 0.05, **p , 0.01, ***p , 0.001.
 International Gambling Studies 221

Table 4. 2SLS – alternative model specifications.

Model 1 Model 2 Model 3 Model 4 Model 5 Model 6

Dependent variable: PGSI Sc. DSM Pg. PGSI Gr. PGSI Gr. PGSI Gr. PGSI Gr.

Main model (second stage)

Online gambling past 23.336*** 2 1.632***
12 months
Number of gambling 0.370*** 0.310*** 0.303*** 0.293*** 0.285*** 0.272***
activities
Online slots past 12 months 21.160***
Online casino past 21.199***
12 months
Online sports bet past 2 1.112***
12 months
2 0.839*
Downloaded by [Georgian Court University] at 09:51 20 February 2015

Online bingo past

12 months
Online gambling past 12 months (first stage)
Browsing the Internet 0.284**
Shop online 0.300***
Number of gambling 0.315***
activities
Online slots past 12 months (first stage)
Browsing the Internet 0.256***
Shop online 0.247**
Number of gambling 0.295***
activities
Online casino past 12 months (first stage)
Browsing the Internet 0.144
Shop online 0.209
Number of gambling 0.274***
activities
Online sports bet past 12 months (first stage)
Browsing the Internet 0.355**
Shop online 0.371***
Number of gambling 0.246***
activities
Online bingo past 12 months (first stage)
Browsing the Internet 0.264*
Shop online 0.323**
Number of gambling 0.244***
activities
Gender Yes Yes Yes Yes Yes Yes
Age group Yes Yes Yes Yes Yes Yes
Ethnicity Yes Yes Yes Yes Yes Yes
Economic activity Yes Yes Yes Yes Yes Yes
Education Yes Yes Yes Yes Yes Yes
General health Yes Yes Yes Yes Yes Yes
Smoking Yes Yes Yes Yes Yes Yes
Alcohol Yes Yes Yes Yes Yes Yes
Observations 7,671 7,674 7,672 7,673 7,672 7,672

Note: Perfectly predicted observations are dropped from the estimation samples; Sc. ¼ Score; Pg. ¼ Problem
gambler; Gr. ¼ Group; *p , 0.05, **p , 0.01, ***p , 0.001.

DSM-IV-only. The results appear to hold for online sports betting, but do not appear to
hold for online bingo, as that coefficient is not statistically significant. We note that the
maximum likelihood estimation method used here may not be fully efficient, although it is
 222 K.S. Philander and Terri-Lynn MacKay

consistent, which may have reduced some power to detect significant relationship
(Roodman, 2011) Although the coefficient on the casino games variable is significant,
those results should be interpreted with caution as the instruments appear to be weakly
related to online casino game participation. Similar models were estimated using male-
only and female-only observations (not shown), also producing similar coefficient sizes.

Model robustness on secondary data set

The secondary data set in Ontario, Canada, is sufficiently similar in question design to be
used as a test of robustness, but minor variations in survey response items required use of
proxy variables to replicate the analysis. In the Ontario data set, ‘Shop online’ and
‘Personal use of the Internet’ leisure activities responses on a Likert scale were used as
instruments for online gambling participation, income was used as a proxy for economic
Downloaded by [Georgian Court University] at 09:51 20 February 2015

activity and no health variables were available for inclusion, though their exclusion in the
primary data set did not meaningfully affect the estimation results (see models 4 and 5 in
Table 1 as a means of comparison).
As shown in Table 5, the ‘Shop online’ variable appears to be a valid instrument for
Internet gambling but the ‘Personal Internet use’ categories are not statistically different
from one another. The lack of statistical significance appears to be attributable to the
online panel survey administration method, where the entire sample was self-selected in
choosing an Internet-based method of delivery. It is likely that all participants had a

Table 5. Two-stage least square ordered probit – robustness test.

PGSI group (second-stage) (Model 1) (Model 2) (Model 3) (Model 4)
Online gambling past 12 months 2.114*** 21.250*** 2.161*** 21.308***
(20.40) (2 15.86) (20.13) (2 16.12)
Number of gambling activities 0.380*** 0.384***
(46.30) (42.78)

Any online gambling past 12 months (First-stage)

Shop online (Base ¼ Daily):

Weekly 20.401 0.378** 20.482 0.314**
Monthly 20.780** 0.225** 20.850** 0.215*
Less than monthly 2 1.054*** 0.287*** 2 1.124*** 0.284***
Never 20.962** 0.296* 21.072** 0.278*

Personal Internet use (Base ¼ 1 hour or less):

1 – 5 hours 20.102 0.122 20.016 0.126

6 – 10 hours 20.078 0.226 0.009 0.237
11 – 20 hours 20.033 0.233 0.045 0.247
. 20 hours 0.081 0.275 0.173 0.263
Number of gambling activities 0.398*** 0.400***
Gender Yes Yes Yes Yes
Age group Yes Yes Yes Yes
Ethnicity No No Yes Yes
Income No No Yes Yes
Education No No Yes Yes
Observations 3,041 3,041 3,041 3,041
Note: *p , 0.05, **p , 0.01, ***p , 0.001.
 International Gambling Studies 223

relatively high comfort level with the Internet (given that they are willing to engage in
online surveys on an ongoing basis), causing the ‘Personal Internet use’ variable to not
differentiate between respondents.
While the loss of ‘Personal Internet use’ as a differentiating factor reduces the number
of available instruments, one valid instrument is sufficient to produce a consistent estimate
(Wooldridge, 2010) and the UK model established ‘Shop online’ as a valid instrument.
Therefore, the model should still continue to be consistent overall, albeit with weaker
power. This weaker power effect is evident when comparing models without ‘Number of
gambling activities’ (Models 1 and 3) to the models where it is included (Models 2 and 4).
In the former models, the endogenous correlation remains strong and the results show a
positive correlation with problem gambling severity. In the latter models that include the
important ‘involvement’ proxy variable, the estimates are much more in line with the
results in Table 1 and further support the robustness of this result. The ‘Number of
Downloaded by [Georgian Court University] at 09:51 20 February 2015

gambling activities’ coefficient size is also similar in magnitude (roughly 0.31 in Table 1
as compared to roughly 0.38 in Table 5), though caution should be used when interpreting
relative coefficient sizes in ordered probit models.

Discussion
The results from this study provide a strong indication that previous ideas about the
relationship between online gambling participation and problem gambling severity may be
misguided. When endogenous correlation in online gambling participation is corrected,
participation appears to be negatively related to problem gambling severity. This is the
opposite of the effect that is observed in less robust models, and is indicative of some sort
of spurious relationship. This finding was noted to be robust in another secondary data set
from a jurisdiction without regulated online gambling, it was robust to many online
gambling variants, and it was robust to use of either the PGSI or DSM-IV-based metrics to
measure problem gambling.
The absence of a positive causal link between online gambling and problem gambling
is an important finding for policymakers. A lack of regulatory standards has perpetuated
grey-market business practices and leaves players vulnerable without appropriate
consumer protection. A primary force against widespread adoption of Internet gambling
has been public health concerns over problem gambling, and this study provides evidence
that such behaviour should be reconsidered by policymakers. Legal adoption of online
gambling would support economic expansion agendas, reduce the potential for money
laundering and improve player security through formal regulation, potentially without the
previously assumed public health concerns.
While this study does not provide a clear explanation of how this relationship manifests,
some explanations seem plausible. First, the change in the online gambling participation
coefficient towards a more negative value when the ‘Number of gambling activities’
variable is added to the model is informative. Once online gambling participation and
‘Number of gambling activities’ are both included in the model, the size of the change in the
participation variable’s coefficient should be indicative of online gambling’s risk relative to
other forms of gambling. If online gambling is less harmful than other forms of gambling,
then the predicted risk from the overall model should decrease, and this is consistent with
what was observed. However, the effect of online gambling participation relative to
participation in other gambling variants is not the entire explanation, otherwise we would
observe an insignificant relationship in our instrumental variable model, rather than a
negative relationship. If the model is estimating the relationship as intended, which the
 224 K.S. Philander and Terri-Lynn MacKay

empirical tests support, it is worthwhile to consider how online gambling may be different
from offline gambling to explain the remainder of the variable’s effect size.
Online games create minimal variable costs to the operator for each additional
wager, whereas offline gambling requires significant labour inputs (e.g. slot cashiers,
table game dealers, pit bosses). This makes low-denomination games more widely
offered online, compared to an offline blackjack game where minimum bet sizes below
$5 a hand would effectively be unprofitable for a casino (Lucas & Kilby, 2008).
Consider poker, for example. The lowest denomination No Limit Hold’em game
typically offered in a casino is $1/$2 blind stakes. Online, 92% of players play below
this level, with 47% of players at $0.05/$0.10 blind stakes or less (Fiedler, 2012).
Indeed, three of the nine questions that form the PGSI questionnaire directly relate to
financial problems, including:
Downloaded by [Georgian Court University] at 09:51 20 February 2015

. Have you bet more than you could really afford to lose?
. Have you borrowed money or sold anything to get money to gamble?
. Has your gambling caused any financial problems for you or your household?
The ability to make smaller wagers (or to adjust to smaller wagers after an unfavourable
period of losses) may allow players to more easily manage any financial hardship from
gambling, while still maintaining similar time-on-device lengths.
The convenience of access to online gambling is often noted as a source of risk, but
gambling in the home or another comfortable location may actually cause players to end
playing sessions more quickly due to the availability of alternate activities. If a player
reaches his/her predetermined money limit early in a session at home, he/she would be
able to quickly perform many non-gambling activities at home – surf the Internet, watch
TV, cook a meal, talk to family/room-mates, etc. The same convenience of free
alternatives is not present at a casino, where amenities are typically provided as profit
centres or to encourage further gambling. Eadington (1975) has previously described a
phenomenon whereby individuals gambling in a casino allocate a time and money budget
to each casino visit. If the consumer loses more than expected, he/she may re-evaluate his/
her budget to account for the void of activities to perform in the time period previously
allocated to the visit. This effect may be lessened when gambling at home. With many
casinos located in remote locations, this ‘isolation effect’ may be understated in related
gambling literature.
The presence of the gambler at an Internet enabled device also implies quick and easy
access to a wide range of online help materials for responsible gambling and problem
gambling. Gamblers with concerns about their play, or simply seeking to improve their
informed decision-making, are typically no more than a couple of clicks away from
responsible gambling materials and contact information, which may or may not be easily
available (or evident) in a physical gaming location. Online gamblers are also not
restricted to accessing the materials provided by the operator, but can also immediately
seek out information from non-profit and health providers.
Some limitations of this research are worthwhile to note. The measures of effect size
from this study may not be relevant or consistent for all levels of gambling severity. For
example, it may be the case that the protective nature of online gambling participation is
mostly (or only) effective for certain types of gamblers. The ability to play small stakes
online or more easily transition to other recreational activities in the home may enable
gamblers in the low-risk categories to avoid accumulating associated financial risks, but it
also may trigger binging periods by higher-risk gamblers. However, the Internet may also
provide higher-risk gamblers with an opportunity to separate from more isolating and
 International Gambling Studies 225

financially stressful forms of gambling. Even though this study provided epidemiological
evidence about the aggregate relationship between online gambling and problem
gambling, it certainly does not show that online gambling participation will not increase
problem gambling severity for some players. Individual impacts are clearly different from
average impacts. A common question posed to gamblers in a clinical environment is to
identify the form of gambling that is most problematic, and future epidemiological
research could focus on identifying whether Internet gamblers self-report online play as a
less pathological form of gambling activity. Clinical treatment and public policy involve
very different considerations, and the findings from this study certainly warrant further
investigation in terms of its application to each domain.
While this study provides a consistent estimate of the relationship between online
gambling participation and problem gambling severity, and provides plausible
explanations for this relationship, a full explanation of the causal mechanism is still
Downloaded by [Georgian Court University] at 09:51 20 February 2015

needed. As more evidence of online and offline gambler behaviour becomes available
through longitudinal research, future studies should be able to enlighten this relationship
more fully.

Notes
1. Additionally, the correction of potential bias from the interpretation of non-gamblers as non-
problem gamblers would only make this negative relationship even larger in absolute value.
2. We thank an anonymous reviewer for pointing out this relationship.

Notes on contributors
Kahlil Philander is a Visiting Assistant Professor at the William F. Harrah College of Hotel
Administration, University of Nevada, Las Vegas. He is also the Director of Research at the
International Gambling Institute, University of Nevada, Las Vegas. His research interests include the
economics of gambling, gambling policy, and responsible gambling.
Terri-Lynn Mackay is a Visiting Assistant Research Professor in the Department of Educational and
Clinical Studies, University of Nevada, Las Vegas. Her research and clinical interests include
understanding the psychological factors that lead to addictive behaviours and the motivational
factors that influence change.

References
American Psychiatric Association. (2000). Diagnostic and statistical manual of mental disorders
(4th ed.). Washington, DC: American Psychiatric Publishing.
Eadington, W. R. (1975). Economic implications of legalizing casino gambling. Journal of
Behavioral Economics, 4, 55 –77.
Ferris, J., & Wynne, H. (2001). The Canadian problem gambling index. Ottawa.
Fiedler, I. (2012). The online poker database of the university of Hamburg. Presentation for
Quebec’s Working Group on Online Gambling.
Gainsbury, S. M., Russell, A., Hing, N., Wood, R., & Blaszczynski, A. (2013). The impact of
Internet gambling on gambling problems: A comparison of moderate-risk and problem Internet
and non-Internet gamblers. Psychology of Addictive Behaviors, 27, 1092–1101. doi:10.1037/
a0031475.
Gainsbury, S., Russell, A., Wood, R., Hing, N., & Blaszczynski, A. (in press). How risky is Internet
gambling? A comparison of subgroups of Internet gamblers based on problem gambling status.
New Media & Society. Published OnlineFirst Jan 15, 2014. doi:10.1177/1461444813518185.
Griffiths, M., & Barnes, A. (2008). Internet gambling: An online empirical study among student
gamblers. International Journal of Mental Health and Addiction, 6, 194– 204.
 226 K.S. Philander and Terri-Lynn MacKay

Griffiths, M., Parke, J., Wood, R., & Rigbye, J. (2010). Online poker gambling in university students:
Further findings from an online survey. International Journal of Mental Health and Addiction, 8,
82 – 89.
Griffiths, M., Wardle, H., Orford, J., Sproston, K., & Erens, B. (2009). Sociodemographic correlates
of Internet gambling: Findings from the 2007 British gambling prevalence survey.
CyberPsychology & Behavior, 12, 199– 202.
Ladd, G. T., & Petry, N. M. (2002). Disordered gambling among university-based medical and
dental patients: A focus on Internet gambling. Psychology of Addictive Behaviors, 16, 76 – 79.
The Land of eGaming Opportunity. (2012). Retrieved from http://www.h2gc.com
LaPlante, D. A., Nelson, S. E., LaBrie, R. A., & Shaffer, H. J. (2011). Disordered gambling, type of
gambling and gambling involvement in the British gambling prevalence survey 2007. doi:10.
1093/eurpub/ckp177 The European Journal of Public Health, 21, 532– 537.
Lloyd, J., Doll, H., Hawton, K., Dutton, W. H., Geddes, J. R., Goodwin, G. M., & Rogers, R. D.
(2010). How psychological symptoms relate to different motivations for gambling: An online
study of Internet gamblers. Biological Psychiatry, 68, 733– 740.
Downloaded by [Georgian Court University] at 09:51 20 February 2015

Lucas, A., & Kilby, J. (2008). Principles of casino marketing. Escondido, CA: Okie International.
McBride, J., & Derevensky, J. (2009). Internet gambling behavior in a sample of online gamblers.
International Journal of Mental Health and Addiction, 7, 149– 167.
McCormack, A., & Griffiths, M. D. (2012). Motivating and inhibiting factors in online gambling
behaviour: A grounded theory study. International Journal of Mental Health and Addiction, 10,
39 – 53.
Petry, N. M. (2006). Internet gambling: An emerging concern in family practice medicine? Family
Practice, 23, 421– 426.
Petry, N. M., & Weinstock, J. (2007). Internet gambling is common in college students and
associated with poor mental health. The American Journal on Addictions, 16, 325– 330.
Rogers, W. (1994). Regression standard errors in clustered samples. Stata Technical Bulletin,
3, 19 – 23.
Roodman, D. (2011). Estimating fully observed recursive mixed-process models with cmp. Stata
Journal, 11, 159– 206.
Sargan, J. D. (1958). The estimation of economic relationships using instrumental variables.
Econometrica, 26, 393–415.
Shaffer, H. J. (1996). Understanding the means and objects of addiction: Technology, the Internet,
and gambling. Journal of Gambling Studies, 12, 461– 469.
Shaffer, H. J., Hall, M. N., & Bilt, J. V. (2000). “Computer addiction”: A critical consideration.
American Journal of Orthopsychiatry, 70, 162– 168.
Sovey, A. J., & Green, D. P. (2011). Instrumental variables estimation in political science: A readers
guide. American Journal of Political Science, 55, 188– 200.
Staiger, D., & Stock, J. H. (1997). Instrumental variables regression with weak instruments.
Econometrica, 65, 557–586.
Trynacity, H. (2011). Alberta hedges bet on online gambling. CBC News. Retrieved from http://
www.cbc.ca/news/canada/edmonton/alberta-hedges-bet-on-online-gambling-1.1089332
Vardi, N. (2013). Sheldon Adelson says he is ‘willing to spend whatever it takes’ to stop online
gambling. Forbes. Retrieved from http://www.forbes.com/sites/nathanvardi/2013/11/22/shel
don-adelson-says-he-is-willing-to-spend-whatever-it-takes-to-stop-online-gambling/
Wardle, H., Moody, A., Griffiths, M., Orford, J., & Volberg, R. (2011). Defining the online gambler
and patterns of behaviour integration: Evidence from the British gambling prevalence survey
2010. International Gambling Studies, 11, 339– 356.
Wardle, H., Moody, A., Spence, S., Orford, J., Volberg, R., Jotangia, D., & . . . Dobbie, F. (2011).
British gambling prevalence study 2010. Retrieved from http://www.gamblingcommission.gov.
uk/researchconsultations/research/bgps.aspx
Welte, J. W., Barnes, G. M., Tidwell, M. O., & Hoffman, J. H. (2009). The association of form of
gambling with problem gambling among American youth. Psychology of Addictive Behaviors,
23, 105– 112.
Wiebe, J., Mun, P., & Kauffman, N. (2006). Gambling and problem gambling in Ontario 2005.
Ontario: Responsible Gaming Council.
Williams, R. J., Volberg, R. A., & Stevens, R. M. (2012). The population prevalence of problem
gambling: Methodological influences, standardized rates, jurisdictional differences, and
worldwide trends. Guelph: Ontario Problem Gambling Research Centre.
 International Gambling Studies 227

Wood, R. T., & Williams, R. J. (2007). Problem gambling on the internet: Implications for Internet
gambling policy in North America. New Media & Society, 9, 520– 542.
Wood, R. T., & Williams, R. J. (2009). Internet gambling: Prevalence, patterns, problems, and
policy options. Guelph: Ontario Problem Gambling Research Centre.
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,
1123– 1141. doi:10.1177/1461444810397650. Retrieved from http://nms.sagepub.com/content/
13/7/1123.full.pdfþhtml
Wooldridge, J. M. (2010). Econometric analysis of cross section and panel data. Cambridge, MA:
The MIT Press.
Downloaded by [Georgian Court University] at 09:51 20 February 2015