ISSN: 1445-9795 (Print) 1479-4276 (Online) Journal homepage: https://www.tandfonline.com/loi/rigs20

Consumer spending in the gaming industry:
evidence of complementary demand in casino and
online venues
Kahlil S. Philander, Brett L.L. Abarbanel & Toni Repetti
To cite this article: Kahlil S. Philander, Brett L.L. Abarbanel & Toni Repetti (2015) Consumer
spending in the gaming industry: evidence of complementary demand in casino and online venues,
International Gambling Studies, 15:2, 256-272, DOI: 10.1080/14459795.2015.1042002
To link to this article: https://doi.org/10.1080/14459795.2015.1042002

Published online: 02 Jun 2015.

Submit your article to this journal

Article views: 525

View related articles

View Crossmark data

Citing articles: 5 View citing articles

Full Terms & Conditions of access and use can be found at
https://www.tandfonline.com/action/journalInformation?journalCode=rigs20

International Gambling Studies, 2015
Vol. 15, No. 2, 256–272, http://dx.doi.org/10.1080/14459795.2015.1042002

Consumer spending in the gaming industry: evidence of
complementary demand in casino and online venues
Kahlil S. Philandera*, Brett L.L. Abarbanelb and Toni Repettic
a

Social Responsibility, BCLC, Vancouver, Canada; bUCLA Gambling Studies Program, University
of California, Los Angeles, USA; cWilliam F. Harrah College of Hotel Administration, University of
Nevada Las Vegas, Las Vegas, USA
(Received 28 October 2014; accepted 31 March 2015)
Brick and mortar gambling stakeholders have scrutinized Internet gambling sites over
concern that online operators may serve as substitutes for their products. In some cases,
this has led to regulatory protection to prevent or restrict entry by online providers.
However, many industry observers have remarked that the two gambling modalities
may serve different consumers/consumer needs, or even serve as complementary
goods. Policymakers, who look to gambling as an important source of tax revenue,
must determine how expansion of Internet gambling will affect overall economic
welfare. Using self-reported consumer gambling behaviour data from the United
Kingdom, the net effect of Internet-based gambling activity on land-based demand is
estimated in this study. A robust complementary (positive) relationship between online
and offline gambling is found, using ordinary least squares, two-stage least squares, and
two-part modeling techniques. These particular findings suggest that economic
concerns around the cannibalization of traditional gambling industries should be
reconsidered, and provide support for prior research showing that Internet based firms
can be complementary to brick and mortar businesses.
Keywords: casino gambling; Internet gambling; cannibalization; consumer behaviour;
policy

Introduction
In most of the service sector, the Internet provides no viable threat to brick and mortar
consumption. While Internet-based channels can be used as an alternate point of purchase
(e.g. purchasing a hotel room with a third-party online travel agent [OTA] instead of
directly), they cannot replace a service – consumption must still occur where it always
has, at the brick and mortar location. The gambling industry therefore has a relatively
unique challenge. Consumption of gambling no longer must occur in multimillion (or
billion) dollar gaming venues, but can instead occur on nearly any device with a working
Internet connection.
Worldwide online gambling has seen substantial market growth over the past decade.
Global gross gaming win (online and offline) was an estimated USD $529 billion in 2012,
and online gambling accounted for an estimated USD $30 billion (5.6%) of that amount
(H2 Gambling Capital, 2012). In 2003, the online gambling market was estimated at
, USD $8.6 billion and between 2003 and 2012, the compound annual growth rate for the
online gambling market was 14.1% (H2 Gambling Capital, & Odobo, 2013).

*Corresponding author. Email: KPhilander@BCLC.com
q 2015 Taylor & Francis

In some respects, online gambling appears to provide a direct competitor to brick and
mortar gaming, from bingo rooms to casino-style table games to slot machines. This
makes it important to investigate the online – offline relationship in the changing industry.
Not only can gambling consumption occur online, but software-based innovation has also
enabled the delivery of new products that are not feasible in a brick and mortar casino
setting – for example, some poker sites allow a single player to play up to 24 tables at
once. Key elements of hospitality service management, like atmospherics and
servicescape (Abarbanel, 2013), service reliability (Richard Wood & Griffiths, 2008)
and customer service (Richard Wood & Griffiths, 2008) have all been demonstrated as
important components of the online gambling product. This suggests that the online
domain may have more in common with brick and mortar gambling than may initially be
expected. Perhaps most importantly to brick and mortar casino stakeholders (and
integrated resort stakeholders more generally), online gambling is a relevant industry
because of its potential to change consumption patterns of gamblers and/or create new
players for traditional venues. The potential threat of online gambling’s cannibalization of
the offline gambling market is an industry hot topic at conferences and in the media. The
impacts of online gambling on existing offline markets are often discussed in parallel with
the impacts of new offline casinos on those same existing markets (Martin, 2012). At the
most recent Global iGaming Summit & Expo (GiGSE), North America’s largest online
gambling industry conference, four sessions debated the possibility of online gambling’s
cannibalization of offline gambling and how gambling operators can effectively converge
an online and offline business (Bakun, 2014; GambleID, 2014; OpenBet, 2014; Ryan et al.,
2014).
As online gambling becomes legal in more jurisdictions and a larger variety of games
are offered online, stakeholders of land-based gambling companies need to understand the
potential link between offline and online gambling operations. Most importantly,
stakeholders must understand if the new medium is a complement to ongoing operations or
whether it is a substitute good, cannibalizing casino revenue. More formally, is the crosselasticity of demand negative (complement) or positive (substitute). In addition,
government policymakers must understand the link: without a clear grasp of their
relationship, policymakers will be unable to design legislation that takes into account how
Internet gambling expansion will affect tourism activity, public revenue, or employment
in brick and mortar gaming locations.
To empirically examine the relationship between online and brick and mortar gaming
demand, a simple theoretical model to understand the potential impact of online gambling
expansion is first developed. Data from the British Gambling Prevalence Survey is then
used to empirically estimate the relationship. The United Kingdom (UK) has permitted
and regulated online gambling since the implementation of the Gambling Act 2005,
making it a useful market in which to analyse the potential association between gamblers’
behaviour in online and offline gambling products. It is of note that the UK gambling
market is much different from the United States and other markets that include integrated
resorts. The overall UK gambling market serves as an useful foundation for market
analysis due to its history as a regulated market offering numerous game types and
varieties in online and offline settings and the large prevalence of gambling – 73% of
British adults have gambled in some form, either online or offline, in the past year
(Gambling Commission, 2014; National Centre for Social Research, 2011; National
Lottery Commission, 2013).
Gambling in Britain is comprised of six primary, sometimes cross-categorical, sectors:
amusement arcades, betting, bingo, casinos, electronic games, lotteries and remote/online

K.S. Philander et al.

gambling (Gambling Commission, 2014). Remote/online gambling encompasses any
gambling (including many of the games above) that is placed via electronic
communication, such as through a computer, mobile phone or Internet-capable television.
The most recent published data (September 2013) from the UK Gambling Commission
shows that remote/online gross gaming yield (GGY) represents approximately 16% of the
regulated gambling market.1
Literature review
Online and offline consumer behaviour
Outside the gaming industry, researchers have noted potential differences in online and instore consumer motivations. Online and offline shopping are described as different
experientially. Online shopping typically provides consumers with time and effort savings,
as well as better search and compare abilities, while offline shopping offers expert advice
and social interaction, a tactile experience and a shorter delivery time (Kollmann,
Kuckertz, & Kayser, 2012). Online shopping may also not provide the same outcome as
offline shopping since consumers are multi-channel shoppers who prefer online and offline
shopping mediums at different times (y Monsuwé, Dellaert, & de Ruyter, 2004). In the
gaming industry, the comparative advantage of offline gambling in terms of delivery time
and expert advice would appear to be less clear. For example, gambling is consumed as
quickly online as offline (that is, there is little difference in delay in waiting for delivery);
and expert advice may be easier to offer online, since gamblers can learn how to play the
games for free (Griffiths, Parke, Wood, & Rigbye, 2010).
An important advantage for brick and mortar gambling venues may be a higher level of
service. Kollmann et al. (2012) analysed consumers’ shopping motives and found that
convenience is positively related to online customers, but a customer’s desire for service
will move shoppers from online to offline experiences. Binde (2013) has noted the reduced
social dimension within some games played online (e.g. poker and bingo); in-person
gambling attracts gamblers for social interaction possibilities (a night out with friends, for
example), while online gamblers seek the activity for means other than social purposes.
Overall, evidence supports a hypothesis that offline and online consumers have different
shopping motives. Price is therefore not likely to be the only important characteristic for
determining demand across these mediums, and online goods will not be perfect
substitutes to offline goods.
The online gambling consumer
Since gambling is a service (as opposed to a tangible good), there may be reason to think
that consumer behaviour findings from general retail literature – such as the importance of
a tactile experience or shorter delivery time – are less applicable to the gambling industry.
Gambling motivations have been relatively well studied from a psychological perspective.
Lloyd et al. (2010) identified mood regulation, obtaining money, and enjoyment as the
primary motivations for gambling online. McCormack and Griffiths (2012) identified one
major theme motivating online gambling (greater opportunity to gamble) and four
subthemes (convenience, value for money, greater game variety, and anonymity). We also
identified one major inhibiting theme of online gambling (reduced authenticity of
gambling) and four sub-themes (reduced realism, asocial nature of the Internet, use of
electronic money and concerns about the safety of sites). Mowen, Fang, and Scott (2009)
noted that general gambling propensity and introversion – perhaps due to the isolation of

gambling alone in one’s home – are predictors of online gambling proneness, while
financial conservatism and emotional instability serve as predictors of online gambling
avoidance.
When considering the individuals that make up the online gambling market, there is no
homogenous population across jurisdictions. In the United Kingdom, online gambler
demographics tend to skew toward males in younger age groups, with higher income and
levels of education at A-levels or higher2 (National Centre for Social Research, 2011;
Robert Wood & Williams, 2012). In 2010, 14% of the British adult population participated
in a least one online gambling variant, including betting, lotteries, bingo, football pools,
casino-style games, and online slots (National Centre for Social Research, 2011).
Approximately half of this online gambling participation can be attributed to wagers
placed on the National Lottery, with the remaining online gambling population
participating in poker, bingo, slot games and/or casino games.
While the online gambling population is heterogeneous at the jurisdictional level, there
are some international composite data that allow comparisons to the worldwide mean. The
British online gambler demographic profile reflects the typical international player
description, which is likely to be male, younger, employed full-time and/or a student, with
a higher average income than non-gamblers and less likely to be married (Robert Wood &
Williams, 2012). Specific games split the international gambling population. Females
between the ages of 46 and 55 tend to prefer online casino-style games, while males
between the ages of 26 and 35 tend to prefer online poker (Parke, Parke, Rigbye, Suhonen,
& Williams, 2012). In the National Centre for Social Research (2011) Britain-specific
data, men have a higher prevalence of gambling in all reported online games. Prior
research has also suggested that technology anxiety, in addition to demographics, impacts
participation in new technologies like online gambling (Meuter, Ostrom, Bitner, &
Roundtree, 2003).
Online gamblers have the flexibility to gamble anywhere Internet access is present,
which carries both benefits and drawbacks. While many gamblers may view this
flexibility as a benefit akin to the ability to access other information and activities from
anywhere, the isolation of gambling alone in one’s home may bring with it asocial
behaviours. Gambling sometimes replaces participation in social circles, which may
lead to gambling addiction and other health issues (Putnam, 2000; Shaffer & Kidman,
2003).
Empirical research on the relationship between gamblers’ online gambling activity and
offline gambling behaviour is relatively undeveloped. While most online gamblers also
gamble offline (Wardle, Moody, Griffiths, Orford, & Volberg, 2011), it remains unclear
how consumption is linked. Some studies have investigated empirical relationships
between online and offline gambling, but the conclusions drawn in these studies were
based on aggregate macroeconomic data from jurisdictions where online games were not
fully legalized and regulated. Philander (2012) examined time series data of online
gambling revenue in the US from grey market sites, estimating a small negative
relationship with offline gambling revenue, but noted that this relationship may not be
indicative of the present relationship. Philander and Fiedler (2012) examined the postUnlawful Internet Gambling Enforcement Act (UIGEA) North American poker market
using a cross-sectional analysis of US states and Canadian provinces. The authors found
that online poker revenue was positively related to commercial casino revenue. Leal,
López-Laborda, and Rodrigo (2014) found results indicative of a negative impact of online
gambling on casinos, but the authors used a panel estimation procedure that is likely to
have been biased by endogenous macroeconomic events.

K.S. Philander et al.

Consumer theory
Based on the evidence in prior literature that suggests motivation for online consumption
differs from offline consumption, a simple two-good theoretical model of consumer
behaviour is proposed to describe the actions of gambling consumers (Mas-Colell,
Whinston, & Green, 1995). As shown in Figure 1, a representative consumer chooses
between out-of-home (social) leisure activities and in-home (private) leisure activities
along a budget line constraint – the constraint can be thought of as the number of nonworking hours available in the day, and/or the income available to spend on leisure
activities.3 The budget line intersects the y- and x-axes where only social leisure or private
leisure is respectively consumed. The set of combinations of social and private leisure
consumption that lead to the same level of utility are represented by indifference curves –
that is, consumers are indifferent about consuming any point along the same curve.
Indifference curves farther from the origin represent higher levels of utility and are
therefore preferred over curves closer to the origin. Consumers choose the combination of
social and private leisure that maximize utility, subject to the budget constraint.
Maximization therefore occurs at the point where the budget line is tangent to the highest
possible indifference curve (Mas-Colell et al., 1995).
The ability to consume online causes an increase in the maximum consumption of
private leisure, as technology reduces costs of consumption.4 This shifts the point where
the budget line intersects with the x-axis, but the y-axis intersection does not change, since
the cost of social leisure remains unchanged. However, the change in technology has
another effect. The shape of the indifference curve becomes steeper as private
consumption becomes more desirable than was previously the case – the available range
of private consumption options has improved.
As shown in Figure 2, the net effect of the change in consumption of social leisure – in
this case, casino gambling – depends on how these two factors compete against one
another. The left panel shows the case where no noticeable change in preferences occurs
and consumption of social leisure ends up increasing. This would be the case if Internet
gambling simply substituted other (more expensive) private leisure activities but had no
influence on social leisure activities. The right panel illustrates a case where there is a large

Social
Leisure
(Casino
Gambling)

S0

Budget
Line

Indifference
Curves

a
I0

Private
Leisure

Figure 1. Two-good model of social leisure and private leisure.

Figure 2. Alternate cases of the effect of the availability of Internet based consumption on the
market for social and private leisure.

change in preferences towards private leisure. The availability of Internet-based
consumption in this case leads to a corner solution – there is no tangent point and only
private leisure consumption occurs. This describes a case where Internet gambling
substitutes away all casino consumption. In general, the larger the budget line and smaller
the preference effect, the more likely that online gambling expansion will have a
complementary relationship, while the smaller the budget line and the larger the
preference effect, the more likely a substitutionary (or cannibalistic) relationship will
occur.
The outcomes predicted from this theoretical model appear to be reasonable
descriptions of what could occur in a gambling market. In the next section, we describe an
empirical model that is used to estimate aggregate effects of Internet gambling consumption
on offline gambling consumption. We pay particular attention to the directional effects, to
identify whether the two mediums are net complements or net substitutes.

Methodology
Data for this analysis is taken from the 2010 British Gambling Prevalence Study (BPGS;
National Centre for Social Research, 2011). This data set is a nationally representative
survey of 7756 adults aged 16 and older living in private households in England, Scotland
and Wales. The BGPS aims to measure the gambling behaviour of Great Britain residents
and assess the prevalence of problem gambling, factors associated with problem gambling,
and general attitudes toward gambling.

K.S. Philander et al.

Data collection and description
Researchers for the BGPS collected data both in-person with computer-assisted
interviewing and via telephone when interviewers were unable to make direct in-person
contact with randomly sampled households. Gambling behaviour data was self-reported
and interviewers were trained to ensure uniform terminology definitions were provided to
all respondents.
Respondents were asked about their participation in various gambling activities in the
past 12 months (in person, online or both) and their respective levels of involvement in
those activities. Due to data availability issues, this investigation looks at online and
offline participation levels for only slot machines, casino-style games and bingo. Other
forms of gambling were examined in the survey, but those other forms did not have similar
frequency data as was available for casino, slot and bingo and thus were not included in the
relationship analysis. While this survey does include questions on average spending, selfreported spend data has been shown to be unreliable in gambling (Shaffer, Peller,
LaPlante, Nelson, & LaBrie, 2010). Therefore, this analysis focuses on the frequency with
which people gamble for money, offline and online. Frequency of play response options
were divided into 8 ordinal categories of involvement, ranging from no involvement (1) to
everyday/almost every day (8).
Table 1 summarizes the level of participation in each of the gambling categories used
in this study. When examining slot data, 34 observations were removed since the offline
slot behaviour questions asked if respondents had participated in fruit machines in ‘other’
locations, and it may have been ambiguous if ‘other’ could include online slots.
Other variables used in this study include problem gambling status, gender, age range,
ethnicity, economic status, highest level of education, smoking status and level of alcohol
use. Problem gambling was measured with criteria from the Problem Gambling Severity
Index (PGSI; Ferris & Wynne, 2001). PGSI scores are categorized into non-problem, lowrisk, medium-risk and problem gambling categories. The categories are ordinally ranked,
with non-problem as the lowest rank and problem gambling as the highest rank. A more
detailed breakdown of demographic characteristics and gambling participation rates can
be found in National Centre for Social Research (2011).
Model
Equation 1 defines the linear regression model for the underlying relationship between
online gambling frequency and offline gambling frequency:
Offlinej ¼ b0 þ b1 Onlinej þ b2 x2j þ · · · þ bk xkj þ uj

ð1Þ

Where Onlinej and Offlinej are variables indicating past-year frequency of participation in
online and offline gambling for respondent j, respectively, and are treated as continuous in
Table 1. Gambling participation summary statistics.
Casino
participation

Count
Percentage of total sample

Slot
participation

Bingo
participation

Offline

Online

Offline

Online

Offline

Online

Total sample

3.46%

1.80%

12.32%

2.59%

7.83%

1.73%

100.00%

a linear model; x2j . . . xkj are other explanatory variables believed to affect offline gambling
frequency; and uj is the standard model error term.
Not all k 2 1 variables that explain offline gambling frequency can be fully identified.
For example, data on the respondent’s travel distance to a gambling venue is not available.
The effect of these m number of unidentified variables will be captured in the error term,
bk2m xk2mj þ · · · þ bk xkj þ uj ¼ vj . The presence of these missing variables in the error
term can create measurement bias if they are correlated with our variable of interest, online
gambling frequency: Cov(Onlinej,vj) – 0. For example, if an individual who does not live
in a casino catchment area is more likely to gamble online if the goods are substitutes and
less likely if they are complements, then our coefficient estimate of the effect of online
gambling may be biased. To remedy this potential source of bias, a two-stage least squares
(2SLS) instrumental variable approach is employed, which produces a consistent
coefficient estimate for the (endogenous) variable of interest, online gambling frequency
(Wooldridge, 2010). Instead of Equation 1, the following system of Equations (for each
gambling variant) is estimated:
d j þ b2 x2j þ · · · þ bk xkj þ vj
Offline ¼ b0 þ b1 Online

ð2Þ

Onlinej ¼ a0 þ a1 z1j þ a2 z2j þ a3 z2j þ · · · þ ak2nþ1 xðk2nÞj þ 1j

ð3Þ

d j in Equation 2 is the predicted value from Equation 3, n is the number of
Where Online
missing explanatory variables, and z1j and z2j are exogenous variables that must satisfy two
necessary conditions: (1) have zero correlation with vj, that is, they provide no more
explanatory power for offline gambling frequency than the variables already included in
the model; and (2) they must have a non-zero correlation with online gambling frequency.
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 that explains the
relationship between offline and online gambling participation (Wooldridge, 2010).
Separate models are produced for each of the different forms of gambling (casino, slot and
bingo).
Two variables are suggested as potential instruments for online gambling frequency,
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 ‘Online chat rooms/social networking sites’ (Social Networking)
and another potential response is ‘Play Computer Games’ (Computer Games). The two
variables are binary, where ‘1’ indicates participation and ‘0’ indicates non-participation.
Both of these measures describe familiarity and comfort in interacting with computers
and playing games in a virtual setting. Comfort with the Internet/computers is likely to be
important to choosing to gamble online, therefore the instruments should be positively
related to online gambling, and satisfy necessary condition (2): a nonzero correlation with
online gambling participation. These variables have no direct or established empirical
relationship to offline gambling and it would seem unlikely that familiarity with computers
would have a meaningful effect on offline gambling – at the very least, no more so than
online gambling does directly – satisfying necessary condition (1). These necessary
conditions are tested empirically in the next section.
Lastly, we estimate a two-part model for each of casino, slot and bingo. A two-part
model is used when a data set is characterized as having many outcomes of the dependent
variable reported as zero (Humphreys, 2013). As previously described in Table 1, many
BGPS respondents reported that they did not participate in given forms of brick and mortar

K.S. Philander et al.

gaming. A related question to ask with this analysis therefore would be, given that a player
is willing to gamble in a brick and mortar location, what effect does online consumption
have on his or her frequency of offline consumption? The two-part model empirically
addresses this query by estimating a participation decision model (probit) and then a
consumption level model (OLS). As noted by Humphreys (2013), the two-part model
recognizes that participation decisions will differ from quantity decision.
Other independent variables used during the estimation procedure include: number of
gambling activities participated in during the past year (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), spending time with friends and family (Friends and
Family), government office region of residence (Gov. Region), 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.

Results
In the interest of readability, only coefficients for the variables of particular interest and an
indication of the presence of other control variables (and their significance level) that were
created from categorical responses are provided. The number of categorical variables is
noted in brackets next to the variable name. Spending time with friends and family, a
control variable to measure levels of social interaction, was not found to be significant in
any model, nor was it of theoretical value in prior literature, and it was removed from
analysis.
Table 2 provides ordinary least squares (OLS) model results for each of casino, slot
and bingo. After controlling for demographic information, government office regions and
Table 2. Linear model (OLS).

Online casino freq.
Online slot freq
Online bingo freq.
PGSI Low
PGSI Moderate
PGSI Problem
Gender (2 cat.)
Age group (7 cat.)
Economic activity (7 cat.)
Education (6 cat.)
Ethnicity (4 cat.)
Gov. Region (12 cat.)
Smoking (2 cat.)
Alcohol (6 cat.)
Constant
Observations

(1)
Offline casino freq.

(2)
Offline slot freq.

(3)
Offline bingo freq.

Coef.

Std. err.

Coef.

Std. err.

Coef.

Std. err.

0.088***

0.009
0.396***

0.021

0.805***
1.221***
2.283***
Yes***
Yes***
Yes***
Yes**
Yes**
Yes*
Yes***
Yes***
0.695***

0.046
0.081
0.127

0.191***
0.341***
0.406***
0.955***
Yes***
Yes**
Yes*
Yes***
Yes
Yes***
Yes***
Yes
0.260**

0.029
0.053
0.094
0.146

0.255***
0.377***
0.934***
Yes**
Yes*
Yes
Yes**
Yes
Yes
Yes
Yes***
0.034

*p , 0.05, **p , 0.01, ***p , 0.001

0.022
0.039
0.060

0.033

0.069

0.080

health variables (including PGSI categories), the results of this study show that online
gambling frequency is positively related to offline gambling frequency. The effect size
using OLS modelling appears to be fairly small, as a 1-point increase in reported online
frequency (along a scale of 8 options) is related to a 0.088-point increase in casino, 0.396
in slot and 0.191 in bingo. None of the control variables produced results that were
inconsistent with prior literature.
Table 3 shows the first stage of the 2SLS models for each of casino, slot and bingo. The
results generally support the approach taken in this study. The two instrumental variables,
Social Networking and Computer Games, are statistically significant predictors of each
online gambling variant, with a positive sign as expected. Additionally, the F-statistic of
weak identification exceeded the benchmark value of 10 for all 3 variants: casino, F
(2,7642) ¼ 16.689, p , 0.001; slot, F(2,7608) ¼ 11.086, p , 0.001; bingo, F
(2,7642) ¼ 15.960, p , 0.001. These results further support these variables adequacy in
satisfying the second necessary condition of strong instruments (Sovey & Green, 2011;
Staiger & Stock, 1997). The Sargan tests of overidentifying restrictions all failed to reject
the validity of necessary condition 1: casino, x 2(1) ¼ 2.93, p ¼ 0.086; slot, x 2(1) ¼ 1.66,
p ¼ 0.198; and bingo, x 2(1) ¼ 1.29, p ¼ 0.255. None of the other variables provide
unexpected results. For example, the PGSI variables are consistent with prior research
showing high correlation between online gambling participation levels and problem
gambling severity (Robert Wood & Williams, 2007, 2009). Therefore the analysis further
proceeded under the assumption that Social Networking and Computer Games are strong
and valid instruments.
Table 4 displays the results from the second stage of the 2SLS model, which describes
the effects of the online gambling variants on their respective offline variables. In general,
the direction of the effect is consistent with results from the OLS models, which describe
online gambling as a complementary good to offline gambling. However, the results from
the 2SLS models suggest that the effect size is actually larger than those estimated in the
Table 3. First-stage results.
(1)
Online casino freq.

Social Networking
Computer Games
PGSI Low
PGSI Moderate
PGSI Problem
Gender (2 cat.)
Age group (7 cat.)
Economic activity (7 cat.)
Education (6 cat.)
Ethnicity (4 cat.)
Gov. Region (12 cat.)
Smoking (2 cat.)
Alcohol (6 cat.)
Constant
Observations
F-Stat

(2)
Online slot freq.

(3)
Online bingo freq.

Coef.

S.E.

Coef.

S.E.

Coef.

S.E.

0.062***
0.051***
0.214***
0.464***
0.477***
Yes***
Yes
Yes**
Yes
Yes
Yes
Yes**
Yes***
0.085*
16.689

0.016
0.014
0.027
0.047
0.073

0.041**
0.045***
0.231***
0.552***
0.981***
Yes
Yes**
Yes***
Yes**
Yes**
Yes*
Yes
Yes
0.022
11.086

0.015
0.013
0.025
0.044
0.068

0.042***
0.042***
0.175***
0.201***
0.477***
Yes***
Yes
Yes***
Yes*
Yes*
Yes
Yes*
Yes**
0.069*
15.960

0.013
0.011
0.020
0.036
0.056

*p , 0.05, **p , 0.01, ***p , 0.001

0.041

0.038

0.032

K.S. Philander et al.

Table 4. Second-stage results.

Online casino freq.
Online slot freq.
Online bingo Freq.
PGSI Low
PGSI Moderate
PGSI Problem
Gender (2 cat.)
Age group (7 cat.)
Economic activity (7 cat.)
Education (6 cat.)
Ethnicity (4 cat.)
Gov. Region (12 cat.)
Smoking (2 cat.)
Alcohol (6 cat.)
Constant
Observations

(1)
Offline casino freq.

(2)
Offline slot freq.

(3)
Offline bingo freq.

Coef.

S.E.

Coef.

S.E.

Coef.

S.E.

0.365*

0.151
2.543***

0.599

0.297
0.027
0.148
Yes
Yes***
Yes
Yes
Yes
Yes**
Yes**
Yes***
0.558***

0.159
0.355
0.626

0.814
0.233*
0.279*
0.651*
Yes***
Yes**
Yes
Yes***
Yes
Yes***
Yes**
Yes
0.190*

0.471
0.102
0.136
0.274

0.194***
0.246**
0.797***
Yes
Yes
Yes
Yes**
Yes
Yes
Yes
Yes***
2 0.005

0.041
0.082
0.098

0.041

0.112

0.078

*p , 0.05, **p , 0.01, ***p , 0.001

OLS models, creating a much more economically meaningful effect. A 1-point increase in
reported online frequency of gambling (along a scale of 8 options) is related to a 0.365point increase in casino, 2.543 in slot and 0.817 in bingo. Readers should use caution when
interpreting the bingo coefficient, however, as the effect size is significant at the a ¼ 0.1
level but not the a ¼ 0.05 level. Part of the issue with the lack of statistical significance in
the bingo model may be the loss of statistical power when using a 2SLS model. While the
combined large magnitude of the slot frequency coefficient and the absence of significance
of the PGSI variables in the slot model is particularly remarkable, none of those variables
had a variance inflation factor (VIF) above 1.11, suggesting that multicollinearity is not an
issue in the difference between the OLS and 2SLS results. No variable had a VIF above
4.30 in any of the 2SLS models.
In the final test of the findings’ robustness, a two-part model is estimated. This model
treats the level of consumption of gambling as a different decision from the participation in
gambling. This addresses potential bias that could be introduced from a high frequency of
consumers who choose not to gamble at all, and therefore have zero frequency in both
online and offline gambling. The results from this series of models again show that online
casino and slot frequency is positively related to their brick and mortar variants, but bingo
is found to have an insignificant relationship. The casino and slot frequency coefficient
sizes are 0.155 and 0.151 respectively, which is higher than the OLS estimates, but lower
than the findings from the 2SLS models.
Discussion
To date, offline gambling companies offer conflicting views on the value of online
gambling. US-based operators present a microcosmic example of the worldwide debate –
some operators actively seek to ban online gambling while others welcome it to the
market.

Table 5. Two-part model.

Probit
Online casino freq.
Online slot freq.
Online bingo freq.
PGSI Low
PGSI Moderate
PGSI Problem
Gender (2 cat.)
Age group (7 cat.)
Economic activity (7 cat.)
Education (6 cat.)
Ethnicity (4 cat.)
Gov. Region (12 cat.)
Smoking (2 cat.)
Alcohol (6 cat.)
Constant
Regress
Online casino freq.
Online slot freq.
Online bingo Freq.
PGSI Low
PGSI Moderate
PGSI Problem
Gender (2 cat.)
Age group (7 cat.)
Economic activity (7 cat.)
Education (6 cat.)
Ethnicity (4 cat.)
Gov. Region (12 cat.)
Smoking (2 cat.)
Alcohol (6 cat.)
Constant
Observations (probit)
Observations (regress)

(1)
Offline casino freq.

(2)
Offline slot freq.

(3)
Offline bingo freq.

Coef.

Std. err.

Coef.

Std. err.

Coef.

Std. err.

0.110***

0.032
0.336***

0.037

0.815***
1.000***
1.362***
Yes*
Yes***
Yes**
Yes***
Yes***
Yes**
Yes***
Yes***
Yes***

0.074
0.132
0.207

0.143***
0.498***
0.613***
0.857***
Yes***
Yes**
Yes***
Yes***
Yes**
Yes**
Yes***
Yes
Yes***

0.039
0.087
0.153
0.220

0.151**

0.046

0.758***
1.174***
2.454***
Yes***
Yes***
Yes**
Yes***
Yes
Yes**
Yes
Yes**
Yes***

0.143
0.218
0.306

0.144
0.401
0.463
1.812**
Yes
Yes***
Yes
Yes**
Yes
Yes**
Yes
Yes
Yes***

0.096
0.260
0.460
0.607

0.828***
0.978***
1.244***
Yes**
Yes***
Yes*
Yes**
Yes
Yes*
Yes
Yes***
Yes***

0.091
0.145
0.214

0.155*

0.071

0.525*
0.415
2.141***
Yes
Yes
Yes**
Yes
Yes
Yes
Yes
Yes
Yes**

0.227
0.316
0.498

*p , 0.05, **p , 0.01, ***p , 0.001

The results from the analysis in this study generally point to a complementary
(positive) relationship between online gambling activity and offline gambling activity.
In no model did we find evidence of a cannibalistic relationship among the gambling
modes. The results expand on those of Philander and Fiedler (2012), who found that
online poker revenue was positively related to land-based casino revenue. The results,
though, contradict findings in Philander (2012) and Leal, López-Laborda, and Rodrigo
(2014). Philander (2012), however, notes that his results stem from analysis of grey
market sites operating in the US, data that is different from that collected in regulated
markets like the analysis here. Meanwhile, Leal, López-Laborda, and Rodrigo (2014)
comment that the negative relationship they observed between online and offline
gambling may be largely due to non-taxation of online gambling during the years of
their data.

K.S. Philander et al.

In a theoretical model of two-good consumer choice, results from this study suggest
that the effect of reduced costs of consumption from the Internet medium outweigh the
perceived change in preferences towards more private (i.e. Internet based) leisure. This
should perhaps be expected, since prior research on online gamblers has been consistent
with that the theory that online gambling is comprised of a different demographic profile
from offline gamblers.
Online gamblers tend to skew toward males in younger age groups, with higher income
and levels of education, whereas offline gamblers have a much higher proportion of
females with a wider age range and have lower average levels of full-time employment
and income than do online gamblers (Gainsbury, Wood, Russell, Hing, & Blaszczynski,
2012). Online sites’ lower marginal cost of production and economies scale also allow
them to offer smaller stakes,5 making it more attractive for people who previously had not
been able to afford higher stakes gambling at offline venues. For either group, we should
therefore not expect to see much of a change in preferences between social leisure
consumption and private leisure consumption. People who previously participated in
offline gambling continue to have a strong preference for that medium, while people who
did not previously gamble offline would already have indifference curves that reflect that
preference.
In the OLS, 2SLS and two-part models, all the online frequency coefficients were
positive. The 2SLS and two-part models also produced larger coefficient sizes than the
OLS models, suggesting that the relationships between the two gambling media are biased
downwards in the linear models. The largest relationship was between online slots and
offline slots, where a 1-unit increase in the slot frequency variable online was related to a
2.5-unit increase in offline frequency under the 2SLS model. As a means of providing
context, a 2-unit change could be the difference between ‘about once a month’ to ‘about
once a week’.
The nature of the complementary relationship holds implications for both operators
and policymakers in other jurisdictions. For example, in the US the debate to legalize and
regulate or ban online gambling has become quite contentious. US operators should
recognize the positive value that online gambling can have with relation to their offline
operations. Further, these results support a perspective that both offline and online
gambling can coexist in the same market, and that there may be synergistic opportunities.
Rather than resisting change and working actively on the political scene to ban online
gambling, casino operators may be better off developing their own online gambling site or
partnering with existing sites. Their brand recognition has the potential to attract a wide
audience, but continuing to hold out allows other brands a first-mover advantage.
Governments may also benefit from online gambling adoption, since the expansion of
online gambling will lead to new (online) revenue sources and higher revenue within
existing (complementary) products. These results show that local tourism dollars being
generated by offline casinos are enhanced by the addition of online gambling legalization.
The positive relationship between online and offline gambling frequency can reassure
local governments that tax revenue and employment generated from tourism activities of
offline gambling are not likely to decrease due to the legalization of online gambling.
In addition, the significant positive relationship between online and offline frequency may
lead to more non-gambling revenue, such as food and beverage sales, which also increases
tax revenue and employment.
Policymakers also need to consider responsible gambling issues that run in tandem
with new gambling outlets. Generally, new online gambling availability follows a novelty
effect, with a spike of at-risk behaviour with the activity’s introduction, followed by a

steep reduction and eventual low plateau of problematic behaviour (LaPlante & Shaffer,
2007). Most new recreational online gamblers tend to adapt quickly and participate in
gambling as a healthy leisure activity, but heavily involved gamblers maintain their at-risk
status and may need special attention (LaPlante, Schumann, LaBrie, & Shaffer, 2008).
Of particular note in relation to prior research, higher PGSI category involvement was
related to higher offline gambling frequency.
Limitations
While relatively robust evidence of a positive relationship between offline and online
gambling frequencies was found in this study, we note that a limitation of this study is that
this may not directly translate to revenue, since average spend figures may differ. At the
very least, however, it appears unlikely that a substitutionary relationship exists overall (of
course, individual consumers still may respond with substitutionary behaviour). It should
also be noted that these results are from only one western country with a relatively welldeveloped gambling infrastructure, regulatory scheme and Internet connectivity. The
offline/online relationship may certainly be different in another jurisdiction without these,
or other, characteristics. The BPGS also relies on self-reporting data, which can create bias
in the responses. It is possible the respondents did not truthfully answer all questions to
make themselves seem like more or less of a gambler, leading to potentially unreliable
results. Self-report bias is regularly recognized as a limitation in studies on gambling
behaviour (Baumeister, Vohs, & Funder, 2007; Volberg, Gerstein, Christiansen, &
Baldridge, 2001). As more data continues to emerge from newly regulated jurisdictions
such as Nevada, New Jersey, British Columbia and Quebec, the relationship between
online and offline gambling should be investigated for individual regions and for the
composite industry.
Acknowledgements
We wish to thank Terri-Lynn Mackay of the University of Nevada, Las Vegas, and David Paster of
Oklahoma State University for their comments on an earlier version of this paper. All errors are our
own.

Conflicts of Interest
None of the authors received direct funding for research related to this study. Each of the authors has
received direct compensation for gambling related work from government, non-profit, and private
organizations in the past.
Competing interests: The first author is employed by an organization that operates online and offline
gambling.

Constraints on publishing
None of the authors received direct funding for research related to this study. Each of the authors has,
in the past, received direct compensation for gambling related work from government, non-profit,
and private organizations.

Notes
1.
2.

This proportional estimate of GGY does not include British bookmakers serving the population
from an offshore licence, such as Gibraltar.
A-level education levels are approximately equivalent to secondary school, generally completed
at ages 16 – 18.

3.
4.

5.

K.S. Philander et al.
The model describes a representative consumer, while the aggregate outcome is the sum of
individual responses.
In gambling, reduced costs of consumption could be thought of as replacing the need to organize
a home poker game with easy online access. This effect could also be thought of as reduced
uncertainty costs with black market consumption (e.g. replacing private bookies). Finally,
economies of scale from large online operations can also reduce price.
For example, the lowest denomination No Limit Hold’em poker game typically offered in a
casino is USD $1/$2 blind stakes. Online, 92% of hands are played below this level, with 49% of
hands played at USD $0.05/$0.10 blind stakes or less (Fiedler, 2012).

Notes on contributors
Kahlil Philander is the director of social responsibility at BCLC. His research interests include the
economics of gambling, gambling policy and responsible gambling.
Brett L.L. Abarbanel is a research fellow at the UCLA Gambling Studies Program, where her
expertise lies in human – computer interaction in online gambling, and global sociocultural and
policy analysis.
Toni Repetti, PhD, is an assistant professor at the William F. Harrah College of Hotel Administration
at the University of Nevada, Las Vegas. Her research interests are in managerial finance and gaming
operations, with an emphasis on the effect gaming management decisions have on revenue and net
income.

References
Abarbanel, B. (2013). Mapping the online gambling e-servicescape: A conceptual model. UNLV
Gaming Research & Review Journal, 17, 27 – 44.
Bakun, E. (2014, July). Clarifying the Misconceptions of iGaming: iGaming will Cannibalize
Customer Base. Paper presented at the Global iGaming Summit & Expo, San Francisco, CA.
Baumeister, R. F., Vohs, K. D., & Funder, D. C. (2007). Psychology as the science of self-reports and
finger movements: Whatever happened to actual behavior? Perspectives on Psychological
Science, 2, 396– 403. doi:10.1111/j.1745-6916.2007.00051.x
Binde, P. (2013). Why people gamble: A model with five motivational dimensions. International
Gambling Studies, 13, 81 – 97. doi:10.1080/14459795.2012.712150
Ferris, J., & Wynne, H. (2001). The Canadian problem gambling index: Final report: Canadian
Centre on Substance Abuse.
Fiedler, I. The online poker database of the University of Hamburg. Presentation for Quebec’s
Working Group on Online Gambling (2012).
Gainsbury, S., Wood, R., Russell, A., Hing, N., & Blaszczynski, A. (2012). A digital revolution:
Comparison of demographic profiles, attitudes and gambling behavior of internet and noninternet gamblers. Computers in Human Behavior, 28, 1388–1398. doi:10.1016/j.chb.2012.02.
GambleID. Cross-channel gaming in action. Paper presented at the Global iGaming Summit &
Expo, San Francisco, CA (2014, July).
Gambling Commission. (2014). Industry statistics: April 2009 to September 2013. Birmingham, UK:
Gambling Commission.
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. doi:10.1007/s11469-009-9203-7
H2 Gambling Capital, & Odobo (2013). There’s nothing virtual about the opportunity in real-money
gambling: Opportunities for game developers in regulated real-money online gambling.
Gibraltar.
H2 Gambling Capital. (2012, November 14). Leading global gaming nations – Asia and eGaming
continue to outperform. Retrieved from http://www.h2gc.com/article/leading-global-gamblingnations – asia-and-egaming-continue-to-out-perform
Humphreys, B. R. (2013). Dealing with zeros in economic data. Retrieved from http://www.ualberta.
ca/, bhumphre/class/zeros_v1.pdf

Kollmann, T., Kuckertz, A., & Kayser, I. (2012). Cannibalization or synergy? Consumers’ channel
selection in online-offline multichannel systems. Journal of Retailing and Consumer Services,
19, 186– 194. doi:10.1016/j.jretconser.2011.11.008
LaPlante, D. A., Schumann, A., LaBrie, R. A., & Shaffer, H. J. (2008). Population trends in internet
sports gambling. Computers in Human Behavior, 24, 2399– 2414. doi:10.1016/j.chb.2008.02.
Leal, A., López-Laborda, J., & Rodrigo, F. (2014). The inside and outside revenue impact of regional
gambling taxes in Spain. Growth and Change, 45, 79 –97.
LaPlante, D. A., & Shaffer, H. J. (2007). Understanding the influence of gambling opportunities:
Expanding exposure models to include adaptation. American Journal of Orthopsychiatry, 77,
616– 623. doi:10.1037/0002-9432.77.4.616
Lloyd, J., Doll, H., Hawton, K., Dutton, W. H., Geddes, J. R., & Goodwin, G. M. (2010). How
psychological symptoms relate to different motivations for gambling: An online study of internet
gamblers. Biological Psychiatry, 68, 733– 740. doi:10.1016/j.biopsych.2010.03.038
Martin, V. (2012). Investing the hard way: Cannibalization in the US market. CalvinAyre.com.
Retrieved from http://www.calvinayre.com/2012/06/25/business/cannibalization-in-the-UScasino-market
Mas-Colell, A., Whinston, M. D., & Green, J. R. (1995). Microeconomic theory. New York: Oxford
University Press.
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. doi:10.1007/s11469-010-9300-7
Meuter, M. L., Ostrom, A. L., Bitner, M. J., & Roundtree, R. (2003). The influence of technology
anxiety on consumer use and experiences with self-service technologies. Journal of Business
Research, 56, 899– 906. doi:10.1016/S0148-2963(01)00276-4
Mowen, J. C., Fang, X., & Scott, K. (2009). A hierarchical model approach for identifying the trait
antecedents of general gambling propensity and of four gambling-related genres. Journal of
Business Research, 62, 1262– 1268. doi:10.1016/j.jbusres.2008.11.007
National Centre for Social Research. (2011). British Gambling Prevalence Survey 2010. Norwich,
UK: The Stationery Office.
National Lottery Commission. (2013). National Lottery Commission Annual Report and Accounts
2012/2013. Norwich, UK: The Stationery Office.
OpenBet. (2014, July). Land-based/online convergence in action. Paper presented at the Global
iGaming Summit & Expo, San Francisco, CA.
Parke, J., Parke, A., Rigbye, J., Suhonen, N., & Williams, L. V. (2012). The eCOGRA global online
gambler report. In R. J. Williams, R. Wood, & J. Parke (Eds.), Routledge international handbook
of internet gambling (pp. 140– 160). New York: Taylor & Francis Group.
Philander, K. S. (2012). The effect of online gaming on commercial casino revenue. UNLV Gaming
Research & Review Journal, 15, 23 – 34.
Philander, K., & Fiedler, I. (2012). Online poker in North America: Empirical evidence on its
complementary effect on the offline gambling market. Gaming Law Review and Economics, 16,
415– 423.
Putnam, R. D. (2000). Bowling alone: The collapse and revival of American community. New York:
Simon and Schuster.
Ryan, J., Jones, M. J., Woods, L., Quigley, J., Abarbanel, B., Dixon, G., & Buro, F. (2014).
Consumer insight briefing: Presenting and analyzing the findings of the NJ Players’ Survey.
Global iGaming Summit & Expo. San Francisco, CA.
Shaffer, H. J., & Kidman, R. (2003). Shifting perspectives on gambling and addiction. Journal of
Gambling Studies, 19(1), 1 – 6. doi:10.1023/A:1021267028254
Shaffer, H. J., Peller, A. J., LaPlante, D. A., Nelson, S. E., & LaBrie, R. A. (2010). Toward a
paradigm shift in internet gambling research: From opinion and self-report to actual behavior.
Addiction Research & Theory, 18, 270– 283. doi:10.3109/16066350902777974
Sovey, A. J., & Green, D. P. (2011). Instrumental variables estimation in political science: A readers’
guide. American Journal of Political Science, 55, 188– 200. doi:10.111/j.1540-5907.2010.
00477.x
Staiger, D., & Stock, J. H. (1997). Instrumental variables regression with weak instruments.
Econometrica, 65, 557–586. doi:10.2307/2171753

K.S. Philander et al.

Volberg, R. A., Gerstein, D. R., Christiansen, E. M., & Baldridge, J. (2001). Assessing self-reported
expenditures on gambling. Managerial and Decision Economics, 22, 77 – 96. doi:10.1002/mde.
Wardle, H., Moody, A., Griffiths, M., Orford, J., & Volberg, R. A. (2011). Defining the online
gambler and patterns of behaviour integration: Evidence from the British Gambling Prevalence
Survey 2010. International Gambling Studies, 11, 339– 356. doi:10.1080/14459795.2011.
628684
Wood, R. T., & Griffiths, M. D. (2008). Why Swedish people play online poker and factors that can
increase or decrease trust in poker web sites: A qualitative investigation. Journal of Gambling
Issues, 21, 80– 97. doi:10.4309/jgi.2008.21.8
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. doi:10.1177/
1461444807076987
Wood, R., & Williams, R. J. (2009). Internet gambling: Prevalence, patterns, problems, and policy
options. Guelph, Ontario, CA: Ontario Problem Gambling Research Centre.
Wood, R., & Williams, R. J. (2012). The Casino City study: A large-scale international study of
online gamblers. In R. J. Williams, R. Wood, & J. Parke (Eds.), Routledge international
handbook of internet gambling (pp. 103–125). New York: Taylor & Francis Group.
Wooldridge, J. M. (2010). Econometric analysis of cross section and panel data. Cambridge, MA:
MIT Press.
y Monsuwé, T. P., Dellaert, B. G. C., & de Ruyter, K. (2004). What drives consumers to shop online?
A literature review. International Journal of Service Industry Management, 15, 102– 121.
doi:10.1108/09564230410523358